xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision ce32a9e9)
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     riscv_add_satp_mode_properties(cpu_obj);
400 
401     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
402         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
403         int bit = misa_cfg->offset;
404 
405         misa_cfg->name = riscv_get_misa_ext_name(bit);
406         misa_cfg->description = riscv_get_misa_ext_description(bit);
407 
408         object_property_add(cpu_obj, misa_cfg->name, "bool",
409                             NULL,
410                             kvm_cpu_set_misa_ext_cfg,
411                             NULL, misa_cfg);
412         object_property_set_description(cpu_obj, misa_cfg->name,
413                                         misa_cfg->description);
414     }
415 
416     for (i = 0; misa_bits[i] != 0; i++) {
417         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
418         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
419     }
420 
421     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
422         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
423 
424         object_property_add(cpu_obj, multi_cfg->name, "bool",
425                             NULL,
426                             kvm_cpu_set_multi_ext_cfg,
427                             NULL, multi_cfg);
428     }
429 
430     object_property_add(cpu_obj, "cbom_blocksize", "uint16",
431                         NULL, kvm_cpu_set_cbomz_blksize,
432                         NULL, &kvm_cbom_blocksize);
433 
434     object_property_add(cpu_obj, "cboz_blocksize", "uint16",
435                         NULL, kvm_cpu_set_cbomz_blksize,
436                         NULL, &kvm_cboz_blocksize);
437 
438     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
439     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
440     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
441 }
442 
443 static int kvm_riscv_get_regs_core(CPUState *cs)
444 {
445     int ret = 0;
446     int i;
447     target_ulong reg;
448     CPURISCVState *env = &RISCV_CPU(cs)->env;
449 
450     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
451     if (ret) {
452         return ret;
453     }
454     env->pc = reg;
455 
456     for (i = 1; i < 32; i++) {
457         uint64_t id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, i);
458         ret = kvm_get_one_reg(cs, id, &reg);
459         if (ret) {
460             return ret;
461         }
462         env->gpr[i] = reg;
463     }
464 
465     return ret;
466 }
467 
468 static int kvm_riscv_put_regs_core(CPUState *cs)
469 {
470     int ret = 0;
471     int i;
472     target_ulong reg;
473     CPURISCVState *env = &RISCV_CPU(cs)->env;
474 
475     reg = env->pc;
476     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
477     if (ret) {
478         return ret;
479     }
480 
481     for (i = 1; i < 32; i++) {
482         uint64_t id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, i);
483         reg = env->gpr[i];
484         ret = kvm_set_one_reg(cs, id, &reg);
485         if (ret) {
486             return ret;
487         }
488     }
489 
490     return ret;
491 }
492 
493 static int kvm_riscv_get_regs_csr(CPUState *cs)
494 {
495     int ret = 0;
496     CPURISCVState *env = &RISCV_CPU(cs)->env;
497 
498     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
499     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
500     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
501     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
502     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
503     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
504     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
505     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
506     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
507     return ret;
508 }
509 
510 static int kvm_riscv_put_regs_csr(CPUState *cs)
511 {
512     int ret = 0;
513     CPURISCVState *env = &RISCV_CPU(cs)->env;
514 
515     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
516     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
517     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
518     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
519     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
520     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
521     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
522     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
523     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
524 
525     return ret;
526 }
527 
528 static int kvm_riscv_get_regs_fp(CPUState *cs)
529 {
530     int ret = 0;
531     int i;
532     CPURISCVState *env = &RISCV_CPU(cs)->env;
533 
534     if (riscv_has_ext(env, RVD)) {
535         uint64_t reg;
536         for (i = 0; i < 32; i++) {
537             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(env, i), &reg);
538             if (ret) {
539                 return ret;
540             }
541             env->fpr[i] = reg;
542         }
543         return ret;
544     }
545 
546     if (riscv_has_ext(env, RVF)) {
547         uint32_t reg;
548         for (i = 0; i < 32; i++) {
549             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(env, i), &reg);
550             if (ret) {
551                 return ret;
552             }
553             env->fpr[i] = reg;
554         }
555         return ret;
556     }
557 
558     return ret;
559 }
560 
561 static int kvm_riscv_put_regs_fp(CPUState *cs)
562 {
563     int ret = 0;
564     int i;
565     CPURISCVState *env = &RISCV_CPU(cs)->env;
566 
567     if (riscv_has_ext(env, RVD)) {
568         uint64_t reg;
569         for (i = 0; i < 32; i++) {
570             reg = env->fpr[i];
571             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(env, i), &reg);
572             if (ret) {
573                 return ret;
574             }
575         }
576         return ret;
577     }
578 
579     if (riscv_has_ext(env, RVF)) {
580         uint32_t reg;
581         for (i = 0; i < 32; i++) {
582             reg = env->fpr[i];
583             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(env, i), &reg);
584             if (ret) {
585                 return ret;
586             }
587         }
588         return ret;
589     }
590 
591     return ret;
592 }
593 
594 static void kvm_riscv_get_regs_timer(CPUState *cs)
595 {
596     CPURISCVState *env = &RISCV_CPU(cs)->env;
597 
598     if (env->kvm_timer_dirty) {
599         return;
600     }
601 
602     KVM_RISCV_GET_TIMER(cs, env, time, env->kvm_timer_time);
603     KVM_RISCV_GET_TIMER(cs, env, compare, env->kvm_timer_compare);
604     KVM_RISCV_GET_TIMER(cs, env, state, env->kvm_timer_state);
605     KVM_RISCV_GET_TIMER(cs, env, frequency, env->kvm_timer_frequency);
606 
607     env->kvm_timer_dirty = true;
608 }
609 
610 static void kvm_riscv_put_regs_timer(CPUState *cs)
611 {
612     uint64_t reg;
613     CPURISCVState *env = &RISCV_CPU(cs)->env;
614 
615     if (!env->kvm_timer_dirty) {
616         return;
617     }
618 
619     KVM_RISCV_SET_TIMER(cs, env, time, env->kvm_timer_time);
620     KVM_RISCV_SET_TIMER(cs, env, compare, env->kvm_timer_compare);
621 
622     /*
623      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
624      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
625      * doesn't matter that adaping in QEMU now.
626      * TODO If KVM changes, adapt here.
627      */
628     if (env->kvm_timer_state) {
629         KVM_RISCV_SET_TIMER(cs, env, state, env->kvm_timer_state);
630     }
631 
632     /*
633      * For now, migration will not work between Hosts with different timer
634      * frequency. Therefore, we should check whether they are the same here
635      * during the migration.
636      */
637     if (migration_is_running(migrate_get_current()->state)) {
638         KVM_RISCV_GET_TIMER(cs, env, frequency, reg);
639         if (reg != env->kvm_timer_frequency) {
640             error_report("Dst Hosts timer frequency != Src Hosts");
641         }
642     }
643 
644     env->kvm_timer_dirty = false;
645 }
646 
647 typedef struct KVMScratchCPU {
648     int kvmfd;
649     int vmfd;
650     int cpufd;
651 } KVMScratchCPU;
652 
653 /*
654  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
655  * from target/arm/kvm.c.
656  */
657 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
658 {
659     int kvmfd = -1, vmfd = -1, cpufd = -1;
660 
661     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
662     if (kvmfd < 0) {
663         goto err;
664     }
665     do {
666         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
667     } while (vmfd == -1 && errno == EINTR);
668     if (vmfd < 0) {
669         goto err;
670     }
671     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
672     if (cpufd < 0) {
673         goto err;
674     }
675 
676     scratch->kvmfd =  kvmfd;
677     scratch->vmfd = vmfd;
678     scratch->cpufd = cpufd;
679 
680     return true;
681 
682  err:
683     if (cpufd >= 0) {
684         close(cpufd);
685     }
686     if (vmfd >= 0) {
687         close(vmfd);
688     }
689     if (kvmfd >= 0) {
690         close(kvmfd);
691     }
692 
693     return false;
694 }
695 
696 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
697 {
698     close(scratch->cpufd);
699     close(scratch->vmfd);
700     close(scratch->kvmfd);
701 }
702 
703 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
704 {
705     CPURISCVState *env = &cpu->env;
706     struct kvm_one_reg reg;
707     int ret;
708 
709     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
710                               KVM_REG_RISCV_CONFIG_REG(mvendorid));
711     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
712     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
713     if (ret != 0) {
714         error_report("Unable to retrieve mvendorid from host, error %d", ret);
715     }
716 
717     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
718                               KVM_REG_RISCV_CONFIG_REG(marchid));
719     reg.addr = (uint64_t)&cpu->cfg.marchid;
720     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
721     if (ret != 0) {
722         error_report("Unable to retrieve marchid from host, error %d", ret);
723     }
724 
725     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
726                               KVM_REG_RISCV_CONFIG_REG(mimpid));
727     reg.addr = (uint64_t)&cpu->cfg.mimpid;
728     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
729     if (ret != 0) {
730         error_report("Unable to retrieve mimpid from host, error %d", ret);
731     }
732 }
733 
734 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
735                                          KVMScratchCPU *kvmcpu)
736 {
737     CPURISCVState *env = &cpu->env;
738     struct kvm_one_reg reg;
739     int ret;
740 
741     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
742                               KVM_REG_RISCV_CONFIG_REG(isa));
743     reg.addr = (uint64_t)&env->misa_ext_mask;
744     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
745 
746     if (ret) {
747         error_report("Unable to fetch ISA register from KVM, "
748                      "error %d", ret);
749         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
750         exit(EXIT_FAILURE);
751     }
752 
753     env->misa_ext = env->misa_ext_mask;
754 }
755 
756 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
757                                          KVMCPUConfig *cbomz_cfg)
758 {
759     CPURISCVState *env = &cpu->env;
760     struct kvm_one_reg reg;
761     int ret;
762 
763     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
764                               cbomz_cfg->kvm_reg_id);
765     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
766     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
767     if (ret != 0) {
768         error_report("Unable to read KVM reg %s, error %d",
769                      cbomz_cfg->name, ret);
770         exit(EXIT_FAILURE);
771     }
772 }
773 
774 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
775                                            KVMScratchCPU *kvmcpu)
776 {
777     CPURISCVState *env = &cpu->env;
778     uint64_t val;
779     int i, ret;
780 
781     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
782         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
783         struct kvm_one_reg reg;
784 
785         reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
786                                   multi_ext_cfg->kvm_reg_id);
787         reg.addr = (uint64_t)&val;
788         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
789         if (ret != 0) {
790             if (errno == EINVAL) {
791                 /* Silently default to 'false' if KVM does not support it. */
792                 multi_ext_cfg->supported = false;
793                 val = false;
794             } else {
795                 error_report("Unable to read ISA_EXT KVM register %s, "
796                              "error code: %s", multi_ext_cfg->name,
797                              strerrorname_np(errno));
798                 exit(EXIT_FAILURE);
799             }
800         } else {
801             multi_ext_cfg->supported = true;
802         }
803 
804         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
805     }
806 
807     if (cpu->cfg.ext_icbom) {
808         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
809     }
810 
811     if (cpu->cfg.ext_icboz) {
812         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
813     }
814 }
815 
816 static int uint64_cmp(const void *a, const void *b)
817 {
818     uint64_t val1 = *(const uint64_t *)a;
819     uint64_t val2 = *(const uint64_t *)b;
820 
821     if (val1 < val2) {
822         return -1;
823     }
824 
825     if (val1 > val2) {
826         return 1;
827     }
828 
829     return 0;
830 }
831 
832 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
833 {
834     KVMCPUConfig *multi_ext_cfg;
835     struct kvm_one_reg reg;
836     struct kvm_reg_list rl_struct;
837     struct kvm_reg_list *reglist;
838     uint64_t val, reg_id, *reg_search;
839     int i, ret;
840 
841     rl_struct.n = 0;
842     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
843 
844     /*
845      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
846      * (EINVAL). Use read_legacy() in this case.
847      */
848     if (errno == EINVAL) {
849         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
850     } else if (errno != E2BIG) {
851         /*
852          * E2BIG is an expected error message for the API since we
853          * don't know the number of registers. The right amount will
854          * be written in rl_struct.n.
855          *
856          * Error out if we get any other errno.
857          */
858         error_report("Error when accessing get-reg-list, code: %s",
859                      strerrorname_np(errno));
860         exit(EXIT_FAILURE);
861     }
862 
863     reglist = g_malloc(sizeof(struct kvm_reg_list) +
864                        rl_struct.n * sizeof(uint64_t));
865     reglist->n = rl_struct.n;
866     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
867     if (ret) {
868         error_report("Error when reading KVM_GET_REG_LIST, code %s ",
869                      strerrorname_np(errno));
870         exit(EXIT_FAILURE);
871     }
872 
873     /* sort reglist to use bsearch() */
874     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
875 
876     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
877         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
878         reg_id = kvm_riscv_reg_id(&cpu->env, KVM_REG_RISCV_ISA_EXT,
879                                   multi_ext_cfg->kvm_reg_id);
880         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
881                              sizeof(uint64_t), uint64_cmp);
882         if (!reg_search) {
883             continue;
884         }
885 
886         reg.id = reg_id;
887         reg.addr = (uint64_t)&val;
888         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
889         if (ret != 0) {
890             error_report("Unable to read ISA_EXT KVM register %s, "
891                          "error code: %s", multi_ext_cfg->name,
892                          strerrorname_np(errno));
893             exit(EXIT_FAILURE);
894         }
895 
896         multi_ext_cfg->supported = true;
897         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
898     }
899 
900     if (cpu->cfg.ext_icbom) {
901         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
902     }
903 
904     if (cpu->cfg.ext_icboz) {
905         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
906     }
907 }
908 
909 static void riscv_init_kvm_registers(Object *cpu_obj)
910 {
911     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
912     KVMScratchCPU kvmcpu;
913 
914     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
915         return;
916     }
917 
918     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
919     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
920     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
921 
922     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
923 }
924 
925 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
926     KVM_CAP_LAST_INFO
927 };
928 
929 int kvm_arch_get_registers(CPUState *cs)
930 {
931     int ret = 0;
932 
933     ret = kvm_riscv_get_regs_core(cs);
934     if (ret) {
935         return ret;
936     }
937 
938     ret = kvm_riscv_get_regs_csr(cs);
939     if (ret) {
940         return ret;
941     }
942 
943     ret = kvm_riscv_get_regs_fp(cs);
944     if (ret) {
945         return ret;
946     }
947 
948     return ret;
949 }
950 
951 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
952 {
953     if (cap_has_mp_state) {
954         struct kvm_mp_state mp_state = {
955             .mp_state = state
956         };
957 
958         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
959         if (ret) {
960             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
961                     __func__, ret, strerror(-ret));
962             return -1;
963         }
964     }
965 
966     return 0;
967 }
968 
969 int kvm_arch_put_registers(CPUState *cs, int level)
970 {
971     int ret = 0;
972 
973     ret = kvm_riscv_put_regs_core(cs);
974     if (ret) {
975         return ret;
976     }
977 
978     ret = kvm_riscv_put_regs_csr(cs);
979     if (ret) {
980         return ret;
981     }
982 
983     ret = kvm_riscv_put_regs_fp(cs);
984     if (ret) {
985         return ret;
986     }
987 
988     if (KVM_PUT_RESET_STATE == level) {
989         RISCVCPU *cpu = RISCV_CPU(cs);
990         if (cs->cpu_index == 0) {
991             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
992         } else {
993             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
994         }
995         if (ret) {
996             return ret;
997         }
998     }
999 
1000     return ret;
1001 }
1002 
1003 int kvm_arch_release_virq_post(int virq)
1004 {
1005     return 0;
1006 }
1007 
1008 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1009                              uint64_t address, uint32_t data, PCIDevice *dev)
1010 {
1011     return 0;
1012 }
1013 
1014 int kvm_arch_destroy_vcpu(CPUState *cs)
1015 {
1016     return 0;
1017 }
1018 
1019 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1020 {
1021     return cpu->cpu_index;
1022 }
1023 
1024 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1025                                       RunState state)
1026 {
1027     CPUState *cs = opaque;
1028 
1029     if (running) {
1030         kvm_riscv_put_regs_timer(cs);
1031     } else {
1032         kvm_riscv_get_regs_timer(cs);
1033     }
1034 }
1035 
1036 void kvm_arch_init_irq_routing(KVMState *s)
1037 {
1038 }
1039 
1040 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1041 {
1042     CPURISCVState *env = &cpu->env;
1043     target_ulong reg;
1044     uint64_t id;
1045     int ret;
1046 
1047     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1048                           KVM_REG_RISCV_CONFIG_REG(mvendorid));
1049     /*
1050      * cfg.mvendorid is an uint32 but a target_ulong will
1051      * be written. Assign it to a target_ulong var to avoid
1052      * writing pieces of other cpu->cfg fields in the reg.
1053      */
1054     reg = cpu->cfg.mvendorid;
1055     ret = kvm_set_one_reg(cs, id, &reg);
1056     if (ret != 0) {
1057         return ret;
1058     }
1059 
1060     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1061                           KVM_REG_RISCV_CONFIG_REG(marchid));
1062     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1063     if (ret != 0) {
1064         return ret;
1065     }
1066 
1067     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1068                           KVM_REG_RISCV_CONFIG_REG(mimpid));
1069     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1070 
1071     return ret;
1072 }
1073 
1074 int kvm_arch_init_vcpu(CPUState *cs)
1075 {
1076     int ret = 0;
1077     RISCVCPU *cpu = RISCV_CPU(cs);
1078 
1079     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1080 
1081     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1082         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1083         if (ret != 0) {
1084             return ret;
1085         }
1086     }
1087 
1088     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1089     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1090 
1091     return ret;
1092 }
1093 
1094 int kvm_arch_msi_data_to_gsi(uint32_t data)
1095 {
1096     abort();
1097 }
1098 
1099 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1100                                 int vector, PCIDevice *dev)
1101 {
1102     return 0;
1103 }
1104 
1105 int kvm_arch_get_default_type(MachineState *ms)
1106 {
1107     return 0;
1108 }
1109 
1110 int kvm_arch_init(MachineState *ms, KVMState *s)
1111 {
1112     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1113     return 0;
1114 }
1115 
1116 int kvm_arch_irqchip_create(KVMState *s)
1117 {
1118     if (kvm_kernel_irqchip_split()) {
1119         error_report("-machine kernel_irqchip=split is not supported on RISC-V.");
1120         exit(1);
1121     }
1122 
1123     /*
1124      * We can create the VAIA using the newer device control API.
1125      */
1126     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1127 }
1128 
1129 int kvm_arch_process_async_events(CPUState *cs)
1130 {
1131     return 0;
1132 }
1133 
1134 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1135 {
1136 }
1137 
1138 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1139 {
1140     return MEMTXATTRS_UNSPECIFIED;
1141 }
1142 
1143 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1144 {
1145     return true;
1146 }
1147 
1148 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1149 {
1150     int ret = 0;
1151     unsigned char ch;
1152     switch (run->riscv_sbi.extension_id) {
1153     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1154         ch = run->riscv_sbi.args[0];
1155         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1156         break;
1157     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1158         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1159         if (ret == sizeof(ch)) {
1160             run->riscv_sbi.ret[0] = ch;
1161         } else {
1162             run->riscv_sbi.ret[0] = -1;
1163         }
1164         ret = 0;
1165         break;
1166     default:
1167         qemu_log_mask(LOG_UNIMP,
1168                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1169                       __func__, run->riscv_sbi.extension_id);
1170         ret = -1;
1171         break;
1172     }
1173     return ret;
1174 }
1175 
1176 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1177 {
1178     int ret = 0;
1179     switch (run->exit_reason) {
1180     case KVM_EXIT_RISCV_SBI:
1181         ret = kvm_riscv_handle_sbi(cs, run);
1182         break;
1183     default:
1184         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1185                       __func__, run->exit_reason);
1186         ret = -1;
1187         break;
1188     }
1189     return ret;
1190 }
1191 
1192 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1193 {
1194     CPURISCVState *env = &cpu->env;
1195     int i;
1196 
1197     if (!kvm_enabled()) {
1198         return;
1199     }
1200     for (i = 0; i < 32; i++) {
1201         env->gpr[i] = 0;
1202     }
1203     env->pc = cpu->env.kernel_addr;
1204     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1205     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1206     env->satp = 0;
1207     env->mie = 0;
1208     env->stvec = 0;
1209     env->sscratch = 0;
1210     env->sepc = 0;
1211     env->scause = 0;
1212     env->stval = 0;
1213     env->mip = 0;
1214 }
1215 
1216 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1217 {
1218     int ret;
1219     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1220 
1221     if (irq != IRQ_S_EXT) {
1222         perror("kvm riscv set irq != IRQ_S_EXT\n");
1223         abort();
1224     }
1225 
1226     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1227     if (ret < 0) {
1228         perror("Set irq failed");
1229         abort();
1230     }
1231 }
1232 
1233 bool kvm_arch_cpu_check_are_resettable(void)
1234 {
1235     return true;
1236 }
1237 
1238 static int aia_mode;
1239 
1240 static const char *kvm_aia_mode_str(uint64_t mode)
1241 {
1242     switch (mode) {
1243     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1244         return "emul";
1245     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1246         return "hwaccel";
1247     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1248     default:
1249         return "auto";
1250     };
1251 }
1252 
1253 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1254 {
1255     return g_strdup(kvm_aia_mode_str(aia_mode));
1256 }
1257 
1258 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1259 {
1260     if (!strcmp(val, "emul")) {
1261         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1262     } else if (!strcmp(val, "hwaccel")) {
1263         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1264     } else if (!strcmp(val, "auto")) {
1265         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1266     } else {
1267         error_setg(errp, "Invalid KVM AIA mode");
1268         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1269     }
1270 }
1271 
1272 void kvm_arch_accel_class_init(ObjectClass *oc)
1273 {
1274     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1275                                   riscv_set_kvm_aia);
1276     object_class_property_set_description(oc, "riscv-aia",
1277                                           "Set KVM AIA mode. Valid values are "
1278                                           "emul, hwaccel, and auto. Default "
1279                                           "is auto.");
1280     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1281                                     "auto");
1282 }
1283 
1284 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1285                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1286                           uint64_t aplic_base, uint64_t imsic_base,
1287                           uint64_t guest_num)
1288 {
1289     int ret, i;
1290     int aia_fd = -1;
1291     uint64_t default_aia_mode;
1292     uint64_t socket_count = riscv_socket_count(machine);
1293     uint64_t max_hart_per_socket = 0;
1294     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1295     uint64_t socket_bits, hart_bits, guest_bits;
1296 
1297     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1298 
1299     if (aia_fd < 0) {
1300         error_report("Unable to create in-kernel irqchip");
1301         exit(1);
1302     }
1303 
1304     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1305                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1306                             &default_aia_mode, false, NULL);
1307     if (ret < 0) {
1308         error_report("KVM AIA: failed to get current KVM AIA mode");
1309         exit(1);
1310     }
1311     qemu_log("KVM AIA: default mode is %s\n",
1312              kvm_aia_mode_str(default_aia_mode));
1313 
1314     if (default_aia_mode != aia_mode) {
1315         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1316                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1317                                 &aia_mode, true, NULL);
1318         if (ret < 0)
1319             warn_report("KVM AIA: failed to set KVM AIA mode");
1320         else
1321             qemu_log("KVM AIA: set current mode to %s\n",
1322                      kvm_aia_mode_str(aia_mode));
1323     }
1324 
1325     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1326                             KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1327                             &aia_irq_num, true, NULL);
1328     if (ret < 0) {
1329         error_report("KVM AIA: failed to set number of input irq lines");
1330         exit(1);
1331     }
1332 
1333     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1334                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1335                             &aia_msi_num, true, NULL);
1336     if (ret < 0) {
1337         error_report("KVM AIA: failed to set number of msi");
1338         exit(1);
1339     }
1340 
1341     socket_bits = find_last_bit(&socket_count, BITS_PER_LONG) + 1;
1342     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1343                             KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1344                             &socket_bits, true, NULL);
1345     if (ret < 0) {
1346         error_report("KVM AIA: failed to set group_bits");
1347         exit(1);
1348     }
1349 
1350     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1351                             KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1352                             &group_shift, true, NULL);
1353     if (ret < 0) {
1354         error_report("KVM AIA: failed to set group_shift");
1355         exit(1);
1356     }
1357 
1358     guest_bits = guest_num == 0 ? 0 :
1359                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1360     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1361                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1362                             &guest_bits, true, NULL);
1363     if (ret < 0) {
1364         error_report("KVM AIA: failed to set guest_bits");
1365         exit(1);
1366     }
1367 
1368     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1369                             KVM_DEV_RISCV_AIA_ADDR_APLIC,
1370                             &aplic_base, true, NULL);
1371     if (ret < 0) {
1372         error_report("KVM AIA: failed to set the base address of APLIC");
1373         exit(1);
1374     }
1375 
1376     for (socket = 0; socket < socket_count; socket++) {
1377         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1378         hart_count = riscv_socket_hart_count(machine, socket);
1379         base_hart = riscv_socket_first_hartid(machine, socket);
1380 
1381         if (max_hart_per_socket < hart_count) {
1382             max_hart_per_socket = hart_count;
1383         }
1384 
1385         for (i = 0; i < hart_count; i++) {
1386             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1387             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1388                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1389                                     &imsic_addr, true, NULL);
1390             if (ret < 0) {
1391                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1392                 exit(1);
1393             }
1394         }
1395     }
1396 
1397     hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1398     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1399                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1400                             &hart_bits, true, NULL);
1401     if (ret < 0) {
1402         error_report("KVM AIA: failed to set hart_bits");
1403         exit(1);
1404     }
1405 
1406     if (kvm_has_gsi_routing()) {
1407         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1408             /* KVM AIA only has one APLIC instance */
1409             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1410         }
1411         kvm_gsi_routing_allowed = true;
1412         kvm_irqchip_commit_routes(kvm_state);
1413     }
1414 
1415     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1416                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1417                             NULL, true, NULL);
1418     if (ret < 0) {
1419         error_report("KVM AIA: initialized fail");
1420         exit(1);
1421     }
1422 
1423     kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
1424 }
1425 
1426 static void kvm_cpu_instance_init(CPUState *cs)
1427 {
1428     Object *obj = OBJECT(RISCV_CPU(cs));
1429     DeviceState *dev = DEVICE(obj);
1430 
1431     riscv_init_kvm_registers(obj);
1432 
1433     kvm_riscv_add_cpu_user_properties(obj);
1434 
1435     for (Property *prop = riscv_cpu_options; prop && prop->name; prop++) {
1436         /* Check if we have a specific KVM handler for the option */
1437         if (object_property_find(obj, prop->name)) {
1438             continue;
1439         }
1440         qdev_property_add_static(dev, prop);
1441     }
1442 }
1443 
1444 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1445 {
1446     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1447 
1448     acc->cpu_instance_init = kvm_cpu_instance_init;
1449 }
1450 
1451 static const TypeInfo kvm_cpu_accel_type_info = {
1452     .name = ACCEL_CPU_NAME("kvm"),
1453 
1454     .parent = TYPE_ACCEL_CPU,
1455     .class_init = kvm_cpu_accel_class_init,
1456     .abstract = true,
1457 };
1458 static void kvm_cpu_accel_register_types(void)
1459 {
1460     type_register_static(&kvm_cpu_accel_type_info);
1461 }
1462 type_init(kvm_cpu_accel_register_types);
1463 
1464 static void riscv_host_cpu_init(Object *obj)
1465 {
1466     CPURISCVState *env = &RISCV_CPU(obj)->env;
1467 
1468 #if defined(TARGET_RISCV32)
1469     env->misa_mxl_max = env->misa_mxl = MXL_RV32;
1470 #elif defined(TARGET_RISCV64)
1471     env->misa_mxl_max = env->misa_mxl = MXL_RV64;
1472 #endif
1473 }
1474 
1475 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
1476     {
1477         .name = TYPE_RISCV_CPU_HOST,
1478         .parent = TYPE_RISCV_CPU,
1479         .instance_init = riscv_host_cpu_init,
1480     }
1481 };
1482 
1483 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
1484