xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision a1e61fc44b1a5fdad08206cbd7f015d1cc146713)
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("smnpm", ext_smnpm, KVM_RISCV_ISA_EXT_SMNPM),
325     KVM_EXT_CFG("smstateen", ext_smstateen, KVM_RISCV_ISA_EXT_SMSTATEEN),
326     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
327     KVM_EXT_CFG("sscofpmf", ext_sscofpmf, KVM_RISCV_ISA_EXT_SSCOFPMF),
328     KVM_EXT_CFG("ssnpm", ext_ssnpm, KVM_RISCV_ISA_EXT_SSNPM),
329     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
330     KVM_EXT_CFG("svade", ext_svade, KVM_RISCV_ISA_EXT_SVADE),
331     KVM_EXT_CFG("svadu", ext_svadu, KVM_RISCV_ISA_EXT_SVADU),
332     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
333     KVM_EXT_CFG("svnapot", ext_svnapot, KVM_RISCV_ISA_EXT_SVNAPOT),
334     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
335     KVM_EXT_CFG("svvptc", ext_svvptc, KVM_RISCV_ISA_EXT_SVVPTC),
336 };
337 
338 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
339 {
340     return (void *)&cpu->cfg + kvmcfg->offset;
341 }
342 
343 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
344                             uint32_t val)
345 {
346     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
347 
348     *ext_enabled = val;
349 }
350 
351 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
352                                 KVMCPUConfig *multi_ext)
353 {
354     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
355 
356     return *ext_enabled;
357 }
358 
359 static void kvm_cpu_get_multi_ext_cfg(Object *obj, Visitor *v,
360                                       const char *name,
361                                       void *opaque, Error **errp)
362 {
363     KVMCPUConfig *multi_ext_cfg = opaque;
364     RISCVCPU *cpu = RISCV_CPU(obj);
365     bool value = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
366 
367     visit_type_bool(v, name, &value, errp);
368 }
369 
370 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
371                                       const char *name,
372                                       void *opaque, Error **errp)
373 {
374     KVMCPUConfig *multi_ext_cfg = opaque;
375     RISCVCPU *cpu = RISCV_CPU(obj);
376     bool value, host_val;
377 
378     if (!visit_type_bool(v, name, &value, errp)) {
379         return;
380     }
381 
382     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
383 
384     /*
385      * Ignore if the user is setting the same value
386      * as the host.
387      */
388     if (value == host_val) {
389         return;
390     }
391 
392     if (!multi_ext_cfg->supported) {
393         /*
394          * Error out if the user is trying to enable an
395          * extension that KVM doesn't support. Ignore
396          * option otherwise.
397          */
398         if (value) {
399             error_setg(errp, "KVM does not support disabling extension %s",
400                        multi_ext_cfg->name);
401         }
402 
403         return;
404     }
405 
406     multi_ext_cfg->user_set = true;
407     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
408 }
409 
410 static KVMCPUConfig kvm_cbom_blocksize = {
411     .name = "cbom_blocksize",
412     .offset = CPU_CFG_OFFSET(cbom_blocksize),
413     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
414 };
415 
416 static KVMCPUConfig kvm_cboz_blocksize = {
417     .name = "cboz_blocksize",
418     .offset = CPU_CFG_OFFSET(cboz_blocksize),
419     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
420 };
421 
422 static KVMCPUConfig kvm_v_vlenb = {
423     .name = "vlenb",
424     .offset = CPU_CFG_OFFSET(vlenb),
425     .kvm_reg_id =  KVM_REG_RISCV | KVM_REG_SIZE_U64 | KVM_REG_RISCV_VECTOR |
426                    KVM_REG_RISCV_VECTOR_CSR_REG(vlenb)
427 };
428 
429 static KVMCPUConfig kvm_sbi_dbcn = {
430     .name = "sbi_dbcn",
431     .kvm_reg_id = KVM_REG_RISCV | KVM_REG_SIZE_U64 |
432                   KVM_REG_RISCV_SBI_EXT | KVM_RISCV_SBI_EXT_DBCN
433 };
434 
435 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
436 {
437     CPURISCVState *env = &cpu->env;
438     uint64_t id, reg;
439     int i, ret;
440 
441     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
442         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
443 
444         if (!multi_ext_cfg->user_set) {
445             continue;
446         }
447 
448         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
449                                     multi_ext_cfg->kvm_reg_id);
450         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
451         ret = kvm_set_one_reg(cs, id, &reg);
452         if (ret != 0) {
453             if (!reg && ret == -EINVAL) {
454                 warn_report("KVM cannot disable extension %s",
455                             multi_ext_cfg->name);
456             } else {
457                 error_report("Unable to enable extension %s in KVM, error %d",
458                              multi_ext_cfg->name, ret);
459                 exit(EXIT_FAILURE);
460             }
461         }
462     }
463 }
464 
465 static void cpu_get_cfg_unavailable(Object *obj, Visitor *v,
466                                     const char *name,
467                                     void *opaque, Error **errp)
468 {
469     bool value = false;
470 
471     visit_type_bool(v, name, &value, errp);
472 }
473 
474 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
475                                     const char *name,
476                                     void *opaque, Error **errp)
477 {
478     const char *propname = opaque;
479     bool value;
480 
481     if (!visit_type_bool(v, name, &value, errp)) {
482         return;
483     }
484 
485     if (value) {
486         error_setg(errp, "'%s' is not available with KVM",
487                    propname);
488     }
489 }
490 
491 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
492 {
493     /* Check if KVM created the property already */
494     if (object_property_find(obj, prop_name)) {
495         return;
496     }
497 
498     /*
499      * Set the default to disabled for every extension
500      * unknown to KVM and error out if the user attempts
501      * to enable any of them.
502      */
503     object_property_add(obj, prop_name, "bool",
504                         cpu_get_cfg_unavailable,
505                         cpu_set_cfg_unavailable,
506                         NULL, (void *)prop_name);
507 }
508 
509 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
510                                         const RISCVCPUMultiExtConfig *array)
511 {
512     const RISCVCPUMultiExtConfig *prop;
513 
514     g_assert(array);
515 
516     for (prop = array; prop && prop->name; prop++) {
517         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
518     }
519 }
520 
521 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
522 {
523     int i;
524 
525     riscv_add_satp_mode_properties(cpu_obj);
526 
527     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
528         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
529         int bit = misa_cfg->offset;
530 
531         misa_cfg->name = riscv_get_misa_ext_name(bit);
532         misa_cfg->description = riscv_get_misa_ext_description(bit);
533 
534         object_property_add(cpu_obj, misa_cfg->name, "bool",
535                             kvm_cpu_get_misa_ext_cfg,
536                             kvm_cpu_set_misa_ext_cfg,
537                             NULL, misa_cfg);
538         object_property_set_description(cpu_obj, misa_cfg->name,
539                                         misa_cfg->description);
540     }
541 
542     for (i = 0; misa_bits[i] != 0; i++) {
543         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
544         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
545     }
546 
547     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
548         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
549 
550         object_property_add(cpu_obj, multi_cfg->name, "bool",
551                             kvm_cpu_get_multi_ext_cfg,
552                             kvm_cpu_set_multi_ext_cfg,
553                             NULL, multi_cfg);
554     }
555 
556     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
557     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
558     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
559 
560    /* We don't have the needed KVM support for profiles */
561     for (i = 0; riscv_profiles[i] != NULL; i++) {
562         riscv_cpu_add_kvm_unavail_prop(cpu_obj, riscv_profiles[i]->name);
563     }
564 }
565 
566 static int kvm_riscv_get_regs_core(CPUState *cs)
567 {
568     int ret = 0;
569     int i;
570     target_ulong reg;
571     CPURISCVState *env = &RISCV_CPU(cs)->env;
572 
573     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
574     if (ret) {
575         return ret;
576     }
577     env->pc = reg;
578 
579     for (i = 1; i < 32; i++) {
580         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
581         ret = kvm_get_one_reg(cs, id, &reg);
582         if (ret) {
583             return ret;
584         }
585         env->gpr[i] = reg;
586     }
587 
588     return ret;
589 }
590 
591 static int kvm_riscv_put_regs_core(CPUState *cs)
592 {
593     int ret = 0;
594     int i;
595     target_ulong reg;
596     CPURISCVState *env = &RISCV_CPU(cs)->env;
597 
598     reg = env->pc;
599     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
600     if (ret) {
601         return ret;
602     }
603 
604     for (i = 1; i < 32; i++) {
605         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
606         reg = env->gpr[i];
607         ret = kvm_set_one_reg(cs, id, &reg);
608         if (ret) {
609             return ret;
610         }
611     }
612 
613     return ret;
614 }
615 
616 static void kvm_riscv_reset_regs_csr(CPURISCVState *env)
617 {
618     env->mstatus = 0;
619     env->mie = 0;
620     env->stvec = 0;
621     env->sscratch = 0;
622     env->sepc = 0;
623     env->scause = 0;
624     env->stval = 0;
625     env->mip = 0;
626     env->satp = 0;
627 }
628 
629 static int kvm_riscv_get_regs_csr(CPUState *cs)
630 {
631     CPURISCVState *env = &RISCV_CPU(cs)->env;
632 
633     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
634     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
635     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
636     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
637     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
638     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
639     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
640     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
641     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
642 
643     return 0;
644 }
645 
646 static int kvm_riscv_put_regs_csr(CPUState *cs)
647 {
648     CPURISCVState *env = &RISCV_CPU(cs)->env;
649 
650     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
651     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
652     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
653     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
654     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
655     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
656     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
657     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
658     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
659 
660     return 0;
661 }
662 
663 static int kvm_riscv_get_regs_fp(CPUState *cs)
664 {
665     int ret = 0;
666     int i;
667     CPURISCVState *env = &RISCV_CPU(cs)->env;
668 
669     if (riscv_has_ext(env, RVD)) {
670         uint64_t reg;
671         for (i = 0; i < 32; i++) {
672             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
673             if (ret) {
674                 return ret;
675             }
676             env->fpr[i] = reg;
677         }
678         return ret;
679     }
680 
681     if (riscv_has_ext(env, RVF)) {
682         uint32_t reg;
683         for (i = 0; i < 32; i++) {
684             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
685             if (ret) {
686                 return ret;
687             }
688             env->fpr[i] = reg;
689         }
690         return ret;
691     }
692 
693     return ret;
694 }
695 
696 static int kvm_riscv_put_regs_fp(CPUState *cs)
697 {
698     int ret = 0;
699     int i;
700     CPURISCVState *env = &RISCV_CPU(cs)->env;
701 
702     if (riscv_has_ext(env, RVD)) {
703         uint64_t reg;
704         for (i = 0; i < 32; i++) {
705             reg = env->fpr[i];
706             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
707             if (ret) {
708                 return ret;
709             }
710         }
711         return ret;
712     }
713 
714     if (riscv_has_ext(env, RVF)) {
715         uint32_t reg;
716         for (i = 0; i < 32; i++) {
717             reg = env->fpr[i];
718             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
719             if (ret) {
720                 return ret;
721             }
722         }
723         return ret;
724     }
725 
726     return ret;
727 }
728 
729 static void kvm_riscv_get_regs_timer(CPUState *cs)
730 {
731     CPURISCVState *env = &RISCV_CPU(cs)->env;
732 
733     if (env->kvm_timer_dirty) {
734         return;
735     }
736 
737     KVM_RISCV_GET_TIMER(cs, time, env->kvm_timer_time);
738     KVM_RISCV_GET_TIMER(cs, compare, env->kvm_timer_compare);
739     KVM_RISCV_GET_TIMER(cs, state, env->kvm_timer_state);
740     KVM_RISCV_GET_TIMER(cs, frequency, env->kvm_timer_frequency);
741 
742     env->kvm_timer_dirty = true;
743 }
744 
745 static void kvm_riscv_put_regs_timer(CPUState *cs)
746 {
747     uint64_t reg;
748     CPURISCVState *env = &RISCV_CPU(cs)->env;
749 
750     if (!env->kvm_timer_dirty) {
751         return;
752     }
753 
754     KVM_RISCV_SET_TIMER(cs, time, env->kvm_timer_time);
755     KVM_RISCV_SET_TIMER(cs, compare, env->kvm_timer_compare);
756 
757     /*
758      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
759      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
760      * doesn't matter that adaping in QEMU now.
761      * TODO If KVM changes, adapt here.
762      */
763     if (env->kvm_timer_state) {
764         KVM_RISCV_SET_TIMER(cs, state, env->kvm_timer_state);
765     }
766 
767     /*
768      * For now, migration will not work between Hosts with different timer
769      * frequency. Therefore, we should check whether they are the same here
770      * during the migration.
771      */
772     if (migration_is_running()) {
773         KVM_RISCV_GET_TIMER(cs, frequency, reg);
774         if (reg != env->kvm_timer_frequency) {
775             error_report("Dst Hosts timer frequency != Src Hosts");
776         }
777     }
778 
779     env->kvm_timer_dirty = false;
780 }
781 
782 uint64_t kvm_riscv_get_timebase_frequency(RISCVCPU *cpu)
783 {
784     uint64_t reg;
785 
786     KVM_RISCV_GET_TIMER(CPU(cpu), frequency, reg);
787 
788     return reg;
789 }
790 
791 static int kvm_riscv_get_regs_vector(CPUState *cs)
792 {
793     RISCVCPU *cpu = RISCV_CPU(cs);
794     CPURISCVState *env = &cpu->env;
795     target_ulong reg;
796     uint64_t vreg_id;
797     int vreg_idx, ret = 0;
798 
799     if (!riscv_has_ext(env, RVV)) {
800         return 0;
801     }
802 
803     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
804     if (ret) {
805         return ret;
806     }
807     env->vstart = reg;
808 
809     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
810     if (ret) {
811         return ret;
812     }
813     env->vl = reg;
814 
815     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
816     if (ret) {
817         return ret;
818     }
819     env->vtype = reg;
820 
821     if (kvm_v_vlenb.supported) {
822         ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
823         if (ret) {
824             return ret;
825         }
826         cpu->cfg.vlenb = reg;
827 
828         for (int i = 0; i < 32; i++) {
829             /*
830              * vreg[] is statically allocated using RV_VLEN_MAX.
831              * Use it instead of vlenb to calculate vreg_idx for
832              * simplicity.
833              */
834             vreg_idx = i * RV_VLEN_MAX / 64;
835             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
836 
837             ret = kvm_get_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
838             if (ret) {
839                 return ret;
840             }
841         }
842     }
843 
844     return 0;
845 }
846 
847 static int kvm_riscv_put_regs_vector(CPUState *cs)
848 {
849     RISCVCPU *cpu = RISCV_CPU(cs);
850     CPURISCVState *env = &cpu->env;
851     target_ulong reg;
852     uint64_t vreg_id;
853     int vreg_idx, ret = 0;
854 
855     if (!riscv_has_ext(env, RVV)) {
856         return 0;
857     }
858 
859     reg = env->vstart;
860     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
861     if (ret) {
862         return ret;
863     }
864 
865     reg = env->vl;
866     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
867     if (ret) {
868         return ret;
869     }
870 
871     reg = env->vtype;
872     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
873     if (ret) {
874         return ret;
875     }
876 
877     if (kvm_v_vlenb.supported) {
878         reg = cpu->cfg.vlenb;
879         ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
880 
881         for (int i = 0; i < 32; i++) {
882             /*
883              * vreg[] is statically allocated using RV_VLEN_MAX.
884              * Use it instead of vlenb to calculate vreg_idx for
885              * simplicity.
886              */
887             vreg_idx = i * RV_VLEN_MAX / 64;
888             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
889 
890             ret = kvm_set_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
891             if (ret) {
892                 return ret;
893             }
894         }
895     }
896 
897     return ret;
898 }
899 
900 typedef struct KVMScratchCPU {
901     int kvmfd;
902     int vmfd;
903     int cpufd;
904 } KVMScratchCPU;
905 
906 /*
907  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
908  * from target/arm/kvm.c.
909  */
910 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
911 {
912     int kvmfd = -1, vmfd = -1, cpufd = -1;
913 
914     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
915     if (kvmfd < 0) {
916         goto err;
917     }
918     do {
919         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
920     } while (vmfd == -1 && errno == EINTR);
921     if (vmfd < 0) {
922         goto err;
923     }
924     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
925     if (cpufd < 0) {
926         goto err;
927     }
928 
929     scratch->kvmfd =  kvmfd;
930     scratch->vmfd = vmfd;
931     scratch->cpufd = cpufd;
932 
933     return true;
934 
935  err:
936     if (cpufd >= 0) {
937         close(cpufd);
938     }
939     if (vmfd >= 0) {
940         close(vmfd);
941     }
942     if (kvmfd >= 0) {
943         close(kvmfd);
944     }
945 
946     return false;
947 }
948 
949 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
950 {
951     close(scratch->cpufd);
952     close(scratch->vmfd);
953     close(scratch->kvmfd);
954 }
955 
956 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
957 {
958     CPURISCVState *env = &cpu->env;
959     struct kvm_one_reg reg;
960     int ret;
961 
962     reg.id = RISCV_CONFIG_REG(env, mvendorid);
963     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
964     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
965     if (ret != 0) {
966         error_report("Unable to retrieve mvendorid from host, error %d", ret);
967     }
968 
969     reg.id = RISCV_CONFIG_REG(env, marchid);
970     reg.addr = (uint64_t)&cpu->cfg.marchid;
971     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
972     if (ret != 0) {
973         error_report("Unable to retrieve marchid from host, error %d", ret);
974     }
975 
976     reg.id = RISCV_CONFIG_REG(env, mimpid);
977     reg.addr = (uint64_t)&cpu->cfg.mimpid;
978     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
979     if (ret != 0) {
980         error_report("Unable to retrieve mimpid from host, error %d", ret);
981     }
982 }
983 
984 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
985                                          KVMScratchCPU *kvmcpu)
986 {
987     CPURISCVState *env = &cpu->env;
988     struct kvm_one_reg reg;
989     int ret;
990 
991     reg.id = RISCV_CONFIG_REG(env, isa);
992     reg.addr = (uint64_t)&env->misa_ext_mask;
993     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
994 
995     if (ret) {
996         error_report("Unable to fetch ISA register from KVM, "
997                      "error %d", ret);
998         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
999         exit(EXIT_FAILURE);
1000     }
1001 
1002     env->misa_ext = env->misa_ext_mask;
1003 }
1004 
1005 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
1006                                          KVMCPUConfig *cbomz_cfg)
1007 {
1008     CPURISCVState *env = &cpu->env;
1009     struct kvm_one_reg reg;
1010     int ret;
1011 
1012     reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1013                                     cbomz_cfg->kvm_reg_id);
1014     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
1015     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1016     if (ret != 0) {
1017         error_report("Unable to read KVM reg %s, error %d",
1018                      cbomz_cfg->name, ret);
1019         exit(EXIT_FAILURE);
1020     }
1021 }
1022 
1023 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
1024                                            KVMScratchCPU *kvmcpu)
1025 {
1026     CPURISCVState *env = &cpu->env;
1027     uint64_t val;
1028     int i, ret;
1029 
1030     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1031         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1032         struct kvm_one_reg reg;
1033 
1034         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
1035                                         multi_ext_cfg->kvm_reg_id);
1036         reg.addr = (uint64_t)&val;
1037         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1038         if (ret != 0) {
1039             if (errno == EINVAL) {
1040                 /* Silently default to 'false' if KVM does not support it. */
1041                 multi_ext_cfg->supported = false;
1042                 val = false;
1043             } else {
1044                 error_report("Unable to read ISA_EXT KVM register %s: %s",
1045                              multi_ext_cfg->name, strerror(errno));
1046                 exit(EXIT_FAILURE);
1047             }
1048         } else {
1049             multi_ext_cfg->supported = true;
1050         }
1051 
1052         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1053     }
1054 
1055     if (cpu->cfg.ext_zicbom) {
1056         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1057     }
1058 
1059     if (cpu->cfg.ext_zicboz) {
1060         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1061     }
1062 }
1063 
1064 static int uint64_cmp(const void *a, const void *b)
1065 {
1066     uint64_t val1 = *(const uint64_t *)a;
1067     uint64_t val2 = *(const uint64_t *)b;
1068 
1069     if (val1 < val2) {
1070         return -1;
1071     }
1072 
1073     if (val1 > val2) {
1074         return 1;
1075     }
1076 
1077     return 0;
1078 }
1079 
1080 static void kvm_riscv_check_sbi_dbcn_support(RISCVCPU *cpu,
1081                                              KVMScratchCPU *kvmcpu,
1082                                              struct kvm_reg_list *reglist)
1083 {
1084     struct kvm_reg_list *reg_search;
1085 
1086     reg_search = bsearch(&kvm_sbi_dbcn.kvm_reg_id, reglist->reg, reglist->n,
1087                          sizeof(uint64_t), uint64_cmp);
1088 
1089     if (reg_search) {
1090         kvm_sbi_dbcn.supported = true;
1091     }
1092 }
1093 
1094 static void kvm_riscv_read_vlenb(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
1095                                  struct kvm_reg_list *reglist)
1096 {
1097     struct kvm_one_reg reg;
1098     struct kvm_reg_list *reg_search;
1099     uint64_t val;
1100     int ret;
1101 
1102     reg_search = bsearch(&kvm_v_vlenb.kvm_reg_id, reglist->reg, reglist->n,
1103                          sizeof(uint64_t), uint64_cmp);
1104 
1105     if (reg_search) {
1106         reg.id = kvm_v_vlenb.kvm_reg_id;
1107         reg.addr = (uint64_t)&val;
1108 
1109         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1110         if (ret != 0) {
1111             error_report("Unable to read vlenb register, error code: %d",
1112                          errno);
1113             exit(EXIT_FAILURE);
1114         }
1115 
1116         kvm_v_vlenb.supported = true;
1117         cpu->cfg.vlenb = val;
1118     }
1119 }
1120 
1121 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
1122 {
1123     KVMCPUConfig *multi_ext_cfg;
1124     struct kvm_one_reg reg;
1125     struct kvm_reg_list rl_struct;
1126     struct kvm_reg_list *reglist;
1127     uint64_t val, reg_id, *reg_search;
1128     int i, ret;
1129 
1130     rl_struct.n = 0;
1131     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
1132 
1133     /*
1134      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
1135      * (EINVAL). Use read_legacy() in this case.
1136      */
1137     if (errno == EINVAL) {
1138         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
1139     } else if (errno != E2BIG) {
1140         /*
1141          * E2BIG is an expected error message for the API since we
1142          * don't know the number of registers. The right amount will
1143          * be written in rl_struct.n.
1144          *
1145          * Error out if we get any other errno.
1146          */
1147         error_report("Error when accessing get-reg-list: %s",
1148                      strerror(errno));
1149         exit(EXIT_FAILURE);
1150     }
1151 
1152     reglist = g_malloc(sizeof(struct kvm_reg_list) +
1153                        rl_struct.n * sizeof(uint64_t));
1154     reglist->n = rl_struct.n;
1155     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
1156     if (ret) {
1157         error_report("Error when reading KVM_GET_REG_LIST: %s",
1158                      strerror(errno));
1159         exit(EXIT_FAILURE);
1160     }
1161 
1162     /* sort reglist to use bsearch() */
1163     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
1164 
1165     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1166         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1167         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
1168                                         multi_ext_cfg->kvm_reg_id);
1169         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
1170                              sizeof(uint64_t), uint64_cmp);
1171         if (!reg_search) {
1172             continue;
1173         }
1174 
1175         reg.id = reg_id;
1176         reg.addr = (uint64_t)&val;
1177         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1178         if (ret != 0) {
1179             error_report("Unable to read ISA_EXT KVM register %s: %s",
1180                          multi_ext_cfg->name, strerror(errno));
1181             exit(EXIT_FAILURE);
1182         }
1183 
1184         multi_ext_cfg->supported = true;
1185         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1186     }
1187 
1188     if (cpu->cfg.ext_zicbom) {
1189         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1190     }
1191 
1192     if (cpu->cfg.ext_zicboz) {
1193         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1194     }
1195 
1196     if (riscv_has_ext(&cpu->env, RVV)) {
1197         kvm_riscv_read_vlenb(cpu, kvmcpu, reglist);
1198     }
1199 
1200     kvm_riscv_check_sbi_dbcn_support(cpu, kvmcpu, reglist);
1201 }
1202 
1203 static void riscv_init_kvm_registers(Object *cpu_obj)
1204 {
1205     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
1206     KVMScratchCPU kvmcpu;
1207 
1208     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1209         return;
1210     }
1211 
1212     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
1213     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
1214     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
1215 
1216     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1217 }
1218 
1219 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
1220     KVM_CAP_LAST_INFO
1221 };
1222 
1223 int kvm_arch_get_registers(CPUState *cs, Error **errp)
1224 {
1225     int ret = 0;
1226 
1227     ret = kvm_riscv_get_regs_core(cs);
1228     if (ret) {
1229         return ret;
1230     }
1231 
1232     ret = kvm_riscv_get_regs_csr(cs);
1233     if (ret) {
1234         return ret;
1235     }
1236 
1237     ret = kvm_riscv_get_regs_fp(cs);
1238     if (ret) {
1239         return ret;
1240     }
1241 
1242     ret = kvm_riscv_get_regs_vector(cs);
1243     if (ret) {
1244         return ret;
1245     }
1246 
1247     return ret;
1248 }
1249 
1250 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1251 {
1252     if (cap_has_mp_state) {
1253         struct kvm_mp_state mp_state = {
1254             .mp_state = state
1255         };
1256 
1257         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1258         if (ret) {
1259             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1260                     __func__, ret, strerror(-ret));
1261             return -1;
1262         }
1263     }
1264 
1265     return 0;
1266 }
1267 
1268 int kvm_arch_put_registers(CPUState *cs, int level, Error **errp)
1269 {
1270     int ret = 0;
1271 
1272     ret = kvm_riscv_put_regs_core(cs);
1273     if (ret) {
1274         return ret;
1275     }
1276 
1277     ret = kvm_riscv_put_regs_csr(cs);
1278     if (ret) {
1279         return ret;
1280     }
1281 
1282     ret = kvm_riscv_put_regs_fp(cs);
1283     if (ret) {
1284         return ret;
1285     }
1286 
1287     ret = kvm_riscv_put_regs_vector(cs);
1288     if (ret) {
1289         return ret;
1290     }
1291 
1292     if (KVM_PUT_RESET_STATE == level) {
1293         RISCVCPU *cpu = RISCV_CPU(cs);
1294         if (cs->cpu_index == 0) {
1295             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1296         } else {
1297             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1298         }
1299         if (ret) {
1300             return ret;
1301         }
1302     }
1303 
1304     return ret;
1305 }
1306 
1307 int kvm_arch_release_virq_post(int virq)
1308 {
1309     return 0;
1310 }
1311 
1312 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1313                              uint64_t address, uint32_t data, PCIDevice *dev)
1314 {
1315     return 0;
1316 }
1317 
1318 int kvm_arch_destroy_vcpu(CPUState *cs)
1319 {
1320     return 0;
1321 }
1322 
1323 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1324 {
1325     return cpu->cpu_index;
1326 }
1327 
1328 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1329                                       RunState state)
1330 {
1331     CPUState *cs = opaque;
1332 
1333     if (running) {
1334         kvm_riscv_put_regs_timer(cs);
1335     } else {
1336         kvm_riscv_get_regs_timer(cs);
1337     }
1338 }
1339 
1340 void kvm_arch_init_irq_routing(KVMState *s)
1341 {
1342 }
1343 
1344 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1345 {
1346     CPURISCVState *env = &cpu->env;
1347     target_ulong reg;
1348     uint64_t id;
1349     int ret;
1350 
1351     id = RISCV_CONFIG_REG(env, mvendorid);
1352     /*
1353      * cfg.mvendorid is an uint32 but a target_ulong will
1354      * be written. Assign it to a target_ulong var to avoid
1355      * writing pieces of other cpu->cfg fields in the reg.
1356      */
1357     reg = cpu->cfg.mvendorid;
1358     ret = kvm_set_one_reg(cs, id, &reg);
1359     if (ret != 0) {
1360         return ret;
1361     }
1362 
1363     id = RISCV_CONFIG_REG(env, marchid);
1364     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1365     if (ret != 0) {
1366         return ret;
1367     }
1368 
1369     id = RISCV_CONFIG_REG(env, mimpid);
1370     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1371 
1372     return ret;
1373 }
1374 
1375 static int kvm_vcpu_enable_sbi_dbcn(RISCVCPU *cpu, CPUState *cs)
1376 {
1377     target_ulong reg = 1;
1378 
1379     if (!kvm_sbi_dbcn.supported) {
1380         return 0;
1381     }
1382 
1383     return kvm_set_one_reg(cs, kvm_sbi_dbcn.kvm_reg_id, &reg);
1384 }
1385 
1386 int kvm_arch_init_vcpu(CPUState *cs)
1387 {
1388     int ret = 0;
1389     RISCVCPU *cpu = RISCV_CPU(cs);
1390 
1391     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1392 
1393     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1394         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1395         if (ret != 0) {
1396             return ret;
1397         }
1398     }
1399 
1400     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1401     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1402 
1403     ret = kvm_vcpu_enable_sbi_dbcn(cpu, cs);
1404 
1405     return ret;
1406 }
1407 
1408 int kvm_arch_msi_data_to_gsi(uint32_t data)
1409 {
1410     abort();
1411 }
1412 
1413 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1414                                 int vector, PCIDevice *dev)
1415 {
1416     return 0;
1417 }
1418 
1419 int kvm_arch_get_default_type(MachineState *ms)
1420 {
1421     return 0;
1422 }
1423 
1424 int kvm_arch_init(MachineState *ms, KVMState *s)
1425 {
1426     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1427     return 0;
1428 }
1429 
1430 int kvm_arch_irqchip_create(KVMState *s)
1431 {
1432     /*
1433      * We can create the VAIA using the newer device control API.
1434      */
1435     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1436 }
1437 
1438 int kvm_arch_process_async_events(CPUState *cs)
1439 {
1440     return 0;
1441 }
1442 
1443 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1444 {
1445 }
1446 
1447 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1448 {
1449     return MEMTXATTRS_UNSPECIFIED;
1450 }
1451 
1452 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1453 {
1454     return true;
1455 }
1456 
1457 static void kvm_riscv_handle_sbi_dbcn(CPUState *cs, struct kvm_run *run)
1458 {
1459     g_autofree uint8_t *buf = NULL;
1460     RISCVCPU *cpu = RISCV_CPU(cs);
1461     target_ulong num_bytes;
1462     uint64_t addr;
1463     unsigned char ch;
1464     int ret;
1465 
1466     switch (run->riscv_sbi.function_id) {
1467     case SBI_EXT_DBCN_CONSOLE_READ:
1468     case SBI_EXT_DBCN_CONSOLE_WRITE:
1469         num_bytes = run->riscv_sbi.args[0];
1470 
1471         if (num_bytes == 0) {
1472             run->riscv_sbi.ret[0] = SBI_SUCCESS;
1473             run->riscv_sbi.ret[1] = 0;
1474             break;
1475         }
1476 
1477         addr = run->riscv_sbi.args[1];
1478 
1479         /*
1480          * Handle the case where a 32 bit CPU is running in a
1481          * 64 bit addressing env.
1482          */
1483         if (riscv_cpu_mxl(&cpu->env) == MXL_RV32) {
1484             addr |= (uint64_t)run->riscv_sbi.args[2] << 32;
1485         }
1486 
1487         buf = g_malloc0(num_bytes);
1488 
1489         if (run->riscv_sbi.function_id == SBI_EXT_DBCN_CONSOLE_READ) {
1490             ret = qemu_chr_fe_read_all(serial_hd(0)->be, buf, num_bytes);
1491             if (ret < 0) {
1492                 error_report("SBI_EXT_DBCN_CONSOLE_READ: error when "
1493                              "reading chardev");
1494                 exit(1);
1495             }
1496 
1497             cpu_physical_memory_write(addr, buf, ret);
1498         } else {
1499             cpu_physical_memory_read(addr, buf, num_bytes);
1500 
1501             ret = qemu_chr_fe_write_all(serial_hd(0)->be, buf, num_bytes);
1502             if (ret < 0) {
1503                 error_report("SBI_EXT_DBCN_CONSOLE_WRITE: error when "
1504                              "writing chardev");
1505                 exit(1);
1506             }
1507         }
1508 
1509         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1510         run->riscv_sbi.ret[1] = ret;
1511         break;
1512     case SBI_EXT_DBCN_CONSOLE_WRITE_BYTE:
1513         ch = run->riscv_sbi.args[0];
1514         ret = qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1515 
1516         if (ret < 0) {
1517             error_report("SBI_EXT_DBCN_CONSOLE_WRITE_BYTE: error when "
1518                          "writing chardev");
1519             exit(1);
1520         }
1521 
1522         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1523         run->riscv_sbi.ret[1] = 0;
1524         break;
1525     default:
1526         run->riscv_sbi.ret[0] = SBI_ERR_NOT_SUPPORTED;
1527     }
1528 }
1529 
1530 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1531 {
1532     int ret = 0;
1533     unsigned char ch;
1534     switch (run->riscv_sbi.extension_id) {
1535     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1536         ch = run->riscv_sbi.args[0];
1537         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1538         break;
1539     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1540         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1541         if (ret == sizeof(ch)) {
1542             run->riscv_sbi.ret[0] = ch;
1543         } else {
1544             run->riscv_sbi.ret[0] = -1;
1545         }
1546         ret = 0;
1547         break;
1548     case SBI_EXT_DBCN:
1549         kvm_riscv_handle_sbi_dbcn(cs, run);
1550         break;
1551     default:
1552         qemu_log_mask(LOG_UNIMP,
1553                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1554                       __func__, run->riscv_sbi.extension_id);
1555         ret = -1;
1556         break;
1557     }
1558     return ret;
1559 }
1560 
1561 static int kvm_riscv_handle_csr(CPUState *cs, struct kvm_run *run)
1562 {
1563     target_ulong csr_num = run->riscv_csr.csr_num;
1564     target_ulong new_value = run->riscv_csr.new_value;
1565     target_ulong write_mask = run->riscv_csr.write_mask;
1566     int ret = 0;
1567 
1568     switch (csr_num) {
1569     case CSR_SEED:
1570         run->riscv_csr.ret_value = riscv_new_csr_seed(new_value, write_mask);
1571         break;
1572     default:
1573         qemu_log_mask(LOG_UNIMP,
1574                       "%s: un-handled CSR EXIT for CSR %lx\n",
1575                       __func__, csr_num);
1576         ret = -1;
1577         break;
1578     }
1579 
1580     return ret;
1581 }
1582 
1583 static bool kvm_riscv_handle_debug(CPUState *cs)
1584 {
1585     RISCVCPU *cpu = RISCV_CPU(cs);
1586     CPURISCVState *env = &cpu->env;
1587 
1588     /* Ensure PC is synchronised */
1589     kvm_cpu_synchronize_state(cs);
1590 
1591     if (kvm_find_sw_breakpoint(cs, env->pc)) {
1592         return true;
1593     }
1594 
1595     return false;
1596 }
1597 
1598 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1599 {
1600     int ret = 0;
1601     switch (run->exit_reason) {
1602     case KVM_EXIT_RISCV_SBI:
1603         ret = kvm_riscv_handle_sbi(cs, run);
1604         break;
1605     case KVM_EXIT_RISCV_CSR:
1606         ret = kvm_riscv_handle_csr(cs, run);
1607         break;
1608     case KVM_EXIT_DEBUG:
1609         if (kvm_riscv_handle_debug(cs)) {
1610             ret = EXCP_DEBUG;
1611         }
1612         break;
1613     default:
1614         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1615                       __func__, run->exit_reason);
1616         ret = -1;
1617         break;
1618     }
1619     return ret;
1620 }
1621 
1622 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1623 {
1624     CPURISCVState *env = &cpu->env;
1625     int i;
1626 
1627     for (i = 0; i < 32; i++) {
1628         env->gpr[i] = 0;
1629     }
1630     env->pc = cpu->env.kernel_addr;
1631     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1632     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1633 
1634     kvm_riscv_reset_regs_csr(env);
1635 }
1636 
1637 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1638 {
1639     int ret;
1640     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1641 
1642     if (irq != IRQ_S_EXT) {
1643         perror("kvm riscv set irq != IRQ_S_EXT\n");
1644         abort();
1645     }
1646 
1647     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1648     if (ret < 0) {
1649         perror("Set irq failed");
1650         abort();
1651     }
1652 }
1653 
1654 static int aia_mode;
1655 
1656 static const char *kvm_aia_mode_str(uint64_t mode)
1657 {
1658     switch (mode) {
1659     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1660         return "emul";
1661     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1662         return "hwaccel";
1663     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1664     default:
1665         return "auto";
1666     };
1667 }
1668 
1669 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1670 {
1671     return g_strdup(kvm_aia_mode_str(aia_mode));
1672 }
1673 
1674 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1675 {
1676     if (!strcmp(val, "emul")) {
1677         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1678     } else if (!strcmp(val, "hwaccel")) {
1679         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1680     } else if (!strcmp(val, "auto")) {
1681         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1682     } else {
1683         error_setg(errp, "Invalid KVM AIA mode");
1684         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1685     }
1686 }
1687 
1688 void kvm_arch_accel_class_init(ObjectClass *oc)
1689 {
1690     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1691                                   riscv_set_kvm_aia);
1692     object_class_property_set_description(oc, "riscv-aia",
1693         "Set KVM AIA mode. Valid values are 'emul', 'hwaccel' and 'auto'. "
1694         "Changing KVM AIA modes relies on host support. Defaults to 'auto' "
1695         "if the host supports it");
1696     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1697                                     "auto");
1698 }
1699 
1700 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1701                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1702                           uint64_t aplic_base, uint64_t imsic_base,
1703                           uint64_t guest_num)
1704 {
1705     int ret, i;
1706     int aia_fd = -1;
1707     uint64_t default_aia_mode;
1708     uint64_t socket_count = riscv_socket_count(machine);
1709     uint64_t max_hart_per_socket = 0;
1710     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1711     uint64_t socket_bits, hart_bits, guest_bits;
1712     uint64_t max_group_id;
1713 
1714     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1715 
1716     if (aia_fd < 0) {
1717         error_report("Unable to create in-kernel irqchip");
1718         exit(1);
1719     }
1720 
1721     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1722                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1723                             &default_aia_mode, false, NULL);
1724     if (ret < 0) {
1725         error_report("KVM AIA: failed to get current KVM AIA mode");
1726         exit(1);
1727     }
1728 
1729     if (default_aia_mode != aia_mode) {
1730         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1731                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1732                                 &aia_mode, true, NULL);
1733         if (ret < 0) {
1734             warn_report("KVM AIA: failed to set KVM AIA mode '%s', using "
1735                         "default host mode '%s'",
1736                         kvm_aia_mode_str(aia_mode),
1737                         kvm_aia_mode_str(default_aia_mode));
1738 
1739             /* failed to change AIA mode, use default */
1740             aia_mode = default_aia_mode;
1741         }
1742     }
1743 
1744     /*
1745      * Skip APLIC creation in KVM if we're running split mode.
1746      * This is done by leaving KVM_DEV_RISCV_AIA_CONFIG_SRCS
1747      * unset. We can also skip KVM_DEV_RISCV_AIA_ADDR_APLIC
1748      * since KVM won't be using it.
1749      */
1750     if (!kvm_kernel_irqchip_split()) {
1751         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1752                                 KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1753                                 &aia_irq_num, true, NULL);
1754         if (ret < 0) {
1755             error_report("KVM AIA: failed to set number of input irq lines");
1756             exit(1);
1757         }
1758 
1759         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1760                                 KVM_DEV_RISCV_AIA_ADDR_APLIC,
1761                                 &aplic_base, true, NULL);
1762         if (ret < 0) {
1763             error_report("KVM AIA: failed to set the base address of APLIC");
1764             exit(1);
1765         }
1766      }
1767 
1768     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1769                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1770                             &aia_msi_num, true, NULL);
1771     if (ret < 0) {
1772         error_report("KVM AIA: failed to set number of msi");
1773         exit(1);
1774     }
1775 
1776 
1777     if (socket_count > 1) {
1778         max_group_id = socket_count - 1;
1779         socket_bits = find_last_bit(&max_group_id, BITS_PER_LONG) + 1;
1780         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1781                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1782                                 &socket_bits, true, NULL);
1783         if (ret < 0) {
1784             error_report("KVM AIA: failed to set group_bits");
1785             exit(1);
1786         }
1787 
1788         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1789                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1790                                 &group_shift, true, NULL);
1791         if (ret < 0) {
1792             error_report("KVM AIA: failed to set group_shift");
1793             exit(1);
1794         }
1795     }
1796 
1797     guest_bits = guest_num == 0 ? 0 :
1798                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1799     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1800                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1801                             &guest_bits, true, NULL);
1802     if (ret < 0) {
1803         error_report("KVM AIA: failed to set guest_bits");
1804         exit(1);
1805     }
1806 
1807     for (socket = 0; socket < socket_count; socket++) {
1808         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1809         hart_count = riscv_socket_hart_count(machine, socket);
1810         base_hart = riscv_socket_first_hartid(machine, socket);
1811 
1812         if (max_hart_per_socket < hart_count) {
1813             max_hart_per_socket = hart_count;
1814         }
1815 
1816         for (i = 0; i < hart_count; i++) {
1817             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1818             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1819                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1820                                     &imsic_addr, true, NULL);
1821             if (ret < 0) {
1822                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1823                 exit(1);
1824             }
1825         }
1826     }
1827 
1828 
1829     if (max_hart_per_socket > 1) {
1830         max_hart_per_socket--;
1831         hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1832     } else {
1833         hart_bits = 0;
1834     }
1835 
1836     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1837                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1838                             &hart_bits, true, NULL);
1839     if (ret < 0) {
1840         error_report("KVM AIA: failed to set hart_bits");
1841         exit(1);
1842     }
1843 
1844     if (kvm_has_gsi_routing()) {
1845         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1846             /* KVM AIA only has one APLIC instance */
1847             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1848         }
1849         kvm_gsi_routing_allowed = true;
1850         kvm_irqchip_commit_routes(kvm_state);
1851     }
1852 
1853     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1854                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1855                             NULL, true, NULL);
1856     if (ret < 0) {
1857         error_report("KVM AIA: initialized fail");
1858         exit(1);
1859     }
1860 
1861     kvm_msi_via_irqfd_allowed = true;
1862 }
1863 
1864 static void kvm_cpu_instance_init(CPUState *cs)
1865 {
1866     Object *obj = OBJECT(RISCV_CPU(cs));
1867 
1868     riscv_init_kvm_registers(obj);
1869 
1870     kvm_riscv_add_cpu_user_properties(obj);
1871 }
1872 
1873 /*
1874  * We'll get here via the following path:
1875  *
1876  * riscv_cpu_realize()
1877  *   -> cpu_exec_realizefn()
1878  *      -> kvm_cpu_realize() (via accel_cpu_common_realize())
1879  */
1880 static bool kvm_cpu_realize(CPUState *cs, Error **errp)
1881 {
1882     RISCVCPU *cpu = RISCV_CPU(cs);
1883     int ret;
1884 
1885     if (riscv_has_ext(&cpu->env, RVV)) {
1886         ret = prctl(PR_RISCV_V_SET_CONTROL, PR_RISCV_V_VSTATE_CTRL_ON);
1887         if (ret) {
1888             error_setg(errp, "Error in prctl PR_RISCV_V_SET_CONTROL, code: %s",
1889                        strerrorname_np(errno));
1890             return false;
1891         }
1892     }
1893 
1894    return true;
1895 }
1896 
1897 void riscv_kvm_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
1898 {
1899     CPURISCVState *env = &cpu->env;
1900     KVMScratchCPU kvmcpu;
1901     struct kvm_one_reg reg;
1902     uint64_t val;
1903     int ret;
1904 
1905     /* short-circuit without spinning the scratch CPU */
1906     if (!cpu->cfg.ext_zicbom && !cpu->cfg.ext_zicboz &&
1907         !riscv_has_ext(env, RVV)) {
1908         return;
1909     }
1910 
1911     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1912         error_setg(errp, "Unable to create scratch KVM cpu");
1913         return;
1914     }
1915 
1916     if (cpu->cfg.ext_zicbom &&
1917         riscv_cpu_option_set(kvm_cbom_blocksize.name)) {
1918 
1919         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1920                                         kvm_cbom_blocksize.kvm_reg_id);
1921         reg.addr = (uint64_t)&val;
1922         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1923         if (ret != 0) {
1924             error_setg(errp, "Unable to read cbom_blocksize, error %d", errno);
1925             return;
1926         }
1927 
1928         if (cpu->cfg.cbom_blocksize != val) {
1929             error_setg(errp, "Unable to set cbom_blocksize to a different "
1930                        "value than the host (%lu)", val);
1931             return;
1932         }
1933     }
1934 
1935     if (cpu->cfg.ext_zicboz &&
1936         riscv_cpu_option_set(kvm_cboz_blocksize.name)) {
1937 
1938         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1939                                         kvm_cboz_blocksize.kvm_reg_id);
1940         reg.addr = (uint64_t)&val;
1941         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1942         if (ret != 0) {
1943             error_setg(errp, "Unable to read cboz_blocksize, error %d", errno);
1944             return;
1945         }
1946 
1947         if (cpu->cfg.cboz_blocksize != val) {
1948             error_setg(errp, "Unable to set cboz_blocksize to a different "
1949                        "value than the host (%lu)", val);
1950             return;
1951         }
1952     }
1953 
1954     /* Users are setting vlen, not vlenb */
1955     if (riscv_has_ext(env, RVV) && riscv_cpu_option_set("vlen")) {
1956         if (!kvm_v_vlenb.supported) {
1957             error_setg(errp, "Unable to set 'vlenb': register not supported");
1958             return;
1959         }
1960 
1961         reg.id = kvm_v_vlenb.kvm_reg_id;
1962         reg.addr = (uint64_t)&val;
1963         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1964         if (ret != 0) {
1965             error_setg(errp, "Unable to read vlenb register, error %d", errno);
1966             return;
1967         }
1968 
1969         if (cpu->cfg.vlenb != val) {
1970             error_setg(errp, "Unable to set 'vlen' to a different "
1971                        "value than the host (%lu)", val * 8);
1972             return;
1973         }
1974     }
1975 
1976     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1977 }
1978 
1979 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1980 {
1981     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1982 
1983     acc->cpu_instance_init = kvm_cpu_instance_init;
1984     acc->cpu_target_realize = kvm_cpu_realize;
1985 }
1986 
1987 static const TypeInfo kvm_cpu_accel_type_info = {
1988     .name = ACCEL_CPU_NAME("kvm"),
1989 
1990     .parent = TYPE_ACCEL_CPU,
1991     .class_init = kvm_cpu_accel_class_init,
1992     .abstract = true,
1993 };
1994 static void kvm_cpu_accel_register_types(void)
1995 {
1996     type_register_static(&kvm_cpu_accel_type_info);
1997 }
1998 type_init(kvm_cpu_accel_register_types);
1999 
2000 static void riscv_host_cpu_class_init(ObjectClass *c, void *data)
2001 {
2002     RISCVCPUClass *mcc = RISCV_CPU_CLASS(c);
2003 
2004 #if defined(TARGET_RISCV32)
2005     mcc->misa_mxl_max = MXL_RV32;
2006 #elif defined(TARGET_RISCV64)
2007     mcc->misa_mxl_max = MXL_RV64;
2008 #endif
2009 }
2010 
2011 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
2012     {
2013         .name = TYPE_RISCV_CPU_HOST,
2014         .parent = TYPE_RISCV_CPU,
2015         .class_init = riscv_host_cpu_class_init,
2016     }
2017 };
2018 
2019 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
2020 
2021 static const uint32_t ebreak_insn = 0x00100073;
2022 static const uint16_t c_ebreak_insn = 0x9002;
2023 
2024 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2025 {
2026     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 0)) {
2027         return -EINVAL;
2028     }
2029 
2030     if ((bp->saved_insn & 0x3) == 0x3) {
2031         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 0)
2032             || cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak_insn, 4, 1)) {
2033             return -EINVAL;
2034         }
2035     } else {
2036         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak_insn, 2, 1)) {
2037             return -EINVAL;
2038         }
2039     }
2040 
2041     return 0;
2042 }
2043 
2044 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2045 {
2046     uint32_t ebreak;
2047     uint16_t c_ebreak;
2048 
2049     if ((bp->saved_insn & 0x3) == 0x3) {
2050         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak, 4, 0) ||
2051             ebreak != ebreak_insn ||
2052             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 1)) {
2053             return -EINVAL;
2054         }
2055     } else {
2056         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak, 2, 0) ||
2057             c_ebreak != c_ebreak_insn ||
2058             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 1)) {
2059             return -EINVAL;
2060         }
2061     }
2062 
2063     return 0;
2064 }
2065 
2066 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type)
2067 {
2068     /* TODO; To be implemented later. */
2069     return -EINVAL;
2070 }
2071 
2072 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type)
2073 {
2074     /* TODO; To be implemented later. */
2075     return -EINVAL;
2076 }
2077 
2078 void kvm_arch_remove_all_hw_breakpoints(void)
2079 {
2080     /* TODO; To be implemented later. */
2081 }
2082 
2083 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
2084 {
2085     if (kvm_sw_breakpoints_active(cs)) {
2086         dbg->control |= KVM_GUESTDBG_ENABLE;
2087     }
2088 }
2089