xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision 513823e7521a09ed7ad1e32e6454bac3b2cbf52d)
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 #include <sys/prctl.h>
22 
23 #include <linux/kvm.h>
24 
25 #include "qemu/timer.h"
26 #include "qapi/error.h"
27 #include "qemu/error-report.h"
28 #include "qemu/main-loop.h"
29 #include "qapi/visitor.h"
30 #include "system/system.h"
31 #include "system/kvm.h"
32 #include "system/kvm_int.h"
33 #include "cpu.h"
34 #include "trace.h"
35 #include "hw/core/accel-cpu.h"
36 #include "hw/pci/pci.h"
37 #include "exec/memattrs.h"
38 #include "exec/address-spaces.h"
39 #include "hw/boards.h"
40 #include "hw/irq.h"
41 #include "hw/intc/riscv_imsic.h"
42 #include "qemu/log.h"
43 #include "hw/loader.h"
44 #include "kvm_riscv.h"
45 #include "sbi_ecall_interface.h"
46 #include "chardev/char-fe.h"
47 #include "migration/misc.h"
48 #include "system/runstate.h"
49 #include "hw/riscv/numa.h"
50 
51 #define PR_RISCV_V_SET_CONTROL            69
52 #define PR_RISCV_V_VSTATE_CTRL_ON          2
53 
54 void riscv_kvm_aplic_request(void *opaque, int irq, int level)
55 {
56     kvm_set_irq(kvm_state, irq, !!level);
57 }
58 
59 static bool cap_has_mp_state;
60 
61 static uint64_t kvm_riscv_reg_id_ulong(CPURISCVState *env, uint64_t type,
62                                  uint64_t idx)
63 {
64     uint64_t id = KVM_REG_RISCV | type | idx;
65 
66     switch (riscv_cpu_mxl(env)) {
67     case MXL_RV32:
68         id |= KVM_REG_SIZE_U32;
69         break;
70     case MXL_RV64:
71         id |= KVM_REG_SIZE_U64;
72         break;
73     default:
74         g_assert_not_reached();
75     }
76     return id;
77 }
78 
79 static uint64_t kvm_riscv_reg_id_u32(uint64_t type, uint64_t idx)
80 {
81     return KVM_REG_RISCV | KVM_REG_SIZE_U32 | type | idx;
82 }
83 
84 static uint64_t kvm_riscv_reg_id_u64(uint64_t type, uint64_t idx)
85 {
86     return KVM_REG_RISCV | KVM_REG_SIZE_U64 | type | idx;
87 }
88 
89 static uint64_t kvm_encode_reg_size_id(uint64_t id, size_t size_b)
90 {
91     uint64_t size_ctz = __builtin_ctz(size_b);
92 
93     return id | (size_ctz << KVM_REG_SIZE_SHIFT);
94 }
95 
96 static uint64_t kvm_riscv_vector_reg_id(RISCVCPU *cpu,
97                                         uint64_t idx)
98 {
99     uint64_t id;
100     size_t size_b;
101 
102     g_assert(idx < 32);
103 
104     id = KVM_REG_RISCV | KVM_REG_RISCV_VECTOR | KVM_REG_RISCV_VECTOR_REG(idx);
105     size_b = cpu->cfg.vlenb;
106 
107     return kvm_encode_reg_size_id(id, size_b);
108 }
109 
110 #define RISCV_CORE_REG(env, name) \
111     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, \
112                            KVM_REG_RISCV_CORE_REG(name))
113 
114 #define RISCV_CSR_REG(env, name) \
115     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CSR, \
116                            KVM_REG_RISCV_CSR_REG(name))
117 
118 #define RISCV_CONFIG_REG(env, name) \
119     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG, \
120                            KVM_REG_RISCV_CONFIG_REG(name))
121 
122 #define RISCV_TIMER_REG(name)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_TIMER, \
123                  KVM_REG_RISCV_TIMER_REG(name))
124 
125 #define RISCV_FP_F_REG(idx)  kvm_riscv_reg_id_u32(KVM_REG_RISCV_FP_F, idx)
126 
127 #define RISCV_FP_D_REG(idx)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_FP_D, idx)
128 
129 #define RISCV_VECTOR_CSR_REG(env, name) \
130     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_VECTOR, \
131                            KVM_REG_RISCV_VECTOR_CSR_REG(name))
132 
133 #define KVM_RISCV_GET_CSR(cs, env, csr, reg) \
134     do { \
135         int _ret = kvm_get_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
136         if (_ret) { \
137             return _ret; \
138         } \
139     } while (0)
140 
141 #define KVM_RISCV_SET_CSR(cs, env, csr, reg) \
142     do { \
143         int _ret = kvm_set_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
144         if (_ret) { \
145             return _ret; \
146         } \
147     } while (0)
148 
149 #define KVM_RISCV_GET_TIMER(cs, name, reg) \
150     do { \
151         int ret = kvm_get_one_reg(cs, RISCV_TIMER_REG(name), &reg); \
152         if (ret) { \
153             abort(); \
154         } \
155     } while (0)
156 
157 #define KVM_RISCV_SET_TIMER(cs, name, reg) \
158     do { \
159         int ret = kvm_set_one_reg(cs, RISCV_TIMER_REG(name), &reg); \
160         if (ret) { \
161             abort(); \
162         } \
163     } while (0)
164 
165 typedef struct KVMCPUConfig {
166     const char *name;
167     const char *description;
168     target_ulong offset;
169     uint64_t kvm_reg_id;
170     bool user_set;
171     bool supported;
172 } KVMCPUConfig;
173 
174 #define KVM_MISA_CFG(_bit, _reg_id) \
175     {.offset = _bit, .kvm_reg_id = _reg_id}
176 
177 /* KVM ISA extensions */
178 static KVMCPUConfig kvm_misa_ext_cfgs[] = {
179     KVM_MISA_CFG(RVA, KVM_RISCV_ISA_EXT_A),
180     KVM_MISA_CFG(RVC, KVM_RISCV_ISA_EXT_C),
181     KVM_MISA_CFG(RVD, KVM_RISCV_ISA_EXT_D),
182     KVM_MISA_CFG(RVF, KVM_RISCV_ISA_EXT_F),
183     KVM_MISA_CFG(RVH, KVM_RISCV_ISA_EXT_H),
184     KVM_MISA_CFG(RVI, KVM_RISCV_ISA_EXT_I),
185     KVM_MISA_CFG(RVM, KVM_RISCV_ISA_EXT_M),
186     KVM_MISA_CFG(RVV, KVM_RISCV_ISA_EXT_V),
187 };
188 
189 static void kvm_cpu_get_misa_ext_cfg(Object *obj, Visitor *v,
190                                      const char *name,
191                                      void *opaque, Error **errp)
192 {
193     KVMCPUConfig *misa_ext_cfg = opaque;
194     target_ulong misa_bit = misa_ext_cfg->offset;
195     RISCVCPU *cpu = RISCV_CPU(obj);
196     CPURISCVState *env = &cpu->env;
197     bool value = env->misa_ext_mask & misa_bit;
198 
199     visit_type_bool(v, name, &value, errp);
200 }
201 
202 static void kvm_cpu_set_misa_ext_cfg(Object *obj, Visitor *v,
203                                      const char *name,
204                                      void *opaque, Error **errp)
205 {
206     KVMCPUConfig *misa_ext_cfg = opaque;
207     target_ulong misa_bit = misa_ext_cfg->offset;
208     RISCVCPU *cpu = RISCV_CPU(obj);
209     CPURISCVState *env = &cpu->env;
210     bool value, host_bit;
211 
212     if (!visit_type_bool(v, name, &value, errp)) {
213         return;
214     }
215 
216     host_bit = env->misa_ext_mask & misa_bit;
217 
218     if (value == host_bit) {
219         return;
220     }
221 
222     if (!value) {
223         misa_ext_cfg->user_set = true;
224         return;
225     }
226 
227     /*
228      * Forbid users to enable extensions that aren't
229      * available in the hart.
230      */
231     error_setg(errp, "Enabling MISA bit '%s' is not allowed: it's not "
232                "enabled in the host", misa_ext_cfg->name);
233 }
234 
235 static void kvm_riscv_update_cpu_misa_ext(RISCVCPU *cpu, CPUState *cs)
236 {
237     CPURISCVState *env = &cpu->env;
238     uint64_t id, reg;
239     int i, ret;
240 
241     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
242         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
243         target_ulong misa_bit = misa_cfg->offset;
244 
245         if (!misa_cfg->user_set) {
246             continue;
247         }
248 
249         /* If we're here we're going to disable the MISA bit */
250         reg = 0;
251         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
252                                     misa_cfg->kvm_reg_id);
253         ret = kvm_set_one_reg(cs, id, &reg);
254         if (ret != 0) {
255             /*
256              * We're not checking for -EINVAL because if the bit is about
257              * to be disabled, it means that it was already enabled by
258              * KVM. We determined that by fetching the 'isa' register
259              * during init() time. Any error at this point is worth
260              * aborting.
261              */
262             error_report("Unable to set KVM reg %s, error %d",
263                          misa_cfg->name, ret);
264             exit(EXIT_FAILURE);
265         }
266         env->misa_ext &= ~misa_bit;
267     }
268 }
269 
270 #define KVM_EXT_CFG(_name, _prop, _reg_id) \
271     {.name = _name, .offset = CPU_CFG_OFFSET(_prop), \
272      .kvm_reg_id = _reg_id}
273 
274 static KVMCPUConfig kvm_multi_ext_cfgs[] = {
275     KVM_EXT_CFG("zicbom", ext_zicbom, KVM_RISCV_ISA_EXT_ZICBOM),
276     KVM_EXT_CFG("zicboz", ext_zicboz, KVM_RISCV_ISA_EXT_ZICBOZ),
277     KVM_EXT_CFG("zicntr", ext_zicntr, KVM_RISCV_ISA_EXT_ZICNTR),
278     KVM_EXT_CFG("zicond", ext_zicond, KVM_RISCV_ISA_EXT_ZICOND),
279     KVM_EXT_CFG("zicsr", ext_zicsr, KVM_RISCV_ISA_EXT_ZICSR),
280     KVM_EXT_CFG("zifencei", ext_zifencei, KVM_RISCV_ISA_EXT_ZIFENCEI),
281     KVM_EXT_CFG("zihintntl", ext_zihintntl, KVM_RISCV_ISA_EXT_ZIHINTNTL),
282     KVM_EXT_CFG("zihintpause", ext_zihintpause, KVM_RISCV_ISA_EXT_ZIHINTPAUSE),
283     KVM_EXT_CFG("zihpm", ext_zihpm, KVM_RISCV_ISA_EXT_ZIHPM),
284     KVM_EXT_CFG("zimop", ext_zimop, KVM_RISCV_ISA_EXT_ZIMOP),
285     KVM_EXT_CFG("zcmop", ext_zcmop, KVM_RISCV_ISA_EXT_ZCMOP),
286     KVM_EXT_CFG("zacas", ext_zacas, KVM_RISCV_ISA_EXT_ZACAS),
287     KVM_EXT_CFG("zawrs", ext_zawrs, KVM_RISCV_ISA_EXT_ZAWRS),
288     KVM_EXT_CFG("zfa", ext_zfa, KVM_RISCV_ISA_EXT_ZFA),
289     KVM_EXT_CFG("zfh", ext_zfh, KVM_RISCV_ISA_EXT_ZFH),
290     KVM_EXT_CFG("zfhmin", ext_zfhmin, KVM_RISCV_ISA_EXT_ZFHMIN),
291     KVM_EXT_CFG("zba", ext_zba, KVM_RISCV_ISA_EXT_ZBA),
292     KVM_EXT_CFG("zbb", ext_zbb, KVM_RISCV_ISA_EXT_ZBB),
293     KVM_EXT_CFG("zbc", ext_zbc, KVM_RISCV_ISA_EXT_ZBC),
294     KVM_EXT_CFG("zbkb", ext_zbkb, KVM_RISCV_ISA_EXT_ZBKB),
295     KVM_EXT_CFG("zbkc", ext_zbkc, KVM_RISCV_ISA_EXT_ZBKC),
296     KVM_EXT_CFG("zbkx", ext_zbkx, KVM_RISCV_ISA_EXT_ZBKX),
297     KVM_EXT_CFG("zbs", ext_zbs, KVM_RISCV_ISA_EXT_ZBS),
298     KVM_EXT_CFG("zca", ext_zca, KVM_RISCV_ISA_EXT_ZCA),
299     KVM_EXT_CFG("zcb", ext_zcb, KVM_RISCV_ISA_EXT_ZCB),
300     KVM_EXT_CFG("zcd", ext_zcd, KVM_RISCV_ISA_EXT_ZCD),
301     KVM_EXT_CFG("zcf", ext_zcf, KVM_RISCV_ISA_EXT_ZCF),
302     KVM_EXT_CFG("zknd", ext_zknd, KVM_RISCV_ISA_EXT_ZKND),
303     KVM_EXT_CFG("zkne", ext_zkne, KVM_RISCV_ISA_EXT_ZKNE),
304     KVM_EXT_CFG("zknh", ext_zknh, KVM_RISCV_ISA_EXT_ZKNH),
305     KVM_EXT_CFG("zkr", ext_zkr, KVM_RISCV_ISA_EXT_ZKR),
306     KVM_EXT_CFG("zksed", ext_zksed, KVM_RISCV_ISA_EXT_ZKSED),
307     KVM_EXT_CFG("zksh", ext_zksh, KVM_RISCV_ISA_EXT_ZKSH),
308     KVM_EXT_CFG("zkt", ext_zkt, KVM_RISCV_ISA_EXT_ZKT),
309     KVM_EXT_CFG("ztso", ext_ztso, KVM_RISCV_ISA_EXT_ZTSO),
310     KVM_EXT_CFG("zvbb", ext_zvbb, KVM_RISCV_ISA_EXT_ZVBB),
311     KVM_EXT_CFG("zvbc", ext_zvbc, KVM_RISCV_ISA_EXT_ZVBC),
312     KVM_EXT_CFG("zvfh", ext_zvfh, KVM_RISCV_ISA_EXT_ZVFH),
313     KVM_EXT_CFG("zvfhmin", ext_zvfhmin, KVM_RISCV_ISA_EXT_ZVFHMIN),
314     KVM_EXT_CFG("zvkb", ext_zvkb, KVM_RISCV_ISA_EXT_ZVKB),
315     KVM_EXT_CFG("zvkg", ext_zvkg, KVM_RISCV_ISA_EXT_ZVKG),
316     KVM_EXT_CFG("zvkned", ext_zvkned, KVM_RISCV_ISA_EXT_ZVKNED),
317     KVM_EXT_CFG("zvknha", ext_zvknha, KVM_RISCV_ISA_EXT_ZVKNHA),
318     KVM_EXT_CFG("zvknhb", ext_zvknhb, KVM_RISCV_ISA_EXT_ZVKNHB),
319     KVM_EXT_CFG("zvksed", ext_zvksed, KVM_RISCV_ISA_EXT_ZVKSED),
320     KVM_EXT_CFG("zvksh", ext_zvksh, KVM_RISCV_ISA_EXT_ZVKSH),
321     KVM_EXT_CFG("zvkt", ext_zvkt, KVM_RISCV_ISA_EXT_ZVKT),
322     KVM_EXT_CFG("smstateen", ext_smstateen, KVM_RISCV_ISA_EXT_SMSTATEEN),
323     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
324     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
325     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
326     KVM_EXT_CFG("svnapot", ext_svnapot, KVM_RISCV_ISA_EXT_SVNAPOT),
327     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
328 };
329 
330 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
331 {
332     return (void *)&cpu->cfg + kvmcfg->offset;
333 }
334 
335 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
336                             uint32_t val)
337 {
338     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
339 
340     *ext_enabled = val;
341 }
342 
343 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
344                                 KVMCPUConfig *multi_ext)
345 {
346     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
347 
348     return *ext_enabled;
349 }
350 
351 static void kvm_cpu_get_multi_ext_cfg(Object *obj, Visitor *v,
352                                       const char *name,
353                                       void *opaque, Error **errp)
354 {
355     KVMCPUConfig *multi_ext_cfg = opaque;
356     RISCVCPU *cpu = RISCV_CPU(obj);
357     bool value = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
358 
359     visit_type_bool(v, name, &value, errp);
360 }
361 
362 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
363                                       const char *name,
364                                       void *opaque, Error **errp)
365 {
366     KVMCPUConfig *multi_ext_cfg = opaque;
367     RISCVCPU *cpu = RISCV_CPU(obj);
368     bool value, host_val;
369 
370     if (!visit_type_bool(v, name, &value, errp)) {
371         return;
372     }
373 
374     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
375 
376     /*
377      * Ignore if the user is setting the same value
378      * as the host.
379      */
380     if (value == host_val) {
381         return;
382     }
383 
384     if (!multi_ext_cfg->supported) {
385         /*
386          * Error out if the user is trying to enable an
387          * extension that KVM doesn't support. Ignore
388          * option otherwise.
389          */
390         if (value) {
391             error_setg(errp, "KVM does not support disabling extension %s",
392                        multi_ext_cfg->name);
393         }
394 
395         return;
396     }
397 
398     multi_ext_cfg->user_set = true;
399     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
400 }
401 
402 static KVMCPUConfig kvm_cbom_blocksize = {
403     .name = "cbom_blocksize",
404     .offset = CPU_CFG_OFFSET(cbom_blocksize),
405     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
406 };
407 
408 static KVMCPUConfig kvm_cboz_blocksize = {
409     .name = "cboz_blocksize",
410     .offset = CPU_CFG_OFFSET(cboz_blocksize),
411     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
412 };
413 
414 static KVMCPUConfig kvm_v_vlenb = {
415     .name = "vlenb",
416     .offset = CPU_CFG_OFFSET(vlenb),
417     .kvm_reg_id =  KVM_REG_RISCV | KVM_REG_SIZE_U64 | KVM_REG_RISCV_VECTOR |
418                    KVM_REG_RISCV_VECTOR_CSR_REG(vlenb)
419 };
420 
421 static KVMCPUConfig kvm_sbi_dbcn = {
422     .name = "sbi_dbcn",
423     .kvm_reg_id = KVM_REG_RISCV | KVM_REG_SIZE_U64 |
424                   KVM_REG_RISCV_SBI_EXT | KVM_RISCV_SBI_EXT_DBCN
425 };
426 
427 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
428 {
429     CPURISCVState *env = &cpu->env;
430     uint64_t id, reg;
431     int i, ret;
432 
433     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
434         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
435 
436         if (!multi_ext_cfg->user_set) {
437             continue;
438         }
439 
440         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
441                                     multi_ext_cfg->kvm_reg_id);
442         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
443         ret = kvm_set_one_reg(cs, id, &reg);
444         if (ret != 0) {
445             if (!reg && ret == -EINVAL) {
446                 warn_report("KVM cannot disable extension %s",
447                             multi_ext_cfg->name);
448             } else {
449                 error_report("Unable to enable extension %s in KVM, error %d",
450                              multi_ext_cfg->name, ret);
451                 exit(EXIT_FAILURE);
452             }
453         }
454     }
455 }
456 
457 static void cpu_get_cfg_unavailable(Object *obj, Visitor *v,
458                                     const char *name,
459                                     void *opaque, Error **errp)
460 {
461     bool value = false;
462 
463     visit_type_bool(v, name, &value, errp);
464 }
465 
466 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
467                                     const char *name,
468                                     void *opaque, Error **errp)
469 {
470     const char *propname = opaque;
471     bool value;
472 
473     if (!visit_type_bool(v, name, &value, errp)) {
474         return;
475     }
476 
477     if (value) {
478         error_setg(errp, "'%s' is not available with KVM",
479                    propname);
480     }
481 }
482 
483 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
484 {
485     /* Check if KVM created the property already */
486     if (object_property_find(obj, prop_name)) {
487         return;
488     }
489 
490     /*
491      * Set the default to disabled for every extension
492      * unknown to KVM and error out if the user attempts
493      * to enable any of them.
494      */
495     object_property_add(obj, prop_name, "bool",
496                         cpu_get_cfg_unavailable,
497                         cpu_set_cfg_unavailable,
498                         NULL, (void *)prop_name);
499 }
500 
501 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
502                                         const RISCVCPUMultiExtConfig *array)
503 {
504     const RISCVCPUMultiExtConfig *prop;
505 
506     g_assert(array);
507 
508     for (prop = array; prop && prop->name; prop++) {
509         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
510     }
511 }
512 
513 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
514 {
515     int i;
516 
517     riscv_add_satp_mode_properties(cpu_obj);
518 
519     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
520         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
521         int bit = misa_cfg->offset;
522 
523         misa_cfg->name = riscv_get_misa_ext_name(bit);
524         misa_cfg->description = riscv_get_misa_ext_description(bit);
525 
526         object_property_add(cpu_obj, misa_cfg->name, "bool",
527                             kvm_cpu_get_misa_ext_cfg,
528                             kvm_cpu_set_misa_ext_cfg,
529                             NULL, misa_cfg);
530         object_property_set_description(cpu_obj, misa_cfg->name,
531                                         misa_cfg->description);
532     }
533 
534     for (i = 0; misa_bits[i] != 0; i++) {
535         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
536         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
537     }
538 
539     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
540         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
541 
542         object_property_add(cpu_obj, multi_cfg->name, "bool",
543                             kvm_cpu_get_multi_ext_cfg,
544                             kvm_cpu_set_multi_ext_cfg,
545                             NULL, multi_cfg);
546     }
547 
548     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
549     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
550     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
551 
552    /* We don't have the needed KVM support for profiles */
553     for (i = 0; riscv_profiles[i] != NULL; i++) {
554         riscv_cpu_add_kvm_unavail_prop(cpu_obj, riscv_profiles[i]->name);
555     }
556 }
557 
558 static int kvm_riscv_get_regs_core(CPUState *cs)
559 {
560     int ret = 0;
561     int i;
562     target_ulong reg;
563     CPURISCVState *env = &RISCV_CPU(cs)->env;
564 
565     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
566     if (ret) {
567         return ret;
568     }
569     env->pc = reg;
570 
571     for (i = 1; i < 32; i++) {
572         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
573         ret = kvm_get_one_reg(cs, id, &reg);
574         if (ret) {
575             return ret;
576         }
577         env->gpr[i] = reg;
578     }
579 
580     return ret;
581 }
582 
583 static int kvm_riscv_put_regs_core(CPUState *cs)
584 {
585     int ret = 0;
586     int i;
587     target_ulong reg;
588     CPURISCVState *env = &RISCV_CPU(cs)->env;
589 
590     reg = env->pc;
591     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
592     if (ret) {
593         return ret;
594     }
595 
596     for (i = 1; i < 32; i++) {
597         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
598         reg = env->gpr[i];
599         ret = kvm_set_one_reg(cs, id, &reg);
600         if (ret) {
601             return ret;
602         }
603     }
604 
605     return ret;
606 }
607 
608 static int kvm_riscv_get_regs_csr(CPUState *cs)
609 {
610     CPURISCVState *env = &RISCV_CPU(cs)->env;
611 
612     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
613     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
614     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
615     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
616     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
617     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
618     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
619     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
620     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
621 
622     return 0;
623 }
624 
625 static int kvm_riscv_put_regs_csr(CPUState *cs)
626 {
627     CPURISCVState *env = &RISCV_CPU(cs)->env;
628 
629     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
630     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
631     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
632     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
633     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
634     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
635     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
636     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
637     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
638 
639     return 0;
640 }
641 
642 static int kvm_riscv_get_regs_fp(CPUState *cs)
643 {
644     int ret = 0;
645     int i;
646     CPURISCVState *env = &RISCV_CPU(cs)->env;
647 
648     if (riscv_has_ext(env, RVD)) {
649         uint64_t reg;
650         for (i = 0; i < 32; i++) {
651             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
652             if (ret) {
653                 return ret;
654             }
655             env->fpr[i] = reg;
656         }
657         return ret;
658     }
659 
660     if (riscv_has_ext(env, RVF)) {
661         uint32_t reg;
662         for (i = 0; i < 32; i++) {
663             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
664             if (ret) {
665                 return ret;
666             }
667             env->fpr[i] = reg;
668         }
669         return ret;
670     }
671 
672     return ret;
673 }
674 
675 static int kvm_riscv_put_regs_fp(CPUState *cs)
676 {
677     int ret = 0;
678     int i;
679     CPURISCVState *env = &RISCV_CPU(cs)->env;
680 
681     if (riscv_has_ext(env, RVD)) {
682         uint64_t reg;
683         for (i = 0; i < 32; i++) {
684             reg = env->fpr[i];
685             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
686             if (ret) {
687                 return ret;
688             }
689         }
690         return ret;
691     }
692 
693     if (riscv_has_ext(env, RVF)) {
694         uint32_t reg;
695         for (i = 0; i < 32; i++) {
696             reg = env->fpr[i];
697             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
698             if (ret) {
699                 return ret;
700             }
701         }
702         return ret;
703     }
704 
705     return ret;
706 }
707 
708 static void kvm_riscv_get_regs_timer(CPUState *cs)
709 {
710     CPURISCVState *env = &RISCV_CPU(cs)->env;
711 
712     if (env->kvm_timer_dirty) {
713         return;
714     }
715 
716     KVM_RISCV_GET_TIMER(cs, time, env->kvm_timer_time);
717     KVM_RISCV_GET_TIMER(cs, compare, env->kvm_timer_compare);
718     KVM_RISCV_GET_TIMER(cs, state, env->kvm_timer_state);
719     KVM_RISCV_GET_TIMER(cs, frequency, env->kvm_timer_frequency);
720 
721     env->kvm_timer_dirty = true;
722 }
723 
724 static void kvm_riscv_put_regs_timer(CPUState *cs)
725 {
726     uint64_t reg;
727     CPURISCVState *env = &RISCV_CPU(cs)->env;
728 
729     if (!env->kvm_timer_dirty) {
730         return;
731     }
732 
733     KVM_RISCV_SET_TIMER(cs, time, env->kvm_timer_time);
734     KVM_RISCV_SET_TIMER(cs, compare, env->kvm_timer_compare);
735 
736     /*
737      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
738      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
739      * doesn't matter that adaping in QEMU now.
740      * TODO If KVM changes, adapt here.
741      */
742     if (env->kvm_timer_state) {
743         KVM_RISCV_SET_TIMER(cs, state, env->kvm_timer_state);
744     }
745 
746     /*
747      * For now, migration will not work between Hosts with different timer
748      * frequency. Therefore, we should check whether they are the same here
749      * during the migration.
750      */
751     if (migration_is_running()) {
752         KVM_RISCV_GET_TIMER(cs, frequency, reg);
753         if (reg != env->kvm_timer_frequency) {
754             error_report("Dst Hosts timer frequency != Src Hosts");
755         }
756     }
757 
758     env->kvm_timer_dirty = false;
759 }
760 
761 uint64_t kvm_riscv_get_timebase_frequency(RISCVCPU *cpu)
762 {
763     uint64_t reg;
764 
765     KVM_RISCV_GET_TIMER(CPU(cpu), frequency, reg);
766 
767     return reg;
768 }
769 
770 static int kvm_riscv_get_regs_vector(CPUState *cs)
771 {
772     RISCVCPU *cpu = RISCV_CPU(cs);
773     CPURISCVState *env = &cpu->env;
774     target_ulong reg;
775     uint64_t vreg_id;
776     int vreg_idx, ret = 0;
777 
778     if (!riscv_has_ext(env, RVV)) {
779         return 0;
780     }
781 
782     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
783     if (ret) {
784         return ret;
785     }
786     env->vstart = reg;
787 
788     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
789     if (ret) {
790         return ret;
791     }
792     env->vl = reg;
793 
794     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
795     if (ret) {
796         return ret;
797     }
798     env->vtype = reg;
799 
800     if (kvm_v_vlenb.supported) {
801         ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
802         if (ret) {
803             return ret;
804         }
805         cpu->cfg.vlenb = reg;
806 
807         for (int i = 0; i < 32; i++) {
808             /*
809              * vreg[] is statically allocated using RV_VLEN_MAX.
810              * Use it instead of vlenb to calculate vreg_idx for
811              * simplicity.
812              */
813             vreg_idx = i * RV_VLEN_MAX / 64;
814             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
815 
816             ret = kvm_get_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
817             if (ret) {
818                 return ret;
819             }
820         }
821     }
822 
823     return 0;
824 }
825 
826 static int kvm_riscv_put_regs_vector(CPUState *cs)
827 {
828     RISCVCPU *cpu = RISCV_CPU(cs);
829     CPURISCVState *env = &cpu->env;
830     target_ulong reg;
831     uint64_t vreg_id;
832     int vreg_idx, ret = 0;
833 
834     if (!riscv_has_ext(env, RVV)) {
835         return 0;
836     }
837 
838     reg = env->vstart;
839     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
840     if (ret) {
841         return ret;
842     }
843 
844     reg = env->vl;
845     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
846     if (ret) {
847         return ret;
848     }
849 
850     reg = env->vtype;
851     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
852     if (ret) {
853         return ret;
854     }
855 
856     if (kvm_v_vlenb.supported) {
857         reg = cpu->cfg.vlenb;
858         ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
859 
860         for (int i = 0; i < 32; i++) {
861             /*
862              * vreg[] is statically allocated using RV_VLEN_MAX.
863              * Use it instead of vlenb to calculate vreg_idx for
864              * simplicity.
865              */
866             vreg_idx = i * RV_VLEN_MAX / 64;
867             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
868 
869             ret = kvm_set_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
870             if (ret) {
871                 return ret;
872             }
873         }
874     }
875 
876     return ret;
877 }
878 
879 typedef struct KVMScratchCPU {
880     int kvmfd;
881     int vmfd;
882     int cpufd;
883 } KVMScratchCPU;
884 
885 /*
886  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
887  * from target/arm/kvm.c.
888  */
889 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
890 {
891     int kvmfd = -1, vmfd = -1, cpufd = -1;
892 
893     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
894     if (kvmfd < 0) {
895         goto err;
896     }
897     do {
898         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
899     } while (vmfd == -1 && errno == EINTR);
900     if (vmfd < 0) {
901         goto err;
902     }
903     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
904     if (cpufd < 0) {
905         goto err;
906     }
907 
908     scratch->kvmfd =  kvmfd;
909     scratch->vmfd = vmfd;
910     scratch->cpufd = cpufd;
911 
912     return true;
913 
914  err:
915     if (cpufd >= 0) {
916         close(cpufd);
917     }
918     if (vmfd >= 0) {
919         close(vmfd);
920     }
921     if (kvmfd >= 0) {
922         close(kvmfd);
923     }
924 
925     return false;
926 }
927 
928 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
929 {
930     close(scratch->cpufd);
931     close(scratch->vmfd);
932     close(scratch->kvmfd);
933 }
934 
935 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
936 {
937     CPURISCVState *env = &cpu->env;
938     struct kvm_one_reg reg;
939     int ret;
940 
941     reg.id = RISCV_CONFIG_REG(env, mvendorid);
942     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
943     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
944     if (ret != 0) {
945         error_report("Unable to retrieve mvendorid from host, error %d", ret);
946     }
947 
948     reg.id = RISCV_CONFIG_REG(env, marchid);
949     reg.addr = (uint64_t)&cpu->cfg.marchid;
950     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
951     if (ret != 0) {
952         error_report("Unable to retrieve marchid from host, error %d", ret);
953     }
954 
955     reg.id = RISCV_CONFIG_REG(env, mimpid);
956     reg.addr = (uint64_t)&cpu->cfg.mimpid;
957     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
958     if (ret != 0) {
959         error_report("Unable to retrieve mimpid from host, error %d", ret);
960     }
961 }
962 
963 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
964                                          KVMScratchCPU *kvmcpu)
965 {
966     CPURISCVState *env = &cpu->env;
967     struct kvm_one_reg reg;
968     int ret;
969 
970     reg.id = RISCV_CONFIG_REG(env, isa);
971     reg.addr = (uint64_t)&env->misa_ext_mask;
972     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
973 
974     if (ret) {
975         error_report("Unable to fetch ISA register from KVM, "
976                      "error %d", ret);
977         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
978         exit(EXIT_FAILURE);
979     }
980 
981     env->misa_ext = env->misa_ext_mask;
982 }
983 
984 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
985                                          KVMCPUConfig *cbomz_cfg)
986 {
987     CPURISCVState *env = &cpu->env;
988     struct kvm_one_reg reg;
989     int ret;
990 
991     reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
992                                     cbomz_cfg->kvm_reg_id);
993     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
994     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
995     if (ret != 0) {
996         error_report("Unable to read KVM reg %s, error %d",
997                      cbomz_cfg->name, ret);
998         exit(EXIT_FAILURE);
999     }
1000 }
1001 
1002 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
1003                                            KVMScratchCPU *kvmcpu)
1004 {
1005     CPURISCVState *env = &cpu->env;
1006     uint64_t val;
1007     int i, ret;
1008 
1009     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1010         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1011         struct kvm_one_reg reg;
1012 
1013         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
1014                                         multi_ext_cfg->kvm_reg_id);
1015         reg.addr = (uint64_t)&val;
1016         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1017         if (ret != 0) {
1018             if (errno == EINVAL) {
1019                 /* Silently default to 'false' if KVM does not support it. */
1020                 multi_ext_cfg->supported = false;
1021                 val = false;
1022             } else {
1023                 error_report("Unable to read ISA_EXT KVM register %s: %s",
1024                              multi_ext_cfg->name, strerror(errno));
1025                 exit(EXIT_FAILURE);
1026             }
1027         } else {
1028             multi_ext_cfg->supported = true;
1029         }
1030 
1031         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1032     }
1033 
1034     if (cpu->cfg.ext_zicbom) {
1035         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1036     }
1037 
1038     if (cpu->cfg.ext_zicboz) {
1039         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1040     }
1041 }
1042 
1043 static int uint64_cmp(const void *a, const void *b)
1044 {
1045     uint64_t val1 = *(const uint64_t *)a;
1046     uint64_t val2 = *(const uint64_t *)b;
1047 
1048     if (val1 < val2) {
1049         return -1;
1050     }
1051 
1052     if (val1 > val2) {
1053         return 1;
1054     }
1055 
1056     return 0;
1057 }
1058 
1059 static void kvm_riscv_check_sbi_dbcn_support(RISCVCPU *cpu,
1060                                              KVMScratchCPU *kvmcpu,
1061                                              struct kvm_reg_list *reglist)
1062 {
1063     struct kvm_reg_list *reg_search;
1064 
1065     reg_search = bsearch(&kvm_sbi_dbcn.kvm_reg_id, reglist->reg, reglist->n,
1066                          sizeof(uint64_t), uint64_cmp);
1067 
1068     if (reg_search) {
1069         kvm_sbi_dbcn.supported = true;
1070     }
1071 }
1072 
1073 static void kvm_riscv_read_vlenb(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
1074                                  struct kvm_reg_list *reglist)
1075 {
1076     struct kvm_one_reg reg;
1077     struct kvm_reg_list *reg_search;
1078     uint64_t val;
1079     int ret;
1080 
1081     reg_search = bsearch(&kvm_v_vlenb.kvm_reg_id, reglist->reg, reglist->n,
1082                          sizeof(uint64_t), uint64_cmp);
1083 
1084     if (reg_search) {
1085         reg.id = kvm_v_vlenb.kvm_reg_id;
1086         reg.addr = (uint64_t)&val;
1087 
1088         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1089         if (ret != 0) {
1090             error_report("Unable to read vlenb register, error code: %d",
1091                          errno);
1092             exit(EXIT_FAILURE);
1093         }
1094 
1095         kvm_v_vlenb.supported = true;
1096         cpu->cfg.vlenb = val;
1097     }
1098 }
1099 
1100 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
1101 {
1102     KVMCPUConfig *multi_ext_cfg;
1103     struct kvm_one_reg reg;
1104     struct kvm_reg_list rl_struct;
1105     struct kvm_reg_list *reglist;
1106     uint64_t val, reg_id, *reg_search;
1107     int i, ret;
1108 
1109     rl_struct.n = 0;
1110     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
1111 
1112     /*
1113      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
1114      * (EINVAL). Use read_legacy() in this case.
1115      */
1116     if (errno == EINVAL) {
1117         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
1118     } else if (errno != E2BIG) {
1119         /*
1120          * E2BIG is an expected error message for the API since we
1121          * don't know the number of registers. The right amount will
1122          * be written in rl_struct.n.
1123          *
1124          * Error out if we get any other errno.
1125          */
1126         error_report("Error when accessing get-reg-list: %s",
1127                      strerror(errno));
1128         exit(EXIT_FAILURE);
1129     }
1130 
1131     reglist = g_malloc(sizeof(struct kvm_reg_list) +
1132                        rl_struct.n * sizeof(uint64_t));
1133     reglist->n = rl_struct.n;
1134     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
1135     if (ret) {
1136         error_report("Error when reading KVM_GET_REG_LIST: %s",
1137                      strerror(errno));
1138         exit(EXIT_FAILURE);
1139     }
1140 
1141     /* sort reglist to use bsearch() */
1142     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
1143 
1144     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1145         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1146         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
1147                                         multi_ext_cfg->kvm_reg_id);
1148         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
1149                              sizeof(uint64_t), uint64_cmp);
1150         if (!reg_search) {
1151             continue;
1152         }
1153 
1154         reg.id = reg_id;
1155         reg.addr = (uint64_t)&val;
1156         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1157         if (ret != 0) {
1158             error_report("Unable to read ISA_EXT KVM register %s: %s",
1159                          multi_ext_cfg->name, strerror(errno));
1160             exit(EXIT_FAILURE);
1161         }
1162 
1163         multi_ext_cfg->supported = true;
1164         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1165     }
1166 
1167     if (cpu->cfg.ext_zicbom) {
1168         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1169     }
1170 
1171     if (cpu->cfg.ext_zicboz) {
1172         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1173     }
1174 
1175     if (riscv_has_ext(&cpu->env, RVV)) {
1176         kvm_riscv_read_vlenb(cpu, kvmcpu, reglist);
1177     }
1178 
1179     kvm_riscv_check_sbi_dbcn_support(cpu, kvmcpu, reglist);
1180 }
1181 
1182 static void riscv_init_kvm_registers(Object *cpu_obj)
1183 {
1184     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
1185     KVMScratchCPU kvmcpu;
1186 
1187     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1188         return;
1189     }
1190 
1191     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
1192     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
1193     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
1194 
1195     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1196 }
1197 
1198 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
1199     KVM_CAP_LAST_INFO
1200 };
1201 
1202 int kvm_arch_get_registers(CPUState *cs, Error **errp)
1203 {
1204     int ret = 0;
1205 
1206     ret = kvm_riscv_get_regs_core(cs);
1207     if (ret) {
1208         return ret;
1209     }
1210 
1211     ret = kvm_riscv_get_regs_csr(cs);
1212     if (ret) {
1213         return ret;
1214     }
1215 
1216     ret = kvm_riscv_get_regs_fp(cs);
1217     if (ret) {
1218         return ret;
1219     }
1220 
1221     ret = kvm_riscv_get_regs_vector(cs);
1222     if (ret) {
1223         return ret;
1224     }
1225 
1226     return ret;
1227 }
1228 
1229 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1230 {
1231     if (cap_has_mp_state) {
1232         struct kvm_mp_state mp_state = {
1233             .mp_state = state
1234         };
1235 
1236         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1237         if (ret) {
1238             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1239                     __func__, ret, strerror(-ret));
1240             return -1;
1241         }
1242     }
1243 
1244     return 0;
1245 }
1246 
1247 int kvm_arch_put_registers(CPUState *cs, int level, Error **errp)
1248 {
1249     int ret = 0;
1250 
1251     ret = kvm_riscv_put_regs_core(cs);
1252     if (ret) {
1253         return ret;
1254     }
1255 
1256     ret = kvm_riscv_put_regs_csr(cs);
1257     if (ret) {
1258         return ret;
1259     }
1260 
1261     ret = kvm_riscv_put_regs_fp(cs);
1262     if (ret) {
1263         return ret;
1264     }
1265 
1266     ret = kvm_riscv_put_regs_vector(cs);
1267     if (ret) {
1268         return ret;
1269     }
1270 
1271     if (KVM_PUT_RESET_STATE == level) {
1272         RISCVCPU *cpu = RISCV_CPU(cs);
1273         if (cs->cpu_index == 0) {
1274             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1275         } else {
1276             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1277         }
1278         if (ret) {
1279             return ret;
1280         }
1281     }
1282 
1283     return ret;
1284 }
1285 
1286 int kvm_arch_release_virq_post(int virq)
1287 {
1288     return 0;
1289 }
1290 
1291 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1292                              uint64_t address, uint32_t data, PCIDevice *dev)
1293 {
1294     return 0;
1295 }
1296 
1297 int kvm_arch_destroy_vcpu(CPUState *cs)
1298 {
1299     return 0;
1300 }
1301 
1302 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1303 {
1304     return cpu->cpu_index;
1305 }
1306 
1307 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1308                                       RunState state)
1309 {
1310     CPUState *cs = opaque;
1311 
1312     if (running) {
1313         kvm_riscv_put_regs_timer(cs);
1314     } else {
1315         kvm_riscv_get_regs_timer(cs);
1316     }
1317 }
1318 
1319 void kvm_arch_init_irq_routing(KVMState *s)
1320 {
1321 }
1322 
1323 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1324 {
1325     CPURISCVState *env = &cpu->env;
1326     target_ulong reg;
1327     uint64_t id;
1328     int ret;
1329 
1330     id = RISCV_CONFIG_REG(env, mvendorid);
1331     /*
1332      * cfg.mvendorid is an uint32 but a target_ulong will
1333      * be written. Assign it to a target_ulong var to avoid
1334      * writing pieces of other cpu->cfg fields in the reg.
1335      */
1336     reg = cpu->cfg.mvendorid;
1337     ret = kvm_set_one_reg(cs, id, &reg);
1338     if (ret != 0) {
1339         return ret;
1340     }
1341 
1342     id = RISCV_CONFIG_REG(env, marchid);
1343     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1344     if (ret != 0) {
1345         return ret;
1346     }
1347 
1348     id = RISCV_CONFIG_REG(env, mimpid);
1349     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1350 
1351     return ret;
1352 }
1353 
1354 static int kvm_vcpu_enable_sbi_dbcn(RISCVCPU *cpu, CPUState *cs)
1355 {
1356     target_ulong reg = 1;
1357 
1358     if (!kvm_sbi_dbcn.supported) {
1359         return 0;
1360     }
1361 
1362     return kvm_set_one_reg(cs, kvm_sbi_dbcn.kvm_reg_id, &reg);
1363 }
1364 
1365 int kvm_arch_init_vcpu(CPUState *cs)
1366 {
1367     int ret = 0;
1368     RISCVCPU *cpu = RISCV_CPU(cs);
1369 
1370     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1371 
1372     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1373         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1374         if (ret != 0) {
1375             return ret;
1376         }
1377     }
1378 
1379     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1380     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1381 
1382     ret = kvm_vcpu_enable_sbi_dbcn(cpu, cs);
1383 
1384     return ret;
1385 }
1386 
1387 int kvm_arch_msi_data_to_gsi(uint32_t data)
1388 {
1389     abort();
1390 }
1391 
1392 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1393                                 int vector, PCIDevice *dev)
1394 {
1395     return 0;
1396 }
1397 
1398 int kvm_arch_get_default_type(MachineState *ms)
1399 {
1400     return 0;
1401 }
1402 
1403 int kvm_arch_init(MachineState *ms, KVMState *s)
1404 {
1405     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1406     return 0;
1407 }
1408 
1409 int kvm_arch_irqchip_create(KVMState *s)
1410 {
1411     /*
1412      * We can create the VAIA using the newer device control API.
1413      */
1414     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1415 }
1416 
1417 int kvm_arch_process_async_events(CPUState *cs)
1418 {
1419     return 0;
1420 }
1421 
1422 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1423 {
1424 }
1425 
1426 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1427 {
1428     return MEMTXATTRS_UNSPECIFIED;
1429 }
1430 
1431 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1432 {
1433     return true;
1434 }
1435 
1436 static void kvm_riscv_handle_sbi_dbcn(CPUState *cs, struct kvm_run *run)
1437 {
1438     g_autofree uint8_t *buf = NULL;
1439     RISCVCPU *cpu = RISCV_CPU(cs);
1440     target_ulong num_bytes;
1441     uint64_t addr;
1442     unsigned char ch;
1443     int ret;
1444 
1445     switch (run->riscv_sbi.function_id) {
1446     case SBI_EXT_DBCN_CONSOLE_READ:
1447     case SBI_EXT_DBCN_CONSOLE_WRITE:
1448         num_bytes = run->riscv_sbi.args[0];
1449 
1450         if (num_bytes == 0) {
1451             run->riscv_sbi.ret[0] = SBI_SUCCESS;
1452             run->riscv_sbi.ret[1] = 0;
1453             break;
1454         }
1455 
1456         addr = run->riscv_sbi.args[1];
1457 
1458         /*
1459          * Handle the case where a 32 bit CPU is running in a
1460          * 64 bit addressing env.
1461          */
1462         if (riscv_cpu_mxl(&cpu->env) == MXL_RV32) {
1463             addr |= (uint64_t)run->riscv_sbi.args[2] << 32;
1464         }
1465 
1466         buf = g_malloc0(num_bytes);
1467 
1468         if (run->riscv_sbi.function_id == SBI_EXT_DBCN_CONSOLE_READ) {
1469             ret = qemu_chr_fe_read_all(serial_hd(0)->be, buf, num_bytes);
1470             if (ret < 0) {
1471                 error_report("SBI_EXT_DBCN_CONSOLE_READ: error when "
1472                              "reading chardev");
1473                 exit(1);
1474             }
1475 
1476             cpu_physical_memory_write(addr, buf, ret);
1477         } else {
1478             cpu_physical_memory_read(addr, buf, num_bytes);
1479 
1480             ret = qemu_chr_fe_write_all(serial_hd(0)->be, buf, num_bytes);
1481             if (ret < 0) {
1482                 error_report("SBI_EXT_DBCN_CONSOLE_WRITE: error when "
1483                              "writing chardev");
1484                 exit(1);
1485             }
1486         }
1487 
1488         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1489         run->riscv_sbi.ret[1] = ret;
1490         break;
1491     case SBI_EXT_DBCN_CONSOLE_WRITE_BYTE:
1492         ch = run->riscv_sbi.args[0];
1493         ret = qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1494 
1495         if (ret < 0) {
1496             error_report("SBI_EXT_DBCN_CONSOLE_WRITE_BYTE: error when "
1497                          "writing chardev");
1498             exit(1);
1499         }
1500 
1501         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1502         run->riscv_sbi.ret[1] = 0;
1503         break;
1504     default:
1505         run->riscv_sbi.ret[0] = SBI_ERR_NOT_SUPPORTED;
1506     }
1507 }
1508 
1509 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1510 {
1511     int ret = 0;
1512     unsigned char ch;
1513     switch (run->riscv_sbi.extension_id) {
1514     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1515         ch = run->riscv_sbi.args[0];
1516         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1517         break;
1518     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1519         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1520         if (ret == sizeof(ch)) {
1521             run->riscv_sbi.ret[0] = ch;
1522         } else {
1523             run->riscv_sbi.ret[0] = -1;
1524         }
1525         ret = 0;
1526         break;
1527     case SBI_EXT_DBCN:
1528         kvm_riscv_handle_sbi_dbcn(cs, run);
1529         break;
1530     default:
1531         qemu_log_mask(LOG_UNIMP,
1532                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1533                       __func__, run->riscv_sbi.extension_id);
1534         ret = -1;
1535         break;
1536     }
1537     return ret;
1538 }
1539 
1540 static int kvm_riscv_handle_csr(CPUState *cs, struct kvm_run *run)
1541 {
1542     target_ulong csr_num = run->riscv_csr.csr_num;
1543     target_ulong new_value = run->riscv_csr.new_value;
1544     target_ulong write_mask = run->riscv_csr.write_mask;
1545     int ret = 0;
1546 
1547     switch (csr_num) {
1548     case CSR_SEED:
1549         run->riscv_csr.ret_value = riscv_new_csr_seed(new_value, write_mask);
1550         break;
1551     default:
1552         qemu_log_mask(LOG_UNIMP,
1553                       "%s: un-handled CSR EXIT for CSR %lx\n",
1554                       __func__, csr_num);
1555         ret = -1;
1556         break;
1557     }
1558 
1559     return ret;
1560 }
1561 
1562 static bool kvm_riscv_handle_debug(CPUState *cs)
1563 {
1564     RISCVCPU *cpu = RISCV_CPU(cs);
1565     CPURISCVState *env = &cpu->env;
1566 
1567     /* Ensure PC is synchronised */
1568     kvm_cpu_synchronize_state(cs);
1569 
1570     if (kvm_find_sw_breakpoint(cs, env->pc)) {
1571         return true;
1572     }
1573 
1574     return false;
1575 }
1576 
1577 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1578 {
1579     int ret = 0;
1580     switch (run->exit_reason) {
1581     case KVM_EXIT_RISCV_SBI:
1582         ret = kvm_riscv_handle_sbi(cs, run);
1583         break;
1584     case KVM_EXIT_RISCV_CSR:
1585         ret = kvm_riscv_handle_csr(cs, run);
1586         break;
1587     case KVM_EXIT_DEBUG:
1588         if (kvm_riscv_handle_debug(cs)) {
1589             ret = EXCP_DEBUG;
1590         }
1591         break;
1592     default:
1593         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1594                       __func__, run->exit_reason);
1595         ret = -1;
1596         break;
1597     }
1598     return ret;
1599 }
1600 
1601 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1602 {
1603     CPURISCVState *env = &cpu->env;
1604     int i;
1605 
1606     if (!kvm_enabled()) {
1607         return;
1608     }
1609     for (i = 0; i < 32; i++) {
1610         env->gpr[i] = 0;
1611     }
1612     env->pc = cpu->env.kernel_addr;
1613     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1614     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1615     env->satp = 0;
1616     env->mie = 0;
1617     env->stvec = 0;
1618     env->sscratch = 0;
1619     env->sepc = 0;
1620     env->scause = 0;
1621     env->stval = 0;
1622     env->mip = 0;
1623 }
1624 
1625 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1626 {
1627     int ret;
1628     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1629 
1630     if (irq != IRQ_S_EXT) {
1631         perror("kvm riscv set irq != IRQ_S_EXT\n");
1632         abort();
1633     }
1634 
1635     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1636     if (ret < 0) {
1637         perror("Set irq failed");
1638         abort();
1639     }
1640 }
1641 
1642 static int aia_mode;
1643 
1644 static const char *kvm_aia_mode_str(uint64_t mode)
1645 {
1646     switch (mode) {
1647     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1648         return "emul";
1649     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1650         return "hwaccel";
1651     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1652     default:
1653         return "auto";
1654     };
1655 }
1656 
1657 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1658 {
1659     return g_strdup(kvm_aia_mode_str(aia_mode));
1660 }
1661 
1662 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1663 {
1664     if (!strcmp(val, "emul")) {
1665         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1666     } else if (!strcmp(val, "hwaccel")) {
1667         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1668     } else if (!strcmp(val, "auto")) {
1669         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1670     } else {
1671         error_setg(errp, "Invalid KVM AIA mode");
1672         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1673     }
1674 }
1675 
1676 void kvm_arch_accel_class_init(ObjectClass *oc)
1677 {
1678     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1679                                   riscv_set_kvm_aia);
1680     object_class_property_set_description(oc, "riscv-aia",
1681         "Set KVM AIA mode. Valid values are 'emul', 'hwaccel' and 'auto'. "
1682         "Changing KVM AIA modes relies on host support. Defaults to 'auto' "
1683         "if the host supports it");
1684     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1685                                     "auto");
1686 }
1687 
1688 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1689                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1690                           uint64_t aplic_base, uint64_t imsic_base,
1691                           uint64_t guest_num)
1692 {
1693     int ret, i;
1694     int aia_fd = -1;
1695     uint64_t default_aia_mode;
1696     uint64_t socket_count = riscv_socket_count(machine);
1697     uint64_t max_hart_per_socket = 0;
1698     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1699     uint64_t socket_bits, hart_bits, guest_bits;
1700     uint64_t max_group_id;
1701 
1702     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1703 
1704     if (aia_fd < 0) {
1705         error_report("Unable to create in-kernel irqchip");
1706         exit(1);
1707     }
1708 
1709     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1710                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1711                             &default_aia_mode, false, NULL);
1712     if (ret < 0) {
1713         error_report("KVM AIA: failed to get current KVM AIA mode");
1714         exit(1);
1715     }
1716 
1717     if (default_aia_mode != aia_mode) {
1718         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1719                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1720                                 &aia_mode, true, NULL);
1721         if (ret < 0) {
1722             warn_report("KVM AIA: failed to set KVM AIA mode '%s', using "
1723                         "default host mode '%s'",
1724                         kvm_aia_mode_str(aia_mode),
1725                         kvm_aia_mode_str(default_aia_mode));
1726 
1727             /* failed to change AIA mode, use default */
1728             aia_mode = default_aia_mode;
1729         }
1730     }
1731 
1732     /*
1733      * Skip APLIC creation in KVM if we're running split mode.
1734      * This is done by leaving KVM_DEV_RISCV_AIA_CONFIG_SRCS
1735      * unset. We can also skip KVM_DEV_RISCV_AIA_ADDR_APLIC
1736      * since KVM won't be using it.
1737      */
1738     if (!kvm_kernel_irqchip_split()) {
1739         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1740                                 KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1741                                 &aia_irq_num, true, NULL);
1742         if (ret < 0) {
1743             error_report("KVM AIA: failed to set number of input irq lines");
1744             exit(1);
1745         }
1746 
1747         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1748                                 KVM_DEV_RISCV_AIA_ADDR_APLIC,
1749                                 &aplic_base, true, NULL);
1750         if (ret < 0) {
1751             error_report("KVM AIA: failed to set the base address of APLIC");
1752             exit(1);
1753         }
1754      }
1755 
1756     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1757                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1758                             &aia_msi_num, true, NULL);
1759     if (ret < 0) {
1760         error_report("KVM AIA: failed to set number of msi");
1761         exit(1);
1762     }
1763 
1764 
1765     if (socket_count > 1) {
1766         max_group_id = socket_count - 1;
1767         socket_bits = find_last_bit(&max_group_id, BITS_PER_LONG) + 1;
1768         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1769                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1770                                 &socket_bits, true, NULL);
1771         if (ret < 0) {
1772             error_report("KVM AIA: failed to set group_bits");
1773             exit(1);
1774         }
1775 
1776         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1777                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1778                                 &group_shift, true, NULL);
1779         if (ret < 0) {
1780             error_report("KVM AIA: failed to set group_shift");
1781             exit(1);
1782         }
1783     }
1784 
1785     guest_bits = guest_num == 0 ? 0 :
1786                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1787     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1788                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1789                             &guest_bits, true, NULL);
1790     if (ret < 0) {
1791         error_report("KVM AIA: failed to set guest_bits");
1792         exit(1);
1793     }
1794 
1795     for (socket = 0; socket < socket_count; socket++) {
1796         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1797         hart_count = riscv_socket_hart_count(machine, socket);
1798         base_hart = riscv_socket_first_hartid(machine, socket);
1799 
1800         if (max_hart_per_socket < hart_count) {
1801             max_hart_per_socket = hart_count;
1802         }
1803 
1804         for (i = 0; i < hart_count; i++) {
1805             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1806             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1807                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1808                                     &imsic_addr, true, NULL);
1809             if (ret < 0) {
1810                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1811                 exit(1);
1812             }
1813         }
1814     }
1815 
1816 
1817     if (max_hart_per_socket > 1) {
1818         max_hart_per_socket--;
1819         hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1820     } else {
1821         hart_bits = 0;
1822     }
1823 
1824     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1825                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1826                             &hart_bits, true, NULL);
1827     if (ret < 0) {
1828         error_report("KVM AIA: failed to set hart_bits");
1829         exit(1);
1830     }
1831 
1832     if (kvm_has_gsi_routing()) {
1833         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1834             /* KVM AIA only has one APLIC instance */
1835             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1836         }
1837         kvm_gsi_routing_allowed = true;
1838         kvm_irqchip_commit_routes(kvm_state);
1839     }
1840 
1841     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1842                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1843                             NULL, true, NULL);
1844     if (ret < 0) {
1845         error_report("KVM AIA: initialized fail");
1846         exit(1);
1847     }
1848 
1849     kvm_msi_via_irqfd_allowed = true;
1850 }
1851 
1852 static void kvm_cpu_instance_init(CPUState *cs)
1853 {
1854     Object *obj = OBJECT(RISCV_CPU(cs));
1855 
1856     riscv_init_kvm_registers(obj);
1857 
1858     kvm_riscv_add_cpu_user_properties(obj);
1859 }
1860 
1861 /*
1862  * We'll get here via the following path:
1863  *
1864  * riscv_cpu_realize()
1865  *   -> cpu_exec_realizefn()
1866  *      -> kvm_cpu_realize() (via accel_cpu_common_realize())
1867  */
1868 static bool kvm_cpu_realize(CPUState *cs, Error **errp)
1869 {
1870     RISCVCPU *cpu = RISCV_CPU(cs);
1871     int ret;
1872 
1873     if (riscv_has_ext(&cpu->env, RVV)) {
1874         ret = prctl(PR_RISCV_V_SET_CONTROL, PR_RISCV_V_VSTATE_CTRL_ON);
1875         if (ret) {
1876             error_setg(errp, "Error in prctl PR_RISCV_V_SET_CONTROL, code: %s",
1877                        strerrorname_np(errno));
1878             return false;
1879         }
1880     }
1881 
1882    return true;
1883 }
1884 
1885 void riscv_kvm_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
1886 {
1887     CPURISCVState *env = &cpu->env;
1888     KVMScratchCPU kvmcpu;
1889     struct kvm_one_reg reg;
1890     uint64_t val;
1891     int ret;
1892 
1893     /* short-circuit without spinning the scratch CPU */
1894     if (!cpu->cfg.ext_zicbom && !cpu->cfg.ext_zicboz &&
1895         !riscv_has_ext(env, RVV)) {
1896         return;
1897     }
1898 
1899     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1900         error_setg(errp, "Unable to create scratch KVM cpu");
1901         return;
1902     }
1903 
1904     if (cpu->cfg.ext_zicbom &&
1905         riscv_cpu_option_set(kvm_cbom_blocksize.name)) {
1906 
1907         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1908                                         kvm_cbom_blocksize.kvm_reg_id);
1909         reg.addr = (uint64_t)&val;
1910         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1911         if (ret != 0) {
1912             error_setg(errp, "Unable to read cbom_blocksize, error %d", errno);
1913             return;
1914         }
1915 
1916         if (cpu->cfg.cbom_blocksize != val) {
1917             error_setg(errp, "Unable to set cbom_blocksize to a different "
1918                        "value than the host (%lu)", val);
1919             return;
1920         }
1921     }
1922 
1923     if (cpu->cfg.ext_zicboz &&
1924         riscv_cpu_option_set(kvm_cboz_blocksize.name)) {
1925 
1926         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1927                                         kvm_cboz_blocksize.kvm_reg_id);
1928         reg.addr = (uint64_t)&val;
1929         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1930         if (ret != 0) {
1931             error_setg(errp, "Unable to read cboz_blocksize, error %d", errno);
1932             return;
1933         }
1934 
1935         if (cpu->cfg.cboz_blocksize != val) {
1936             error_setg(errp, "Unable to set cboz_blocksize to a different "
1937                        "value than the host (%lu)", val);
1938             return;
1939         }
1940     }
1941 
1942     /* Users are setting vlen, not vlenb */
1943     if (riscv_has_ext(env, RVV) && riscv_cpu_option_set("vlen")) {
1944         if (!kvm_v_vlenb.supported) {
1945             error_setg(errp, "Unable to set 'vlenb': register not supported");
1946             return;
1947         }
1948 
1949         reg.id = kvm_v_vlenb.kvm_reg_id;
1950         reg.addr = (uint64_t)&val;
1951         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1952         if (ret != 0) {
1953             error_setg(errp, "Unable to read vlenb register, error %d", errno);
1954             return;
1955         }
1956 
1957         if (cpu->cfg.vlenb != val) {
1958             error_setg(errp, "Unable to set 'vlen' to a different "
1959                        "value than the host (%lu)", val * 8);
1960             return;
1961         }
1962     }
1963 
1964     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1965 }
1966 
1967 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1968 {
1969     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1970 
1971     acc->cpu_instance_init = kvm_cpu_instance_init;
1972     acc->cpu_target_realize = kvm_cpu_realize;
1973 }
1974 
1975 static const TypeInfo kvm_cpu_accel_type_info = {
1976     .name = ACCEL_CPU_NAME("kvm"),
1977 
1978     .parent = TYPE_ACCEL_CPU,
1979     .class_init = kvm_cpu_accel_class_init,
1980     .abstract = true,
1981 };
1982 static void kvm_cpu_accel_register_types(void)
1983 {
1984     type_register_static(&kvm_cpu_accel_type_info);
1985 }
1986 type_init(kvm_cpu_accel_register_types);
1987 
1988 static void riscv_host_cpu_class_init(ObjectClass *c, void *data)
1989 {
1990     RISCVCPUClass *mcc = RISCV_CPU_CLASS(c);
1991 
1992 #if defined(TARGET_RISCV32)
1993     mcc->misa_mxl_max = MXL_RV32;
1994 #elif defined(TARGET_RISCV64)
1995     mcc->misa_mxl_max = MXL_RV64;
1996 #endif
1997 }
1998 
1999 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
2000     {
2001         .name = TYPE_RISCV_CPU_HOST,
2002         .parent = TYPE_RISCV_CPU,
2003         .class_init = riscv_host_cpu_class_init,
2004     }
2005 };
2006 
2007 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
2008 
2009 static const uint32_t ebreak_insn = 0x00100073;
2010 static const uint16_t c_ebreak_insn = 0x9002;
2011 
2012 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2013 {
2014     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 0)) {
2015         return -EINVAL;
2016     }
2017 
2018     if ((bp->saved_insn & 0x3) == 0x3) {
2019         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 0)
2020             || cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak_insn, 4, 1)) {
2021             return -EINVAL;
2022         }
2023     } else {
2024         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak_insn, 2, 1)) {
2025             return -EINVAL;
2026         }
2027     }
2028 
2029     return 0;
2030 }
2031 
2032 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2033 {
2034     uint32_t ebreak;
2035     uint16_t c_ebreak;
2036 
2037     if ((bp->saved_insn & 0x3) == 0x3) {
2038         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak, 4, 0) ||
2039             ebreak != ebreak_insn ||
2040             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 1)) {
2041             return -EINVAL;
2042         }
2043     } else {
2044         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak, 2, 0) ||
2045             c_ebreak != c_ebreak_insn ||
2046             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 1)) {
2047             return -EINVAL;
2048         }
2049     }
2050 
2051     return 0;
2052 }
2053 
2054 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type)
2055 {
2056     /* TODO; To be implemented later. */
2057     return -EINVAL;
2058 }
2059 
2060 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type)
2061 {
2062     /* TODO; To be implemented later. */
2063     return -EINVAL;
2064 }
2065 
2066 void kvm_arch_remove_all_hw_breakpoints(void)
2067 {
2068     /* TODO; To be implemented later. */
2069 }
2070 
2071 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
2072 {
2073     if (kvm_sw_breakpoints_active(cs)) {
2074         dbg->control |= KVM_GUESTDBG_ENABLE;
2075     }
2076 }
2077