xref: /qemu/target/s390x/kvm/kvm.c (revision 28221f9c999a9b34f58d94599da9c229df9a4fed)
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 "qemu/units.h"
38 #include "qemu/mmap-alloc.h"
39 #include "sysemu/sysemu.h"
40 #include "sysemu/hw_accel.h"
41 #include "hw/hw.h"
42 #include "sysemu/device_tree.h"
43 #include "exec/gdbstub.h"
44 #include "trace.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 int cap_sync_regs;
139 static int cap_async_pf;
140 static int cap_mem_op;
141 static int cap_s390_irq;
142 static int cap_ri;
143 static int cap_gs;
144 static int cap_hpage_1m;
145 
146 static int active_cmma;
147 
148 static void *legacy_s390_alloc(size_t size, uint64_t *align, bool shared);
149 
150 static int kvm_s390_query_mem_limit(uint64_t *memory_limit)
151 {
152     struct kvm_device_attr attr = {
153         .group = KVM_S390_VM_MEM_CTRL,
154         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
155         .addr = (uint64_t) memory_limit,
156     };
157 
158     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
159 }
160 
161 int kvm_s390_set_mem_limit(uint64_t new_limit, uint64_t *hw_limit)
162 {
163     int rc;
164 
165     struct kvm_device_attr attr = {
166         .group = KVM_S390_VM_MEM_CTRL,
167         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
168         .addr = (uint64_t) &new_limit,
169     };
170 
171     if (!kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_LIMIT_SIZE)) {
172         return 0;
173     }
174 
175     rc = kvm_s390_query_mem_limit(hw_limit);
176     if (rc) {
177         return rc;
178     } else if (*hw_limit < new_limit) {
179         return -E2BIG;
180     }
181 
182     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
183 }
184 
185 int kvm_s390_cmma_active(void)
186 {
187     return active_cmma;
188 }
189 
190 static bool kvm_s390_cmma_available(void)
191 {
192     static bool initialized, value;
193 
194     if (!initialized) {
195         initialized = true;
196         value = kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_ENABLE_CMMA) &&
197                 kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_CLR_CMMA);
198     }
199     return value;
200 }
201 
202 void kvm_s390_cmma_reset(void)
203 {
204     int rc;
205     struct kvm_device_attr attr = {
206         .group = KVM_S390_VM_MEM_CTRL,
207         .attr = KVM_S390_VM_MEM_CLR_CMMA,
208     };
209 
210     if (!kvm_s390_cmma_active()) {
211         return;
212     }
213 
214     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
215     trace_kvm_clear_cmma(rc);
216 }
217 
218 static void kvm_s390_enable_cmma(void)
219 {
220     int rc;
221     struct kvm_device_attr attr = {
222         .group = KVM_S390_VM_MEM_CTRL,
223         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
224     };
225 
226     if (cap_hpage_1m) {
227         warn_report("CMM will not be enabled because it is not "
228                     "compatible with huge memory backings.");
229         return;
230     }
231     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
232     active_cmma = !rc;
233     trace_kvm_enable_cmma(rc);
234 }
235 
236 static void kvm_s390_set_attr(uint64_t attr)
237 {
238     struct kvm_device_attr attribute = {
239         .group = KVM_S390_VM_CRYPTO,
240         .attr  = attr,
241     };
242 
243     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
244 
245     if (ret) {
246         error_report("Failed to set crypto device attribute %lu: %s",
247                      attr, strerror(-ret));
248     }
249 }
250 
251 static void kvm_s390_init_aes_kw(void)
252 {
253     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
254 
255     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
256                                  NULL)) {
257             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
258     }
259 
260     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
261             kvm_s390_set_attr(attr);
262     }
263 }
264 
265 static void kvm_s390_init_dea_kw(void)
266 {
267     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
268 
269     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
270                                  NULL)) {
271             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
272     }
273 
274     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
275             kvm_s390_set_attr(attr);
276     }
277 }
278 
279 void kvm_s390_crypto_reset(void)
280 {
281     if (s390_has_feat(S390_FEAT_MSA_EXT_3)) {
282         kvm_s390_init_aes_kw();
283         kvm_s390_init_dea_kw();
284     }
285 }
286 
287 static int kvm_s390_configure_mempath_backing(KVMState *s)
288 {
289     size_t path_psize = qemu_mempath_getpagesize(mem_path);
290 
291     if (path_psize == 4 * KiB) {
292         return 0;
293     }
294 
295     if (!hpage_1m_allowed()) {
296         error_report("This QEMU machine does not support huge page "
297                      "mappings");
298         return -EINVAL;
299     }
300 
301     if (path_psize != 1 * MiB) {
302         error_report("Memory backing with 2G pages was specified, "
303                      "but KVM does not support this memory backing");
304         return -EINVAL;
305     }
306 
307     if (kvm_vm_enable_cap(s, KVM_CAP_S390_HPAGE_1M, 0)) {
308         error_report("Memory backing with 1M pages was specified, "
309                      "but KVM does not support this memory backing");
310         return -EINVAL;
311     }
312 
313     cap_hpage_1m = 1;
314     return 0;
315 }
316 
317 int kvm_arch_init(MachineState *ms, KVMState *s)
318 {
319     MachineClass *mc = MACHINE_GET_CLASS(ms);
320 
321     if (mem_path && kvm_s390_configure_mempath_backing(s)) {
322         return -EINVAL;
323     }
324 
325     mc->default_cpu_type = S390_CPU_TYPE_NAME("host");
326     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
327     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
328     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
329     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
330 
331     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
332         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
333         phys_mem_set_alloc(legacy_s390_alloc);
334     }
335 
336     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
337     kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
338     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
339     if (ri_allowed()) {
340         if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
341             cap_ri = 1;
342         }
343     }
344     if (cpu_model_allowed()) {
345         if (kvm_vm_enable_cap(s, KVM_CAP_S390_GS, 0) == 0) {
346             cap_gs = 1;
347         }
348     }
349 
350     /*
351      * The migration interface for ais was introduced with kernel 4.13
352      * but the capability itself had been active since 4.12. As migration
353      * support is considered necessary let's disable ais in the 2.10
354      * machine.
355      */
356     /* kvm_vm_enable_cap(s, KVM_CAP_S390_AIS, 0); */
357 
358     return 0;
359 }
360 
361 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
362 {
363     return 0;
364 }
365 
366 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
367 {
368     return cpu->cpu_index;
369 }
370 
371 int kvm_arch_init_vcpu(CPUState *cs)
372 {
373     S390CPU *cpu = S390_CPU(cs);
374     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
375     cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
376     return 0;
377 }
378 
379 void kvm_s390_reset_vcpu(S390CPU *cpu)
380 {
381     CPUState *cs = CPU(cpu);
382 
383     /* The initial reset call is needed here to reset in-kernel
384      * vcpu data that we can't access directly from QEMU
385      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
386      * Before this ioctl cpu_synchronize_state() is called in common kvm
387      * code (kvm-all) */
388     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
389         error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
390     }
391 }
392 
393 static int can_sync_regs(CPUState *cs, int regs)
394 {
395     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
396 }
397 
398 int kvm_arch_put_registers(CPUState *cs, int level)
399 {
400     S390CPU *cpu = S390_CPU(cs);
401     CPUS390XState *env = &cpu->env;
402     struct kvm_sregs sregs;
403     struct kvm_regs regs;
404     struct kvm_fpu fpu = {};
405     int r;
406     int i;
407 
408     /* always save the PSW  and the GPRS*/
409     cs->kvm_run->psw_addr = env->psw.addr;
410     cs->kvm_run->psw_mask = env->psw.mask;
411 
412     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
413         for (i = 0; i < 16; i++) {
414             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
415             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
416         }
417     } else {
418         for (i = 0; i < 16; i++) {
419             regs.gprs[i] = env->regs[i];
420         }
421         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
422         if (r < 0) {
423             return r;
424         }
425     }
426 
427     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
428         for (i = 0; i < 32; i++) {
429             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
430             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
431         }
432         cs->kvm_run->s.regs.fpc = env->fpc;
433         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
434     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
435         for (i = 0; i < 16; i++) {
436             cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
437         }
438         cs->kvm_run->s.regs.fpc = env->fpc;
439         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
440     } else {
441         /* Floating point */
442         for (i = 0; i < 16; i++) {
443             fpu.fprs[i] = get_freg(env, i)->ll;
444         }
445         fpu.fpc = env->fpc;
446 
447         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
448         if (r < 0) {
449             return r;
450         }
451     }
452 
453     /* Do we need to save more than that? */
454     if (level == KVM_PUT_RUNTIME_STATE) {
455         return 0;
456     }
457 
458     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
459         cs->kvm_run->s.regs.cputm = env->cputm;
460         cs->kvm_run->s.regs.ckc = env->ckc;
461         cs->kvm_run->s.regs.todpr = env->todpr;
462         cs->kvm_run->s.regs.gbea = env->gbea;
463         cs->kvm_run->s.regs.pp = env->pp;
464         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
465     } else {
466         /*
467          * These ONE_REGS are not protected by a capability. As they are only
468          * necessary for migration we just trace a possible error, but don't
469          * return with an error return code.
470          */
471         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
472         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
473         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
474         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
475         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
476     }
477 
478     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
479         memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
480         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
481     }
482 
483     /* pfault parameters */
484     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
485         cs->kvm_run->s.regs.pft = env->pfault_token;
486         cs->kvm_run->s.regs.pfs = env->pfault_select;
487         cs->kvm_run->s.regs.pfc = env->pfault_compare;
488         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
489     } else if (cap_async_pf) {
490         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
491         if (r < 0) {
492             return r;
493         }
494         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
495         if (r < 0) {
496             return r;
497         }
498         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
499         if (r < 0) {
500             return r;
501         }
502     }
503 
504     /* access registers and control registers*/
505     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
506         for (i = 0; i < 16; i++) {
507             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
508             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
509         }
510         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
511         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
512     } else {
513         for (i = 0; i < 16; i++) {
514             sregs.acrs[i] = env->aregs[i];
515             sregs.crs[i] = env->cregs[i];
516         }
517         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
518         if (r < 0) {
519             return r;
520         }
521     }
522 
523     if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
524         memcpy(cs->kvm_run->s.regs.gscb, env->gscb, 32);
525         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GSCB;
526     }
527 
528     if (can_sync_regs(cs, KVM_SYNC_BPBC)) {
529         cs->kvm_run->s.regs.bpbc = env->bpbc;
530         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_BPBC;
531     }
532 
533     if (can_sync_regs(cs, KVM_SYNC_ETOKEN)) {
534         cs->kvm_run->s.regs.etoken = env->etoken;
535         cs->kvm_run->s.regs.etoken_extension  = env->etoken_extension;
536         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ETOKEN;
537     }
538 
539     /* Finally the prefix */
540     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
541         cs->kvm_run->s.regs.prefix = env->psa;
542         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
543     } else {
544         /* prefix is only supported via sync regs */
545     }
546     return 0;
547 }
548 
549 int kvm_arch_get_registers(CPUState *cs)
550 {
551     S390CPU *cpu = S390_CPU(cs);
552     CPUS390XState *env = &cpu->env;
553     struct kvm_sregs sregs;
554     struct kvm_regs regs;
555     struct kvm_fpu fpu;
556     int i, r;
557 
558     /* get the PSW */
559     env->psw.addr = cs->kvm_run->psw_addr;
560     env->psw.mask = cs->kvm_run->psw_mask;
561 
562     /* the GPRS */
563     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
564         for (i = 0; i < 16; i++) {
565             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
566         }
567     } else {
568         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
569         if (r < 0) {
570             return r;
571         }
572          for (i = 0; i < 16; i++) {
573             env->regs[i] = regs.gprs[i];
574         }
575     }
576 
577     /* The ACRS and CRS */
578     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
579         for (i = 0; i < 16; i++) {
580             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
581             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
582         }
583     } else {
584         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
585         if (r < 0) {
586             return r;
587         }
588          for (i = 0; i < 16; i++) {
589             env->aregs[i] = sregs.acrs[i];
590             env->cregs[i] = sregs.crs[i];
591         }
592     }
593 
594     /* Floating point and vector registers */
595     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
596         for (i = 0; i < 32; i++) {
597             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
598             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
599         }
600         env->fpc = cs->kvm_run->s.regs.fpc;
601     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
602         for (i = 0; i < 16; i++) {
603             get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
604         }
605         env->fpc = cs->kvm_run->s.regs.fpc;
606     } else {
607         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
608         if (r < 0) {
609             return r;
610         }
611         for (i = 0; i < 16; i++) {
612             get_freg(env, i)->ll = fpu.fprs[i];
613         }
614         env->fpc = fpu.fpc;
615     }
616 
617     /* The prefix */
618     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
619         env->psa = cs->kvm_run->s.regs.prefix;
620     }
621 
622     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
623         env->cputm = cs->kvm_run->s.regs.cputm;
624         env->ckc = cs->kvm_run->s.regs.ckc;
625         env->todpr = cs->kvm_run->s.regs.todpr;
626         env->gbea = cs->kvm_run->s.regs.gbea;
627         env->pp = cs->kvm_run->s.regs.pp;
628     } else {
629         /*
630          * These ONE_REGS are not protected by a capability. As they are only
631          * necessary for migration we just trace a possible error, but don't
632          * return with an error return code.
633          */
634         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
635         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
636         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
637         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
638         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
639     }
640 
641     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
642         memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
643     }
644 
645     if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
646         memcpy(env->gscb, cs->kvm_run->s.regs.gscb, 32);
647     }
648 
649     if (can_sync_regs(cs, KVM_SYNC_BPBC)) {
650         env->bpbc = cs->kvm_run->s.regs.bpbc;
651     }
652 
653     if (can_sync_regs(cs, KVM_SYNC_ETOKEN)) {
654         env->etoken = cs->kvm_run->s.regs.etoken;
655         env->etoken_extension = cs->kvm_run->s.regs.etoken_extension;
656     }
657 
658     /* pfault parameters */
659     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
660         env->pfault_token = cs->kvm_run->s.regs.pft;
661         env->pfault_select = cs->kvm_run->s.regs.pfs;
662         env->pfault_compare = cs->kvm_run->s.regs.pfc;
663     } else if (cap_async_pf) {
664         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
665         if (r < 0) {
666             return r;
667         }
668         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
669         if (r < 0) {
670             return r;
671         }
672         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
673         if (r < 0) {
674             return r;
675         }
676     }
677 
678     return 0;
679 }
680 
681 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
682 {
683     int r;
684     struct kvm_device_attr attr = {
685         .group = KVM_S390_VM_TOD,
686         .attr = KVM_S390_VM_TOD_LOW,
687         .addr = (uint64_t)tod_low,
688     };
689 
690     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
691     if (r) {
692         return r;
693     }
694 
695     attr.attr = KVM_S390_VM_TOD_HIGH;
696     attr.addr = (uint64_t)tod_high;
697     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
698 }
699 
700 int kvm_s390_get_clock_ext(uint8_t *tod_high, uint64_t *tod_low)
701 {
702     int r;
703     struct kvm_s390_vm_tod_clock gtod;
704     struct kvm_device_attr attr = {
705         .group = KVM_S390_VM_TOD,
706         .attr = KVM_S390_VM_TOD_EXT,
707         .addr = (uint64_t)&gtod,
708     };
709 
710     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
711     *tod_high = gtod.epoch_idx;
712     *tod_low  = gtod.tod;
713 
714     return r;
715 }
716 
717 int kvm_s390_set_clock(uint8_t tod_high, uint64_t tod_low)
718 {
719     int r;
720     struct kvm_device_attr attr = {
721         .group = KVM_S390_VM_TOD,
722         .attr = KVM_S390_VM_TOD_LOW,
723         .addr = (uint64_t)&tod_low,
724     };
725 
726     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
727     if (r) {
728         return r;
729     }
730 
731     attr.attr = KVM_S390_VM_TOD_HIGH;
732     attr.addr = (uint64_t)&tod_high;
733     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
734 }
735 
736 int kvm_s390_set_clock_ext(uint8_t tod_high, uint64_t tod_low)
737 {
738     struct kvm_s390_vm_tod_clock gtod = {
739         .epoch_idx = tod_high,
740         .tod  = tod_low,
741     };
742     struct kvm_device_attr attr = {
743         .group = KVM_S390_VM_TOD,
744         .attr = KVM_S390_VM_TOD_EXT,
745         .addr = (uint64_t)&gtod,
746     };
747 
748     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
749 }
750 
751 /**
752  * kvm_s390_mem_op:
753  * @addr:      the logical start address in guest memory
754  * @ar:        the access register number
755  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
756  * @len:       length that should be transferred
757  * @is_write:  true = write, false = read
758  * Returns:    0 on success, non-zero if an exception or error occurred
759  *
760  * Use KVM ioctl to read/write from/to guest memory. An access exception
761  * is injected into the vCPU in case of translation errors.
762  */
763 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
764                     int len, bool is_write)
765 {
766     struct kvm_s390_mem_op mem_op = {
767         .gaddr = addr,
768         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
769         .size = len,
770         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
771                        : KVM_S390_MEMOP_LOGICAL_READ,
772         .buf = (uint64_t)hostbuf,
773         .ar = ar,
774     };
775     int ret;
776 
777     if (!cap_mem_op) {
778         return -ENOSYS;
779     }
780     if (!hostbuf) {
781         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
782     }
783 
784     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
785     if (ret < 0) {
786         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
787     }
788     return ret;
789 }
790 
791 /*
792  * Legacy layout for s390:
793  * Older S390 KVM requires the topmost vma of the RAM to be
794  * smaller than an system defined value, which is at least 256GB.
795  * Larger systems have larger values. We put the guest between
796  * the end of data segment (system break) and this value. We
797  * use 32GB as a base to have enough room for the system break
798  * to grow. We also have to use MAP parameters that avoid
799  * read-only mapping of guest pages.
800  */
801 static void *legacy_s390_alloc(size_t size, uint64_t *align, bool shared)
802 {
803     static void *mem;
804 
805     if (mem) {
806         /* we only support one allocation, which is enough for initial ram */
807         return NULL;
808     }
809 
810     mem = mmap((void *) 0x800000000ULL, size,
811                PROT_EXEC|PROT_READ|PROT_WRITE,
812                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
813     if (mem == MAP_FAILED) {
814         mem = NULL;
815     }
816     if (mem && align) {
817         *align = QEMU_VMALLOC_ALIGN;
818     }
819     return mem;
820 }
821 
822 static uint8_t const *sw_bp_inst;
823 static uint8_t sw_bp_ilen;
824 
825 static void determine_sw_breakpoint_instr(void)
826 {
827         /* DIAG 501 is used for sw breakpoints with old kernels */
828         static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
829         /* Instruction 0x0000 is used for sw breakpoints with recent kernels */
830         static const uint8_t instr_0x0000[] = {0x00, 0x00};
831 
832         if (sw_bp_inst) {
833             return;
834         }
835         if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_USER_INSTR0, 0)) {
836             sw_bp_inst = diag_501;
837             sw_bp_ilen = sizeof(diag_501);
838             DPRINTF("KVM: will use 4-byte sw breakpoints.\n");
839         } else {
840             sw_bp_inst = instr_0x0000;
841             sw_bp_ilen = sizeof(instr_0x0000);
842             DPRINTF("KVM: will use 2-byte sw breakpoints.\n");
843         }
844 }
845 
846 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
847 {
848     determine_sw_breakpoint_instr();
849 
850     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
851                             sw_bp_ilen, 0) ||
852         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)sw_bp_inst, sw_bp_ilen, 1)) {
853         return -EINVAL;
854     }
855     return 0;
856 }
857 
858 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
859 {
860     uint8_t t[MAX_ILEN];
861 
862     if (cpu_memory_rw_debug(cs, bp->pc, t, sw_bp_ilen, 0)) {
863         return -EINVAL;
864     } else if (memcmp(t, sw_bp_inst, sw_bp_ilen)) {
865         return -EINVAL;
866     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
867                                    sw_bp_ilen, 1)) {
868         return -EINVAL;
869     }
870 
871     return 0;
872 }
873 
874 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
875                                                     int len, int type)
876 {
877     int n;
878 
879     for (n = 0; n < nb_hw_breakpoints; n++) {
880         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
881             (hw_breakpoints[n].len == len || len == -1)) {
882             return &hw_breakpoints[n];
883         }
884     }
885 
886     return NULL;
887 }
888 
889 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
890 {
891     int size;
892 
893     if (find_hw_breakpoint(addr, len, type)) {
894         return -EEXIST;
895     }
896 
897     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
898 
899     if (!hw_breakpoints) {
900         nb_hw_breakpoints = 0;
901         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
902     } else {
903         hw_breakpoints =
904             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
905     }
906 
907     if (!hw_breakpoints) {
908         nb_hw_breakpoints = 0;
909         return -ENOMEM;
910     }
911 
912     hw_breakpoints[nb_hw_breakpoints].addr = addr;
913     hw_breakpoints[nb_hw_breakpoints].len = len;
914     hw_breakpoints[nb_hw_breakpoints].type = type;
915 
916     nb_hw_breakpoints++;
917 
918     return 0;
919 }
920 
921 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
922                                   target_ulong len, int type)
923 {
924     switch (type) {
925     case GDB_BREAKPOINT_HW:
926         type = KVM_HW_BP;
927         break;
928     case GDB_WATCHPOINT_WRITE:
929         if (len < 1) {
930             return -EINVAL;
931         }
932         type = KVM_HW_WP_WRITE;
933         break;
934     default:
935         return -ENOSYS;
936     }
937     return insert_hw_breakpoint(addr, len, type);
938 }
939 
940 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
941                                   target_ulong len, int type)
942 {
943     int size;
944     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
945 
946     if (bp == NULL) {
947         return -ENOENT;
948     }
949 
950     nb_hw_breakpoints--;
951     if (nb_hw_breakpoints > 0) {
952         /*
953          * In order to trim the array, move the last element to the position to
954          * be removed - if necessary.
955          */
956         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
957             *bp = hw_breakpoints[nb_hw_breakpoints];
958         }
959         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
960         hw_breakpoints =
961              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
962     } else {
963         g_free(hw_breakpoints);
964         hw_breakpoints = NULL;
965     }
966 
967     return 0;
968 }
969 
970 void kvm_arch_remove_all_hw_breakpoints(void)
971 {
972     nb_hw_breakpoints = 0;
973     g_free(hw_breakpoints);
974     hw_breakpoints = NULL;
975 }
976 
977 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
978 {
979     int i;
980 
981     if (nb_hw_breakpoints > 0) {
982         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
983         dbg->arch.hw_bp = hw_breakpoints;
984 
985         for (i = 0; i < nb_hw_breakpoints; ++i) {
986             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
987                                                        hw_breakpoints[i].addr);
988         }
989         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
990     } else {
991         dbg->arch.nr_hw_bp = 0;
992         dbg->arch.hw_bp = NULL;
993     }
994 }
995 
996 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
997 {
998 }
999 
1000 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1001 {
1002     return MEMTXATTRS_UNSPECIFIED;
1003 }
1004 
1005 int kvm_arch_process_async_events(CPUState *cs)
1006 {
1007     return cs->halted;
1008 }
1009 
1010 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
1011                                      struct kvm_s390_interrupt *interrupt)
1012 {
1013     int r = 0;
1014 
1015     interrupt->type = irq->type;
1016     switch (irq->type) {
1017     case KVM_S390_INT_VIRTIO:
1018         interrupt->parm = irq->u.ext.ext_params;
1019         /* fall through */
1020     case KVM_S390_INT_PFAULT_INIT:
1021     case KVM_S390_INT_PFAULT_DONE:
1022         interrupt->parm64 = irq->u.ext.ext_params2;
1023         break;
1024     case KVM_S390_PROGRAM_INT:
1025         interrupt->parm = irq->u.pgm.code;
1026         break;
1027     case KVM_S390_SIGP_SET_PREFIX:
1028         interrupt->parm = irq->u.prefix.address;
1029         break;
1030     case KVM_S390_INT_SERVICE:
1031         interrupt->parm = irq->u.ext.ext_params;
1032         break;
1033     case KVM_S390_MCHK:
1034         interrupt->parm = irq->u.mchk.cr14;
1035         interrupt->parm64 = irq->u.mchk.mcic;
1036         break;
1037     case KVM_S390_INT_EXTERNAL_CALL:
1038         interrupt->parm = irq->u.extcall.code;
1039         break;
1040     case KVM_S390_INT_EMERGENCY:
1041         interrupt->parm = irq->u.emerg.code;
1042         break;
1043     case KVM_S390_SIGP_STOP:
1044     case KVM_S390_RESTART:
1045         break; /* These types have no parameters */
1046     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1047         interrupt->parm = irq->u.io.subchannel_id << 16;
1048         interrupt->parm |= irq->u.io.subchannel_nr;
1049         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
1050         interrupt->parm64 |= irq->u.io.io_int_word;
1051         break;
1052     default:
1053         r = -EINVAL;
1054         break;
1055     }
1056     return r;
1057 }
1058 
1059 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
1060 {
1061     struct kvm_s390_interrupt kvmint = {};
1062     int r;
1063 
1064     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1065     if (r < 0) {
1066         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1067         exit(1);
1068     }
1069 
1070     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
1071     if (r < 0) {
1072         fprintf(stderr, "KVM failed to inject interrupt\n");
1073         exit(1);
1074     }
1075 }
1076 
1077 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
1078 {
1079     CPUState *cs = CPU(cpu);
1080     int r;
1081 
1082     if (cap_s390_irq) {
1083         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
1084         if (!r) {
1085             return;
1086         }
1087         error_report("KVM failed to inject interrupt %llx", irq->type);
1088         exit(1);
1089     }
1090 
1091     inject_vcpu_irq_legacy(cs, irq);
1092 }
1093 
1094 void kvm_s390_floating_interrupt_legacy(struct kvm_s390_irq *irq)
1095 {
1096     struct kvm_s390_interrupt kvmint = {};
1097     int r;
1098 
1099     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1100     if (r < 0) {
1101         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1102         exit(1);
1103     }
1104 
1105     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
1106     if (r < 0) {
1107         fprintf(stderr, "KVM failed to inject interrupt\n");
1108         exit(1);
1109     }
1110 }
1111 
1112 void kvm_s390_program_interrupt(S390CPU *cpu, uint16_t code)
1113 {
1114     struct kvm_s390_irq irq = {
1115         .type = KVM_S390_PROGRAM_INT,
1116         .u.pgm.code = code,
1117     };
1118 
1119     kvm_s390_vcpu_interrupt(cpu, &irq);
1120 }
1121 
1122 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
1123 {
1124     struct kvm_s390_irq irq = {
1125         .type = KVM_S390_PROGRAM_INT,
1126         .u.pgm.code = code,
1127         .u.pgm.trans_exc_code = te_code,
1128         .u.pgm.exc_access_id = te_code & 3,
1129     };
1130 
1131     kvm_s390_vcpu_interrupt(cpu, &irq);
1132 }
1133 
1134 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
1135                                  uint16_t ipbh0)
1136 {
1137     CPUS390XState *env = &cpu->env;
1138     uint64_t sccb;
1139     uint32_t code;
1140     int r = 0;
1141 
1142     sccb = env->regs[ipbh0 & 0xf];
1143     code = env->regs[(ipbh0 & 0xf0) >> 4];
1144 
1145     r = sclp_service_call(env, sccb, code);
1146     if (r < 0) {
1147         kvm_s390_program_interrupt(cpu, -r);
1148     } else {
1149         setcc(cpu, r);
1150     }
1151 
1152     return 0;
1153 }
1154 
1155 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1156 {
1157     CPUS390XState *env = &cpu->env;
1158     int rc = 0;
1159     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1160 
1161     switch (ipa1) {
1162     case PRIV_B2_XSCH:
1163         ioinst_handle_xsch(cpu, env->regs[1], RA_IGNORED);
1164         break;
1165     case PRIV_B2_CSCH:
1166         ioinst_handle_csch(cpu, env->regs[1], RA_IGNORED);
1167         break;
1168     case PRIV_B2_HSCH:
1169         ioinst_handle_hsch(cpu, env->regs[1], RA_IGNORED);
1170         break;
1171     case PRIV_B2_MSCH:
1172         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1173         break;
1174     case PRIV_B2_SSCH:
1175         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1176         break;
1177     case PRIV_B2_STCRW:
1178         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb, RA_IGNORED);
1179         break;
1180     case PRIV_B2_STSCH:
1181         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1182         break;
1183     case PRIV_B2_TSCH:
1184         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1185         fprintf(stderr, "Spurious tsch intercept\n");
1186         break;
1187     case PRIV_B2_CHSC:
1188         ioinst_handle_chsc(cpu, run->s390_sieic.ipb, RA_IGNORED);
1189         break;
1190     case PRIV_B2_TPI:
1191         /* This should have been handled by kvm already. */
1192         fprintf(stderr, "Spurious tpi intercept\n");
1193         break;
1194     case PRIV_B2_SCHM:
1195         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1196                            run->s390_sieic.ipb, RA_IGNORED);
1197         break;
1198     case PRIV_B2_RSCH:
1199         ioinst_handle_rsch(cpu, env->regs[1], RA_IGNORED);
1200         break;
1201     case PRIV_B2_RCHP:
1202         ioinst_handle_rchp(cpu, env->regs[1], RA_IGNORED);
1203         break;
1204     case PRIV_B2_STCPS:
1205         /* We do not provide this instruction, it is suppressed. */
1206         break;
1207     case PRIV_B2_SAL:
1208         ioinst_handle_sal(cpu, env->regs[1], RA_IGNORED);
1209         break;
1210     case PRIV_B2_SIGA:
1211         /* Not provided, set CC = 3 for subchannel not operational */
1212         setcc(cpu, 3);
1213         break;
1214     case PRIV_B2_SCLP_CALL:
1215         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1216         break;
1217     default:
1218         rc = -1;
1219         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1220         break;
1221     }
1222 
1223     return rc;
1224 }
1225 
1226 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1227                                   uint8_t *ar)
1228 {
1229     CPUS390XState *env = &cpu->env;
1230     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1231     uint32_t base2 = run->s390_sieic.ipb >> 28;
1232     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1233                      ((run->s390_sieic.ipb & 0xff00) << 4);
1234 
1235     if (disp2 & 0x80000) {
1236         disp2 += 0xfff00000;
1237     }
1238     if (ar) {
1239         *ar = base2;
1240     }
1241 
1242     return (base2 ? env->regs[base2] : 0) +
1243            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1244 }
1245 
1246 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1247                                   uint8_t *ar)
1248 {
1249     CPUS390XState *env = &cpu->env;
1250     uint32_t base2 = run->s390_sieic.ipb >> 28;
1251     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1252                      ((run->s390_sieic.ipb & 0xff00) << 4);
1253 
1254     if (disp2 & 0x80000) {
1255         disp2 += 0xfff00000;
1256     }
1257     if (ar) {
1258         *ar = base2;
1259     }
1260 
1261     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1262 }
1263 
1264 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1265 {
1266     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1267 
1268     if (s390_has_feat(S390_FEAT_ZPCI)) {
1269         return clp_service_call(cpu, r2, RA_IGNORED);
1270     } else {
1271         return -1;
1272     }
1273 }
1274 
1275 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1276 {
1277     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1278     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1279 
1280     if (s390_has_feat(S390_FEAT_ZPCI)) {
1281         return pcilg_service_call(cpu, r1, r2, RA_IGNORED);
1282     } else {
1283         return -1;
1284     }
1285 }
1286 
1287 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1288 {
1289     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1290     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1291 
1292     if (s390_has_feat(S390_FEAT_ZPCI)) {
1293         return pcistg_service_call(cpu, r1, r2, RA_IGNORED);
1294     } else {
1295         return -1;
1296     }
1297 }
1298 
1299 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1300 {
1301     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1302     uint64_t fiba;
1303     uint8_t ar;
1304 
1305     if (s390_has_feat(S390_FEAT_ZPCI)) {
1306         fiba = get_base_disp_rxy(cpu, run, &ar);
1307 
1308         return stpcifc_service_call(cpu, r1, fiba, ar, RA_IGNORED);
1309     } else {
1310         return -1;
1311     }
1312 }
1313 
1314 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1315 {
1316     CPUS390XState *env = &cpu->env;
1317     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1318     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1319     uint8_t isc;
1320     uint16_t mode;
1321     int r;
1322 
1323     mode = env->regs[r1] & 0xffff;
1324     isc = (env->regs[r3] >> 27) & 0x7;
1325     r = css_do_sic(env, isc, mode);
1326     if (r) {
1327         kvm_s390_program_interrupt(cpu, -r);
1328     }
1329 
1330     return 0;
1331 }
1332 
1333 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1334 {
1335     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1336     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1337 
1338     if (s390_has_feat(S390_FEAT_ZPCI)) {
1339         return rpcit_service_call(cpu, r1, r2, RA_IGNORED);
1340     } else {
1341         return -1;
1342     }
1343 }
1344 
1345 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1346 {
1347     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1348     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1349     uint64_t gaddr;
1350     uint8_t ar;
1351 
1352     if (s390_has_feat(S390_FEAT_ZPCI)) {
1353         gaddr = get_base_disp_rsy(cpu, run, &ar);
1354 
1355         return pcistb_service_call(cpu, r1, r3, gaddr, ar, RA_IGNORED);
1356     } else {
1357         return -1;
1358     }
1359 }
1360 
1361 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1362 {
1363     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1364     uint64_t fiba;
1365     uint8_t ar;
1366 
1367     if (s390_has_feat(S390_FEAT_ZPCI)) {
1368         fiba = get_base_disp_rxy(cpu, run, &ar);
1369 
1370         return mpcifc_service_call(cpu, r1, fiba, ar, RA_IGNORED);
1371     } else {
1372         return -1;
1373     }
1374 }
1375 
1376 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1377 {
1378     int r = 0;
1379 
1380     switch (ipa1) {
1381     case PRIV_B9_CLP:
1382         r = kvm_clp_service_call(cpu, run);
1383         break;
1384     case PRIV_B9_PCISTG:
1385         r = kvm_pcistg_service_call(cpu, run);
1386         break;
1387     case PRIV_B9_PCILG:
1388         r = kvm_pcilg_service_call(cpu, run);
1389         break;
1390     case PRIV_B9_RPCIT:
1391         r = kvm_rpcit_service_call(cpu, run);
1392         break;
1393     case PRIV_B9_EQBS:
1394         /* just inject exception */
1395         r = -1;
1396         break;
1397     default:
1398         r = -1;
1399         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1400         break;
1401     }
1402 
1403     return r;
1404 }
1405 
1406 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1407 {
1408     int r = 0;
1409 
1410     switch (ipbl) {
1411     case PRIV_EB_PCISTB:
1412         r = kvm_pcistb_service_call(cpu, run);
1413         break;
1414     case PRIV_EB_SIC:
1415         r = kvm_sic_service_call(cpu, run);
1416         break;
1417     case PRIV_EB_SQBS:
1418         /* just inject exception */
1419         r = -1;
1420         break;
1421     default:
1422         r = -1;
1423         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1424         break;
1425     }
1426 
1427     return r;
1428 }
1429 
1430 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1431 {
1432     int r = 0;
1433 
1434     switch (ipbl) {
1435     case PRIV_E3_MPCIFC:
1436         r = kvm_mpcifc_service_call(cpu, run);
1437         break;
1438     case PRIV_E3_STPCIFC:
1439         r = kvm_stpcifc_service_call(cpu, run);
1440         break;
1441     default:
1442         r = -1;
1443         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1444         break;
1445     }
1446 
1447     return r;
1448 }
1449 
1450 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1451 {
1452     CPUS390XState *env = &cpu->env;
1453     int ret;
1454 
1455     ret = s390_virtio_hypercall(env);
1456     if (ret == -EINVAL) {
1457         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1458         return 0;
1459     }
1460 
1461     return ret;
1462 }
1463 
1464 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1465 {
1466     uint64_t r1, r3;
1467     int rc;
1468 
1469     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1470     r3 = run->s390_sieic.ipa & 0x000f;
1471     rc = handle_diag_288(&cpu->env, r1, r3);
1472     if (rc) {
1473         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1474     }
1475 }
1476 
1477 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1478 {
1479     uint64_t r1, r3;
1480 
1481     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1482     r3 = run->s390_sieic.ipa & 0x000f;
1483     handle_diag_308(&cpu->env, r1, r3, RA_IGNORED);
1484 }
1485 
1486 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1487 {
1488     CPUS390XState *env = &cpu->env;
1489     unsigned long pc;
1490 
1491     pc = env->psw.addr - sw_bp_ilen;
1492     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1493         env->psw.addr = pc;
1494         return EXCP_DEBUG;
1495     }
1496 
1497     return -ENOENT;
1498 }
1499 
1500 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1501 
1502 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1503 {
1504     int r = 0;
1505     uint16_t func_code;
1506 
1507     /*
1508      * For any diagnose call we support, bits 48-63 of the resulting
1509      * address specify the function code; the remainder is ignored.
1510      */
1511     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1512     switch (func_code) {
1513     case DIAG_TIMEREVENT:
1514         kvm_handle_diag_288(cpu, run);
1515         break;
1516     case DIAG_IPL:
1517         kvm_handle_diag_308(cpu, run);
1518         break;
1519     case DIAG_KVM_HYPERCALL:
1520         r = handle_hypercall(cpu, run);
1521         break;
1522     case DIAG_KVM_BREAKPOINT:
1523         r = handle_sw_breakpoint(cpu, run);
1524         break;
1525     default:
1526         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1527         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1528         break;
1529     }
1530 
1531     return r;
1532 }
1533 
1534 static int kvm_s390_handle_sigp(S390CPU *cpu, uint8_t ipa1, uint32_t ipb)
1535 {
1536     CPUS390XState *env = &cpu->env;
1537     const uint8_t r1 = ipa1 >> 4;
1538     const uint8_t r3 = ipa1 & 0x0f;
1539     int ret;
1540     uint8_t order;
1541 
1542     /* get order code */
1543     order = decode_basedisp_rs(env, ipb, NULL) & SIGP_ORDER_MASK;
1544 
1545     ret = handle_sigp(env, order, r1, r3);
1546     setcc(cpu, ret);
1547     return 0;
1548 }
1549 
1550 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1551 {
1552     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1553     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1554     int r = -1;
1555 
1556     DPRINTF("handle_instruction 0x%x 0x%x\n",
1557             run->s390_sieic.ipa, run->s390_sieic.ipb);
1558     switch (ipa0) {
1559     case IPA0_B2:
1560         r = handle_b2(cpu, run, ipa1);
1561         break;
1562     case IPA0_B9:
1563         r = handle_b9(cpu, run, ipa1);
1564         break;
1565     case IPA0_EB:
1566         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1567         break;
1568     case IPA0_E3:
1569         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1570         break;
1571     case IPA0_DIAG:
1572         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1573         break;
1574     case IPA0_SIGP:
1575         r = kvm_s390_handle_sigp(cpu, ipa1, run->s390_sieic.ipb);
1576         break;
1577     }
1578 
1579     if (r < 0) {
1580         r = 0;
1581         kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1582     }
1583 
1584     return r;
1585 }
1586 
1587 static void unmanageable_intercept(S390CPU *cpu, S390CrashReason reason,
1588                                    int pswoffset)
1589 {
1590     CPUState *cs = CPU(cpu);
1591 
1592     s390_cpu_halt(cpu);
1593     cpu->env.crash_reason = reason;
1594     qemu_system_guest_panicked(cpu_get_crash_info(cs));
1595 }
1596 
1597 /* try to detect pgm check loops */
1598 static int handle_oper_loop(S390CPU *cpu, struct kvm_run *run)
1599 {
1600     CPUState *cs = CPU(cpu);
1601     PSW oldpsw, newpsw;
1602 
1603     newpsw.mask = ldq_phys(cs->as, cpu->env.psa +
1604                            offsetof(LowCore, program_new_psw));
1605     newpsw.addr = ldq_phys(cs->as, cpu->env.psa +
1606                            offsetof(LowCore, program_new_psw) + 8);
1607     oldpsw.mask  = run->psw_mask;
1608     oldpsw.addr  = run->psw_addr;
1609     /*
1610      * Avoid endless loops of operation exceptions, if the pgm new
1611      * PSW will cause a new operation exception.
1612      * The heuristic checks if the pgm new psw is within 6 bytes before
1613      * the faulting psw address (with same DAT, AS settings) and the
1614      * new psw is not a wait psw and the fault was not triggered by
1615      * problem state. In that case go into crashed state.
1616      */
1617 
1618     if (oldpsw.addr - newpsw.addr <= 6 &&
1619         !(newpsw.mask & PSW_MASK_WAIT) &&
1620         !(oldpsw.mask & PSW_MASK_PSTATE) &&
1621         (newpsw.mask & PSW_MASK_ASC) == (oldpsw.mask & PSW_MASK_ASC) &&
1622         (newpsw.mask & PSW_MASK_DAT) == (oldpsw.mask & PSW_MASK_DAT)) {
1623         unmanageable_intercept(cpu, S390_CRASH_REASON_OPINT_LOOP,
1624                                offsetof(LowCore, program_new_psw));
1625         return EXCP_HALTED;
1626     }
1627     return 0;
1628 }
1629 
1630 static int handle_intercept(S390CPU *cpu)
1631 {
1632     CPUState *cs = CPU(cpu);
1633     struct kvm_run *run = cs->kvm_run;
1634     int icpt_code = run->s390_sieic.icptcode;
1635     int r = 0;
1636 
1637     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1638             (long)cs->kvm_run->psw_addr);
1639     switch (icpt_code) {
1640         case ICPT_INSTRUCTION:
1641             r = handle_instruction(cpu, run);
1642             break;
1643         case ICPT_PROGRAM:
1644             unmanageable_intercept(cpu, S390_CRASH_REASON_PGMINT_LOOP,
1645                                    offsetof(LowCore, program_new_psw));
1646             r = EXCP_HALTED;
1647             break;
1648         case ICPT_EXT_INT:
1649             unmanageable_intercept(cpu, S390_CRASH_REASON_EXTINT_LOOP,
1650                                    offsetof(LowCore, external_new_psw));
1651             r = EXCP_HALTED;
1652             break;
1653         case ICPT_WAITPSW:
1654             /* disabled wait, since enabled wait is handled in kernel */
1655             s390_handle_wait(cpu);
1656             r = EXCP_HALTED;
1657             break;
1658         case ICPT_CPU_STOP:
1659             do_stop_interrupt(&cpu->env);
1660             r = EXCP_HALTED;
1661             break;
1662         case ICPT_OPEREXC:
1663             /* check for break points */
1664             r = handle_sw_breakpoint(cpu, run);
1665             if (r == -ENOENT) {
1666                 /* Then check for potential pgm check loops */
1667                 r = handle_oper_loop(cpu, run);
1668                 if (r == 0) {
1669                     kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1670                 }
1671             }
1672             break;
1673         case ICPT_SOFT_INTERCEPT:
1674             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1675             exit(1);
1676             break;
1677         case ICPT_IO:
1678             fprintf(stderr, "KVM unimplemented icpt IO\n");
1679             exit(1);
1680             break;
1681         default:
1682             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1683             exit(1);
1684             break;
1685     }
1686 
1687     return r;
1688 }
1689 
1690 static int handle_tsch(S390CPU *cpu)
1691 {
1692     CPUState *cs = CPU(cpu);
1693     struct kvm_run *run = cs->kvm_run;
1694     int ret;
1695 
1696     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb,
1697                              RA_IGNORED);
1698     if (ret < 0) {
1699         /*
1700          * Failure.
1701          * If an I/O interrupt had been dequeued, we have to reinject it.
1702          */
1703         if (run->s390_tsch.dequeued) {
1704             s390_io_interrupt(run->s390_tsch.subchannel_id,
1705                               run->s390_tsch.subchannel_nr,
1706                               run->s390_tsch.io_int_parm,
1707                               run->s390_tsch.io_int_word);
1708         }
1709         ret = 0;
1710     }
1711     return ret;
1712 }
1713 
1714 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1715 {
1716     SysIB_322 sysib;
1717     int del;
1718 
1719     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1720         return;
1721     }
1722     /* Shift the stack of Extended Names to prepare for our own data */
1723     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1724             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1725     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1726      * assumed it's not capable of managing Extended Names for lower levels.
1727      */
1728     for (del = 1; del < sysib.count; del++) {
1729         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1730             break;
1731         }
1732     }
1733     if (del < sysib.count) {
1734         memset(sysib.ext_names[del], 0,
1735                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1736     }
1737     /* Insert short machine name in EBCDIC, padded with blanks */
1738     if (qemu_name) {
1739         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1740         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1741                                                     strlen(qemu_name)));
1742     }
1743     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1744     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1745     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1746      * considered by s390 as not capable of providing any Extended Name.
1747      * Therefore if no name was specified on qemu invocation, we go with the
1748      * same "KVMguest" default, which KVM has filled into short name field.
1749      */
1750     if (qemu_name) {
1751         strncpy((char *)sysib.ext_names[0], qemu_name,
1752                 sizeof(sysib.ext_names[0]));
1753     } else {
1754         strcpy((char *)sysib.ext_names[0], "KVMguest");
1755     }
1756     /* Insert UUID */
1757     memcpy(sysib.vm[0].uuid, &qemu_uuid, sizeof(sysib.vm[0].uuid));
1758 
1759     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1760 }
1761 
1762 static int handle_stsi(S390CPU *cpu)
1763 {
1764     CPUState *cs = CPU(cpu);
1765     struct kvm_run *run = cs->kvm_run;
1766 
1767     switch (run->s390_stsi.fc) {
1768     case 3:
1769         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1770             return 0;
1771         }
1772         /* Only sysib 3.2.2 needs post-handling for now. */
1773         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1774         return 0;
1775     default:
1776         return 0;
1777     }
1778 }
1779 
1780 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1781 {
1782     CPUState *cs = CPU(cpu);
1783     struct kvm_run *run = cs->kvm_run;
1784 
1785     int ret = 0;
1786     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1787 
1788     switch (arch_info->type) {
1789     case KVM_HW_WP_WRITE:
1790         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1791             cs->watchpoint_hit = &hw_watchpoint;
1792             hw_watchpoint.vaddr = arch_info->addr;
1793             hw_watchpoint.flags = BP_MEM_WRITE;
1794             ret = EXCP_DEBUG;
1795         }
1796         break;
1797     case KVM_HW_BP:
1798         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1799             ret = EXCP_DEBUG;
1800         }
1801         break;
1802     case KVM_SINGLESTEP:
1803         if (cs->singlestep_enabled) {
1804             ret = EXCP_DEBUG;
1805         }
1806         break;
1807     default:
1808         ret = -ENOSYS;
1809     }
1810 
1811     return ret;
1812 }
1813 
1814 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1815 {
1816     S390CPU *cpu = S390_CPU(cs);
1817     int ret = 0;
1818 
1819     qemu_mutex_lock_iothread();
1820 
1821     kvm_cpu_synchronize_state(cs);
1822 
1823     switch (run->exit_reason) {
1824         case KVM_EXIT_S390_SIEIC:
1825             ret = handle_intercept(cpu);
1826             break;
1827         case KVM_EXIT_S390_RESET:
1828             s390_ipl_reset_request(cs, S390_RESET_REIPL);
1829             break;
1830         case KVM_EXIT_S390_TSCH:
1831             ret = handle_tsch(cpu);
1832             break;
1833         case KVM_EXIT_S390_STSI:
1834             ret = handle_stsi(cpu);
1835             break;
1836         case KVM_EXIT_DEBUG:
1837             ret = kvm_arch_handle_debug_exit(cpu);
1838             break;
1839         default:
1840             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1841             break;
1842     }
1843     qemu_mutex_unlock_iothread();
1844 
1845     if (ret == 0) {
1846         ret = EXCP_INTERRUPT;
1847     }
1848     return ret;
1849 }
1850 
1851 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1852 {
1853     return true;
1854 }
1855 
1856 void kvm_s390_enable_css_support(S390CPU *cpu)
1857 {
1858     int r;
1859 
1860     /* Activate host kernel channel subsystem support. */
1861     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1862     assert(r == 0);
1863 }
1864 
1865 void kvm_arch_init_irq_routing(KVMState *s)
1866 {
1867     /*
1868      * Note that while irqchip capabilities generally imply that cpustates
1869      * are handled in-kernel, it is not true for s390 (yet); therefore, we
1870      * have to override the common code kvm_halt_in_kernel_allowed setting.
1871      */
1872     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
1873         kvm_gsi_routing_allowed = true;
1874         kvm_halt_in_kernel_allowed = false;
1875     }
1876 }
1877 
1878 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
1879                                     int vq, bool assign)
1880 {
1881     struct kvm_ioeventfd kick = {
1882         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
1883         KVM_IOEVENTFD_FLAG_DATAMATCH,
1884         .fd = event_notifier_get_fd(notifier),
1885         .datamatch = vq,
1886         .addr = sch,
1887         .len = 8,
1888     };
1889     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
1890         return -ENOSYS;
1891     }
1892     if (!assign) {
1893         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
1894     }
1895     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
1896 }
1897 
1898 int kvm_s390_get_ri(void)
1899 {
1900     return cap_ri;
1901 }
1902 
1903 int kvm_s390_get_gs(void)
1904 {
1905     return cap_gs;
1906 }
1907 
1908 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
1909 {
1910     struct kvm_mp_state mp_state = {};
1911     int ret;
1912 
1913     /* the kvm part might not have been initialized yet */
1914     if (CPU(cpu)->kvm_state == NULL) {
1915         return 0;
1916     }
1917 
1918     switch (cpu_state) {
1919     case S390_CPU_STATE_STOPPED:
1920         mp_state.mp_state = KVM_MP_STATE_STOPPED;
1921         break;
1922     case S390_CPU_STATE_CHECK_STOP:
1923         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
1924         break;
1925     case S390_CPU_STATE_OPERATING:
1926         mp_state.mp_state = KVM_MP_STATE_OPERATING;
1927         break;
1928     case S390_CPU_STATE_LOAD:
1929         mp_state.mp_state = KVM_MP_STATE_LOAD;
1930         break;
1931     default:
1932         error_report("Requested CPU state is not a valid S390 CPU state: %u",
1933                      cpu_state);
1934         exit(1);
1935     }
1936 
1937     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1938     if (ret) {
1939         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
1940                                        strerror(-ret));
1941     }
1942 
1943     return ret;
1944 }
1945 
1946 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
1947 {
1948     struct kvm_s390_irq_state irq_state = {
1949         .buf = (uint64_t) cpu->irqstate,
1950         .len = VCPU_IRQ_BUF_SIZE,
1951     };
1952     CPUState *cs = CPU(cpu);
1953     int32_t bytes;
1954 
1955     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
1956         return;
1957     }
1958 
1959     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
1960     if (bytes < 0) {
1961         cpu->irqstate_saved_size = 0;
1962         error_report("Migration of interrupt state failed");
1963         return;
1964     }
1965 
1966     cpu->irqstate_saved_size = bytes;
1967 }
1968 
1969 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
1970 {
1971     CPUState *cs = CPU(cpu);
1972     struct kvm_s390_irq_state irq_state = {
1973         .buf = (uint64_t) cpu->irqstate,
1974         .len = cpu->irqstate_saved_size,
1975     };
1976     int r;
1977 
1978     if (cpu->irqstate_saved_size == 0) {
1979         return 0;
1980     }
1981 
1982     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
1983         return -ENOSYS;
1984     }
1985 
1986     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
1987     if (r) {
1988         error_report("Setting interrupt state failed %d", r);
1989     }
1990     return r;
1991 }
1992 
1993 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1994                              uint64_t address, uint32_t data, PCIDevice *dev)
1995 {
1996     S390PCIBusDevice *pbdev;
1997     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
1998 
1999     if (!dev) {
2000         DPRINTF("add_msi_route no pci device\n");
2001         return -ENODEV;
2002     }
2003 
2004     pbdev = s390_pci_find_dev_by_target(s390_get_phb(), DEVICE(dev)->id);
2005     if (!pbdev) {
2006         DPRINTF("add_msi_route no zpci device\n");
2007         return -ENODEV;
2008     }
2009 
2010     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2011     route->flags = 0;
2012     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2013     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2014     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2015     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset + vec;
2016     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2017     return 0;
2018 }
2019 
2020 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2021                                 int vector, PCIDevice *dev)
2022 {
2023     return 0;
2024 }
2025 
2026 int kvm_arch_release_virq_post(int virq)
2027 {
2028     return 0;
2029 }
2030 
2031 int kvm_arch_msi_data_to_gsi(uint32_t data)
2032 {
2033     abort();
2034 }
2035 
2036 static int query_cpu_subfunc(S390FeatBitmap features)
2037 {
2038     struct kvm_s390_vm_cpu_subfunc prop;
2039     struct kvm_device_attr attr = {
2040         .group = KVM_S390_VM_CPU_MODEL,
2041         .attr = KVM_S390_VM_CPU_MACHINE_SUBFUNC,
2042         .addr = (uint64_t) &prop,
2043     };
2044     int rc;
2045 
2046     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2047     if (rc) {
2048         return  rc;
2049     }
2050 
2051     /*
2052      * We're going to add all subfunctions now, if the corresponding feature
2053      * is available that unlocks the query functions.
2054      */
2055     s390_add_from_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2056     if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2057         s390_add_from_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2058     }
2059     if (test_bit(S390_FEAT_MSA, features)) {
2060         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2061         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2062         s390_add_from_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2063         s390_add_from_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2064         s390_add_from_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2065     }
2066     if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2067         s390_add_from_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2068     }
2069     if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2070         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2071         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2072         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2073         s390_add_from_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2074     }
2075     if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2076         s390_add_from_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2077     }
2078     if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2079         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2080     }
2081     return 0;
2082 }
2083 
2084 static int configure_cpu_subfunc(const S390FeatBitmap features)
2085 {
2086     struct kvm_s390_vm_cpu_subfunc prop = {};
2087     struct kvm_device_attr attr = {
2088         .group = KVM_S390_VM_CPU_MODEL,
2089         .attr = KVM_S390_VM_CPU_PROCESSOR_SUBFUNC,
2090         .addr = (uint64_t) &prop,
2091     };
2092 
2093     if (!kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2094                            KVM_S390_VM_CPU_PROCESSOR_SUBFUNC)) {
2095         /* hardware support might be missing, IBC will handle most of this */
2096         return 0;
2097     }
2098 
2099     s390_fill_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2100     if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2101         s390_fill_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2102     }
2103     if (test_bit(S390_FEAT_MSA, features)) {
2104         s390_fill_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2105         s390_fill_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2106         s390_fill_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2107         s390_fill_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2108         s390_fill_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2109     }
2110     if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2111         s390_fill_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2112     }
2113     if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2114         s390_fill_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2115         s390_fill_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2116         s390_fill_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2117         s390_fill_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2118     }
2119     if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2120         s390_fill_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2121     }
2122     if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2123         s390_fill_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2124     }
2125     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2126 }
2127 
2128 static int kvm_to_feat[][2] = {
2129     { KVM_S390_VM_CPU_FEAT_ESOP, S390_FEAT_ESOP },
2130     { KVM_S390_VM_CPU_FEAT_SIEF2, S390_FEAT_SIE_F2 },
2131     { KVM_S390_VM_CPU_FEAT_64BSCAO , S390_FEAT_SIE_64BSCAO },
2132     { KVM_S390_VM_CPU_FEAT_SIIF, S390_FEAT_SIE_SIIF },
2133     { KVM_S390_VM_CPU_FEAT_GPERE, S390_FEAT_SIE_GPERE },
2134     { KVM_S390_VM_CPU_FEAT_GSLS, S390_FEAT_SIE_GSLS },
2135     { KVM_S390_VM_CPU_FEAT_IB, S390_FEAT_SIE_IB },
2136     { KVM_S390_VM_CPU_FEAT_CEI, S390_FEAT_SIE_CEI },
2137     { KVM_S390_VM_CPU_FEAT_IBS, S390_FEAT_SIE_IBS },
2138     { KVM_S390_VM_CPU_FEAT_SKEY, S390_FEAT_SIE_SKEY },
2139     { KVM_S390_VM_CPU_FEAT_CMMA, S390_FEAT_SIE_CMMA },
2140     { KVM_S390_VM_CPU_FEAT_PFMFI, S390_FEAT_SIE_PFMFI},
2141     { KVM_S390_VM_CPU_FEAT_SIGPIF, S390_FEAT_SIE_SIGPIF},
2142     { KVM_S390_VM_CPU_FEAT_KSS, S390_FEAT_SIE_KSS},
2143 };
2144 
2145 static int query_cpu_feat(S390FeatBitmap features)
2146 {
2147     struct kvm_s390_vm_cpu_feat prop;
2148     struct kvm_device_attr attr = {
2149         .group = KVM_S390_VM_CPU_MODEL,
2150         .attr = KVM_S390_VM_CPU_MACHINE_FEAT,
2151         .addr = (uint64_t) &prop,
2152     };
2153     int rc;
2154     int i;
2155 
2156     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2157     if (rc) {
2158         return  rc;
2159     }
2160 
2161     for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2162         if (test_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat)) {
2163             set_bit(kvm_to_feat[i][1], features);
2164         }
2165     }
2166     return 0;
2167 }
2168 
2169 static int configure_cpu_feat(const S390FeatBitmap features)
2170 {
2171     struct kvm_s390_vm_cpu_feat prop = {};
2172     struct kvm_device_attr attr = {
2173         .group = KVM_S390_VM_CPU_MODEL,
2174         .attr = KVM_S390_VM_CPU_PROCESSOR_FEAT,
2175         .addr = (uint64_t) &prop,
2176     };
2177     int i;
2178 
2179     for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2180         if (test_bit(kvm_to_feat[i][1], features)) {
2181             set_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat);
2182         }
2183     }
2184     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2185 }
2186 
2187 bool kvm_s390_cpu_models_supported(void)
2188 {
2189     if (!cpu_model_allowed()) {
2190         /* compatibility machines interfere with the cpu model */
2191         return false;
2192     }
2193     return kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2194                              KVM_S390_VM_CPU_MACHINE) &&
2195            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2196                              KVM_S390_VM_CPU_PROCESSOR) &&
2197            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2198                              KVM_S390_VM_CPU_MACHINE_FEAT) &&
2199            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2200                              KVM_S390_VM_CPU_PROCESSOR_FEAT) &&
2201            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2202                              KVM_S390_VM_CPU_MACHINE_SUBFUNC);
2203 }
2204 
2205 void kvm_s390_get_host_cpu_model(S390CPUModel *model, Error **errp)
2206 {
2207     struct kvm_s390_vm_cpu_machine prop = {};
2208     struct kvm_device_attr attr = {
2209         .group = KVM_S390_VM_CPU_MODEL,
2210         .attr = KVM_S390_VM_CPU_MACHINE,
2211         .addr = (uint64_t) &prop,
2212     };
2213     uint16_t unblocked_ibc = 0, cpu_type = 0;
2214     int rc;
2215 
2216     memset(model, 0, sizeof(*model));
2217 
2218     if (!kvm_s390_cpu_models_supported()) {
2219         error_setg(errp, "KVM doesn't support CPU models");
2220         return;
2221     }
2222 
2223     /* query the basic cpu model properties */
2224     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2225     if (rc) {
2226         error_setg(errp, "KVM: Error querying host CPU model: %d", rc);
2227         return;
2228     }
2229 
2230     cpu_type = cpuid_type(prop.cpuid);
2231     if (has_ibc(prop.ibc)) {
2232         model->lowest_ibc = lowest_ibc(prop.ibc);
2233         unblocked_ibc = unblocked_ibc(prop.ibc);
2234     }
2235     model->cpu_id = cpuid_id(prop.cpuid);
2236     model->cpu_id_format = cpuid_format(prop.cpuid);
2237     model->cpu_ver = 0xff;
2238 
2239     /* get supported cpu features indicated via STFL(E) */
2240     s390_add_from_feat_block(model->features, S390_FEAT_TYPE_STFL,
2241                              (uint8_t *) prop.fac_mask);
2242     /* dat-enhancement facility 2 has no bit but was introduced with stfle */
2243     if (test_bit(S390_FEAT_STFLE, model->features)) {
2244         set_bit(S390_FEAT_DAT_ENH_2, model->features);
2245     }
2246     /* get supported cpu features indicated e.g. via SCLP */
2247     rc = query_cpu_feat(model->features);
2248     if (rc) {
2249         error_setg(errp, "KVM: Error querying CPU features: %d", rc);
2250         return;
2251     }
2252     /* get supported cpu subfunctions indicated via query / test bit */
2253     rc = query_cpu_subfunc(model->features);
2254     if (rc) {
2255         error_setg(errp, "KVM: Error querying CPU subfunctions: %d", rc);
2256         return;
2257     }
2258 
2259     /* PTFF subfunctions might be indicated although kernel support missing */
2260     if (!test_bit(S390_FEAT_MULTIPLE_EPOCH, model->features)) {
2261         clear_bit(S390_FEAT_PTFF_QSIE, model->features);
2262         clear_bit(S390_FEAT_PTFF_QTOUE, model->features);
2263         clear_bit(S390_FEAT_PTFF_STOE, model->features);
2264         clear_bit(S390_FEAT_PTFF_STOUE, model->features);
2265     }
2266 
2267     /* with cpu model support, CMM is only indicated if really available */
2268     if (kvm_s390_cmma_available()) {
2269         set_bit(S390_FEAT_CMM, model->features);
2270     } else {
2271         /* no cmm -> no cmm nt */
2272         clear_bit(S390_FEAT_CMM_NT, model->features);
2273     }
2274 
2275     /* bpb needs kernel support for migration, VSIE and reset */
2276     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_BPB)) {
2277         clear_bit(S390_FEAT_BPB, model->features);
2278     }
2279 
2280     /* We emulate a zPCI bus and AEN, therefore we don't need HW support */
2281     if (pci_available) {
2282         set_bit(S390_FEAT_ZPCI, model->features);
2283     }
2284     set_bit(S390_FEAT_ADAPTER_EVENT_NOTIFICATION, model->features);
2285 
2286     if (s390_known_cpu_type(cpu_type)) {
2287         /* we want the exact model, even if some features are missing */
2288         model->def = s390_find_cpu_def(cpu_type, ibc_gen(unblocked_ibc),
2289                                        ibc_ec_ga(unblocked_ibc), NULL);
2290     } else {
2291         /* model unknown, e.g. too new - search using features */
2292         model->def = s390_find_cpu_def(0, ibc_gen(unblocked_ibc),
2293                                        ibc_ec_ga(unblocked_ibc),
2294                                        model->features);
2295     }
2296     if (!model->def) {
2297         error_setg(errp, "KVM: host CPU model could not be identified");
2298         return;
2299     }
2300     /* strip of features that are not part of the maximum model */
2301     bitmap_and(model->features, model->features, model->def->full_feat,
2302                S390_FEAT_MAX);
2303 }
2304 
2305 void kvm_s390_apply_cpu_model(const S390CPUModel *model, Error **errp)
2306 {
2307     struct kvm_s390_vm_cpu_processor prop  = {
2308         .fac_list = { 0 },
2309     };
2310     struct kvm_device_attr attr = {
2311         .group = KVM_S390_VM_CPU_MODEL,
2312         .attr = KVM_S390_VM_CPU_PROCESSOR,
2313         .addr = (uint64_t) &prop,
2314     };
2315     int rc;
2316 
2317     if (!model) {
2318         /* compatibility handling if cpu models are disabled */
2319         if (kvm_s390_cmma_available()) {
2320             kvm_s390_enable_cmma();
2321         }
2322         return;
2323     }
2324     if (!kvm_s390_cpu_models_supported()) {
2325         error_setg(errp, "KVM doesn't support CPU models");
2326         return;
2327     }
2328     prop.cpuid = s390_cpuid_from_cpu_model(model);
2329     prop.ibc = s390_ibc_from_cpu_model(model);
2330     /* configure cpu features indicated via STFL(e) */
2331     s390_fill_feat_block(model->features, S390_FEAT_TYPE_STFL,
2332                          (uint8_t *) prop.fac_list);
2333     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2334     if (rc) {
2335         error_setg(errp, "KVM: Error configuring the CPU model: %d", rc);
2336         return;
2337     }
2338     /* configure cpu features indicated e.g. via SCLP */
2339     rc = configure_cpu_feat(model->features);
2340     if (rc) {
2341         error_setg(errp, "KVM: Error configuring CPU features: %d", rc);
2342         return;
2343     }
2344     /* configure cpu subfunctions indicated via query / test bit */
2345     rc = configure_cpu_subfunc(model->features);
2346     if (rc) {
2347         error_setg(errp, "KVM: Error configuring CPU subfunctions: %d", rc);
2348         return;
2349     }
2350     /* enable CMM via CMMA */
2351     if (test_bit(S390_FEAT_CMM, model->features)) {
2352         kvm_s390_enable_cmma();
2353     }
2354 }
2355 
2356 void kvm_s390_restart_interrupt(S390CPU *cpu)
2357 {
2358     struct kvm_s390_irq irq = {
2359         .type = KVM_S390_RESTART,
2360     };
2361 
2362     kvm_s390_vcpu_interrupt(cpu, &irq);
2363 }
2364 
2365 void kvm_s390_stop_interrupt(S390CPU *cpu)
2366 {
2367     struct kvm_s390_irq irq = {
2368         .type = KVM_S390_SIGP_STOP,
2369     };
2370 
2371     kvm_s390_vcpu_interrupt(cpu, &irq);
2372 }
2373