xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision d86c25b292bf9685615cbb93d7b07f33a9b63104)
1 /*
2  * RISC-V implementation of KVM hooks
3  *
4  * Copyright (c) 2020 Huawei Technologies Co., Ltd
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms and conditions of the GNU General Public License,
8  * version 2 or later, as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along with
16  * this program.  If not, see <http://www.gnu.org/licenses/>.
17  */
18 
19 #include "qemu/osdep.h"
20 #include <sys/ioctl.h>
21 
22 #include <linux/kvm.h>
23 
24 #include "qemu/timer.h"
25 #include "qapi/error.h"
26 #include "qemu/error-report.h"
27 #include "qemu/main-loop.h"
28 #include "qapi/visitor.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/kvm.h"
31 #include "sysemu/kvm_int.h"
32 #include "cpu.h"
33 #include "trace.h"
34 #include "hw/core/accel-cpu.h"
35 #include "hw/pci/pci.h"
36 #include "exec/memattrs.h"
37 #include "exec/address-spaces.h"
38 #include "hw/boards.h"
39 #include "hw/irq.h"
40 #include "hw/intc/riscv_imsic.h"
41 #include "qemu/log.h"
42 #include "hw/loader.h"
43 #include "kvm_riscv.h"
44 #include "sbi_ecall_interface.h"
45 #include "chardev/char-fe.h"
46 #include "migration/migration.h"
47 #include "sysemu/runstate.h"
48 #include "hw/riscv/numa.h"
49 
50 void riscv_kvm_aplic_request(void *opaque, int irq, int level)
51 {
52     kvm_set_irq(kvm_state, irq, !!level);
53 }
54 
55 static bool cap_has_mp_state;
56 
57 static uint64_t kvm_riscv_reg_id(CPURISCVState *env, uint64_t type,
58                                  uint64_t idx)
59 {
60     uint64_t id = KVM_REG_RISCV | type | idx;
61 
62     switch (riscv_cpu_mxl(env)) {
63     case MXL_RV32:
64         id |= KVM_REG_SIZE_U32;
65         break;
66     case MXL_RV64:
67         id |= KVM_REG_SIZE_U64;
68         break;
69     default:
70         g_assert_not_reached();
71     }
72     return id;
73 }
74 
75 #define RISCV_CORE_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, \
76                  KVM_REG_RISCV_CORE_REG(name))
77 
78 #define RISCV_CSR_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_CSR, \
79                  KVM_REG_RISCV_CSR_REG(name))
80 
81 #define RISCV_TIMER_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_TIMER, \
82                  KVM_REG_RISCV_TIMER_REG(name))
83 
84 #define RISCV_FP_F_REG(env, idx)  kvm_riscv_reg_id(env, KVM_REG_RISCV_FP_F, idx)
85 
86 #define RISCV_FP_D_REG(env, idx)  kvm_riscv_reg_id(env, KVM_REG_RISCV_FP_D, idx)
87 
88 #define KVM_RISCV_GET_CSR(cs, env, csr, reg) \
89     do { \
90         int ret = kvm_get_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
91         if (ret) { \
92             return ret; \
93         } \
94     } while (0)
95 
96 #define KVM_RISCV_SET_CSR(cs, env, csr, reg) \
97     do { \
98         int ret = kvm_set_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
99         if (ret) { \
100             return ret; \
101         } \
102     } while (0)
103 
104 #define KVM_RISCV_GET_TIMER(cs, env, name, reg) \
105     do { \
106         int ret = kvm_get_one_reg(cs, RISCV_TIMER_REG(env, name), &reg); \
107         if (ret) { \
108             abort(); \
109         } \
110     } while (0)
111 
112 #define KVM_RISCV_SET_TIMER(cs, env, name, reg) \
113     do { \
114         int ret = kvm_set_one_reg(cs, RISCV_TIMER_REG(env, name), &reg); \
115         if (ret) { \
116             abort(); \
117         } \
118     } while (0)
119 
120 typedef struct KVMCPUConfig {
121     const char *name;
122     const char *description;
123     target_ulong offset;
124     int kvm_reg_id;
125     bool user_set;
126     bool supported;
127 } KVMCPUConfig;
128 
129 #define KVM_MISA_CFG(_bit, _reg_id) \
130     {.offset = _bit, .kvm_reg_id = _reg_id}
131 
132 /* KVM ISA extensions */
133 static KVMCPUConfig kvm_misa_ext_cfgs[] = {
134     KVM_MISA_CFG(RVA, KVM_RISCV_ISA_EXT_A),
135     KVM_MISA_CFG(RVC, KVM_RISCV_ISA_EXT_C),
136     KVM_MISA_CFG(RVD, KVM_RISCV_ISA_EXT_D),
137     KVM_MISA_CFG(RVF, KVM_RISCV_ISA_EXT_F),
138     KVM_MISA_CFG(RVH, KVM_RISCV_ISA_EXT_H),
139     KVM_MISA_CFG(RVI, KVM_RISCV_ISA_EXT_I),
140     KVM_MISA_CFG(RVM, KVM_RISCV_ISA_EXT_M),
141 };
142 
143 static void kvm_cpu_set_misa_ext_cfg(Object *obj, Visitor *v,
144                                      const char *name,
145                                      void *opaque, Error **errp)
146 {
147     KVMCPUConfig *misa_ext_cfg = opaque;
148     target_ulong misa_bit = misa_ext_cfg->offset;
149     RISCVCPU *cpu = RISCV_CPU(obj);
150     CPURISCVState *env = &cpu->env;
151     bool value, host_bit;
152 
153     if (!visit_type_bool(v, name, &value, errp)) {
154         return;
155     }
156 
157     host_bit = env->misa_ext_mask & misa_bit;
158 
159     if (value == host_bit) {
160         return;
161     }
162 
163     if (!value) {
164         misa_ext_cfg->user_set = true;
165         return;
166     }
167 
168     /*
169      * Forbid users to enable extensions that aren't
170      * available in the hart.
171      */
172     error_setg(errp, "Enabling MISA bit '%s' is not allowed: it's not "
173                "enabled in the host", misa_ext_cfg->name);
174 }
175 
176 static void kvm_riscv_update_cpu_misa_ext(RISCVCPU *cpu, CPUState *cs)
177 {
178     CPURISCVState *env = &cpu->env;
179     uint64_t id, reg;
180     int i, ret;
181 
182     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
183         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
184         target_ulong misa_bit = misa_cfg->offset;
185 
186         if (!misa_cfg->user_set) {
187             continue;
188         }
189 
190         /* If we're here we're going to disable the MISA bit */
191         reg = 0;
192         id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
193                               misa_cfg->kvm_reg_id);
194         ret = kvm_set_one_reg(cs, id, &reg);
195         if (ret != 0) {
196             /*
197              * We're not checking for -EINVAL because if the bit is about
198              * to be disabled, it means that it was already enabled by
199              * KVM. We determined that by fetching the 'isa' register
200              * during init() time. Any error at this point is worth
201              * aborting.
202              */
203             error_report("Unable to set KVM reg %s, error %d",
204                          misa_cfg->name, ret);
205             exit(EXIT_FAILURE);
206         }
207         env->misa_ext &= ~misa_bit;
208     }
209 }
210 
211 #define KVM_EXT_CFG(_name, _prop, _reg_id) \
212     {.name = _name, .offset = CPU_CFG_OFFSET(_prop), \
213      .kvm_reg_id = _reg_id}
214 
215 static KVMCPUConfig kvm_multi_ext_cfgs[] = {
216     KVM_EXT_CFG("zicbom", ext_icbom, KVM_RISCV_ISA_EXT_ZICBOM),
217     KVM_EXT_CFG("zicboz", ext_icboz, KVM_RISCV_ISA_EXT_ZICBOZ),
218     KVM_EXT_CFG("zihintpause", ext_zihintpause, KVM_RISCV_ISA_EXT_ZIHINTPAUSE),
219     KVM_EXT_CFG("zbb", ext_zbb, KVM_RISCV_ISA_EXT_ZBB),
220     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
221     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
222     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
223     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
224 };
225 
226 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
227 {
228     return (void *)&cpu->cfg + kvmcfg->offset;
229 }
230 
231 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
232                             uint32_t val)
233 {
234     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
235 
236     *ext_enabled = val;
237 }
238 
239 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
240                                 KVMCPUConfig *multi_ext)
241 {
242     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
243 
244     return *ext_enabled;
245 }
246 
247 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
248                                       const char *name,
249                                       void *opaque, Error **errp)
250 {
251     KVMCPUConfig *multi_ext_cfg = opaque;
252     RISCVCPU *cpu = RISCV_CPU(obj);
253     bool value, host_val;
254 
255     if (!visit_type_bool(v, name, &value, errp)) {
256         return;
257     }
258 
259     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
260 
261     /*
262      * Ignore if the user is setting the same value
263      * as the host.
264      */
265     if (value == host_val) {
266         return;
267     }
268 
269     if (!multi_ext_cfg->supported) {
270         /*
271          * Error out if the user is trying to enable an
272          * extension that KVM doesn't support. Ignore
273          * option otherwise.
274          */
275         if (value) {
276             error_setg(errp, "KVM does not support disabling extension %s",
277                        multi_ext_cfg->name);
278         }
279 
280         return;
281     }
282 
283     multi_ext_cfg->user_set = true;
284     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
285 }
286 
287 static KVMCPUConfig kvm_cbom_blocksize = {
288     .name = "cbom_blocksize",
289     .offset = CPU_CFG_OFFSET(cbom_blocksize),
290     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
291 };
292 
293 static KVMCPUConfig kvm_cboz_blocksize = {
294     .name = "cboz_blocksize",
295     .offset = CPU_CFG_OFFSET(cboz_blocksize),
296     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
297 };
298 
299 static void kvm_cpu_set_cbomz_blksize(Object *obj, Visitor *v,
300                                       const char *name,
301                                       void *opaque, Error **errp)
302 {
303     KVMCPUConfig *cbomz_cfg = opaque;
304     RISCVCPU *cpu = RISCV_CPU(obj);
305     uint16_t value, *host_val;
306 
307     if (!visit_type_uint16(v, name, &value, errp)) {
308         return;
309     }
310 
311     host_val = kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
312 
313     if (value != *host_val) {
314         error_report("Unable to set %s to a different value than "
315                      "the host (%u)",
316                      cbomz_cfg->name, *host_val);
317         exit(EXIT_FAILURE);
318     }
319 
320     cbomz_cfg->user_set = true;
321 }
322 
323 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
324 {
325     CPURISCVState *env = &cpu->env;
326     uint64_t id, reg;
327     int i, ret;
328 
329     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
330         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
331 
332         if (!multi_ext_cfg->user_set) {
333             continue;
334         }
335 
336         id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
337                               multi_ext_cfg->kvm_reg_id);
338         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
339         ret = kvm_set_one_reg(cs, id, &reg);
340         if (ret != 0) {
341             error_report("Unable to %s extension %s in KVM, error %d",
342                          reg ? "enable" : "disable",
343                          multi_ext_cfg->name, ret);
344             exit(EXIT_FAILURE);
345         }
346     }
347 }
348 
349 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
350                                     const char *name,
351                                     void *opaque, Error **errp)
352 {
353     const char *propname = opaque;
354     bool value;
355 
356     if (!visit_type_bool(v, name, &value, errp)) {
357         return;
358     }
359 
360     if (value) {
361         error_setg(errp, "extension %s is not available with KVM",
362                    propname);
363     }
364 }
365 
366 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
367 {
368     /* Check if KVM created the property already */
369     if (object_property_find(obj, prop_name)) {
370         return;
371     }
372 
373     /*
374      * Set the default to disabled for every extension
375      * unknown to KVM and error out if the user attempts
376      * to enable any of them.
377      */
378     object_property_add(obj, prop_name, "bool",
379                         NULL, cpu_set_cfg_unavailable,
380                         NULL, (void *)prop_name);
381 }
382 
383 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
384                                         const RISCVCPUMultiExtConfig *array)
385 {
386     const RISCVCPUMultiExtConfig *prop;
387 
388     g_assert(array);
389 
390     for (prop = array; prop && prop->name; prop++) {
391         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
392     }
393 }
394 
395 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
396 {
397     int i;
398 
399     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
400         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
401         int bit = misa_cfg->offset;
402 
403         misa_cfg->name = riscv_get_misa_ext_name(bit);
404         misa_cfg->description = riscv_get_misa_ext_description(bit);
405 
406         object_property_add(cpu_obj, misa_cfg->name, "bool",
407                             NULL,
408                             kvm_cpu_set_misa_ext_cfg,
409                             NULL, misa_cfg);
410         object_property_set_description(cpu_obj, misa_cfg->name,
411                                         misa_cfg->description);
412     }
413 
414     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
415         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
416 
417         object_property_add(cpu_obj, multi_cfg->name, "bool",
418                             NULL,
419                             kvm_cpu_set_multi_ext_cfg,
420                             NULL, multi_cfg);
421     }
422 
423     object_property_add(cpu_obj, "cbom_blocksize", "uint16",
424                         NULL, kvm_cpu_set_cbomz_blksize,
425                         NULL, &kvm_cbom_blocksize);
426 
427     object_property_add(cpu_obj, "cboz_blocksize", "uint16",
428                         NULL, kvm_cpu_set_cbomz_blksize,
429                         NULL, &kvm_cboz_blocksize);
430 }
431 
432 static int kvm_riscv_get_regs_core(CPUState *cs)
433 {
434     int ret = 0;
435     int i;
436     target_ulong reg;
437     CPURISCVState *env = &RISCV_CPU(cs)->env;
438 
439     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
440     if (ret) {
441         return ret;
442     }
443     env->pc = reg;
444 
445     for (i = 1; i < 32; i++) {
446         uint64_t id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, i);
447         ret = kvm_get_one_reg(cs, id, &reg);
448         if (ret) {
449             return ret;
450         }
451         env->gpr[i] = reg;
452     }
453 
454     return ret;
455 }
456 
457 static int kvm_riscv_put_regs_core(CPUState *cs)
458 {
459     int ret = 0;
460     int i;
461     target_ulong reg;
462     CPURISCVState *env = &RISCV_CPU(cs)->env;
463 
464     reg = env->pc;
465     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
466     if (ret) {
467         return ret;
468     }
469 
470     for (i = 1; i < 32; i++) {
471         uint64_t id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, i);
472         reg = env->gpr[i];
473         ret = kvm_set_one_reg(cs, id, &reg);
474         if (ret) {
475             return ret;
476         }
477     }
478 
479     return ret;
480 }
481 
482 static int kvm_riscv_get_regs_csr(CPUState *cs)
483 {
484     int ret = 0;
485     CPURISCVState *env = &RISCV_CPU(cs)->env;
486 
487     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
488     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
489     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
490     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
491     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
492     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
493     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
494     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
495     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
496     return ret;
497 }
498 
499 static int kvm_riscv_put_regs_csr(CPUState *cs)
500 {
501     int ret = 0;
502     CPURISCVState *env = &RISCV_CPU(cs)->env;
503 
504     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
505     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
506     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
507     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
508     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
509     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
510     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
511     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
512     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
513 
514     return ret;
515 }
516 
517 static int kvm_riscv_get_regs_fp(CPUState *cs)
518 {
519     int ret = 0;
520     int i;
521     CPURISCVState *env = &RISCV_CPU(cs)->env;
522 
523     if (riscv_has_ext(env, RVD)) {
524         uint64_t reg;
525         for (i = 0; i < 32; i++) {
526             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(env, i), &reg);
527             if (ret) {
528                 return ret;
529             }
530             env->fpr[i] = reg;
531         }
532         return ret;
533     }
534 
535     if (riscv_has_ext(env, RVF)) {
536         uint32_t reg;
537         for (i = 0; i < 32; i++) {
538             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(env, i), &reg);
539             if (ret) {
540                 return ret;
541             }
542             env->fpr[i] = reg;
543         }
544         return ret;
545     }
546 
547     return ret;
548 }
549 
550 static int kvm_riscv_put_regs_fp(CPUState *cs)
551 {
552     int ret = 0;
553     int i;
554     CPURISCVState *env = &RISCV_CPU(cs)->env;
555 
556     if (riscv_has_ext(env, RVD)) {
557         uint64_t reg;
558         for (i = 0; i < 32; i++) {
559             reg = env->fpr[i];
560             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(env, i), &reg);
561             if (ret) {
562                 return ret;
563             }
564         }
565         return ret;
566     }
567 
568     if (riscv_has_ext(env, RVF)) {
569         uint32_t reg;
570         for (i = 0; i < 32; i++) {
571             reg = env->fpr[i];
572             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(env, i), &reg);
573             if (ret) {
574                 return ret;
575             }
576         }
577         return ret;
578     }
579 
580     return ret;
581 }
582 
583 static void kvm_riscv_get_regs_timer(CPUState *cs)
584 {
585     CPURISCVState *env = &RISCV_CPU(cs)->env;
586 
587     if (env->kvm_timer_dirty) {
588         return;
589     }
590 
591     KVM_RISCV_GET_TIMER(cs, env, time, env->kvm_timer_time);
592     KVM_RISCV_GET_TIMER(cs, env, compare, env->kvm_timer_compare);
593     KVM_RISCV_GET_TIMER(cs, env, state, env->kvm_timer_state);
594     KVM_RISCV_GET_TIMER(cs, env, frequency, env->kvm_timer_frequency);
595 
596     env->kvm_timer_dirty = true;
597 }
598 
599 static void kvm_riscv_put_regs_timer(CPUState *cs)
600 {
601     uint64_t reg;
602     CPURISCVState *env = &RISCV_CPU(cs)->env;
603 
604     if (!env->kvm_timer_dirty) {
605         return;
606     }
607 
608     KVM_RISCV_SET_TIMER(cs, env, time, env->kvm_timer_time);
609     KVM_RISCV_SET_TIMER(cs, env, compare, env->kvm_timer_compare);
610 
611     /*
612      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
613      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
614      * doesn't matter that adaping in QEMU now.
615      * TODO If KVM changes, adapt here.
616      */
617     if (env->kvm_timer_state) {
618         KVM_RISCV_SET_TIMER(cs, env, state, env->kvm_timer_state);
619     }
620 
621     /*
622      * For now, migration will not work between Hosts with different timer
623      * frequency. Therefore, we should check whether they are the same here
624      * during the migration.
625      */
626     if (migration_is_running(migrate_get_current()->state)) {
627         KVM_RISCV_GET_TIMER(cs, env, frequency, reg);
628         if (reg != env->kvm_timer_frequency) {
629             error_report("Dst Hosts timer frequency != Src Hosts");
630         }
631     }
632 
633     env->kvm_timer_dirty = false;
634 }
635 
636 typedef struct KVMScratchCPU {
637     int kvmfd;
638     int vmfd;
639     int cpufd;
640 } KVMScratchCPU;
641 
642 /*
643  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
644  * from target/arm/kvm.c.
645  */
646 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
647 {
648     int kvmfd = -1, vmfd = -1, cpufd = -1;
649 
650     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
651     if (kvmfd < 0) {
652         goto err;
653     }
654     do {
655         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
656     } while (vmfd == -1 && errno == EINTR);
657     if (vmfd < 0) {
658         goto err;
659     }
660     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
661     if (cpufd < 0) {
662         goto err;
663     }
664 
665     scratch->kvmfd =  kvmfd;
666     scratch->vmfd = vmfd;
667     scratch->cpufd = cpufd;
668 
669     return true;
670 
671  err:
672     if (cpufd >= 0) {
673         close(cpufd);
674     }
675     if (vmfd >= 0) {
676         close(vmfd);
677     }
678     if (kvmfd >= 0) {
679         close(kvmfd);
680     }
681 
682     return false;
683 }
684 
685 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
686 {
687     close(scratch->cpufd);
688     close(scratch->vmfd);
689     close(scratch->kvmfd);
690 }
691 
692 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
693 {
694     CPURISCVState *env = &cpu->env;
695     struct kvm_one_reg reg;
696     int ret;
697 
698     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
699                               KVM_REG_RISCV_CONFIG_REG(mvendorid));
700     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
701     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
702     if (ret != 0) {
703         error_report("Unable to retrieve mvendorid from host, error %d", ret);
704     }
705 
706     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
707                               KVM_REG_RISCV_CONFIG_REG(marchid));
708     reg.addr = (uint64_t)&cpu->cfg.marchid;
709     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
710     if (ret != 0) {
711         error_report("Unable to retrieve marchid from host, error %d", ret);
712     }
713 
714     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
715                               KVM_REG_RISCV_CONFIG_REG(mimpid));
716     reg.addr = (uint64_t)&cpu->cfg.mimpid;
717     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
718     if (ret != 0) {
719         error_report("Unable to retrieve mimpid from host, error %d", ret);
720     }
721 }
722 
723 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
724                                          KVMScratchCPU *kvmcpu)
725 {
726     CPURISCVState *env = &cpu->env;
727     struct kvm_one_reg reg;
728     int ret;
729 
730     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
731                               KVM_REG_RISCV_CONFIG_REG(isa));
732     reg.addr = (uint64_t)&env->misa_ext_mask;
733     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
734 
735     if (ret) {
736         error_report("Unable to fetch ISA register from KVM, "
737                      "error %d", ret);
738         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
739         exit(EXIT_FAILURE);
740     }
741 
742     env->misa_ext = env->misa_ext_mask;
743 }
744 
745 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
746                                          KVMCPUConfig *cbomz_cfg)
747 {
748     CPURISCVState *env = &cpu->env;
749     struct kvm_one_reg reg;
750     int ret;
751 
752     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
753                               cbomz_cfg->kvm_reg_id);
754     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
755     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
756     if (ret != 0) {
757         error_report("Unable to read KVM reg %s, error %d",
758                      cbomz_cfg->name, ret);
759         exit(EXIT_FAILURE);
760     }
761 }
762 
763 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
764 {
765     CPURISCVState *env = &cpu->env;
766     uint64_t val;
767     int i, ret;
768 
769     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
770         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
771         struct kvm_one_reg reg;
772 
773         reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
774                                   multi_ext_cfg->kvm_reg_id);
775         reg.addr = (uint64_t)&val;
776         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
777         if (ret != 0) {
778             if (errno == EINVAL) {
779                 /* Silently default to 'false' if KVM does not support it. */
780                 multi_ext_cfg->supported = false;
781                 val = false;
782             } else {
783                 error_report("Unable to read ISA_EXT KVM register %s, "
784                              "error %d", multi_ext_cfg->name, ret);
785                 kvm_riscv_destroy_scratch_vcpu(kvmcpu);
786                 exit(EXIT_FAILURE);
787             }
788         } else {
789             multi_ext_cfg->supported = true;
790         }
791 
792         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
793     }
794 
795     if (cpu->cfg.ext_icbom) {
796         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
797     }
798 
799     if (cpu->cfg.ext_icboz) {
800         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
801     }
802 }
803 
804 static void riscv_init_user_properties(Object *cpu_obj)
805 {
806     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
807     KVMScratchCPU kvmcpu;
808 
809     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
810         return;
811     }
812 
813     kvm_riscv_add_cpu_user_properties(cpu_obj);
814     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
815     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
816     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
817 
818     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
819 }
820 
821 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
822     KVM_CAP_LAST_INFO
823 };
824 
825 int kvm_arch_get_registers(CPUState *cs)
826 {
827     int ret = 0;
828 
829     ret = kvm_riscv_get_regs_core(cs);
830     if (ret) {
831         return ret;
832     }
833 
834     ret = kvm_riscv_get_regs_csr(cs);
835     if (ret) {
836         return ret;
837     }
838 
839     ret = kvm_riscv_get_regs_fp(cs);
840     if (ret) {
841         return ret;
842     }
843 
844     return ret;
845 }
846 
847 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
848 {
849     if (cap_has_mp_state) {
850         struct kvm_mp_state mp_state = {
851             .mp_state = state
852         };
853 
854         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
855         if (ret) {
856             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
857                     __func__, ret, strerror(-ret));
858             return -1;
859         }
860     }
861 
862     return 0;
863 }
864 
865 int kvm_arch_put_registers(CPUState *cs, int level)
866 {
867     int ret = 0;
868 
869     ret = kvm_riscv_put_regs_core(cs);
870     if (ret) {
871         return ret;
872     }
873 
874     ret = kvm_riscv_put_regs_csr(cs);
875     if (ret) {
876         return ret;
877     }
878 
879     ret = kvm_riscv_put_regs_fp(cs);
880     if (ret) {
881         return ret;
882     }
883 
884     if (KVM_PUT_RESET_STATE == level) {
885         RISCVCPU *cpu = RISCV_CPU(cs);
886         if (cs->cpu_index == 0) {
887             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
888         } else {
889             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
890         }
891         if (ret) {
892             return ret;
893         }
894     }
895 
896     return ret;
897 }
898 
899 int kvm_arch_release_virq_post(int virq)
900 {
901     return 0;
902 }
903 
904 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
905                              uint64_t address, uint32_t data, PCIDevice *dev)
906 {
907     return 0;
908 }
909 
910 int kvm_arch_destroy_vcpu(CPUState *cs)
911 {
912     return 0;
913 }
914 
915 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
916 {
917     return cpu->cpu_index;
918 }
919 
920 static void kvm_riscv_vm_state_change(void *opaque, bool running,
921                                       RunState state)
922 {
923     CPUState *cs = opaque;
924 
925     if (running) {
926         kvm_riscv_put_regs_timer(cs);
927     } else {
928         kvm_riscv_get_regs_timer(cs);
929     }
930 }
931 
932 void kvm_arch_init_irq_routing(KVMState *s)
933 {
934 }
935 
936 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
937 {
938     CPURISCVState *env = &cpu->env;
939     target_ulong reg;
940     uint64_t id;
941     int ret;
942 
943     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
944                           KVM_REG_RISCV_CONFIG_REG(mvendorid));
945     /*
946      * cfg.mvendorid is an uint32 but a target_ulong will
947      * be written. Assign it to a target_ulong var to avoid
948      * writing pieces of other cpu->cfg fields in the reg.
949      */
950     reg = cpu->cfg.mvendorid;
951     ret = kvm_set_one_reg(cs, id, &reg);
952     if (ret != 0) {
953         return ret;
954     }
955 
956     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
957                           KVM_REG_RISCV_CONFIG_REG(marchid));
958     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
959     if (ret != 0) {
960         return ret;
961     }
962 
963     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
964                           KVM_REG_RISCV_CONFIG_REG(mimpid));
965     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
966 
967     return ret;
968 }
969 
970 int kvm_arch_init_vcpu(CPUState *cs)
971 {
972     int ret = 0;
973     RISCVCPU *cpu = RISCV_CPU(cs);
974 
975     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
976 
977     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
978         ret = kvm_vcpu_set_machine_ids(cpu, cs);
979         if (ret != 0) {
980             return ret;
981         }
982     }
983 
984     kvm_riscv_update_cpu_misa_ext(cpu, cs);
985     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
986 
987     return ret;
988 }
989 
990 int kvm_arch_msi_data_to_gsi(uint32_t data)
991 {
992     abort();
993 }
994 
995 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
996                                 int vector, PCIDevice *dev)
997 {
998     return 0;
999 }
1000 
1001 int kvm_arch_get_default_type(MachineState *ms)
1002 {
1003     return 0;
1004 }
1005 
1006 int kvm_arch_init(MachineState *ms, KVMState *s)
1007 {
1008     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1009     return 0;
1010 }
1011 
1012 int kvm_arch_irqchip_create(KVMState *s)
1013 {
1014     if (kvm_kernel_irqchip_split()) {
1015         error_report("-machine kernel_irqchip=split is not supported on RISC-V.");
1016         exit(1);
1017     }
1018 
1019     /*
1020      * We can create the VAIA using the newer device control API.
1021      */
1022     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1023 }
1024 
1025 int kvm_arch_process_async_events(CPUState *cs)
1026 {
1027     return 0;
1028 }
1029 
1030 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1031 {
1032 }
1033 
1034 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1035 {
1036     return MEMTXATTRS_UNSPECIFIED;
1037 }
1038 
1039 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1040 {
1041     return true;
1042 }
1043 
1044 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1045 {
1046     int ret = 0;
1047     unsigned char ch;
1048     switch (run->riscv_sbi.extension_id) {
1049     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1050         ch = run->riscv_sbi.args[0];
1051         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1052         break;
1053     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1054         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1055         if (ret == sizeof(ch)) {
1056             run->riscv_sbi.ret[0] = ch;
1057         } else {
1058             run->riscv_sbi.ret[0] = -1;
1059         }
1060         ret = 0;
1061         break;
1062     default:
1063         qemu_log_mask(LOG_UNIMP,
1064                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1065                       __func__, run->riscv_sbi.extension_id);
1066         ret = -1;
1067         break;
1068     }
1069     return ret;
1070 }
1071 
1072 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1073 {
1074     int ret = 0;
1075     switch (run->exit_reason) {
1076     case KVM_EXIT_RISCV_SBI:
1077         ret = kvm_riscv_handle_sbi(cs, run);
1078         break;
1079     default:
1080         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1081                       __func__, run->exit_reason);
1082         ret = -1;
1083         break;
1084     }
1085     return ret;
1086 }
1087 
1088 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1089 {
1090     CPURISCVState *env = &cpu->env;
1091     int i;
1092 
1093     if (!kvm_enabled()) {
1094         return;
1095     }
1096     for (i = 0; i < 32; i++) {
1097         env->gpr[i] = 0;
1098     }
1099     env->pc = cpu->env.kernel_addr;
1100     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1101     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1102     env->satp = 0;
1103     env->mie = 0;
1104     env->stvec = 0;
1105     env->sscratch = 0;
1106     env->sepc = 0;
1107     env->scause = 0;
1108     env->stval = 0;
1109     env->mip = 0;
1110 }
1111 
1112 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1113 {
1114     int ret;
1115     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1116 
1117     if (irq != IRQ_S_EXT) {
1118         perror("kvm riscv set irq != IRQ_S_EXT\n");
1119         abort();
1120     }
1121 
1122     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1123     if (ret < 0) {
1124         perror("Set irq failed");
1125         abort();
1126     }
1127 }
1128 
1129 bool kvm_arch_cpu_check_are_resettable(void)
1130 {
1131     return true;
1132 }
1133 
1134 static int aia_mode;
1135 
1136 static const char *kvm_aia_mode_str(uint64_t mode)
1137 {
1138     switch (mode) {
1139     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1140         return "emul";
1141     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1142         return "hwaccel";
1143     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1144     default:
1145         return "auto";
1146     };
1147 }
1148 
1149 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1150 {
1151     return g_strdup(kvm_aia_mode_str(aia_mode));
1152 }
1153 
1154 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1155 {
1156     if (!strcmp(val, "emul")) {
1157         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1158     } else if (!strcmp(val, "hwaccel")) {
1159         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1160     } else if (!strcmp(val, "auto")) {
1161         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1162     } else {
1163         error_setg(errp, "Invalid KVM AIA mode");
1164         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1165     }
1166 }
1167 
1168 void kvm_arch_accel_class_init(ObjectClass *oc)
1169 {
1170     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1171                                   riscv_set_kvm_aia);
1172     object_class_property_set_description(oc, "riscv-aia",
1173                                           "Set KVM AIA mode. Valid values are "
1174                                           "emul, hwaccel, and auto. Default "
1175                                           "is auto.");
1176     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1177                                     "auto");
1178 }
1179 
1180 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1181                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1182                           uint64_t aplic_base, uint64_t imsic_base,
1183                           uint64_t guest_num)
1184 {
1185     int ret, i;
1186     int aia_fd = -1;
1187     uint64_t default_aia_mode;
1188     uint64_t socket_count = riscv_socket_count(machine);
1189     uint64_t max_hart_per_socket = 0;
1190     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1191     uint64_t socket_bits, hart_bits, guest_bits;
1192 
1193     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1194 
1195     if (aia_fd < 0) {
1196         error_report("Unable to create in-kernel irqchip");
1197         exit(1);
1198     }
1199 
1200     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1201                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1202                             &default_aia_mode, false, NULL);
1203     if (ret < 0) {
1204         error_report("KVM AIA: failed to get current KVM AIA mode");
1205         exit(1);
1206     }
1207     qemu_log("KVM AIA: default mode is %s\n",
1208              kvm_aia_mode_str(default_aia_mode));
1209 
1210     if (default_aia_mode != aia_mode) {
1211         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1212                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1213                                 &aia_mode, true, NULL);
1214         if (ret < 0)
1215             warn_report("KVM AIA: failed to set KVM AIA mode");
1216         else
1217             qemu_log("KVM AIA: set current mode to %s\n",
1218                      kvm_aia_mode_str(aia_mode));
1219     }
1220 
1221     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1222                             KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1223                             &aia_irq_num, true, NULL);
1224     if (ret < 0) {
1225         error_report("KVM AIA: failed to set number of input irq lines");
1226         exit(1);
1227     }
1228 
1229     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1230                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1231                             &aia_msi_num, true, NULL);
1232     if (ret < 0) {
1233         error_report("KVM AIA: failed to set number of msi");
1234         exit(1);
1235     }
1236 
1237     socket_bits = find_last_bit(&socket_count, BITS_PER_LONG) + 1;
1238     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1239                             KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1240                             &socket_bits, true, NULL);
1241     if (ret < 0) {
1242         error_report("KVM AIA: failed to set group_bits");
1243         exit(1);
1244     }
1245 
1246     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1247                             KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1248                             &group_shift, true, NULL);
1249     if (ret < 0) {
1250         error_report("KVM AIA: failed to set group_shift");
1251         exit(1);
1252     }
1253 
1254     guest_bits = guest_num == 0 ? 0 :
1255                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1256     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1257                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1258                             &guest_bits, true, NULL);
1259     if (ret < 0) {
1260         error_report("KVM AIA: failed to set guest_bits");
1261         exit(1);
1262     }
1263 
1264     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1265                             KVM_DEV_RISCV_AIA_ADDR_APLIC,
1266                             &aplic_base, true, NULL);
1267     if (ret < 0) {
1268         error_report("KVM AIA: failed to set the base address of APLIC");
1269         exit(1);
1270     }
1271 
1272     for (socket = 0; socket < socket_count; socket++) {
1273         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1274         hart_count = riscv_socket_hart_count(machine, socket);
1275         base_hart = riscv_socket_first_hartid(machine, socket);
1276 
1277         if (max_hart_per_socket < hart_count) {
1278             max_hart_per_socket = hart_count;
1279         }
1280 
1281         for (i = 0; i < hart_count; i++) {
1282             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1283             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1284                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1285                                     &imsic_addr, true, NULL);
1286             if (ret < 0) {
1287                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1288                 exit(1);
1289             }
1290         }
1291     }
1292 
1293     hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1294     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1295                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1296                             &hart_bits, true, NULL);
1297     if (ret < 0) {
1298         error_report("KVM AIA: failed to set hart_bits");
1299         exit(1);
1300     }
1301 
1302     if (kvm_has_gsi_routing()) {
1303         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1304             /* KVM AIA only has one APLIC instance */
1305             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1306         }
1307         kvm_gsi_routing_allowed = true;
1308         kvm_irqchip_commit_routes(kvm_state);
1309     }
1310 
1311     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1312                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1313                             NULL, true, NULL);
1314     if (ret < 0) {
1315         error_report("KVM AIA: initialized fail");
1316         exit(1);
1317     }
1318 
1319     kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
1320 }
1321 
1322 static void kvm_cpu_instance_init(CPUState *cs)
1323 {
1324     Object *obj = OBJECT(RISCV_CPU(cs));
1325     DeviceState *dev = DEVICE(obj);
1326 
1327     riscv_init_user_properties(obj);
1328     riscv_add_satp_mode_properties(obj);
1329     riscv_cpu_add_misa_properties(obj);
1330 
1331     riscv_cpu_add_kvm_unavail_prop_array(obj, riscv_cpu_extensions);
1332     riscv_cpu_add_kvm_unavail_prop_array(obj, riscv_cpu_vendor_exts);
1333     riscv_cpu_add_kvm_unavail_prop_array(obj, riscv_cpu_experimental_exts);
1334 
1335     for (Property *prop = riscv_cpu_options; prop && prop->name; prop++) {
1336         /* Check if we have a specific KVM handler for the option */
1337         if (object_property_find(obj, prop->name)) {
1338             continue;
1339         }
1340         qdev_property_add_static(dev, prop);
1341     }
1342 }
1343 
1344 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1345 {
1346     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1347 
1348     acc->cpu_instance_init = kvm_cpu_instance_init;
1349 }
1350 
1351 static const TypeInfo kvm_cpu_accel_type_info = {
1352     .name = ACCEL_CPU_NAME("kvm"),
1353 
1354     .parent = TYPE_ACCEL_CPU,
1355     .class_init = kvm_cpu_accel_class_init,
1356     .abstract = true,
1357 };
1358 static void kvm_cpu_accel_register_types(void)
1359 {
1360     type_register_static(&kvm_cpu_accel_type_info);
1361 }
1362 type_init(kvm_cpu_accel_register_types);
1363 
1364 static void riscv_host_cpu_init(Object *obj)
1365 {
1366     CPURISCVState *env = &RISCV_CPU(obj)->env;
1367 
1368 #if defined(TARGET_RISCV32)
1369     env->misa_mxl_max = env->misa_mxl = MXL_RV32;
1370 #elif defined(TARGET_RISCV64)
1371     env->misa_mxl_max = env->misa_mxl = MXL_RV64;
1372 #endif
1373 }
1374 
1375 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
1376     {
1377         .name = TYPE_RISCV_CPU_HOST,
1378         .parent = TYPE_RISCV_CPU,
1379         .instance_init = riscv_host_cpu_init,
1380     }
1381 };
1382 
1383 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
1384