xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision eaa910b14773403963316cfe3617eecc9e02356a)
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("ziccrse", ext_ziccrse, KVM_RISCV_ISA_EXT_ZICCRSE),
278     KVM_EXT_CFG("zicntr", ext_zicntr, KVM_RISCV_ISA_EXT_ZICNTR),
279     KVM_EXT_CFG("zicond", ext_zicond, KVM_RISCV_ISA_EXT_ZICOND),
280     KVM_EXT_CFG("zicsr", ext_zicsr, KVM_RISCV_ISA_EXT_ZICSR),
281     KVM_EXT_CFG("zifencei", ext_zifencei, KVM_RISCV_ISA_EXT_ZIFENCEI),
282     KVM_EXT_CFG("zihintntl", ext_zihintntl, KVM_RISCV_ISA_EXT_ZIHINTNTL),
283     KVM_EXT_CFG("zihintpause", ext_zihintpause, KVM_RISCV_ISA_EXT_ZIHINTPAUSE),
284     KVM_EXT_CFG("zihpm", ext_zihpm, KVM_RISCV_ISA_EXT_ZIHPM),
285     KVM_EXT_CFG("zimop", ext_zimop, KVM_RISCV_ISA_EXT_ZIMOP),
286     KVM_EXT_CFG("zcmop", ext_zcmop, KVM_RISCV_ISA_EXT_ZCMOP),
287     KVM_EXT_CFG("zabha", ext_zabha, KVM_RISCV_ISA_EXT_ZABHA),
288     KVM_EXT_CFG("zacas", ext_zacas, KVM_RISCV_ISA_EXT_ZACAS),
289     KVM_EXT_CFG("zawrs", ext_zawrs, KVM_RISCV_ISA_EXT_ZAWRS),
290     KVM_EXT_CFG("zfa", ext_zfa, KVM_RISCV_ISA_EXT_ZFA),
291     KVM_EXT_CFG("zfh", ext_zfh, KVM_RISCV_ISA_EXT_ZFH),
292     KVM_EXT_CFG("zfhmin", ext_zfhmin, KVM_RISCV_ISA_EXT_ZFHMIN),
293     KVM_EXT_CFG("zba", ext_zba, KVM_RISCV_ISA_EXT_ZBA),
294     KVM_EXT_CFG("zbb", ext_zbb, KVM_RISCV_ISA_EXT_ZBB),
295     KVM_EXT_CFG("zbc", ext_zbc, KVM_RISCV_ISA_EXT_ZBC),
296     KVM_EXT_CFG("zbkb", ext_zbkb, KVM_RISCV_ISA_EXT_ZBKB),
297     KVM_EXT_CFG("zbkc", ext_zbkc, KVM_RISCV_ISA_EXT_ZBKC),
298     KVM_EXT_CFG("zbkx", ext_zbkx, KVM_RISCV_ISA_EXT_ZBKX),
299     KVM_EXT_CFG("zbs", ext_zbs, KVM_RISCV_ISA_EXT_ZBS),
300     KVM_EXT_CFG("zca", ext_zca, KVM_RISCV_ISA_EXT_ZCA),
301     KVM_EXT_CFG("zcb", ext_zcb, KVM_RISCV_ISA_EXT_ZCB),
302     KVM_EXT_CFG("zcd", ext_zcd, KVM_RISCV_ISA_EXT_ZCD),
303     KVM_EXT_CFG("zcf", ext_zcf, KVM_RISCV_ISA_EXT_ZCF),
304     KVM_EXT_CFG("zknd", ext_zknd, KVM_RISCV_ISA_EXT_ZKND),
305     KVM_EXT_CFG("zkne", ext_zkne, KVM_RISCV_ISA_EXT_ZKNE),
306     KVM_EXT_CFG("zknh", ext_zknh, KVM_RISCV_ISA_EXT_ZKNH),
307     KVM_EXT_CFG("zkr", ext_zkr, KVM_RISCV_ISA_EXT_ZKR),
308     KVM_EXT_CFG("zksed", ext_zksed, KVM_RISCV_ISA_EXT_ZKSED),
309     KVM_EXT_CFG("zksh", ext_zksh, KVM_RISCV_ISA_EXT_ZKSH),
310     KVM_EXT_CFG("zkt", ext_zkt, KVM_RISCV_ISA_EXT_ZKT),
311     KVM_EXT_CFG("ztso", ext_ztso, KVM_RISCV_ISA_EXT_ZTSO),
312     KVM_EXT_CFG("zvbb", ext_zvbb, KVM_RISCV_ISA_EXT_ZVBB),
313     KVM_EXT_CFG("zvbc", ext_zvbc, KVM_RISCV_ISA_EXT_ZVBC),
314     KVM_EXT_CFG("zvfh", ext_zvfh, KVM_RISCV_ISA_EXT_ZVFH),
315     KVM_EXT_CFG("zvfhmin", ext_zvfhmin, KVM_RISCV_ISA_EXT_ZVFHMIN),
316     KVM_EXT_CFG("zvkb", ext_zvkb, KVM_RISCV_ISA_EXT_ZVKB),
317     KVM_EXT_CFG("zvkg", ext_zvkg, KVM_RISCV_ISA_EXT_ZVKG),
318     KVM_EXT_CFG("zvkned", ext_zvkned, KVM_RISCV_ISA_EXT_ZVKNED),
319     KVM_EXT_CFG("zvknha", ext_zvknha, KVM_RISCV_ISA_EXT_ZVKNHA),
320     KVM_EXT_CFG("zvknhb", ext_zvknhb, KVM_RISCV_ISA_EXT_ZVKNHB),
321     KVM_EXT_CFG("zvksed", ext_zvksed, KVM_RISCV_ISA_EXT_ZVKSED),
322     KVM_EXT_CFG("zvksh", ext_zvksh, KVM_RISCV_ISA_EXT_ZVKSH),
323     KVM_EXT_CFG("zvkt", ext_zvkt, KVM_RISCV_ISA_EXT_ZVKT),
324     KVM_EXT_CFG("smstateen", ext_smstateen, KVM_RISCV_ISA_EXT_SMSTATEEN),
325     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
326     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
327     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
328     KVM_EXT_CFG("svnapot", ext_svnapot, KVM_RISCV_ISA_EXT_SVNAPOT),
329     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
330     KVM_EXT_CFG("svvptc", ext_svvptc, KVM_RISCV_ISA_EXT_SVVPTC),
331 };
332 
333 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
334 {
335     return (void *)&cpu->cfg + kvmcfg->offset;
336 }
337 
338 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
339                             uint32_t val)
340 {
341     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
342 
343     *ext_enabled = val;
344 }
345 
346 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
347                                 KVMCPUConfig *multi_ext)
348 {
349     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
350 
351     return *ext_enabled;
352 }
353 
354 static void kvm_cpu_get_multi_ext_cfg(Object *obj, Visitor *v,
355                                       const char *name,
356                                       void *opaque, Error **errp)
357 {
358     KVMCPUConfig *multi_ext_cfg = opaque;
359     RISCVCPU *cpu = RISCV_CPU(obj);
360     bool value = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
361 
362     visit_type_bool(v, name, &value, errp);
363 }
364 
365 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
366                                       const char *name,
367                                       void *opaque, Error **errp)
368 {
369     KVMCPUConfig *multi_ext_cfg = opaque;
370     RISCVCPU *cpu = RISCV_CPU(obj);
371     bool value, host_val;
372 
373     if (!visit_type_bool(v, name, &value, errp)) {
374         return;
375     }
376 
377     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
378 
379     /*
380      * Ignore if the user is setting the same value
381      * as the host.
382      */
383     if (value == host_val) {
384         return;
385     }
386 
387     if (!multi_ext_cfg->supported) {
388         /*
389          * Error out if the user is trying to enable an
390          * extension that KVM doesn't support. Ignore
391          * option otherwise.
392          */
393         if (value) {
394             error_setg(errp, "KVM does not support disabling extension %s",
395                        multi_ext_cfg->name);
396         }
397 
398         return;
399     }
400 
401     multi_ext_cfg->user_set = true;
402     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
403 }
404 
405 static KVMCPUConfig kvm_cbom_blocksize = {
406     .name = "cbom_blocksize",
407     .offset = CPU_CFG_OFFSET(cbom_blocksize),
408     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
409 };
410 
411 static KVMCPUConfig kvm_cboz_blocksize = {
412     .name = "cboz_blocksize",
413     .offset = CPU_CFG_OFFSET(cboz_blocksize),
414     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
415 };
416 
417 static KVMCPUConfig kvm_v_vlenb = {
418     .name = "vlenb",
419     .offset = CPU_CFG_OFFSET(vlenb),
420     .kvm_reg_id =  KVM_REG_RISCV | KVM_REG_SIZE_U64 | KVM_REG_RISCV_VECTOR |
421                    KVM_REG_RISCV_VECTOR_CSR_REG(vlenb)
422 };
423 
424 static KVMCPUConfig kvm_sbi_dbcn = {
425     .name = "sbi_dbcn",
426     .kvm_reg_id = KVM_REG_RISCV | KVM_REG_SIZE_U64 |
427                   KVM_REG_RISCV_SBI_EXT | KVM_RISCV_SBI_EXT_DBCN
428 };
429 
430 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
431 {
432     CPURISCVState *env = &cpu->env;
433     uint64_t id, reg;
434     int i, ret;
435 
436     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
437         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
438 
439         if (!multi_ext_cfg->user_set) {
440             continue;
441         }
442 
443         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
444                                     multi_ext_cfg->kvm_reg_id);
445         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
446         ret = kvm_set_one_reg(cs, id, &reg);
447         if (ret != 0) {
448             if (!reg && ret == -EINVAL) {
449                 warn_report("KVM cannot disable extension %s",
450                             multi_ext_cfg->name);
451             } else {
452                 error_report("Unable to enable extension %s in KVM, error %d",
453                              multi_ext_cfg->name, ret);
454                 exit(EXIT_FAILURE);
455             }
456         }
457     }
458 }
459 
460 static void cpu_get_cfg_unavailable(Object *obj, Visitor *v,
461                                     const char *name,
462                                     void *opaque, Error **errp)
463 {
464     bool value = false;
465 
466     visit_type_bool(v, name, &value, errp);
467 }
468 
469 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
470                                     const char *name,
471                                     void *opaque, Error **errp)
472 {
473     const char *propname = opaque;
474     bool value;
475 
476     if (!visit_type_bool(v, name, &value, errp)) {
477         return;
478     }
479 
480     if (value) {
481         error_setg(errp, "'%s' is not available with KVM",
482                    propname);
483     }
484 }
485 
486 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
487 {
488     /* Check if KVM created the property already */
489     if (object_property_find(obj, prop_name)) {
490         return;
491     }
492 
493     /*
494      * Set the default to disabled for every extension
495      * unknown to KVM and error out if the user attempts
496      * to enable any of them.
497      */
498     object_property_add(obj, prop_name, "bool",
499                         cpu_get_cfg_unavailable,
500                         cpu_set_cfg_unavailable,
501                         NULL, (void *)prop_name);
502 }
503 
504 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
505                                         const RISCVCPUMultiExtConfig *array)
506 {
507     const RISCVCPUMultiExtConfig *prop;
508 
509     g_assert(array);
510 
511     for (prop = array; prop && prop->name; prop++) {
512         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
513     }
514 }
515 
516 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
517 {
518     int i;
519 
520     riscv_add_satp_mode_properties(cpu_obj);
521 
522     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
523         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
524         int bit = misa_cfg->offset;
525 
526         misa_cfg->name = riscv_get_misa_ext_name(bit);
527         misa_cfg->description = riscv_get_misa_ext_description(bit);
528 
529         object_property_add(cpu_obj, misa_cfg->name, "bool",
530                             kvm_cpu_get_misa_ext_cfg,
531                             kvm_cpu_set_misa_ext_cfg,
532                             NULL, misa_cfg);
533         object_property_set_description(cpu_obj, misa_cfg->name,
534                                         misa_cfg->description);
535     }
536 
537     for (i = 0; misa_bits[i] != 0; i++) {
538         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
539         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
540     }
541 
542     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
543         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
544 
545         object_property_add(cpu_obj, multi_cfg->name, "bool",
546                             kvm_cpu_get_multi_ext_cfg,
547                             kvm_cpu_set_multi_ext_cfg,
548                             NULL, multi_cfg);
549     }
550 
551     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
552     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
553     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
554 
555    /* We don't have the needed KVM support for profiles */
556     for (i = 0; riscv_profiles[i] != NULL; i++) {
557         riscv_cpu_add_kvm_unavail_prop(cpu_obj, riscv_profiles[i]->name);
558     }
559 }
560 
561 static int kvm_riscv_get_regs_core(CPUState *cs)
562 {
563     int ret = 0;
564     int i;
565     target_ulong reg;
566     CPURISCVState *env = &RISCV_CPU(cs)->env;
567 
568     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
569     if (ret) {
570         return ret;
571     }
572     env->pc = reg;
573 
574     for (i = 1; i < 32; i++) {
575         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
576         ret = kvm_get_one_reg(cs, id, &reg);
577         if (ret) {
578             return ret;
579         }
580         env->gpr[i] = reg;
581     }
582 
583     return ret;
584 }
585 
586 static int kvm_riscv_put_regs_core(CPUState *cs)
587 {
588     int ret = 0;
589     int i;
590     target_ulong reg;
591     CPURISCVState *env = &RISCV_CPU(cs)->env;
592 
593     reg = env->pc;
594     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
595     if (ret) {
596         return ret;
597     }
598 
599     for (i = 1; i < 32; i++) {
600         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
601         reg = env->gpr[i];
602         ret = kvm_set_one_reg(cs, id, &reg);
603         if (ret) {
604             return ret;
605         }
606     }
607 
608     return ret;
609 }
610 
611 static int kvm_riscv_get_regs_csr(CPUState *cs)
612 {
613     CPURISCVState *env = &RISCV_CPU(cs)->env;
614 
615     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
616     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
617     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
618     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
619     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
620     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
621     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
622     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
623     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
624 
625     return 0;
626 }
627 
628 static int kvm_riscv_put_regs_csr(CPUState *cs)
629 {
630     CPURISCVState *env = &RISCV_CPU(cs)->env;
631 
632     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
633     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
634     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
635     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
636     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
637     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
638     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
639     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
640     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
641 
642     return 0;
643 }
644 
645 static int kvm_riscv_get_regs_fp(CPUState *cs)
646 {
647     int ret = 0;
648     int i;
649     CPURISCVState *env = &RISCV_CPU(cs)->env;
650 
651     if (riscv_has_ext(env, RVD)) {
652         uint64_t reg;
653         for (i = 0; i < 32; i++) {
654             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
655             if (ret) {
656                 return ret;
657             }
658             env->fpr[i] = reg;
659         }
660         return ret;
661     }
662 
663     if (riscv_has_ext(env, RVF)) {
664         uint32_t reg;
665         for (i = 0; i < 32; i++) {
666             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
667             if (ret) {
668                 return ret;
669             }
670             env->fpr[i] = reg;
671         }
672         return ret;
673     }
674 
675     return ret;
676 }
677 
678 static int kvm_riscv_put_regs_fp(CPUState *cs)
679 {
680     int ret = 0;
681     int i;
682     CPURISCVState *env = &RISCV_CPU(cs)->env;
683 
684     if (riscv_has_ext(env, RVD)) {
685         uint64_t reg;
686         for (i = 0; i < 32; i++) {
687             reg = env->fpr[i];
688             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
689             if (ret) {
690                 return ret;
691             }
692         }
693         return ret;
694     }
695 
696     if (riscv_has_ext(env, RVF)) {
697         uint32_t reg;
698         for (i = 0; i < 32; i++) {
699             reg = env->fpr[i];
700             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
701             if (ret) {
702                 return ret;
703             }
704         }
705         return ret;
706     }
707 
708     return ret;
709 }
710 
711 static void kvm_riscv_get_regs_timer(CPUState *cs)
712 {
713     CPURISCVState *env = &RISCV_CPU(cs)->env;
714 
715     if (env->kvm_timer_dirty) {
716         return;
717     }
718 
719     KVM_RISCV_GET_TIMER(cs, time, env->kvm_timer_time);
720     KVM_RISCV_GET_TIMER(cs, compare, env->kvm_timer_compare);
721     KVM_RISCV_GET_TIMER(cs, state, env->kvm_timer_state);
722     KVM_RISCV_GET_TIMER(cs, frequency, env->kvm_timer_frequency);
723 
724     env->kvm_timer_dirty = true;
725 }
726 
727 static void kvm_riscv_put_regs_timer(CPUState *cs)
728 {
729     uint64_t reg;
730     CPURISCVState *env = &RISCV_CPU(cs)->env;
731 
732     if (!env->kvm_timer_dirty) {
733         return;
734     }
735 
736     KVM_RISCV_SET_TIMER(cs, time, env->kvm_timer_time);
737     KVM_RISCV_SET_TIMER(cs, compare, env->kvm_timer_compare);
738 
739     /*
740      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
741      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
742      * doesn't matter that adaping in QEMU now.
743      * TODO If KVM changes, adapt here.
744      */
745     if (env->kvm_timer_state) {
746         KVM_RISCV_SET_TIMER(cs, state, env->kvm_timer_state);
747     }
748 
749     /*
750      * For now, migration will not work between Hosts with different timer
751      * frequency. Therefore, we should check whether they are the same here
752      * during the migration.
753      */
754     if (migration_is_running()) {
755         KVM_RISCV_GET_TIMER(cs, frequency, reg);
756         if (reg != env->kvm_timer_frequency) {
757             error_report("Dst Hosts timer frequency != Src Hosts");
758         }
759     }
760 
761     env->kvm_timer_dirty = false;
762 }
763 
764 uint64_t kvm_riscv_get_timebase_frequency(RISCVCPU *cpu)
765 {
766     uint64_t reg;
767 
768     KVM_RISCV_GET_TIMER(CPU(cpu), frequency, reg);
769 
770     return reg;
771 }
772 
773 static int kvm_riscv_get_regs_vector(CPUState *cs)
774 {
775     RISCVCPU *cpu = RISCV_CPU(cs);
776     CPURISCVState *env = &cpu->env;
777     target_ulong reg;
778     uint64_t vreg_id;
779     int vreg_idx, ret = 0;
780 
781     if (!riscv_has_ext(env, RVV)) {
782         return 0;
783     }
784 
785     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
786     if (ret) {
787         return ret;
788     }
789     env->vstart = reg;
790 
791     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
792     if (ret) {
793         return ret;
794     }
795     env->vl = reg;
796 
797     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
798     if (ret) {
799         return ret;
800     }
801     env->vtype = reg;
802 
803     if (kvm_v_vlenb.supported) {
804         ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
805         if (ret) {
806             return ret;
807         }
808         cpu->cfg.vlenb = reg;
809 
810         for (int i = 0; i < 32; i++) {
811             /*
812              * vreg[] is statically allocated using RV_VLEN_MAX.
813              * Use it instead of vlenb to calculate vreg_idx for
814              * simplicity.
815              */
816             vreg_idx = i * RV_VLEN_MAX / 64;
817             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
818 
819             ret = kvm_get_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
820             if (ret) {
821                 return ret;
822             }
823         }
824     }
825 
826     return 0;
827 }
828 
829 static int kvm_riscv_put_regs_vector(CPUState *cs)
830 {
831     RISCVCPU *cpu = RISCV_CPU(cs);
832     CPURISCVState *env = &cpu->env;
833     target_ulong reg;
834     uint64_t vreg_id;
835     int vreg_idx, ret = 0;
836 
837     if (!riscv_has_ext(env, RVV)) {
838         return 0;
839     }
840 
841     reg = env->vstart;
842     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
843     if (ret) {
844         return ret;
845     }
846 
847     reg = env->vl;
848     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
849     if (ret) {
850         return ret;
851     }
852 
853     reg = env->vtype;
854     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
855     if (ret) {
856         return ret;
857     }
858 
859     if (kvm_v_vlenb.supported) {
860         reg = cpu->cfg.vlenb;
861         ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
862 
863         for (int i = 0; i < 32; i++) {
864             /*
865              * vreg[] is statically allocated using RV_VLEN_MAX.
866              * Use it instead of vlenb to calculate vreg_idx for
867              * simplicity.
868              */
869             vreg_idx = i * RV_VLEN_MAX / 64;
870             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
871 
872             ret = kvm_set_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
873             if (ret) {
874                 return ret;
875             }
876         }
877     }
878 
879     return ret;
880 }
881 
882 typedef struct KVMScratchCPU {
883     int kvmfd;
884     int vmfd;
885     int cpufd;
886 } KVMScratchCPU;
887 
888 /*
889  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
890  * from target/arm/kvm.c.
891  */
892 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
893 {
894     int kvmfd = -1, vmfd = -1, cpufd = -1;
895 
896     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
897     if (kvmfd < 0) {
898         goto err;
899     }
900     do {
901         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
902     } while (vmfd == -1 && errno == EINTR);
903     if (vmfd < 0) {
904         goto err;
905     }
906     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
907     if (cpufd < 0) {
908         goto err;
909     }
910 
911     scratch->kvmfd =  kvmfd;
912     scratch->vmfd = vmfd;
913     scratch->cpufd = cpufd;
914 
915     return true;
916 
917  err:
918     if (cpufd >= 0) {
919         close(cpufd);
920     }
921     if (vmfd >= 0) {
922         close(vmfd);
923     }
924     if (kvmfd >= 0) {
925         close(kvmfd);
926     }
927 
928     return false;
929 }
930 
931 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
932 {
933     close(scratch->cpufd);
934     close(scratch->vmfd);
935     close(scratch->kvmfd);
936 }
937 
938 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
939 {
940     CPURISCVState *env = &cpu->env;
941     struct kvm_one_reg reg;
942     int ret;
943 
944     reg.id = RISCV_CONFIG_REG(env, mvendorid);
945     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
946     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
947     if (ret != 0) {
948         error_report("Unable to retrieve mvendorid from host, error %d", ret);
949     }
950 
951     reg.id = RISCV_CONFIG_REG(env, marchid);
952     reg.addr = (uint64_t)&cpu->cfg.marchid;
953     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
954     if (ret != 0) {
955         error_report("Unable to retrieve marchid from host, error %d", ret);
956     }
957 
958     reg.id = RISCV_CONFIG_REG(env, mimpid);
959     reg.addr = (uint64_t)&cpu->cfg.mimpid;
960     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
961     if (ret != 0) {
962         error_report("Unable to retrieve mimpid from host, error %d", ret);
963     }
964 }
965 
966 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
967                                          KVMScratchCPU *kvmcpu)
968 {
969     CPURISCVState *env = &cpu->env;
970     struct kvm_one_reg reg;
971     int ret;
972 
973     reg.id = RISCV_CONFIG_REG(env, isa);
974     reg.addr = (uint64_t)&env->misa_ext_mask;
975     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
976 
977     if (ret) {
978         error_report("Unable to fetch ISA register from KVM, "
979                      "error %d", ret);
980         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
981         exit(EXIT_FAILURE);
982     }
983 
984     env->misa_ext = env->misa_ext_mask;
985 }
986 
987 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
988                                          KVMCPUConfig *cbomz_cfg)
989 {
990     CPURISCVState *env = &cpu->env;
991     struct kvm_one_reg reg;
992     int ret;
993 
994     reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
995                                     cbomz_cfg->kvm_reg_id);
996     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
997     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
998     if (ret != 0) {
999         error_report("Unable to read KVM reg %s, error %d",
1000                      cbomz_cfg->name, ret);
1001         exit(EXIT_FAILURE);
1002     }
1003 }
1004 
1005 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
1006                                            KVMScratchCPU *kvmcpu)
1007 {
1008     CPURISCVState *env = &cpu->env;
1009     uint64_t val;
1010     int i, ret;
1011 
1012     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1013         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1014         struct kvm_one_reg reg;
1015 
1016         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
1017                                         multi_ext_cfg->kvm_reg_id);
1018         reg.addr = (uint64_t)&val;
1019         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1020         if (ret != 0) {
1021             if (errno == EINVAL) {
1022                 /* Silently default to 'false' if KVM does not support it. */
1023                 multi_ext_cfg->supported = false;
1024                 val = false;
1025             } else {
1026                 error_report("Unable to read ISA_EXT KVM register %s: %s",
1027                              multi_ext_cfg->name, strerror(errno));
1028                 exit(EXIT_FAILURE);
1029             }
1030         } else {
1031             multi_ext_cfg->supported = true;
1032         }
1033 
1034         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1035     }
1036 
1037     if (cpu->cfg.ext_zicbom) {
1038         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1039     }
1040 
1041     if (cpu->cfg.ext_zicboz) {
1042         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1043     }
1044 }
1045 
1046 static int uint64_cmp(const void *a, const void *b)
1047 {
1048     uint64_t val1 = *(const uint64_t *)a;
1049     uint64_t val2 = *(const uint64_t *)b;
1050 
1051     if (val1 < val2) {
1052         return -1;
1053     }
1054 
1055     if (val1 > val2) {
1056         return 1;
1057     }
1058 
1059     return 0;
1060 }
1061 
1062 static void kvm_riscv_check_sbi_dbcn_support(RISCVCPU *cpu,
1063                                              KVMScratchCPU *kvmcpu,
1064                                              struct kvm_reg_list *reglist)
1065 {
1066     struct kvm_reg_list *reg_search;
1067 
1068     reg_search = bsearch(&kvm_sbi_dbcn.kvm_reg_id, reglist->reg, reglist->n,
1069                          sizeof(uint64_t), uint64_cmp);
1070 
1071     if (reg_search) {
1072         kvm_sbi_dbcn.supported = true;
1073     }
1074 }
1075 
1076 static void kvm_riscv_read_vlenb(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
1077                                  struct kvm_reg_list *reglist)
1078 {
1079     struct kvm_one_reg reg;
1080     struct kvm_reg_list *reg_search;
1081     uint64_t val;
1082     int ret;
1083 
1084     reg_search = bsearch(&kvm_v_vlenb.kvm_reg_id, reglist->reg, reglist->n,
1085                          sizeof(uint64_t), uint64_cmp);
1086 
1087     if (reg_search) {
1088         reg.id = kvm_v_vlenb.kvm_reg_id;
1089         reg.addr = (uint64_t)&val;
1090 
1091         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1092         if (ret != 0) {
1093             error_report("Unable to read vlenb register, error code: %d",
1094                          errno);
1095             exit(EXIT_FAILURE);
1096         }
1097 
1098         kvm_v_vlenb.supported = true;
1099         cpu->cfg.vlenb = val;
1100     }
1101 }
1102 
1103 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
1104 {
1105     KVMCPUConfig *multi_ext_cfg;
1106     struct kvm_one_reg reg;
1107     struct kvm_reg_list rl_struct;
1108     struct kvm_reg_list *reglist;
1109     uint64_t val, reg_id, *reg_search;
1110     int i, ret;
1111 
1112     rl_struct.n = 0;
1113     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
1114 
1115     /*
1116      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
1117      * (EINVAL). Use read_legacy() in this case.
1118      */
1119     if (errno == EINVAL) {
1120         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
1121     } else if (errno != E2BIG) {
1122         /*
1123          * E2BIG is an expected error message for the API since we
1124          * don't know the number of registers. The right amount will
1125          * be written in rl_struct.n.
1126          *
1127          * Error out if we get any other errno.
1128          */
1129         error_report("Error when accessing get-reg-list: %s",
1130                      strerror(errno));
1131         exit(EXIT_FAILURE);
1132     }
1133 
1134     reglist = g_malloc(sizeof(struct kvm_reg_list) +
1135                        rl_struct.n * sizeof(uint64_t));
1136     reglist->n = rl_struct.n;
1137     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
1138     if (ret) {
1139         error_report("Error when reading KVM_GET_REG_LIST: %s",
1140                      strerror(errno));
1141         exit(EXIT_FAILURE);
1142     }
1143 
1144     /* sort reglist to use bsearch() */
1145     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
1146 
1147     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1148         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1149         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
1150                                         multi_ext_cfg->kvm_reg_id);
1151         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
1152                              sizeof(uint64_t), uint64_cmp);
1153         if (!reg_search) {
1154             continue;
1155         }
1156 
1157         reg.id = reg_id;
1158         reg.addr = (uint64_t)&val;
1159         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1160         if (ret != 0) {
1161             error_report("Unable to read ISA_EXT KVM register %s: %s",
1162                          multi_ext_cfg->name, strerror(errno));
1163             exit(EXIT_FAILURE);
1164         }
1165 
1166         multi_ext_cfg->supported = true;
1167         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1168     }
1169 
1170     if (cpu->cfg.ext_zicbom) {
1171         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1172     }
1173 
1174     if (cpu->cfg.ext_zicboz) {
1175         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1176     }
1177 
1178     if (riscv_has_ext(&cpu->env, RVV)) {
1179         kvm_riscv_read_vlenb(cpu, kvmcpu, reglist);
1180     }
1181 
1182     kvm_riscv_check_sbi_dbcn_support(cpu, kvmcpu, reglist);
1183 }
1184 
1185 static void riscv_init_kvm_registers(Object *cpu_obj)
1186 {
1187     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
1188     KVMScratchCPU kvmcpu;
1189 
1190     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1191         return;
1192     }
1193 
1194     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
1195     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
1196     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
1197 
1198     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1199 }
1200 
1201 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
1202     KVM_CAP_LAST_INFO
1203 };
1204 
1205 int kvm_arch_get_registers(CPUState *cs, Error **errp)
1206 {
1207     int ret = 0;
1208 
1209     ret = kvm_riscv_get_regs_core(cs);
1210     if (ret) {
1211         return ret;
1212     }
1213 
1214     ret = kvm_riscv_get_regs_csr(cs);
1215     if (ret) {
1216         return ret;
1217     }
1218 
1219     ret = kvm_riscv_get_regs_fp(cs);
1220     if (ret) {
1221         return ret;
1222     }
1223 
1224     ret = kvm_riscv_get_regs_vector(cs);
1225     if (ret) {
1226         return ret;
1227     }
1228 
1229     return ret;
1230 }
1231 
1232 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1233 {
1234     if (cap_has_mp_state) {
1235         struct kvm_mp_state mp_state = {
1236             .mp_state = state
1237         };
1238 
1239         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1240         if (ret) {
1241             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1242                     __func__, ret, strerror(-ret));
1243             return -1;
1244         }
1245     }
1246 
1247     return 0;
1248 }
1249 
1250 int kvm_arch_put_registers(CPUState *cs, int level, Error **errp)
1251 {
1252     int ret = 0;
1253 
1254     ret = kvm_riscv_put_regs_core(cs);
1255     if (ret) {
1256         return ret;
1257     }
1258 
1259     ret = kvm_riscv_put_regs_csr(cs);
1260     if (ret) {
1261         return ret;
1262     }
1263 
1264     ret = kvm_riscv_put_regs_fp(cs);
1265     if (ret) {
1266         return ret;
1267     }
1268 
1269     ret = kvm_riscv_put_regs_vector(cs);
1270     if (ret) {
1271         return ret;
1272     }
1273 
1274     if (KVM_PUT_RESET_STATE == level) {
1275         RISCVCPU *cpu = RISCV_CPU(cs);
1276         if (cs->cpu_index == 0) {
1277             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1278         } else {
1279             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1280         }
1281         if (ret) {
1282             return ret;
1283         }
1284     }
1285 
1286     return ret;
1287 }
1288 
1289 int kvm_arch_release_virq_post(int virq)
1290 {
1291     return 0;
1292 }
1293 
1294 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1295                              uint64_t address, uint32_t data, PCIDevice *dev)
1296 {
1297     return 0;
1298 }
1299 
1300 int kvm_arch_destroy_vcpu(CPUState *cs)
1301 {
1302     return 0;
1303 }
1304 
1305 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1306 {
1307     return cpu->cpu_index;
1308 }
1309 
1310 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1311                                       RunState state)
1312 {
1313     CPUState *cs = opaque;
1314 
1315     if (running) {
1316         kvm_riscv_put_regs_timer(cs);
1317     } else {
1318         kvm_riscv_get_regs_timer(cs);
1319     }
1320 }
1321 
1322 void kvm_arch_init_irq_routing(KVMState *s)
1323 {
1324 }
1325 
1326 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1327 {
1328     CPURISCVState *env = &cpu->env;
1329     target_ulong reg;
1330     uint64_t id;
1331     int ret;
1332 
1333     id = RISCV_CONFIG_REG(env, mvendorid);
1334     /*
1335      * cfg.mvendorid is an uint32 but a target_ulong will
1336      * be written. Assign it to a target_ulong var to avoid
1337      * writing pieces of other cpu->cfg fields in the reg.
1338      */
1339     reg = cpu->cfg.mvendorid;
1340     ret = kvm_set_one_reg(cs, id, &reg);
1341     if (ret != 0) {
1342         return ret;
1343     }
1344 
1345     id = RISCV_CONFIG_REG(env, marchid);
1346     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1347     if (ret != 0) {
1348         return ret;
1349     }
1350 
1351     id = RISCV_CONFIG_REG(env, mimpid);
1352     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1353 
1354     return ret;
1355 }
1356 
1357 static int kvm_vcpu_enable_sbi_dbcn(RISCVCPU *cpu, CPUState *cs)
1358 {
1359     target_ulong reg = 1;
1360 
1361     if (!kvm_sbi_dbcn.supported) {
1362         return 0;
1363     }
1364 
1365     return kvm_set_one_reg(cs, kvm_sbi_dbcn.kvm_reg_id, &reg);
1366 }
1367 
1368 int kvm_arch_init_vcpu(CPUState *cs)
1369 {
1370     int ret = 0;
1371     RISCVCPU *cpu = RISCV_CPU(cs);
1372 
1373     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1374 
1375     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1376         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1377         if (ret != 0) {
1378             return ret;
1379         }
1380     }
1381 
1382     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1383     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1384 
1385     ret = kvm_vcpu_enable_sbi_dbcn(cpu, cs);
1386 
1387     return ret;
1388 }
1389 
1390 int kvm_arch_msi_data_to_gsi(uint32_t data)
1391 {
1392     abort();
1393 }
1394 
1395 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1396                                 int vector, PCIDevice *dev)
1397 {
1398     return 0;
1399 }
1400 
1401 int kvm_arch_get_default_type(MachineState *ms)
1402 {
1403     return 0;
1404 }
1405 
1406 int kvm_arch_init(MachineState *ms, KVMState *s)
1407 {
1408     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1409     return 0;
1410 }
1411 
1412 int kvm_arch_irqchip_create(KVMState *s)
1413 {
1414     /*
1415      * We can create the VAIA using the newer device control API.
1416      */
1417     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1418 }
1419 
1420 int kvm_arch_process_async_events(CPUState *cs)
1421 {
1422     return 0;
1423 }
1424 
1425 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1426 {
1427 }
1428 
1429 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1430 {
1431     return MEMTXATTRS_UNSPECIFIED;
1432 }
1433 
1434 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1435 {
1436     return true;
1437 }
1438 
1439 static void kvm_riscv_handle_sbi_dbcn(CPUState *cs, struct kvm_run *run)
1440 {
1441     g_autofree uint8_t *buf = NULL;
1442     RISCVCPU *cpu = RISCV_CPU(cs);
1443     target_ulong num_bytes;
1444     uint64_t addr;
1445     unsigned char ch;
1446     int ret;
1447 
1448     switch (run->riscv_sbi.function_id) {
1449     case SBI_EXT_DBCN_CONSOLE_READ:
1450     case SBI_EXT_DBCN_CONSOLE_WRITE:
1451         num_bytes = run->riscv_sbi.args[0];
1452 
1453         if (num_bytes == 0) {
1454             run->riscv_sbi.ret[0] = SBI_SUCCESS;
1455             run->riscv_sbi.ret[1] = 0;
1456             break;
1457         }
1458 
1459         addr = run->riscv_sbi.args[1];
1460 
1461         /*
1462          * Handle the case where a 32 bit CPU is running in a
1463          * 64 bit addressing env.
1464          */
1465         if (riscv_cpu_mxl(&cpu->env) == MXL_RV32) {
1466             addr |= (uint64_t)run->riscv_sbi.args[2] << 32;
1467         }
1468 
1469         buf = g_malloc0(num_bytes);
1470 
1471         if (run->riscv_sbi.function_id == SBI_EXT_DBCN_CONSOLE_READ) {
1472             ret = qemu_chr_fe_read_all(serial_hd(0)->be, buf, num_bytes);
1473             if (ret < 0) {
1474                 error_report("SBI_EXT_DBCN_CONSOLE_READ: error when "
1475                              "reading chardev");
1476                 exit(1);
1477             }
1478 
1479             cpu_physical_memory_write(addr, buf, ret);
1480         } else {
1481             cpu_physical_memory_read(addr, buf, num_bytes);
1482 
1483             ret = qemu_chr_fe_write_all(serial_hd(0)->be, buf, num_bytes);
1484             if (ret < 0) {
1485                 error_report("SBI_EXT_DBCN_CONSOLE_WRITE: error when "
1486                              "writing chardev");
1487                 exit(1);
1488             }
1489         }
1490 
1491         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1492         run->riscv_sbi.ret[1] = ret;
1493         break;
1494     case SBI_EXT_DBCN_CONSOLE_WRITE_BYTE:
1495         ch = run->riscv_sbi.args[0];
1496         ret = qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1497 
1498         if (ret < 0) {
1499             error_report("SBI_EXT_DBCN_CONSOLE_WRITE_BYTE: error when "
1500                          "writing chardev");
1501             exit(1);
1502         }
1503 
1504         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1505         run->riscv_sbi.ret[1] = 0;
1506         break;
1507     default:
1508         run->riscv_sbi.ret[0] = SBI_ERR_NOT_SUPPORTED;
1509     }
1510 }
1511 
1512 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1513 {
1514     int ret = 0;
1515     unsigned char ch;
1516     switch (run->riscv_sbi.extension_id) {
1517     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1518         ch = run->riscv_sbi.args[0];
1519         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1520         break;
1521     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1522         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1523         if (ret == sizeof(ch)) {
1524             run->riscv_sbi.ret[0] = ch;
1525         } else {
1526             run->riscv_sbi.ret[0] = -1;
1527         }
1528         ret = 0;
1529         break;
1530     case SBI_EXT_DBCN:
1531         kvm_riscv_handle_sbi_dbcn(cs, run);
1532         break;
1533     default:
1534         qemu_log_mask(LOG_UNIMP,
1535                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1536                       __func__, run->riscv_sbi.extension_id);
1537         ret = -1;
1538         break;
1539     }
1540     return ret;
1541 }
1542 
1543 static int kvm_riscv_handle_csr(CPUState *cs, struct kvm_run *run)
1544 {
1545     target_ulong csr_num = run->riscv_csr.csr_num;
1546     target_ulong new_value = run->riscv_csr.new_value;
1547     target_ulong write_mask = run->riscv_csr.write_mask;
1548     int ret = 0;
1549 
1550     switch (csr_num) {
1551     case CSR_SEED:
1552         run->riscv_csr.ret_value = riscv_new_csr_seed(new_value, write_mask);
1553         break;
1554     default:
1555         qemu_log_mask(LOG_UNIMP,
1556                       "%s: un-handled CSR EXIT for CSR %lx\n",
1557                       __func__, csr_num);
1558         ret = -1;
1559         break;
1560     }
1561 
1562     return ret;
1563 }
1564 
1565 static bool kvm_riscv_handle_debug(CPUState *cs)
1566 {
1567     RISCVCPU *cpu = RISCV_CPU(cs);
1568     CPURISCVState *env = &cpu->env;
1569 
1570     /* Ensure PC is synchronised */
1571     kvm_cpu_synchronize_state(cs);
1572 
1573     if (kvm_find_sw_breakpoint(cs, env->pc)) {
1574         return true;
1575     }
1576 
1577     return false;
1578 }
1579 
1580 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1581 {
1582     int ret = 0;
1583     switch (run->exit_reason) {
1584     case KVM_EXIT_RISCV_SBI:
1585         ret = kvm_riscv_handle_sbi(cs, run);
1586         break;
1587     case KVM_EXIT_RISCV_CSR:
1588         ret = kvm_riscv_handle_csr(cs, run);
1589         break;
1590     case KVM_EXIT_DEBUG:
1591         if (kvm_riscv_handle_debug(cs)) {
1592             ret = EXCP_DEBUG;
1593         }
1594         break;
1595     default:
1596         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1597                       __func__, run->exit_reason);
1598         ret = -1;
1599         break;
1600     }
1601     return ret;
1602 }
1603 
1604 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1605 {
1606     CPURISCVState *env = &cpu->env;
1607     int i;
1608 
1609     if (!kvm_enabled()) {
1610         return;
1611     }
1612     for (i = 0; i < 32; i++) {
1613         env->gpr[i] = 0;
1614     }
1615     env->pc = cpu->env.kernel_addr;
1616     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1617     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1618     env->satp = 0;
1619     env->mie = 0;
1620     env->stvec = 0;
1621     env->sscratch = 0;
1622     env->sepc = 0;
1623     env->scause = 0;
1624     env->stval = 0;
1625     env->mip = 0;
1626 }
1627 
1628 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1629 {
1630     int ret;
1631     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1632 
1633     if (irq != IRQ_S_EXT) {
1634         perror("kvm riscv set irq != IRQ_S_EXT\n");
1635         abort();
1636     }
1637 
1638     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1639     if (ret < 0) {
1640         perror("Set irq failed");
1641         abort();
1642     }
1643 }
1644 
1645 static int aia_mode;
1646 
1647 static const char *kvm_aia_mode_str(uint64_t mode)
1648 {
1649     switch (mode) {
1650     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1651         return "emul";
1652     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1653         return "hwaccel";
1654     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1655     default:
1656         return "auto";
1657     };
1658 }
1659 
1660 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1661 {
1662     return g_strdup(kvm_aia_mode_str(aia_mode));
1663 }
1664 
1665 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1666 {
1667     if (!strcmp(val, "emul")) {
1668         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1669     } else if (!strcmp(val, "hwaccel")) {
1670         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1671     } else if (!strcmp(val, "auto")) {
1672         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1673     } else {
1674         error_setg(errp, "Invalid KVM AIA mode");
1675         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1676     }
1677 }
1678 
1679 void kvm_arch_accel_class_init(ObjectClass *oc)
1680 {
1681     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1682                                   riscv_set_kvm_aia);
1683     object_class_property_set_description(oc, "riscv-aia",
1684         "Set KVM AIA mode. Valid values are 'emul', 'hwaccel' and 'auto'. "
1685         "Changing KVM AIA modes relies on host support. Defaults to 'auto' "
1686         "if the host supports it");
1687     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1688                                     "auto");
1689 }
1690 
1691 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1692                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1693                           uint64_t aplic_base, uint64_t imsic_base,
1694                           uint64_t guest_num)
1695 {
1696     int ret, i;
1697     int aia_fd = -1;
1698     uint64_t default_aia_mode;
1699     uint64_t socket_count = riscv_socket_count(machine);
1700     uint64_t max_hart_per_socket = 0;
1701     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1702     uint64_t socket_bits, hart_bits, guest_bits;
1703     uint64_t max_group_id;
1704 
1705     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1706 
1707     if (aia_fd < 0) {
1708         error_report("Unable to create in-kernel irqchip");
1709         exit(1);
1710     }
1711 
1712     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1713                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1714                             &default_aia_mode, false, NULL);
1715     if (ret < 0) {
1716         error_report("KVM AIA: failed to get current KVM AIA mode");
1717         exit(1);
1718     }
1719 
1720     if (default_aia_mode != aia_mode) {
1721         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1722                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1723                                 &aia_mode, true, NULL);
1724         if (ret < 0) {
1725             warn_report("KVM AIA: failed to set KVM AIA mode '%s', using "
1726                         "default host mode '%s'",
1727                         kvm_aia_mode_str(aia_mode),
1728                         kvm_aia_mode_str(default_aia_mode));
1729 
1730             /* failed to change AIA mode, use default */
1731             aia_mode = default_aia_mode;
1732         }
1733     }
1734 
1735     /*
1736      * Skip APLIC creation in KVM if we're running split mode.
1737      * This is done by leaving KVM_DEV_RISCV_AIA_CONFIG_SRCS
1738      * unset. We can also skip KVM_DEV_RISCV_AIA_ADDR_APLIC
1739      * since KVM won't be using it.
1740      */
1741     if (!kvm_kernel_irqchip_split()) {
1742         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1743                                 KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1744                                 &aia_irq_num, true, NULL);
1745         if (ret < 0) {
1746             error_report("KVM AIA: failed to set number of input irq lines");
1747             exit(1);
1748         }
1749 
1750         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1751                                 KVM_DEV_RISCV_AIA_ADDR_APLIC,
1752                                 &aplic_base, true, NULL);
1753         if (ret < 0) {
1754             error_report("KVM AIA: failed to set the base address of APLIC");
1755             exit(1);
1756         }
1757      }
1758 
1759     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1760                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1761                             &aia_msi_num, true, NULL);
1762     if (ret < 0) {
1763         error_report("KVM AIA: failed to set number of msi");
1764         exit(1);
1765     }
1766 
1767 
1768     if (socket_count > 1) {
1769         max_group_id = socket_count - 1;
1770         socket_bits = find_last_bit(&max_group_id, BITS_PER_LONG) + 1;
1771         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1772                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1773                                 &socket_bits, true, NULL);
1774         if (ret < 0) {
1775             error_report("KVM AIA: failed to set group_bits");
1776             exit(1);
1777         }
1778 
1779         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1780                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1781                                 &group_shift, true, NULL);
1782         if (ret < 0) {
1783             error_report("KVM AIA: failed to set group_shift");
1784             exit(1);
1785         }
1786     }
1787 
1788     guest_bits = guest_num == 0 ? 0 :
1789                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1790     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1791                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1792                             &guest_bits, true, NULL);
1793     if (ret < 0) {
1794         error_report("KVM AIA: failed to set guest_bits");
1795         exit(1);
1796     }
1797 
1798     for (socket = 0; socket < socket_count; socket++) {
1799         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1800         hart_count = riscv_socket_hart_count(machine, socket);
1801         base_hart = riscv_socket_first_hartid(machine, socket);
1802 
1803         if (max_hart_per_socket < hart_count) {
1804             max_hart_per_socket = hart_count;
1805         }
1806 
1807         for (i = 0; i < hart_count; i++) {
1808             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1809             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1810                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1811                                     &imsic_addr, true, NULL);
1812             if (ret < 0) {
1813                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1814                 exit(1);
1815             }
1816         }
1817     }
1818 
1819 
1820     if (max_hart_per_socket > 1) {
1821         max_hart_per_socket--;
1822         hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1823     } else {
1824         hart_bits = 0;
1825     }
1826 
1827     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1828                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1829                             &hart_bits, true, NULL);
1830     if (ret < 0) {
1831         error_report("KVM AIA: failed to set hart_bits");
1832         exit(1);
1833     }
1834 
1835     if (kvm_has_gsi_routing()) {
1836         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1837             /* KVM AIA only has one APLIC instance */
1838             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1839         }
1840         kvm_gsi_routing_allowed = true;
1841         kvm_irqchip_commit_routes(kvm_state);
1842     }
1843 
1844     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1845                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1846                             NULL, true, NULL);
1847     if (ret < 0) {
1848         error_report("KVM AIA: initialized fail");
1849         exit(1);
1850     }
1851 
1852     kvm_msi_via_irqfd_allowed = true;
1853 }
1854 
1855 static void kvm_cpu_instance_init(CPUState *cs)
1856 {
1857     Object *obj = OBJECT(RISCV_CPU(cs));
1858 
1859     riscv_init_kvm_registers(obj);
1860 
1861     kvm_riscv_add_cpu_user_properties(obj);
1862 }
1863 
1864 /*
1865  * We'll get here via the following path:
1866  *
1867  * riscv_cpu_realize()
1868  *   -> cpu_exec_realizefn()
1869  *      -> kvm_cpu_realize() (via accel_cpu_common_realize())
1870  */
1871 static bool kvm_cpu_realize(CPUState *cs, Error **errp)
1872 {
1873     RISCVCPU *cpu = RISCV_CPU(cs);
1874     int ret;
1875 
1876     if (riscv_has_ext(&cpu->env, RVV)) {
1877         ret = prctl(PR_RISCV_V_SET_CONTROL, PR_RISCV_V_VSTATE_CTRL_ON);
1878         if (ret) {
1879             error_setg(errp, "Error in prctl PR_RISCV_V_SET_CONTROL, code: %s",
1880                        strerrorname_np(errno));
1881             return false;
1882         }
1883     }
1884 
1885    return true;
1886 }
1887 
1888 void riscv_kvm_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
1889 {
1890     CPURISCVState *env = &cpu->env;
1891     KVMScratchCPU kvmcpu;
1892     struct kvm_one_reg reg;
1893     uint64_t val;
1894     int ret;
1895 
1896     /* short-circuit without spinning the scratch CPU */
1897     if (!cpu->cfg.ext_zicbom && !cpu->cfg.ext_zicboz &&
1898         !riscv_has_ext(env, RVV)) {
1899         return;
1900     }
1901 
1902     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1903         error_setg(errp, "Unable to create scratch KVM cpu");
1904         return;
1905     }
1906 
1907     if (cpu->cfg.ext_zicbom &&
1908         riscv_cpu_option_set(kvm_cbom_blocksize.name)) {
1909 
1910         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1911                                         kvm_cbom_blocksize.kvm_reg_id);
1912         reg.addr = (uint64_t)&val;
1913         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1914         if (ret != 0) {
1915             error_setg(errp, "Unable to read cbom_blocksize, error %d", errno);
1916             return;
1917         }
1918 
1919         if (cpu->cfg.cbom_blocksize != val) {
1920             error_setg(errp, "Unable to set cbom_blocksize to a different "
1921                        "value than the host (%lu)", val);
1922             return;
1923         }
1924     }
1925 
1926     if (cpu->cfg.ext_zicboz &&
1927         riscv_cpu_option_set(kvm_cboz_blocksize.name)) {
1928 
1929         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1930                                         kvm_cboz_blocksize.kvm_reg_id);
1931         reg.addr = (uint64_t)&val;
1932         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1933         if (ret != 0) {
1934             error_setg(errp, "Unable to read cboz_blocksize, error %d", errno);
1935             return;
1936         }
1937 
1938         if (cpu->cfg.cboz_blocksize != val) {
1939             error_setg(errp, "Unable to set cboz_blocksize to a different "
1940                        "value than the host (%lu)", val);
1941             return;
1942         }
1943     }
1944 
1945     /* Users are setting vlen, not vlenb */
1946     if (riscv_has_ext(env, RVV) && riscv_cpu_option_set("vlen")) {
1947         if (!kvm_v_vlenb.supported) {
1948             error_setg(errp, "Unable to set 'vlenb': register not supported");
1949             return;
1950         }
1951 
1952         reg.id = kvm_v_vlenb.kvm_reg_id;
1953         reg.addr = (uint64_t)&val;
1954         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1955         if (ret != 0) {
1956             error_setg(errp, "Unable to read vlenb register, error %d", errno);
1957             return;
1958         }
1959 
1960         if (cpu->cfg.vlenb != val) {
1961             error_setg(errp, "Unable to set 'vlen' to a different "
1962                        "value than the host (%lu)", val * 8);
1963             return;
1964         }
1965     }
1966 
1967     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1968 }
1969 
1970 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1971 {
1972     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1973 
1974     acc->cpu_instance_init = kvm_cpu_instance_init;
1975     acc->cpu_target_realize = kvm_cpu_realize;
1976 }
1977 
1978 static const TypeInfo kvm_cpu_accel_type_info = {
1979     .name = ACCEL_CPU_NAME("kvm"),
1980 
1981     .parent = TYPE_ACCEL_CPU,
1982     .class_init = kvm_cpu_accel_class_init,
1983     .abstract = true,
1984 };
1985 static void kvm_cpu_accel_register_types(void)
1986 {
1987     type_register_static(&kvm_cpu_accel_type_info);
1988 }
1989 type_init(kvm_cpu_accel_register_types);
1990 
1991 static void riscv_host_cpu_class_init(ObjectClass *c, void *data)
1992 {
1993     RISCVCPUClass *mcc = RISCV_CPU_CLASS(c);
1994 
1995 #if defined(TARGET_RISCV32)
1996     mcc->misa_mxl_max = MXL_RV32;
1997 #elif defined(TARGET_RISCV64)
1998     mcc->misa_mxl_max = MXL_RV64;
1999 #endif
2000 }
2001 
2002 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
2003     {
2004         .name = TYPE_RISCV_CPU_HOST,
2005         .parent = TYPE_RISCV_CPU,
2006         .class_init = riscv_host_cpu_class_init,
2007     }
2008 };
2009 
2010 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
2011 
2012 static const uint32_t ebreak_insn = 0x00100073;
2013 static const uint16_t c_ebreak_insn = 0x9002;
2014 
2015 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2016 {
2017     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 0)) {
2018         return -EINVAL;
2019     }
2020 
2021     if ((bp->saved_insn & 0x3) == 0x3) {
2022         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 0)
2023             || cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak_insn, 4, 1)) {
2024             return -EINVAL;
2025         }
2026     } else {
2027         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak_insn, 2, 1)) {
2028             return -EINVAL;
2029         }
2030     }
2031 
2032     return 0;
2033 }
2034 
2035 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2036 {
2037     uint32_t ebreak;
2038     uint16_t c_ebreak;
2039 
2040     if ((bp->saved_insn & 0x3) == 0x3) {
2041         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak, 4, 0) ||
2042             ebreak != ebreak_insn ||
2043             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 1)) {
2044             return -EINVAL;
2045         }
2046     } else {
2047         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak, 2, 0) ||
2048             c_ebreak != c_ebreak_insn ||
2049             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 1)) {
2050             return -EINVAL;
2051         }
2052     }
2053 
2054     return 0;
2055 }
2056 
2057 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type)
2058 {
2059     /* TODO; To be implemented later. */
2060     return -EINVAL;
2061 }
2062 
2063 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type)
2064 {
2065     /* TODO; To be implemented later. */
2066     return -EINVAL;
2067 }
2068 
2069 void kvm_arch_remove_all_hw_breakpoints(void)
2070 {
2071     /* TODO; To be implemented later. */
2072 }
2073 
2074 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
2075 {
2076     if (kvm_sw_breakpoints_active(cs)) {
2077         dbg->control |= KVM_GUESTDBG_ENABLE;
2078     }
2079 }
2080