xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision 450bd6618fda3d2e2ab02b2fce1c79efd5b66084)
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 static uint64_t kvm_riscv_reg_id_u32(uint64_t type, uint64_t idx)
76 {
77     return KVM_REG_RISCV | KVM_REG_SIZE_U32 | type | idx;
78 }
79 
80 static uint64_t kvm_riscv_reg_id_u64(uint64_t type, uint64_t idx)
81 {
82     return KVM_REG_RISCV | KVM_REG_SIZE_U64 | type | idx;
83 }
84 
85 #define RISCV_CORE_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, \
86                  KVM_REG_RISCV_CORE_REG(name))
87 
88 #define RISCV_CSR_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_CSR, \
89                  KVM_REG_RISCV_CSR_REG(name))
90 
91 #define RISCV_TIMER_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_TIMER, \
92                  KVM_REG_RISCV_TIMER_REG(name))
93 
94 #define RISCV_FP_F_REG(idx)  kvm_riscv_reg_id_u32(KVM_REG_RISCV_FP_F, idx)
95 
96 #define RISCV_FP_D_REG(idx)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_FP_D, idx)
97 
98 #define KVM_RISCV_GET_CSR(cs, env, csr, reg) \
99     do { \
100         int _ret = kvm_get_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
101         if (_ret) { \
102             return _ret; \
103         } \
104     } while (0)
105 
106 #define KVM_RISCV_SET_CSR(cs, env, csr, reg) \
107     do { \
108         int _ret = kvm_set_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
109         if (_ret) { \
110             return _ret; \
111         } \
112     } while (0)
113 
114 #define KVM_RISCV_GET_TIMER(cs, env, name, reg) \
115     do { \
116         int ret = kvm_get_one_reg(cs, RISCV_TIMER_REG(env, name), &reg); \
117         if (ret) { \
118             abort(); \
119         } \
120     } while (0)
121 
122 #define KVM_RISCV_SET_TIMER(cs, env, name, reg) \
123     do { \
124         int ret = kvm_set_one_reg(cs, RISCV_TIMER_REG(env, name), &reg); \
125         if (ret) { \
126             abort(); \
127         } \
128     } while (0)
129 
130 typedef struct KVMCPUConfig {
131     const char *name;
132     const char *description;
133     target_ulong offset;
134     int kvm_reg_id;
135     bool user_set;
136     bool supported;
137 } KVMCPUConfig;
138 
139 #define KVM_MISA_CFG(_bit, _reg_id) \
140     {.offset = _bit, .kvm_reg_id = _reg_id}
141 
142 /* KVM ISA extensions */
143 static KVMCPUConfig kvm_misa_ext_cfgs[] = {
144     KVM_MISA_CFG(RVA, KVM_RISCV_ISA_EXT_A),
145     KVM_MISA_CFG(RVC, KVM_RISCV_ISA_EXT_C),
146     KVM_MISA_CFG(RVD, KVM_RISCV_ISA_EXT_D),
147     KVM_MISA_CFG(RVF, KVM_RISCV_ISA_EXT_F),
148     KVM_MISA_CFG(RVH, KVM_RISCV_ISA_EXT_H),
149     KVM_MISA_CFG(RVI, KVM_RISCV_ISA_EXT_I),
150     KVM_MISA_CFG(RVM, KVM_RISCV_ISA_EXT_M),
151 };
152 
153 static void kvm_cpu_get_misa_ext_cfg(Object *obj, Visitor *v,
154                                      const char *name,
155                                      void *opaque, Error **errp)
156 {
157     KVMCPUConfig *misa_ext_cfg = opaque;
158     target_ulong misa_bit = misa_ext_cfg->offset;
159     RISCVCPU *cpu = RISCV_CPU(obj);
160     CPURISCVState *env = &cpu->env;
161     bool value = env->misa_ext_mask & misa_bit;
162 
163     visit_type_bool(v, name, &value, errp);
164 }
165 
166 static void kvm_cpu_set_misa_ext_cfg(Object *obj, Visitor *v,
167                                      const char *name,
168                                      void *opaque, Error **errp)
169 {
170     KVMCPUConfig *misa_ext_cfg = opaque;
171     target_ulong misa_bit = misa_ext_cfg->offset;
172     RISCVCPU *cpu = RISCV_CPU(obj);
173     CPURISCVState *env = &cpu->env;
174     bool value, host_bit;
175 
176     if (!visit_type_bool(v, name, &value, errp)) {
177         return;
178     }
179 
180     host_bit = env->misa_ext_mask & misa_bit;
181 
182     if (value == host_bit) {
183         return;
184     }
185 
186     if (!value) {
187         misa_ext_cfg->user_set = true;
188         return;
189     }
190 
191     /*
192      * Forbid users to enable extensions that aren't
193      * available in the hart.
194      */
195     error_setg(errp, "Enabling MISA bit '%s' is not allowed: it's not "
196                "enabled in the host", misa_ext_cfg->name);
197 }
198 
199 static void kvm_riscv_update_cpu_misa_ext(RISCVCPU *cpu, CPUState *cs)
200 {
201     CPURISCVState *env = &cpu->env;
202     uint64_t id, reg;
203     int i, ret;
204 
205     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
206         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
207         target_ulong misa_bit = misa_cfg->offset;
208 
209         if (!misa_cfg->user_set) {
210             continue;
211         }
212 
213         /* If we're here we're going to disable the MISA bit */
214         reg = 0;
215         id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
216                               misa_cfg->kvm_reg_id);
217         ret = kvm_set_one_reg(cs, id, &reg);
218         if (ret != 0) {
219             /*
220              * We're not checking for -EINVAL because if the bit is about
221              * to be disabled, it means that it was already enabled by
222              * KVM. We determined that by fetching the 'isa' register
223              * during init() time. Any error at this point is worth
224              * aborting.
225              */
226             error_report("Unable to set KVM reg %s, error %d",
227                          misa_cfg->name, ret);
228             exit(EXIT_FAILURE);
229         }
230         env->misa_ext &= ~misa_bit;
231     }
232 }
233 
234 #define KVM_EXT_CFG(_name, _prop, _reg_id) \
235     {.name = _name, .offset = CPU_CFG_OFFSET(_prop), \
236      .kvm_reg_id = _reg_id}
237 
238 static KVMCPUConfig kvm_multi_ext_cfgs[] = {
239     KVM_EXT_CFG("zicbom", ext_zicbom, KVM_RISCV_ISA_EXT_ZICBOM),
240     KVM_EXT_CFG("zicboz", ext_zicboz, KVM_RISCV_ISA_EXT_ZICBOZ),
241     KVM_EXT_CFG("zicntr", ext_zicntr, KVM_RISCV_ISA_EXT_ZICNTR),
242     KVM_EXT_CFG("zicsr", ext_zicsr, KVM_RISCV_ISA_EXT_ZICSR),
243     KVM_EXT_CFG("zifencei", ext_zifencei, KVM_RISCV_ISA_EXT_ZIFENCEI),
244     KVM_EXT_CFG("zihintpause", ext_zihintpause, KVM_RISCV_ISA_EXT_ZIHINTPAUSE),
245     KVM_EXT_CFG("zihpm", ext_zihpm, KVM_RISCV_ISA_EXT_ZIHPM),
246     KVM_EXT_CFG("zba", ext_zba, KVM_RISCV_ISA_EXT_ZBA),
247     KVM_EXT_CFG("zbb", ext_zbb, KVM_RISCV_ISA_EXT_ZBB),
248     KVM_EXT_CFG("zbs", ext_zbs, KVM_RISCV_ISA_EXT_ZBS),
249     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
250     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
251     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
252     KVM_EXT_CFG("svnapot", ext_svnapot, KVM_RISCV_ISA_EXT_SVNAPOT),
253     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
254 };
255 
256 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
257 {
258     return (void *)&cpu->cfg + kvmcfg->offset;
259 }
260 
261 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
262                             uint32_t val)
263 {
264     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
265 
266     *ext_enabled = val;
267 }
268 
269 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
270                                 KVMCPUConfig *multi_ext)
271 {
272     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
273 
274     return *ext_enabled;
275 }
276 
277 static void kvm_cpu_get_multi_ext_cfg(Object *obj, Visitor *v,
278                                       const char *name,
279                                       void *opaque, Error **errp)
280 {
281     KVMCPUConfig *multi_ext_cfg = opaque;
282     RISCVCPU *cpu = RISCV_CPU(obj);
283     bool value = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
284 
285     visit_type_bool(v, name, &value, errp);
286 }
287 
288 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
289                                       const char *name,
290                                       void *opaque, Error **errp)
291 {
292     KVMCPUConfig *multi_ext_cfg = opaque;
293     RISCVCPU *cpu = RISCV_CPU(obj);
294     bool value, host_val;
295 
296     if (!visit_type_bool(v, name, &value, errp)) {
297         return;
298     }
299 
300     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
301 
302     /*
303      * Ignore if the user is setting the same value
304      * as the host.
305      */
306     if (value == host_val) {
307         return;
308     }
309 
310     if (!multi_ext_cfg->supported) {
311         /*
312          * Error out if the user is trying to enable an
313          * extension that KVM doesn't support. Ignore
314          * option otherwise.
315          */
316         if (value) {
317             error_setg(errp, "KVM does not support disabling extension %s",
318                        multi_ext_cfg->name);
319         }
320 
321         return;
322     }
323 
324     multi_ext_cfg->user_set = true;
325     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
326 }
327 
328 static KVMCPUConfig kvm_cbom_blocksize = {
329     .name = "cbom_blocksize",
330     .offset = CPU_CFG_OFFSET(cbom_blocksize),
331     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
332 };
333 
334 static KVMCPUConfig kvm_cboz_blocksize = {
335     .name = "cboz_blocksize",
336     .offset = CPU_CFG_OFFSET(cboz_blocksize),
337     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
338 };
339 
340 static void kvm_cpu_set_cbomz_blksize(Object *obj, Visitor *v,
341                                       const char *name,
342                                       void *opaque, Error **errp)
343 {
344     KVMCPUConfig *cbomz_cfg = opaque;
345     RISCVCPU *cpu = RISCV_CPU(obj);
346     uint16_t value, *host_val;
347 
348     if (!visit_type_uint16(v, name, &value, errp)) {
349         return;
350     }
351 
352     host_val = kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
353 
354     if (value != *host_val) {
355         error_report("Unable to set %s to a different value than "
356                      "the host (%u)",
357                      cbomz_cfg->name, *host_val);
358         exit(EXIT_FAILURE);
359     }
360 
361     cbomz_cfg->user_set = true;
362 }
363 
364 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
365 {
366     CPURISCVState *env = &cpu->env;
367     uint64_t id, reg;
368     int i, ret;
369 
370     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
371         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
372 
373         if (!multi_ext_cfg->user_set) {
374             continue;
375         }
376 
377         id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
378                               multi_ext_cfg->kvm_reg_id);
379         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
380         ret = kvm_set_one_reg(cs, id, &reg);
381         if (ret != 0) {
382             error_report("Unable to %s extension %s in KVM, error %d",
383                          reg ? "enable" : "disable",
384                          multi_ext_cfg->name, ret);
385             exit(EXIT_FAILURE);
386         }
387     }
388 }
389 
390 static void cpu_get_cfg_unavailable(Object *obj, Visitor *v,
391                                     const char *name,
392                                     void *opaque, Error **errp)
393 {
394     bool value = false;
395 
396     visit_type_bool(v, name, &value, errp);
397 }
398 
399 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
400                                     const char *name,
401                                     void *opaque, Error **errp)
402 {
403     const char *propname = opaque;
404     bool value;
405 
406     if (!visit_type_bool(v, name, &value, errp)) {
407         return;
408     }
409 
410     if (value) {
411         error_setg(errp, "extension %s is not available with KVM",
412                    propname);
413     }
414 }
415 
416 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
417 {
418     /* Check if KVM created the property already */
419     if (object_property_find(obj, prop_name)) {
420         return;
421     }
422 
423     /*
424      * Set the default to disabled for every extension
425      * unknown to KVM and error out if the user attempts
426      * to enable any of them.
427      */
428     object_property_add(obj, prop_name, "bool",
429                         cpu_get_cfg_unavailable,
430                         cpu_set_cfg_unavailable,
431                         NULL, (void *)prop_name);
432 }
433 
434 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
435                                         const RISCVCPUMultiExtConfig *array)
436 {
437     const RISCVCPUMultiExtConfig *prop;
438 
439     g_assert(array);
440 
441     for (prop = array; prop && prop->name; prop++) {
442         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
443     }
444 }
445 
446 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
447 {
448     int i;
449 
450     riscv_add_satp_mode_properties(cpu_obj);
451 
452     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
453         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
454         int bit = misa_cfg->offset;
455 
456         misa_cfg->name = riscv_get_misa_ext_name(bit);
457         misa_cfg->description = riscv_get_misa_ext_description(bit);
458 
459         object_property_add(cpu_obj, misa_cfg->name, "bool",
460                             kvm_cpu_get_misa_ext_cfg,
461                             kvm_cpu_set_misa_ext_cfg,
462                             NULL, misa_cfg);
463         object_property_set_description(cpu_obj, misa_cfg->name,
464                                         misa_cfg->description);
465     }
466 
467     for (i = 0; misa_bits[i] != 0; i++) {
468         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
469         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
470     }
471 
472     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
473         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
474 
475         object_property_add(cpu_obj, multi_cfg->name, "bool",
476                             kvm_cpu_get_multi_ext_cfg,
477                             kvm_cpu_set_multi_ext_cfg,
478                             NULL, multi_cfg);
479     }
480 
481     object_property_add(cpu_obj, "cbom_blocksize", "uint16",
482                         NULL, kvm_cpu_set_cbomz_blksize,
483                         NULL, &kvm_cbom_blocksize);
484 
485     object_property_add(cpu_obj, "cboz_blocksize", "uint16",
486                         NULL, kvm_cpu_set_cbomz_blksize,
487                         NULL, &kvm_cboz_blocksize);
488 
489     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
490     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
491     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
492 }
493 
494 static int kvm_riscv_get_regs_core(CPUState *cs)
495 {
496     int ret = 0;
497     int i;
498     target_ulong reg;
499     CPURISCVState *env = &RISCV_CPU(cs)->env;
500 
501     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
502     if (ret) {
503         return ret;
504     }
505     env->pc = reg;
506 
507     for (i = 1; i < 32; i++) {
508         uint64_t id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, i);
509         ret = kvm_get_one_reg(cs, id, &reg);
510         if (ret) {
511             return ret;
512         }
513         env->gpr[i] = reg;
514     }
515 
516     return ret;
517 }
518 
519 static int kvm_riscv_put_regs_core(CPUState *cs)
520 {
521     int ret = 0;
522     int i;
523     target_ulong reg;
524     CPURISCVState *env = &RISCV_CPU(cs)->env;
525 
526     reg = env->pc;
527     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
528     if (ret) {
529         return ret;
530     }
531 
532     for (i = 1; i < 32; i++) {
533         uint64_t id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, i);
534         reg = env->gpr[i];
535         ret = kvm_set_one_reg(cs, id, &reg);
536         if (ret) {
537             return ret;
538         }
539     }
540 
541     return ret;
542 }
543 
544 static int kvm_riscv_get_regs_csr(CPUState *cs)
545 {
546     CPURISCVState *env = &RISCV_CPU(cs)->env;
547 
548     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
549     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
550     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
551     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
552     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
553     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
554     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
555     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
556     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
557 
558     return 0;
559 }
560 
561 static int kvm_riscv_put_regs_csr(CPUState *cs)
562 {
563     CPURISCVState *env = &RISCV_CPU(cs)->env;
564 
565     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
566     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
567     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
568     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
569     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
570     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
571     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
572     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
573     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
574 
575     return 0;
576 }
577 
578 static int kvm_riscv_get_regs_fp(CPUState *cs)
579 {
580     int ret = 0;
581     int i;
582     CPURISCVState *env = &RISCV_CPU(cs)->env;
583 
584     if (riscv_has_ext(env, RVD)) {
585         uint64_t reg;
586         for (i = 0; i < 32; i++) {
587             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
588             if (ret) {
589                 return ret;
590             }
591             env->fpr[i] = reg;
592         }
593         return ret;
594     }
595 
596     if (riscv_has_ext(env, RVF)) {
597         uint32_t reg;
598         for (i = 0; i < 32; i++) {
599             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
600             if (ret) {
601                 return ret;
602             }
603             env->fpr[i] = reg;
604         }
605         return ret;
606     }
607 
608     return ret;
609 }
610 
611 static int kvm_riscv_put_regs_fp(CPUState *cs)
612 {
613     int ret = 0;
614     int i;
615     CPURISCVState *env = &RISCV_CPU(cs)->env;
616 
617     if (riscv_has_ext(env, RVD)) {
618         uint64_t reg;
619         for (i = 0; i < 32; i++) {
620             reg = env->fpr[i];
621             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
622             if (ret) {
623                 return ret;
624             }
625         }
626         return ret;
627     }
628 
629     if (riscv_has_ext(env, RVF)) {
630         uint32_t reg;
631         for (i = 0; i < 32; i++) {
632             reg = env->fpr[i];
633             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
634             if (ret) {
635                 return ret;
636             }
637         }
638         return ret;
639     }
640 
641     return ret;
642 }
643 
644 static void kvm_riscv_get_regs_timer(CPUState *cs)
645 {
646     CPURISCVState *env = &RISCV_CPU(cs)->env;
647 
648     if (env->kvm_timer_dirty) {
649         return;
650     }
651 
652     KVM_RISCV_GET_TIMER(cs, env, time, env->kvm_timer_time);
653     KVM_RISCV_GET_TIMER(cs, env, compare, env->kvm_timer_compare);
654     KVM_RISCV_GET_TIMER(cs, env, state, env->kvm_timer_state);
655     KVM_RISCV_GET_TIMER(cs, env, frequency, env->kvm_timer_frequency);
656 
657     env->kvm_timer_dirty = true;
658 }
659 
660 static void kvm_riscv_put_regs_timer(CPUState *cs)
661 {
662     uint64_t reg;
663     CPURISCVState *env = &RISCV_CPU(cs)->env;
664 
665     if (!env->kvm_timer_dirty) {
666         return;
667     }
668 
669     KVM_RISCV_SET_TIMER(cs, env, time, env->kvm_timer_time);
670     KVM_RISCV_SET_TIMER(cs, env, compare, env->kvm_timer_compare);
671 
672     /*
673      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
674      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
675      * doesn't matter that adaping in QEMU now.
676      * TODO If KVM changes, adapt here.
677      */
678     if (env->kvm_timer_state) {
679         KVM_RISCV_SET_TIMER(cs, env, state, env->kvm_timer_state);
680     }
681 
682     /*
683      * For now, migration will not work between Hosts with different timer
684      * frequency. Therefore, we should check whether they are the same here
685      * during the migration.
686      */
687     if (migration_is_running(migrate_get_current()->state)) {
688         KVM_RISCV_GET_TIMER(cs, env, frequency, reg);
689         if (reg != env->kvm_timer_frequency) {
690             error_report("Dst Hosts timer frequency != Src Hosts");
691         }
692     }
693 
694     env->kvm_timer_dirty = false;
695 }
696 
697 typedef struct KVMScratchCPU {
698     int kvmfd;
699     int vmfd;
700     int cpufd;
701 } KVMScratchCPU;
702 
703 /*
704  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
705  * from target/arm/kvm.c.
706  */
707 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
708 {
709     int kvmfd = -1, vmfd = -1, cpufd = -1;
710 
711     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
712     if (kvmfd < 0) {
713         goto err;
714     }
715     do {
716         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
717     } while (vmfd == -1 && errno == EINTR);
718     if (vmfd < 0) {
719         goto err;
720     }
721     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
722     if (cpufd < 0) {
723         goto err;
724     }
725 
726     scratch->kvmfd =  kvmfd;
727     scratch->vmfd = vmfd;
728     scratch->cpufd = cpufd;
729 
730     return true;
731 
732  err:
733     if (cpufd >= 0) {
734         close(cpufd);
735     }
736     if (vmfd >= 0) {
737         close(vmfd);
738     }
739     if (kvmfd >= 0) {
740         close(kvmfd);
741     }
742 
743     return false;
744 }
745 
746 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
747 {
748     close(scratch->cpufd);
749     close(scratch->vmfd);
750     close(scratch->kvmfd);
751 }
752 
753 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
754 {
755     CPURISCVState *env = &cpu->env;
756     struct kvm_one_reg reg;
757     int ret;
758 
759     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
760                               KVM_REG_RISCV_CONFIG_REG(mvendorid));
761     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
762     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
763     if (ret != 0) {
764         error_report("Unable to retrieve mvendorid from host, error %d", ret);
765     }
766 
767     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
768                               KVM_REG_RISCV_CONFIG_REG(marchid));
769     reg.addr = (uint64_t)&cpu->cfg.marchid;
770     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
771     if (ret != 0) {
772         error_report("Unable to retrieve marchid from host, error %d", ret);
773     }
774 
775     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
776                               KVM_REG_RISCV_CONFIG_REG(mimpid));
777     reg.addr = (uint64_t)&cpu->cfg.mimpid;
778     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
779     if (ret != 0) {
780         error_report("Unable to retrieve mimpid from host, error %d", ret);
781     }
782 }
783 
784 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
785                                          KVMScratchCPU *kvmcpu)
786 {
787     CPURISCVState *env = &cpu->env;
788     struct kvm_one_reg reg;
789     int ret;
790 
791     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
792                               KVM_REG_RISCV_CONFIG_REG(isa));
793     reg.addr = (uint64_t)&env->misa_ext_mask;
794     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
795 
796     if (ret) {
797         error_report("Unable to fetch ISA register from KVM, "
798                      "error %d", ret);
799         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
800         exit(EXIT_FAILURE);
801     }
802 
803     env->misa_ext = env->misa_ext_mask;
804 }
805 
806 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
807                                          KVMCPUConfig *cbomz_cfg)
808 {
809     CPURISCVState *env = &cpu->env;
810     struct kvm_one_reg reg;
811     int ret;
812 
813     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
814                               cbomz_cfg->kvm_reg_id);
815     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
816     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
817     if (ret != 0) {
818         error_report("Unable to read KVM reg %s, error %d",
819                      cbomz_cfg->name, ret);
820         exit(EXIT_FAILURE);
821     }
822 }
823 
824 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
825                                            KVMScratchCPU *kvmcpu)
826 {
827     CPURISCVState *env = &cpu->env;
828     uint64_t val;
829     int i, ret;
830 
831     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
832         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
833         struct kvm_one_reg reg;
834 
835         reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
836                                   multi_ext_cfg->kvm_reg_id);
837         reg.addr = (uint64_t)&val;
838         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
839         if (ret != 0) {
840             if (errno == EINVAL) {
841                 /* Silently default to 'false' if KVM does not support it. */
842                 multi_ext_cfg->supported = false;
843                 val = false;
844             } else {
845                 error_report("Unable to read ISA_EXT KVM register %s: %s",
846                              multi_ext_cfg->name, strerror(errno));
847                 exit(EXIT_FAILURE);
848             }
849         } else {
850             multi_ext_cfg->supported = true;
851         }
852 
853         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
854     }
855 
856     if (cpu->cfg.ext_zicbom) {
857         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
858     }
859 
860     if (cpu->cfg.ext_zicboz) {
861         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
862     }
863 }
864 
865 static int uint64_cmp(const void *a, const void *b)
866 {
867     uint64_t val1 = *(const uint64_t *)a;
868     uint64_t val2 = *(const uint64_t *)b;
869 
870     if (val1 < val2) {
871         return -1;
872     }
873 
874     if (val1 > val2) {
875         return 1;
876     }
877 
878     return 0;
879 }
880 
881 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
882 {
883     KVMCPUConfig *multi_ext_cfg;
884     struct kvm_one_reg reg;
885     struct kvm_reg_list rl_struct;
886     struct kvm_reg_list *reglist;
887     uint64_t val, reg_id, *reg_search;
888     int i, ret;
889 
890     rl_struct.n = 0;
891     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
892 
893     /*
894      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
895      * (EINVAL). Use read_legacy() in this case.
896      */
897     if (errno == EINVAL) {
898         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
899     } else if (errno != E2BIG) {
900         /*
901          * E2BIG is an expected error message for the API since we
902          * don't know the number of registers. The right amount will
903          * be written in rl_struct.n.
904          *
905          * Error out if we get any other errno.
906          */
907         error_report("Error when accessing get-reg-list: %s",
908                      strerror(errno));
909         exit(EXIT_FAILURE);
910     }
911 
912     reglist = g_malloc(sizeof(struct kvm_reg_list) +
913                        rl_struct.n * sizeof(uint64_t));
914     reglist->n = rl_struct.n;
915     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
916     if (ret) {
917         error_report("Error when reading KVM_GET_REG_LIST: %s",
918                      strerror(errno));
919         exit(EXIT_FAILURE);
920     }
921 
922     /* sort reglist to use bsearch() */
923     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
924 
925     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
926         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
927         reg_id = kvm_riscv_reg_id(&cpu->env, KVM_REG_RISCV_ISA_EXT,
928                                   multi_ext_cfg->kvm_reg_id);
929         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
930                              sizeof(uint64_t), uint64_cmp);
931         if (!reg_search) {
932             continue;
933         }
934 
935         reg.id = reg_id;
936         reg.addr = (uint64_t)&val;
937         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
938         if (ret != 0) {
939             error_report("Unable to read ISA_EXT KVM register %s: %s",
940                          multi_ext_cfg->name, strerror(errno));
941             exit(EXIT_FAILURE);
942         }
943 
944         multi_ext_cfg->supported = true;
945         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
946     }
947 
948     if (cpu->cfg.ext_zicbom) {
949         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
950     }
951 
952     if (cpu->cfg.ext_zicboz) {
953         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
954     }
955 }
956 
957 static void riscv_init_kvm_registers(Object *cpu_obj)
958 {
959     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
960     KVMScratchCPU kvmcpu;
961 
962     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
963         return;
964     }
965 
966     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
967     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
968     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
969 
970     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
971 }
972 
973 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
974     KVM_CAP_LAST_INFO
975 };
976 
977 int kvm_arch_get_registers(CPUState *cs)
978 {
979     int ret = 0;
980 
981     ret = kvm_riscv_get_regs_core(cs);
982     if (ret) {
983         return ret;
984     }
985 
986     ret = kvm_riscv_get_regs_csr(cs);
987     if (ret) {
988         return ret;
989     }
990 
991     ret = kvm_riscv_get_regs_fp(cs);
992     if (ret) {
993         return ret;
994     }
995 
996     return ret;
997 }
998 
999 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1000 {
1001     if (cap_has_mp_state) {
1002         struct kvm_mp_state mp_state = {
1003             .mp_state = state
1004         };
1005 
1006         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1007         if (ret) {
1008             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1009                     __func__, ret, strerror(-ret));
1010             return -1;
1011         }
1012     }
1013 
1014     return 0;
1015 }
1016 
1017 int kvm_arch_put_registers(CPUState *cs, int level)
1018 {
1019     int ret = 0;
1020 
1021     ret = kvm_riscv_put_regs_core(cs);
1022     if (ret) {
1023         return ret;
1024     }
1025 
1026     ret = kvm_riscv_put_regs_csr(cs);
1027     if (ret) {
1028         return ret;
1029     }
1030 
1031     ret = kvm_riscv_put_regs_fp(cs);
1032     if (ret) {
1033         return ret;
1034     }
1035 
1036     if (KVM_PUT_RESET_STATE == level) {
1037         RISCVCPU *cpu = RISCV_CPU(cs);
1038         if (cs->cpu_index == 0) {
1039             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1040         } else {
1041             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1042         }
1043         if (ret) {
1044             return ret;
1045         }
1046     }
1047 
1048     return ret;
1049 }
1050 
1051 int kvm_arch_release_virq_post(int virq)
1052 {
1053     return 0;
1054 }
1055 
1056 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1057                              uint64_t address, uint32_t data, PCIDevice *dev)
1058 {
1059     return 0;
1060 }
1061 
1062 int kvm_arch_destroy_vcpu(CPUState *cs)
1063 {
1064     return 0;
1065 }
1066 
1067 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1068 {
1069     return cpu->cpu_index;
1070 }
1071 
1072 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1073                                       RunState state)
1074 {
1075     CPUState *cs = opaque;
1076 
1077     if (running) {
1078         kvm_riscv_put_regs_timer(cs);
1079     } else {
1080         kvm_riscv_get_regs_timer(cs);
1081     }
1082 }
1083 
1084 void kvm_arch_init_irq_routing(KVMState *s)
1085 {
1086 }
1087 
1088 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1089 {
1090     CPURISCVState *env = &cpu->env;
1091     target_ulong reg;
1092     uint64_t id;
1093     int ret;
1094 
1095     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1096                           KVM_REG_RISCV_CONFIG_REG(mvendorid));
1097     /*
1098      * cfg.mvendorid is an uint32 but a target_ulong will
1099      * be written. Assign it to a target_ulong var to avoid
1100      * writing pieces of other cpu->cfg fields in the reg.
1101      */
1102     reg = cpu->cfg.mvendorid;
1103     ret = kvm_set_one_reg(cs, id, &reg);
1104     if (ret != 0) {
1105         return ret;
1106     }
1107 
1108     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1109                           KVM_REG_RISCV_CONFIG_REG(marchid));
1110     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1111     if (ret != 0) {
1112         return ret;
1113     }
1114 
1115     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1116                           KVM_REG_RISCV_CONFIG_REG(mimpid));
1117     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1118 
1119     return ret;
1120 }
1121 
1122 int kvm_arch_init_vcpu(CPUState *cs)
1123 {
1124     int ret = 0;
1125     RISCVCPU *cpu = RISCV_CPU(cs);
1126 
1127     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1128 
1129     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1130         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1131         if (ret != 0) {
1132             return ret;
1133         }
1134     }
1135 
1136     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1137     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1138 
1139     return ret;
1140 }
1141 
1142 int kvm_arch_msi_data_to_gsi(uint32_t data)
1143 {
1144     abort();
1145 }
1146 
1147 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1148                                 int vector, PCIDevice *dev)
1149 {
1150     return 0;
1151 }
1152 
1153 int kvm_arch_get_default_type(MachineState *ms)
1154 {
1155     return 0;
1156 }
1157 
1158 int kvm_arch_init(MachineState *ms, KVMState *s)
1159 {
1160     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1161     return 0;
1162 }
1163 
1164 int kvm_arch_irqchip_create(KVMState *s)
1165 {
1166     if (kvm_kernel_irqchip_split()) {
1167         error_report("-machine kernel_irqchip=split is not supported on RISC-V.");
1168         exit(1);
1169     }
1170 
1171     /*
1172      * We can create the VAIA using the newer device control API.
1173      */
1174     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1175 }
1176 
1177 int kvm_arch_process_async_events(CPUState *cs)
1178 {
1179     return 0;
1180 }
1181 
1182 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1183 {
1184 }
1185 
1186 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1187 {
1188     return MEMTXATTRS_UNSPECIFIED;
1189 }
1190 
1191 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1192 {
1193     return true;
1194 }
1195 
1196 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1197 {
1198     int ret = 0;
1199     unsigned char ch;
1200     switch (run->riscv_sbi.extension_id) {
1201     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1202         ch = run->riscv_sbi.args[0];
1203         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1204         break;
1205     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1206         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1207         if (ret == sizeof(ch)) {
1208             run->riscv_sbi.ret[0] = ch;
1209         } else {
1210             run->riscv_sbi.ret[0] = -1;
1211         }
1212         ret = 0;
1213         break;
1214     default:
1215         qemu_log_mask(LOG_UNIMP,
1216                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1217                       __func__, run->riscv_sbi.extension_id);
1218         ret = -1;
1219         break;
1220     }
1221     return ret;
1222 }
1223 
1224 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1225 {
1226     int ret = 0;
1227     switch (run->exit_reason) {
1228     case KVM_EXIT_RISCV_SBI:
1229         ret = kvm_riscv_handle_sbi(cs, run);
1230         break;
1231     default:
1232         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1233                       __func__, run->exit_reason);
1234         ret = -1;
1235         break;
1236     }
1237     return ret;
1238 }
1239 
1240 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1241 {
1242     CPURISCVState *env = &cpu->env;
1243     int i;
1244 
1245     if (!kvm_enabled()) {
1246         return;
1247     }
1248     for (i = 0; i < 32; i++) {
1249         env->gpr[i] = 0;
1250     }
1251     env->pc = cpu->env.kernel_addr;
1252     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1253     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1254     env->satp = 0;
1255     env->mie = 0;
1256     env->stvec = 0;
1257     env->sscratch = 0;
1258     env->sepc = 0;
1259     env->scause = 0;
1260     env->stval = 0;
1261     env->mip = 0;
1262 }
1263 
1264 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1265 {
1266     int ret;
1267     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1268 
1269     if (irq != IRQ_S_EXT) {
1270         perror("kvm riscv set irq != IRQ_S_EXT\n");
1271         abort();
1272     }
1273 
1274     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1275     if (ret < 0) {
1276         perror("Set irq failed");
1277         abort();
1278     }
1279 }
1280 
1281 bool kvm_arch_cpu_check_are_resettable(void)
1282 {
1283     return true;
1284 }
1285 
1286 static int aia_mode;
1287 
1288 static const char *kvm_aia_mode_str(uint64_t mode)
1289 {
1290     switch (mode) {
1291     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1292         return "emul";
1293     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1294         return "hwaccel";
1295     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1296     default:
1297         return "auto";
1298     };
1299 }
1300 
1301 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1302 {
1303     return g_strdup(kvm_aia_mode_str(aia_mode));
1304 }
1305 
1306 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1307 {
1308     if (!strcmp(val, "emul")) {
1309         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1310     } else if (!strcmp(val, "hwaccel")) {
1311         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1312     } else if (!strcmp(val, "auto")) {
1313         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1314     } else {
1315         error_setg(errp, "Invalid KVM AIA mode");
1316         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1317     }
1318 }
1319 
1320 void kvm_arch_accel_class_init(ObjectClass *oc)
1321 {
1322     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1323                                   riscv_set_kvm_aia);
1324     object_class_property_set_description(oc, "riscv-aia",
1325                                           "Set KVM AIA mode. Valid values are "
1326                                           "emul, hwaccel, and auto. Default "
1327                                           "is auto.");
1328     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1329                                     "auto");
1330 }
1331 
1332 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1333                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1334                           uint64_t aplic_base, uint64_t imsic_base,
1335                           uint64_t guest_num)
1336 {
1337     int ret, i;
1338     int aia_fd = -1;
1339     uint64_t default_aia_mode;
1340     uint64_t socket_count = riscv_socket_count(machine);
1341     uint64_t max_hart_per_socket = 0;
1342     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1343     uint64_t socket_bits, hart_bits, guest_bits;
1344 
1345     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1346 
1347     if (aia_fd < 0) {
1348         error_report("Unable to create in-kernel irqchip");
1349         exit(1);
1350     }
1351 
1352     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1353                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1354                             &default_aia_mode, false, NULL);
1355     if (ret < 0) {
1356         error_report("KVM AIA: failed to get current KVM AIA mode");
1357         exit(1);
1358     }
1359     qemu_log("KVM AIA: default mode is %s\n",
1360              kvm_aia_mode_str(default_aia_mode));
1361 
1362     if (default_aia_mode != aia_mode) {
1363         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1364                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1365                                 &aia_mode, true, NULL);
1366         if (ret < 0)
1367             warn_report("KVM AIA: failed to set KVM AIA mode");
1368         else
1369             qemu_log("KVM AIA: set current mode to %s\n",
1370                      kvm_aia_mode_str(aia_mode));
1371     }
1372 
1373     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1374                             KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1375                             &aia_irq_num, true, NULL);
1376     if (ret < 0) {
1377         error_report("KVM AIA: failed to set number of input irq lines");
1378         exit(1);
1379     }
1380 
1381     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1382                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1383                             &aia_msi_num, true, NULL);
1384     if (ret < 0) {
1385         error_report("KVM AIA: failed to set number of msi");
1386         exit(1);
1387     }
1388 
1389     socket_bits = find_last_bit(&socket_count, BITS_PER_LONG) + 1;
1390     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1391                             KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1392                             &socket_bits, true, NULL);
1393     if (ret < 0) {
1394         error_report("KVM AIA: failed to set group_bits");
1395         exit(1);
1396     }
1397 
1398     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1399                             KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1400                             &group_shift, true, NULL);
1401     if (ret < 0) {
1402         error_report("KVM AIA: failed to set group_shift");
1403         exit(1);
1404     }
1405 
1406     guest_bits = guest_num == 0 ? 0 :
1407                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1408     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1409                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1410                             &guest_bits, true, NULL);
1411     if (ret < 0) {
1412         error_report("KVM AIA: failed to set guest_bits");
1413         exit(1);
1414     }
1415 
1416     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1417                             KVM_DEV_RISCV_AIA_ADDR_APLIC,
1418                             &aplic_base, true, NULL);
1419     if (ret < 0) {
1420         error_report("KVM AIA: failed to set the base address of APLIC");
1421         exit(1);
1422     }
1423 
1424     for (socket = 0; socket < socket_count; socket++) {
1425         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1426         hart_count = riscv_socket_hart_count(machine, socket);
1427         base_hart = riscv_socket_first_hartid(machine, socket);
1428 
1429         if (max_hart_per_socket < hart_count) {
1430             max_hart_per_socket = hart_count;
1431         }
1432 
1433         for (i = 0; i < hart_count; i++) {
1434             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1435             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1436                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1437                                     &imsic_addr, true, NULL);
1438             if (ret < 0) {
1439                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1440                 exit(1);
1441             }
1442         }
1443     }
1444 
1445     hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1446     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1447                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1448                             &hart_bits, true, NULL);
1449     if (ret < 0) {
1450         error_report("KVM AIA: failed to set hart_bits");
1451         exit(1);
1452     }
1453 
1454     if (kvm_has_gsi_routing()) {
1455         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1456             /* KVM AIA only has one APLIC instance */
1457             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1458         }
1459         kvm_gsi_routing_allowed = true;
1460         kvm_irqchip_commit_routes(kvm_state);
1461     }
1462 
1463     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1464                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1465                             NULL, true, NULL);
1466     if (ret < 0) {
1467         error_report("KVM AIA: initialized fail");
1468         exit(1);
1469     }
1470 
1471     kvm_msi_via_irqfd_allowed = true;
1472 }
1473 
1474 static void kvm_cpu_instance_init(CPUState *cs)
1475 {
1476     Object *obj = OBJECT(RISCV_CPU(cs));
1477     DeviceState *dev = DEVICE(obj);
1478 
1479     riscv_init_kvm_registers(obj);
1480 
1481     kvm_riscv_add_cpu_user_properties(obj);
1482 
1483     for (Property *prop = riscv_cpu_options; prop && prop->name; prop++) {
1484         /* Check if we have a specific KVM handler for the option */
1485         if (object_property_find(obj, prop->name)) {
1486             continue;
1487         }
1488         qdev_property_add_static(dev, prop);
1489     }
1490 }
1491 
1492 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1493 {
1494     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1495 
1496     acc->cpu_instance_init = kvm_cpu_instance_init;
1497 }
1498 
1499 static const TypeInfo kvm_cpu_accel_type_info = {
1500     .name = ACCEL_CPU_NAME("kvm"),
1501 
1502     .parent = TYPE_ACCEL_CPU,
1503     .class_init = kvm_cpu_accel_class_init,
1504     .abstract = true,
1505 };
1506 static void kvm_cpu_accel_register_types(void)
1507 {
1508     type_register_static(&kvm_cpu_accel_type_info);
1509 }
1510 type_init(kvm_cpu_accel_register_types);
1511 
1512 static void riscv_host_cpu_init(Object *obj)
1513 {
1514     CPURISCVState *env = &RISCV_CPU(obj)->env;
1515 
1516 #if defined(TARGET_RISCV32)
1517     env->misa_mxl_max = env->misa_mxl = MXL_RV32;
1518 #elif defined(TARGET_RISCV64)
1519     env->misa_mxl_max = env->misa_mxl = MXL_RV64;
1520 #endif
1521 }
1522 
1523 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
1524     {
1525         .name = TYPE_RISCV_CPU_HOST,
1526         .parent = TYPE_RISCV_CPU,
1527         .instance_init = riscv_host_cpu_init,
1528     }
1529 };
1530 
1531 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
1532