xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision 871dad3a19c660e17c412b1bc969195f814933c8)
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_ulong(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) \
86     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, \
87                            KVM_REG_RISCV_CORE_REG(name))
88 
89 #define RISCV_CSR_REG(env, name) \
90     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CSR, \
91                            KVM_REG_RISCV_CSR_REG(name))
92 
93 #define RISCV_CONFIG_REG(env, name) \
94     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG, \
95                            KVM_REG_RISCV_CONFIG_REG(name))
96 
97 #define RISCV_TIMER_REG(name)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_TIMER, \
98                  KVM_REG_RISCV_TIMER_REG(name))
99 
100 #define RISCV_FP_F_REG(idx)  kvm_riscv_reg_id_u32(KVM_REG_RISCV_FP_F, idx)
101 
102 #define RISCV_FP_D_REG(idx)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_FP_D, idx)
103 
104 #define KVM_RISCV_GET_CSR(cs, env, csr, reg) \
105     do { \
106         int _ret = kvm_get_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
107         if (_ret) { \
108             return _ret; \
109         } \
110     } while (0)
111 
112 #define KVM_RISCV_SET_CSR(cs, env, csr, reg) \
113     do { \
114         int _ret = kvm_set_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
115         if (_ret) { \
116             return _ret; \
117         } \
118     } while (0)
119 
120 #define KVM_RISCV_GET_TIMER(cs, name, reg) \
121     do { \
122         int ret = kvm_get_one_reg(cs, RISCV_TIMER_REG(name), &reg); \
123         if (ret) { \
124             abort(); \
125         } \
126     } while (0)
127 
128 #define KVM_RISCV_SET_TIMER(cs, name, reg) \
129     do { \
130         int ret = kvm_set_one_reg(cs, RISCV_TIMER_REG(name), &reg); \
131         if (ret) { \
132             abort(); \
133         } \
134     } while (0)
135 
136 typedef struct KVMCPUConfig {
137     const char *name;
138     const char *description;
139     target_ulong offset;
140     int kvm_reg_id;
141     bool user_set;
142     bool supported;
143 } KVMCPUConfig;
144 
145 #define KVM_MISA_CFG(_bit, _reg_id) \
146     {.offset = _bit, .kvm_reg_id = _reg_id}
147 
148 /* KVM ISA extensions */
149 static KVMCPUConfig kvm_misa_ext_cfgs[] = {
150     KVM_MISA_CFG(RVA, KVM_RISCV_ISA_EXT_A),
151     KVM_MISA_CFG(RVC, KVM_RISCV_ISA_EXT_C),
152     KVM_MISA_CFG(RVD, KVM_RISCV_ISA_EXT_D),
153     KVM_MISA_CFG(RVF, KVM_RISCV_ISA_EXT_F),
154     KVM_MISA_CFG(RVH, KVM_RISCV_ISA_EXT_H),
155     KVM_MISA_CFG(RVI, KVM_RISCV_ISA_EXT_I),
156     KVM_MISA_CFG(RVM, KVM_RISCV_ISA_EXT_M),
157 };
158 
159 static void kvm_cpu_get_misa_ext_cfg(Object *obj, Visitor *v,
160                                      const char *name,
161                                      void *opaque, Error **errp)
162 {
163     KVMCPUConfig *misa_ext_cfg = opaque;
164     target_ulong misa_bit = misa_ext_cfg->offset;
165     RISCVCPU *cpu = RISCV_CPU(obj);
166     CPURISCVState *env = &cpu->env;
167     bool value = env->misa_ext_mask & misa_bit;
168 
169     visit_type_bool(v, name, &value, errp);
170 }
171 
172 static void kvm_cpu_set_misa_ext_cfg(Object *obj, Visitor *v,
173                                      const char *name,
174                                      void *opaque, Error **errp)
175 {
176     KVMCPUConfig *misa_ext_cfg = opaque;
177     target_ulong misa_bit = misa_ext_cfg->offset;
178     RISCVCPU *cpu = RISCV_CPU(obj);
179     CPURISCVState *env = &cpu->env;
180     bool value, host_bit;
181 
182     if (!visit_type_bool(v, name, &value, errp)) {
183         return;
184     }
185 
186     host_bit = env->misa_ext_mask & misa_bit;
187 
188     if (value == host_bit) {
189         return;
190     }
191 
192     if (!value) {
193         misa_ext_cfg->user_set = true;
194         return;
195     }
196 
197     /*
198      * Forbid users to enable extensions that aren't
199      * available in the hart.
200      */
201     error_setg(errp, "Enabling MISA bit '%s' is not allowed: it's not "
202                "enabled in the host", misa_ext_cfg->name);
203 }
204 
205 static void kvm_riscv_update_cpu_misa_ext(RISCVCPU *cpu, CPUState *cs)
206 {
207     CPURISCVState *env = &cpu->env;
208     uint64_t id, reg;
209     int i, ret;
210 
211     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
212         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
213         target_ulong misa_bit = misa_cfg->offset;
214 
215         if (!misa_cfg->user_set) {
216             continue;
217         }
218 
219         /* If we're here we're going to disable the MISA bit */
220         reg = 0;
221         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
222                                     misa_cfg->kvm_reg_id);
223         ret = kvm_set_one_reg(cs, id, &reg);
224         if (ret != 0) {
225             /*
226              * We're not checking for -EINVAL because if the bit is about
227              * to be disabled, it means that it was already enabled by
228              * KVM. We determined that by fetching the 'isa' register
229              * during init() time. Any error at this point is worth
230              * aborting.
231              */
232             error_report("Unable to set KVM reg %s, error %d",
233                          misa_cfg->name, ret);
234             exit(EXIT_FAILURE);
235         }
236         env->misa_ext &= ~misa_bit;
237     }
238 }
239 
240 #define KVM_EXT_CFG(_name, _prop, _reg_id) \
241     {.name = _name, .offset = CPU_CFG_OFFSET(_prop), \
242      .kvm_reg_id = _reg_id}
243 
244 static KVMCPUConfig kvm_multi_ext_cfgs[] = {
245     KVM_EXT_CFG("zicbom", ext_zicbom, KVM_RISCV_ISA_EXT_ZICBOM),
246     KVM_EXT_CFG("zicboz", ext_zicboz, KVM_RISCV_ISA_EXT_ZICBOZ),
247     KVM_EXT_CFG("zicntr", ext_zicntr, KVM_RISCV_ISA_EXT_ZICNTR),
248     KVM_EXT_CFG("zicsr", ext_zicsr, KVM_RISCV_ISA_EXT_ZICSR),
249     KVM_EXT_CFG("zifencei", ext_zifencei, KVM_RISCV_ISA_EXT_ZIFENCEI),
250     KVM_EXT_CFG("zihintpause", ext_zihintpause, KVM_RISCV_ISA_EXT_ZIHINTPAUSE),
251     KVM_EXT_CFG("zihpm", ext_zihpm, KVM_RISCV_ISA_EXT_ZIHPM),
252     KVM_EXT_CFG("zba", ext_zba, KVM_RISCV_ISA_EXT_ZBA),
253     KVM_EXT_CFG("zbb", ext_zbb, KVM_RISCV_ISA_EXT_ZBB),
254     KVM_EXT_CFG("zbs", ext_zbs, KVM_RISCV_ISA_EXT_ZBS),
255     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
256     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
257     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
258     KVM_EXT_CFG("svnapot", ext_svnapot, KVM_RISCV_ISA_EXT_SVNAPOT),
259     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
260 };
261 
262 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
263 {
264     return (void *)&cpu->cfg + kvmcfg->offset;
265 }
266 
267 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
268                             uint32_t val)
269 {
270     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
271 
272     *ext_enabled = val;
273 }
274 
275 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
276                                 KVMCPUConfig *multi_ext)
277 {
278     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
279 
280     return *ext_enabled;
281 }
282 
283 static void kvm_cpu_get_multi_ext_cfg(Object *obj, Visitor *v,
284                                       const char *name,
285                                       void *opaque, Error **errp)
286 {
287     KVMCPUConfig *multi_ext_cfg = opaque;
288     RISCVCPU *cpu = RISCV_CPU(obj);
289     bool value = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
290 
291     visit_type_bool(v, name, &value, errp);
292 }
293 
294 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
295                                       const char *name,
296                                       void *opaque, Error **errp)
297 {
298     KVMCPUConfig *multi_ext_cfg = opaque;
299     RISCVCPU *cpu = RISCV_CPU(obj);
300     bool value, host_val;
301 
302     if (!visit_type_bool(v, name, &value, errp)) {
303         return;
304     }
305 
306     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
307 
308     /*
309      * Ignore if the user is setting the same value
310      * as the host.
311      */
312     if (value == host_val) {
313         return;
314     }
315 
316     if (!multi_ext_cfg->supported) {
317         /*
318          * Error out if the user is trying to enable an
319          * extension that KVM doesn't support. Ignore
320          * option otherwise.
321          */
322         if (value) {
323             error_setg(errp, "KVM does not support disabling extension %s",
324                        multi_ext_cfg->name);
325         }
326 
327         return;
328     }
329 
330     multi_ext_cfg->user_set = true;
331     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
332 }
333 
334 static KVMCPUConfig kvm_cbom_blocksize = {
335     .name = "cbom_blocksize",
336     .offset = CPU_CFG_OFFSET(cbom_blocksize),
337     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
338 };
339 
340 static KVMCPUConfig kvm_cboz_blocksize = {
341     .name = "cboz_blocksize",
342     .offset = CPU_CFG_OFFSET(cboz_blocksize),
343     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
344 };
345 
346 static void kvm_cpu_set_cbomz_blksize(Object *obj, Visitor *v,
347                                       const char *name,
348                                       void *opaque, Error **errp)
349 {
350     KVMCPUConfig *cbomz_cfg = opaque;
351     RISCVCPU *cpu = RISCV_CPU(obj);
352     uint16_t value, *host_val;
353 
354     if (!visit_type_uint16(v, name, &value, errp)) {
355         return;
356     }
357 
358     host_val = kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
359 
360     if (value != *host_val) {
361         error_report("Unable to set %s to a different value than "
362                      "the host (%u)",
363                      cbomz_cfg->name, *host_val);
364         exit(EXIT_FAILURE);
365     }
366 
367     cbomz_cfg->user_set = true;
368 }
369 
370 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
371 {
372     CPURISCVState *env = &cpu->env;
373     uint64_t id, reg;
374     int i, ret;
375 
376     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
377         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
378 
379         if (!multi_ext_cfg->user_set) {
380             continue;
381         }
382 
383         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
384                                     multi_ext_cfg->kvm_reg_id);
385         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
386         ret = kvm_set_one_reg(cs, id, &reg);
387         if (ret != 0) {
388             error_report("Unable to %s extension %s in KVM, error %d",
389                          reg ? "enable" : "disable",
390                          multi_ext_cfg->name, ret);
391             exit(EXIT_FAILURE);
392         }
393     }
394 }
395 
396 static void cpu_get_cfg_unavailable(Object *obj, Visitor *v,
397                                     const char *name,
398                                     void *opaque, Error **errp)
399 {
400     bool value = false;
401 
402     visit_type_bool(v, name, &value, errp);
403 }
404 
405 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
406                                     const char *name,
407                                     void *opaque, Error **errp)
408 {
409     const char *propname = opaque;
410     bool value;
411 
412     if (!visit_type_bool(v, name, &value, errp)) {
413         return;
414     }
415 
416     if (value) {
417         error_setg(errp, "'%s' is not available with KVM",
418                    propname);
419     }
420 }
421 
422 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
423 {
424     /* Check if KVM created the property already */
425     if (object_property_find(obj, prop_name)) {
426         return;
427     }
428 
429     /*
430      * Set the default to disabled for every extension
431      * unknown to KVM and error out if the user attempts
432      * to enable any of them.
433      */
434     object_property_add(obj, prop_name, "bool",
435                         cpu_get_cfg_unavailable,
436                         cpu_set_cfg_unavailable,
437                         NULL, (void *)prop_name);
438 }
439 
440 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
441                                         const RISCVCPUMultiExtConfig *array)
442 {
443     const RISCVCPUMultiExtConfig *prop;
444 
445     g_assert(array);
446 
447     for (prop = array; prop && prop->name; prop++) {
448         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
449     }
450 }
451 
452 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
453 {
454     int i;
455 
456     riscv_add_satp_mode_properties(cpu_obj);
457 
458     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
459         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
460         int bit = misa_cfg->offset;
461 
462         misa_cfg->name = riscv_get_misa_ext_name(bit);
463         misa_cfg->description = riscv_get_misa_ext_description(bit);
464 
465         object_property_add(cpu_obj, misa_cfg->name, "bool",
466                             kvm_cpu_get_misa_ext_cfg,
467                             kvm_cpu_set_misa_ext_cfg,
468                             NULL, misa_cfg);
469         object_property_set_description(cpu_obj, misa_cfg->name,
470                                         misa_cfg->description);
471     }
472 
473     for (i = 0; misa_bits[i] != 0; i++) {
474         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
475         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
476     }
477 
478     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
479         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
480 
481         object_property_add(cpu_obj, multi_cfg->name, "bool",
482                             kvm_cpu_get_multi_ext_cfg,
483                             kvm_cpu_set_multi_ext_cfg,
484                             NULL, multi_cfg);
485     }
486 
487     object_property_add(cpu_obj, "cbom_blocksize", "uint16",
488                         NULL, kvm_cpu_set_cbomz_blksize,
489                         NULL, &kvm_cbom_blocksize);
490 
491     object_property_add(cpu_obj, "cboz_blocksize", "uint16",
492                         NULL, kvm_cpu_set_cbomz_blksize,
493                         NULL, &kvm_cboz_blocksize);
494 
495     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
496     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
497     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
498 
499    /* We don't have the needed KVM support for profiles */
500     for (i = 0; riscv_profiles[i] != NULL; i++) {
501         riscv_cpu_add_kvm_unavail_prop(cpu_obj, riscv_profiles[i]->name);
502     }
503 }
504 
505 static int kvm_riscv_get_regs_core(CPUState *cs)
506 {
507     int ret = 0;
508     int i;
509     target_ulong reg;
510     CPURISCVState *env = &RISCV_CPU(cs)->env;
511 
512     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
513     if (ret) {
514         return ret;
515     }
516     env->pc = reg;
517 
518     for (i = 1; i < 32; i++) {
519         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
520         ret = kvm_get_one_reg(cs, id, &reg);
521         if (ret) {
522             return ret;
523         }
524         env->gpr[i] = reg;
525     }
526 
527     return ret;
528 }
529 
530 static int kvm_riscv_put_regs_core(CPUState *cs)
531 {
532     int ret = 0;
533     int i;
534     target_ulong reg;
535     CPURISCVState *env = &RISCV_CPU(cs)->env;
536 
537     reg = env->pc;
538     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
539     if (ret) {
540         return ret;
541     }
542 
543     for (i = 1; i < 32; i++) {
544         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
545         reg = env->gpr[i];
546         ret = kvm_set_one_reg(cs, id, &reg);
547         if (ret) {
548             return ret;
549         }
550     }
551 
552     return ret;
553 }
554 
555 static int kvm_riscv_get_regs_csr(CPUState *cs)
556 {
557     CPURISCVState *env = &RISCV_CPU(cs)->env;
558 
559     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
560     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
561     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
562     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
563     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
564     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
565     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
566     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
567     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
568 
569     return 0;
570 }
571 
572 static int kvm_riscv_put_regs_csr(CPUState *cs)
573 {
574     CPURISCVState *env = &RISCV_CPU(cs)->env;
575 
576     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
577     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
578     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
579     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
580     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
581     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
582     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
583     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
584     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
585 
586     return 0;
587 }
588 
589 static int kvm_riscv_get_regs_fp(CPUState *cs)
590 {
591     int ret = 0;
592     int i;
593     CPURISCVState *env = &RISCV_CPU(cs)->env;
594 
595     if (riscv_has_ext(env, RVD)) {
596         uint64_t reg;
597         for (i = 0; i < 32; i++) {
598             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
599             if (ret) {
600                 return ret;
601             }
602             env->fpr[i] = reg;
603         }
604         return ret;
605     }
606 
607     if (riscv_has_ext(env, RVF)) {
608         uint32_t reg;
609         for (i = 0; i < 32; i++) {
610             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
611             if (ret) {
612                 return ret;
613             }
614             env->fpr[i] = reg;
615         }
616         return ret;
617     }
618 
619     return ret;
620 }
621 
622 static int kvm_riscv_put_regs_fp(CPUState *cs)
623 {
624     int ret = 0;
625     int i;
626     CPURISCVState *env = &RISCV_CPU(cs)->env;
627 
628     if (riscv_has_ext(env, RVD)) {
629         uint64_t reg;
630         for (i = 0; i < 32; i++) {
631             reg = env->fpr[i];
632             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
633             if (ret) {
634                 return ret;
635             }
636         }
637         return ret;
638     }
639 
640     if (riscv_has_ext(env, RVF)) {
641         uint32_t reg;
642         for (i = 0; i < 32; i++) {
643             reg = env->fpr[i];
644             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
645             if (ret) {
646                 return ret;
647             }
648         }
649         return ret;
650     }
651 
652     return ret;
653 }
654 
655 static void kvm_riscv_get_regs_timer(CPUState *cs)
656 {
657     CPURISCVState *env = &RISCV_CPU(cs)->env;
658 
659     if (env->kvm_timer_dirty) {
660         return;
661     }
662 
663     KVM_RISCV_GET_TIMER(cs, time, env->kvm_timer_time);
664     KVM_RISCV_GET_TIMER(cs, compare, env->kvm_timer_compare);
665     KVM_RISCV_GET_TIMER(cs, state, env->kvm_timer_state);
666     KVM_RISCV_GET_TIMER(cs, frequency, env->kvm_timer_frequency);
667 
668     env->kvm_timer_dirty = true;
669 }
670 
671 static void kvm_riscv_put_regs_timer(CPUState *cs)
672 {
673     uint64_t reg;
674     CPURISCVState *env = &RISCV_CPU(cs)->env;
675 
676     if (!env->kvm_timer_dirty) {
677         return;
678     }
679 
680     KVM_RISCV_SET_TIMER(cs, time, env->kvm_timer_time);
681     KVM_RISCV_SET_TIMER(cs, compare, env->kvm_timer_compare);
682 
683     /*
684      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
685      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
686      * doesn't matter that adaping in QEMU now.
687      * TODO If KVM changes, adapt here.
688      */
689     if (env->kvm_timer_state) {
690         KVM_RISCV_SET_TIMER(cs, state, env->kvm_timer_state);
691     }
692 
693     /*
694      * For now, migration will not work between Hosts with different timer
695      * frequency. Therefore, we should check whether they are the same here
696      * during the migration.
697      */
698     if (migration_is_running(migrate_get_current()->state)) {
699         KVM_RISCV_GET_TIMER(cs, frequency, reg);
700         if (reg != env->kvm_timer_frequency) {
701             error_report("Dst Hosts timer frequency != Src Hosts");
702         }
703     }
704 
705     env->kvm_timer_dirty = false;
706 }
707 
708 typedef struct KVMScratchCPU {
709     int kvmfd;
710     int vmfd;
711     int cpufd;
712 } KVMScratchCPU;
713 
714 /*
715  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
716  * from target/arm/kvm.c.
717  */
718 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
719 {
720     int kvmfd = -1, vmfd = -1, cpufd = -1;
721 
722     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
723     if (kvmfd < 0) {
724         goto err;
725     }
726     do {
727         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
728     } while (vmfd == -1 && errno == EINTR);
729     if (vmfd < 0) {
730         goto err;
731     }
732     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
733     if (cpufd < 0) {
734         goto err;
735     }
736 
737     scratch->kvmfd =  kvmfd;
738     scratch->vmfd = vmfd;
739     scratch->cpufd = cpufd;
740 
741     return true;
742 
743  err:
744     if (cpufd >= 0) {
745         close(cpufd);
746     }
747     if (vmfd >= 0) {
748         close(vmfd);
749     }
750     if (kvmfd >= 0) {
751         close(kvmfd);
752     }
753 
754     return false;
755 }
756 
757 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
758 {
759     close(scratch->cpufd);
760     close(scratch->vmfd);
761     close(scratch->kvmfd);
762 }
763 
764 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
765 {
766     CPURISCVState *env = &cpu->env;
767     struct kvm_one_reg reg;
768     int ret;
769 
770     reg.id = RISCV_CONFIG_REG(env, mvendorid);
771     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
772     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
773     if (ret != 0) {
774         error_report("Unable to retrieve mvendorid from host, error %d", ret);
775     }
776 
777     reg.id = RISCV_CONFIG_REG(env, marchid);
778     reg.addr = (uint64_t)&cpu->cfg.marchid;
779     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
780     if (ret != 0) {
781         error_report("Unable to retrieve marchid from host, error %d", ret);
782     }
783 
784     reg.id = RISCV_CONFIG_REG(env, mimpid);
785     reg.addr = (uint64_t)&cpu->cfg.mimpid;
786     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
787     if (ret != 0) {
788         error_report("Unable to retrieve mimpid from host, error %d", ret);
789     }
790 }
791 
792 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
793                                          KVMScratchCPU *kvmcpu)
794 {
795     CPURISCVState *env = &cpu->env;
796     struct kvm_one_reg reg;
797     int ret;
798 
799     reg.id = RISCV_CONFIG_REG(env, isa);
800     reg.addr = (uint64_t)&env->misa_ext_mask;
801     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
802 
803     if (ret) {
804         error_report("Unable to fetch ISA register from KVM, "
805                      "error %d", ret);
806         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
807         exit(EXIT_FAILURE);
808     }
809 
810     env->misa_ext = env->misa_ext_mask;
811 }
812 
813 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
814                                          KVMCPUConfig *cbomz_cfg)
815 {
816     CPURISCVState *env = &cpu->env;
817     struct kvm_one_reg reg;
818     int ret;
819 
820     reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
821                                     cbomz_cfg->kvm_reg_id);
822     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
823     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
824     if (ret != 0) {
825         error_report("Unable to read KVM reg %s, error %d",
826                      cbomz_cfg->name, ret);
827         exit(EXIT_FAILURE);
828     }
829 }
830 
831 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
832                                            KVMScratchCPU *kvmcpu)
833 {
834     CPURISCVState *env = &cpu->env;
835     uint64_t val;
836     int i, ret;
837 
838     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
839         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
840         struct kvm_one_reg reg;
841 
842         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
843                                         multi_ext_cfg->kvm_reg_id);
844         reg.addr = (uint64_t)&val;
845         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
846         if (ret != 0) {
847             if (errno == EINVAL) {
848                 /* Silently default to 'false' if KVM does not support it. */
849                 multi_ext_cfg->supported = false;
850                 val = false;
851             } else {
852                 error_report("Unable to read ISA_EXT KVM register %s: %s",
853                              multi_ext_cfg->name, strerror(errno));
854                 exit(EXIT_FAILURE);
855             }
856         } else {
857             multi_ext_cfg->supported = true;
858         }
859 
860         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
861     }
862 
863     if (cpu->cfg.ext_zicbom) {
864         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
865     }
866 
867     if (cpu->cfg.ext_zicboz) {
868         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
869     }
870 }
871 
872 static int uint64_cmp(const void *a, const void *b)
873 {
874     uint64_t val1 = *(const uint64_t *)a;
875     uint64_t val2 = *(const uint64_t *)b;
876 
877     if (val1 < val2) {
878         return -1;
879     }
880 
881     if (val1 > val2) {
882         return 1;
883     }
884 
885     return 0;
886 }
887 
888 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
889 {
890     KVMCPUConfig *multi_ext_cfg;
891     struct kvm_one_reg reg;
892     struct kvm_reg_list rl_struct;
893     struct kvm_reg_list *reglist;
894     uint64_t val, reg_id, *reg_search;
895     int i, ret;
896 
897     rl_struct.n = 0;
898     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
899 
900     /*
901      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
902      * (EINVAL). Use read_legacy() in this case.
903      */
904     if (errno == EINVAL) {
905         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
906     } else if (errno != E2BIG) {
907         /*
908          * E2BIG is an expected error message for the API since we
909          * don't know the number of registers. The right amount will
910          * be written in rl_struct.n.
911          *
912          * Error out if we get any other errno.
913          */
914         error_report("Error when accessing get-reg-list: %s",
915                      strerror(errno));
916         exit(EXIT_FAILURE);
917     }
918 
919     reglist = g_malloc(sizeof(struct kvm_reg_list) +
920                        rl_struct.n * sizeof(uint64_t));
921     reglist->n = rl_struct.n;
922     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
923     if (ret) {
924         error_report("Error when reading KVM_GET_REG_LIST: %s",
925                      strerror(errno));
926         exit(EXIT_FAILURE);
927     }
928 
929     /* sort reglist to use bsearch() */
930     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
931 
932     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
933         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
934         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
935                                         multi_ext_cfg->kvm_reg_id);
936         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
937                              sizeof(uint64_t), uint64_cmp);
938         if (!reg_search) {
939             continue;
940         }
941 
942         reg.id = reg_id;
943         reg.addr = (uint64_t)&val;
944         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
945         if (ret != 0) {
946             error_report("Unable to read ISA_EXT KVM register %s: %s",
947                          multi_ext_cfg->name, strerror(errno));
948             exit(EXIT_FAILURE);
949         }
950 
951         multi_ext_cfg->supported = true;
952         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
953     }
954 
955     if (cpu->cfg.ext_zicbom) {
956         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
957     }
958 
959     if (cpu->cfg.ext_zicboz) {
960         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
961     }
962 }
963 
964 static void riscv_init_kvm_registers(Object *cpu_obj)
965 {
966     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
967     KVMScratchCPU kvmcpu;
968 
969     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
970         return;
971     }
972 
973     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
974     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
975     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
976 
977     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
978 }
979 
980 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
981     KVM_CAP_LAST_INFO
982 };
983 
984 int kvm_arch_get_registers(CPUState *cs)
985 {
986     int ret = 0;
987 
988     ret = kvm_riscv_get_regs_core(cs);
989     if (ret) {
990         return ret;
991     }
992 
993     ret = kvm_riscv_get_regs_csr(cs);
994     if (ret) {
995         return ret;
996     }
997 
998     ret = kvm_riscv_get_regs_fp(cs);
999     if (ret) {
1000         return ret;
1001     }
1002 
1003     return ret;
1004 }
1005 
1006 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1007 {
1008     if (cap_has_mp_state) {
1009         struct kvm_mp_state mp_state = {
1010             .mp_state = state
1011         };
1012 
1013         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1014         if (ret) {
1015             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1016                     __func__, ret, strerror(-ret));
1017             return -1;
1018         }
1019     }
1020 
1021     return 0;
1022 }
1023 
1024 int kvm_arch_put_registers(CPUState *cs, int level)
1025 {
1026     int ret = 0;
1027 
1028     ret = kvm_riscv_put_regs_core(cs);
1029     if (ret) {
1030         return ret;
1031     }
1032 
1033     ret = kvm_riscv_put_regs_csr(cs);
1034     if (ret) {
1035         return ret;
1036     }
1037 
1038     ret = kvm_riscv_put_regs_fp(cs);
1039     if (ret) {
1040         return ret;
1041     }
1042 
1043     if (KVM_PUT_RESET_STATE == level) {
1044         RISCVCPU *cpu = RISCV_CPU(cs);
1045         if (cs->cpu_index == 0) {
1046             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1047         } else {
1048             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1049         }
1050         if (ret) {
1051             return ret;
1052         }
1053     }
1054 
1055     return ret;
1056 }
1057 
1058 int kvm_arch_release_virq_post(int virq)
1059 {
1060     return 0;
1061 }
1062 
1063 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1064                              uint64_t address, uint32_t data, PCIDevice *dev)
1065 {
1066     return 0;
1067 }
1068 
1069 int kvm_arch_destroy_vcpu(CPUState *cs)
1070 {
1071     return 0;
1072 }
1073 
1074 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1075 {
1076     return cpu->cpu_index;
1077 }
1078 
1079 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1080                                       RunState state)
1081 {
1082     CPUState *cs = opaque;
1083 
1084     if (running) {
1085         kvm_riscv_put_regs_timer(cs);
1086     } else {
1087         kvm_riscv_get_regs_timer(cs);
1088     }
1089 }
1090 
1091 void kvm_arch_init_irq_routing(KVMState *s)
1092 {
1093 }
1094 
1095 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1096 {
1097     CPURISCVState *env = &cpu->env;
1098     target_ulong reg;
1099     uint64_t id;
1100     int ret;
1101 
1102     id = RISCV_CONFIG_REG(env, mvendorid);
1103     /*
1104      * cfg.mvendorid is an uint32 but a target_ulong will
1105      * be written. Assign it to a target_ulong var to avoid
1106      * writing pieces of other cpu->cfg fields in the reg.
1107      */
1108     reg = cpu->cfg.mvendorid;
1109     ret = kvm_set_one_reg(cs, id, &reg);
1110     if (ret != 0) {
1111         return ret;
1112     }
1113 
1114     id = RISCV_CONFIG_REG(env, marchid);
1115     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1116     if (ret != 0) {
1117         return ret;
1118     }
1119 
1120     id = RISCV_CONFIG_REG(env, mimpid);
1121     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1122 
1123     return ret;
1124 }
1125 
1126 int kvm_arch_init_vcpu(CPUState *cs)
1127 {
1128     int ret = 0;
1129     RISCVCPU *cpu = RISCV_CPU(cs);
1130 
1131     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1132 
1133     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1134         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1135         if (ret != 0) {
1136             return ret;
1137         }
1138     }
1139 
1140     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1141     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1142 
1143     return ret;
1144 }
1145 
1146 int kvm_arch_msi_data_to_gsi(uint32_t data)
1147 {
1148     abort();
1149 }
1150 
1151 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1152                                 int vector, PCIDevice *dev)
1153 {
1154     return 0;
1155 }
1156 
1157 int kvm_arch_get_default_type(MachineState *ms)
1158 {
1159     return 0;
1160 }
1161 
1162 int kvm_arch_init(MachineState *ms, KVMState *s)
1163 {
1164     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1165     return 0;
1166 }
1167 
1168 int kvm_arch_irqchip_create(KVMState *s)
1169 {
1170     if (kvm_kernel_irqchip_split()) {
1171         error_report("-machine kernel_irqchip=split is not supported on RISC-V.");
1172         exit(1);
1173     }
1174 
1175     /*
1176      * We can create the VAIA using the newer device control API.
1177      */
1178     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1179 }
1180 
1181 int kvm_arch_process_async_events(CPUState *cs)
1182 {
1183     return 0;
1184 }
1185 
1186 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1187 {
1188 }
1189 
1190 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1191 {
1192     return MEMTXATTRS_UNSPECIFIED;
1193 }
1194 
1195 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1196 {
1197     return true;
1198 }
1199 
1200 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1201 {
1202     int ret = 0;
1203     unsigned char ch;
1204     switch (run->riscv_sbi.extension_id) {
1205     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1206         ch = run->riscv_sbi.args[0];
1207         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1208         break;
1209     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1210         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1211         if (ret == sizeof(ch)) {
1212             run->riscv_sbi.ret[0] = ch;
1213         } else {
1214             run->riscv_sbi.ret[0] = -1;
1215         }
1216         ret = 0;
1217         break;
1218     default:
1219         qemu_log_mask(LOG_UNIMP,
1220                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1221                       __func__, run->riscv_sbi.extension_id);
1222         ret = -1;
1223         break;
1224     }
1225     return ret;
1226 }
1227 
1228 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1229 {
1230     int ret = 0;
1231     switch (run->exit_reason) {
1232     case KVM_EXIT_RISCV_SBI:
1233         ret = kvm_riscv_handle_sbi(cs, run);
1234         break;
1235     default:
1236         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1237                       __func__, run->exit_reason);
1238         ret = -1;
1239         break;
1240     }
1241     return ret;
1242 }
1243 
1244 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1245 {
1246     CPURISCVState *env = &cpu->env;
1247     int i;
1248 
1249     if (!kvm_enabled()) {
1250         return;
1251     }
1252     for (i = 0; i < 32; i++) {
1253         env->gpr[i] = 0;
1254     }
1255     env->pc = cpu->env.kernel_addr;
1256     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1257     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1258     env->satp = 0;
1259     env->mie = 0;
1260     env->stvec = 0;
1261     env->sscratch = 0;
1262     env->sepc = 0;
1263     env->scause = 0;
1264     env->stval = 0;
1265     env->mip = 0;
1266 }
1267 
1268 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1269 {
1270     int ret;
1271     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1272 
1273     if (irq != IRQ_S_EXT) {
1274         perror("kvm riscv set irq != IRQ_S_EXT\n");
1275         abort();
1276     }
1277 
1278     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1279     if (ret < 0) {
1280         perror("Set irq failed");
1281         abort();
1282     }
1283 }
1284 
1285 bool kvm_arch_cpu_check_are_resettable(void)
1286 {
1287     return true;
1288 }
1289 
1290 static int aia_mode;
1291 
1292 static const char *kvm_aia_mode_str(uint64_t mode)
1293 {
1294     switch (mode) {
1295     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1296         return "emul";
1297     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1298         return "hwaccel";
1299     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1300     default:
1301         return "auto";
1302     };
1303 }
1304 
1305 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1306 {
1307     return g_strdup(kvm_aia_mode_str(aia_mode));
1308 }
1309 
1310 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1311 {
1312     if (!strcmp(val, "emul")) {
1313         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1314     } else if (!strcmp(val, "hwaccel")) {
1315         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1316     } else if (!strcmp(val, "auto")) {
1317         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1318     } else {
1319         error_setg(errp, "Invalid KVM AIA mode");
1320         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1321     }
1322 }
1323 
1324 void kvm_arch_accel_class_init(ObjectClass *oc)
1325 {
1326     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1327                                   riscv_set_kvm_aia);
1328     object_class_property_set_description(oc, "riscv-aia",
1329                                           "Set KVM AIA mode. Valid values are "
1330                                           "emul, hwaccel, and auto. Default "
1331                                           "is auto.");
1332     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1333                                     "auto");
1334 }
1335 
1336 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1337                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1338                           uint64_t aplic_base, uint64_t imsic_base,
1339                           uint64_t guest_num)
1340 {
1341     int ret, i;
1342     int aia_fd = -1;
1343     uint64_t default_aia_mode;
1344     uint64_t socket_count = riscv_socket_count(machine);
1345     uint64_t max_hart_per_socket = 0;
1346     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1347     uint64_t socket_bits, hart_bits, guest_bits;
1348 
1349     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1350 
1351     if (aia_fd < 0) {
1352         error_report("Unable to create in-kernel irqchip");
1353         exit(1);
1354     }
1355 
1356     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1357                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1358                             &default_aia_mode, false, NULL);
1359     if (ret < 0) {
1360         error_report("KVM AIA: failed to get current KVM AIA mode");
1361         exit(1);
1362     }
1363     qemu_log("KVM AIA: default mode is %s\n",
1364              kvm_aia_mode_str(default_aia_mode));
1365 
1366     if (default_aia_mode != aia_mode) {
1367         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1368                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1369                                 &aia_mode, true, NULL);
1370         if (ret < 0)
1371             warn_report("KVM AIA: failed to set KVM AIA mode");
1372         else
1373             qemu_log("KVM AIA: set current mode to %s\n",
1374                      kvm_aia_mode_str(aia_mode));
1375     }
1376 
1377     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1378                             KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1379                             &aia_irq_num, true, NULL);
1380     if (ret < 0) {
1381         error_report("KVM AIA: failed to set number of input irq lines");
1382         exit(1);
1383     }
1384 
1385     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1386                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1387                             &aia_msi_num, true, NULL);
1388     if (ret < 0) {
1389         error_report("KVM AIA: failed to set number of msi");
1390         exit(1);
1391     }
1392 
1393 
1394     if (socket_count > 1) {
1395         socket_bits = find_last_bit(&socket_count, BITS_PER_LONG) + 1;
1396         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1397                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1398                                 &socket_bits, true, NULL);
1399         if (ret < 0) {
1400             error_report("KVM AIA: failed to set group_bits");
1401             exit(1);
1402         }
1403 
1404         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1405                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1406                                 &group_shift, true, NULL);
1407         if (ret < 0) {
1408             error_report("KVM AIA: failed to set group_shift");
1409             exit(1);
1410         }
1411     }
1412 
1413     guest_bits = guest_num == 0 ? 0 :
1414                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1415     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1416                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1417                             &guest_bits, true, NULL);
1418     if (ret < 0) {
1419         error_report("KVM AIA: failed to set guest_bits");
1420         exit(1);
1421     }
1422 
1423     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1424                             KVM_DEV_RISCV_AIA_ADDR_APLIC,
1425                             &aplic_base, true, NULL);
1426     if (ret < 0) {
1427         error_report("KVM AIA: failed to set the base address of APLIC");
1428         exit(1);
1429     }
1430 
1431     for (socket = 0; socket < socket_count; socket++) {
1432         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1433         hart_count = riscv_socket_hart_count(machine, socket);
1434         base_hart = riscv_socket_first_hartid(machine, socket);
1435 
1436         if (max_hart_per_socket < hart_count) {
1437             max_hart_per_socket = hart_count;
1438         }
1439 
1440         for (i = 0; i < hart_count; i++) {
1441             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1442             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1443                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1444                                     &imsic_addr, true, NULL);
1445             if (ret < 0) {
1446                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1447                 exit(1);
1448             }
1449         }
1450     }
1451 
1452     hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1453     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1454                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1455                             &hart_bits, true, NULL);
1456     if (ret < 0) {
1457         error_report("KVM AIA: failed to set hart_bits");
1458         exit(1);
1459     }
1460 
1461     if (kvm_has_gsi_routing()) {
1462         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1463             /* KVM AIA only has one APLIC instance */
1464             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1465         }
1466         kvm_gsi_routing_allowed = true;
1467         kvm_irqchip_commit_routes(kvm_state);
1468     }
1469 
1470     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1471                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1472                             NULL, true, NULL);
1473     if (ret < 0) {
1474         error_report("KVM AIA: initialized fail");
1475         exit(1);
1476     }
1477 
1478     kvm_msi_via_irqfd_allowed = true;
1479 }
1480 
1481 static void kvm_cpu_instance_init(CPUState *cs)
1482 {
1483     Object *obj = OBJECT(RISCV_CPU(cs));
1484     DeviceState *dev = DEVICE(obj);
1485 
1486     riscv_init_kvm_registers(obj);
1487 
1488     kvm_riscv_add_cpu_user_properties(obj);
1489 
1490     for (Property *prop = riscv_cpu_options; prop && prop->name; prop++) {
1491         /* Check if we have a specific KVM handler for the option */
1492         if (object_property_find(obj, prop->name)) {
1493             continue;
1494         }
1495         qdev_property_add_static(dev, prop);
1496     }
1497 }
1498 
1499 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1500 {
1501     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1502 
1503     acc->cpu_instance_init = kvm_cpu_instance_init;
1504 }
1505 
1506 static const TypeInfo kvm_cpu_accel_type_info = {
1507     .name = ACCEL_CPU_NAME("kvm"),
1508 
1509     .parent = TYPE_ACCEL_CPU,
1510     .class_init = kvm_cpu_accel_class_init,
1511     .abstract = true,
1512 };
1513 static void kvm_cpu_accel_register_types(void)
1514 {
1515     type_register_static(&kvm_cpu_accel_type_info);
1516 }
1517 type_init(kvm_cpu_accel_register_types);
1518 
1519 static void riscv_host_cpu_init(Object *obj)
1520 {
1521     CPURISCVState *env = &RISCV_CPU(obj)->env;
1522 
1523 #if defined(TARGET_RISCV32)
1524     env->misa_mxl_max = env->misa_mxl = MXL_RV32;
1525 #elif defined(TARGET_RISCV64)
1526     env->misa_mxl_max = env->misa_mxl = MXL_RV64;
1527 #endif
1528 }
1529 
1530 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
1531     {
1532         .name = TYPE_RISCV_CPU_HOST,
1533         .parent = TYPE_RISCV_CPU,
1534         .instance_init = riscv_host_cpu_init,
1535     }
1536 };
1537 
1538 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
1539