xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision 1c17df6fc4bbddf55c8a64a3db7fb1115ecd30f5)
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 int kvm_riscv_get_regs_csr(CPUState *cs)
617 {
618     CPURISCVState *env = &RISCV_CPU(cs)->env;
619 
620     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
621     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
622     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
623     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
624     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
625     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
626     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
627     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
628     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
629 
630     return 0;
631 }
632 
633 static int kvm_riscv_put_regs_csr(CPUState *cs)
634 {
635     CPURISCVState *env = &RISCV_CPU(cs)->env;
636 
637     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
638     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
639     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
640     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
641     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
642     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
643     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
644     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
645     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
646 
647     return 0;
648 }
649 
650 static int kvm_riscv_get_regs_fp(CPUState *cs)
651 {
652     int ret = 0;
653     int i;
654     CPURISCVState *env = &RISCV_CPU(cs)->env;
655 
656     if (riscv_has_ext(env, RVD)) {
657         uint64_t reg;
658         for (i = 0; i < 32; i++) {
659             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
660             if (ret) {
661                 return ret;
662             }
663             env->fpr[i] = reg;
664         }
665         return ret;
666     }
667 
668     if (riscv_has_ext(env, RVF)) {
669         uint32_t reg;
670         for (i = 0; i < 32; i++) {
671             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
672             if (ret) {
673                 return ret;
674             }
675             env->fpr[i] = reg;
676         }
677         return ret;
678     }
679 
680     return ret;
681 }
682 
683 static int kvm_riscv_put_regs_fp(CPUState *cs)
684 {
685     int ret = 0;
686     int i;
687     CPURISCVState *env = &RISCV_CPU(cs)->env;
688 
689     if (riscv_has_ext(env, RVD)) {
690         uint64_t reg;
691         for (i = 0; i < 32; i++) {
692             reg = env->fpr[i];
693             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
694             if (ret) {
695                 return ret;
696             }
697         }
698         return ret;
699     }
700 
701     if (riscv_has_ext(env, RVF)) {
702         uint32_t reg;
703         for (i = 0; i < 32; i++) {
704             reg = env->fpr[i];
705             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
706             if (ret) {
707                 return ret;
708             }
709         }
710         return ret;
711     }
712 
713     return ret;
714 }
715 
716 static void kvm_riscv_get_regs_timer(CPUState *cs)
717 {
718     CPURISCVState *env = &RISCV_CPU(cs)->env;
719 
720     if (env->kvm_timer_dirty) {
721         return;
722     }
723 
724     KVM_RISCV_GET_TIMER(cs, time, env->kvm_timer_time);
725     KVM_RISCV_GET_TIMER(cs, compare, env->kvm_timer_compare);
726     KVM_RISCV_GET_TIMER(cs, state, env->kvm_timer_state);
727     KVM_RISCV_GET_TIMER(cs, frequency, env->kvm_timer_frequency);
728 
729     env->kvm_timer_dirty = true;
730 }
731 
732 static void kvm_riscv_put_regs_timer(CPUState *cs)
733 {
734     uint64_t reg;
735     CPURISCVState *env = &RISCV_CPU(cs)->env;
736 
737     if (!env->kvm_timer_dirty) {
738         return;
739     }
740 
741     KVM_RISCV_SET_TIMER(cs, time, env->kvm_timer_time);
742     KVM_RISCV_SET_TIMER(cs, compare, env->kvm_timer_compare);
743 
744     /*
745      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
746      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
747      * doesn't matter that adaping in QEMU now.
748      * TODO If KVM changes, adapt here.
749      */
750     if (env->kvm_timer_state) {
751         KVM_RISCV_SET_TIMER(cs, state, env->kvm_timer_state);
752     }
753 
754     /*
755      * For now, migration will not work between Hosts with different timer
756      * frequency. Therefore, we should check whether they are the same here
757      * during the migration.
758      */
759     if (migration_is_running()) {
760         KVM_RISCV_GET_TIMER(cs, frequency, reg);
761         if (reg != env->kvm_timer_frequency) {
762             error_report("Dst Hosts timer frequency != Src Hosts");
763         }
764     }
765 
766     env->kvm_timer_dirty = false;
767 }
768 
769 uint64_t kvm_riscv_get_timebase_frequency(RISCVCPU *cpu)
770 {
771     uint64_t reg;
772 
773     KVM_RISCV_GET_TIMER(CPU(cpu), frequency, reg);
774 
775     return reg;
776 }
777 
778 static int kvm_riscv_get_regs_vector(CPUState *cs)
779 {
780     RISCVCPU *cpu = RISCV_CPU(cs);
781     CPURISCVState *env = &cpu->env;
782     target_ulong reg;
783     uint64_t vreg_id;
784     int vreg_idx, ret = 0;
785 
786     if (!riscv_has_ext(env, RVV)) {
787         return 0;
788     }
789 
790     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
791     if (ret) {
792         return ret;
793     }
794     env->vstart = reg;
795 
796     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
797     if (ret) {
798         return ret;
799     }
800     env->vl = reg;
801 
802     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
803     if (ret) {
804         return ret;
805     }
806     env->vtype = reg;
807 
808     if (kvm_v_vlenb.supported) {
809         ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
810         if (ret) {
811             return ret;
812         }
813         cpu->cfg.vlenb = reg;
814 
815         for (int i = 0; i < 32; i++) {
816             /*
817              * vreg[] is statically allocated using RV_VLEN_MAX.
818              * Use it instead of vlenb to calculate vreg_idx for
819              * simplicity.
820              */
821             vreg_idx = i * RV_VLEN_MAX / 64;
822             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
823 
824             ret = kvm_get_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
825             if (ret) {
826                 return ret;
827             }
828         }
829     }
830 
831     return 0;
832 }
833 
834 static int kvm_riscv_put_regs_vector(CPUState *cs)
835 {
836     RISCVCPU *cpu = RISCV_CPU(cs);
837     CPURISCVState *env = &cpu->env;
838     target_ulong reg;
839     uint64_t vreg_id;
840     int vreg_idx, ret = 0;
841 
842     if (!riscv_has_ext(env, RVV)) {
843         return 0;
844     }
845 
846     reg = env->vstart;
847     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
848     if (ret) {
849         return ret;
850     }
851 
852     reg = env->vl;
853     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
854     if (ret) {
855         return ret;
856     }
857 
858     reg = env->vtype;
859     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
860     if (ret) {
861         return ret;
862     }
863 
864     if (kvm_v_vlenb.supported) {
865         reg = cpu->cfg.vlenb;
866         ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
867 
868         for (int i = 0; i < 32; i++) {
869             /*
870              * vreg[] is statically allocated using RV_VLEN_MAX.
871              * Use it instead of vlenb to calculate vreg_idx for
872              * simplicity.
873              */
874             vreg_idx = i * RV_VLEN_MAX / 64;
875             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
876 
877             ret = kvm_set_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
878             if (ret) {
879                 return ret;
880             }
881         }
882     }
883 
884     return ret;
885 }
886 
887 typedef struct KVMScratchCPU {
888     int kvmfd;
889     int vmfd;
890     int cpufd;
891 } KVMScratchCPU;
892 
893 /*
894  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
895  * from target/arm/kvm.c.
896  */
897 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
898 {
899     int kvmfd = -1, vmfd = -1, cpufd = -1;
900 
901     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
902     if (kvmfd < 0) {
903         goto err;
904     }
905     do {
906         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
907     } while (vmfd == -1 && errno == EINTR);
908     if (vmfd < 0) {
909         goto err;
910     }
911     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
912     if (cpufd < 0) {
913         goto err;
914     }
915 
916     scratch->kvmfd =  kvmfd;
917     scratch->vmfd = vmfd;
918     scratch->cpufd = cpufd;
919 
920     return true;
921 
922  err:
923     if (cpufd >= 0) {
924         close(cpufd);
925     }
926     if (vmfd >= 0) {
927         close(vmfd);
928     }
929     if (kvmfd >= 0) {
930         close(kvmfd);
931     }
932 
933     return false;
934 }
935 
936 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
937 {
938     close(scratch->cpufd);
939     close(scratch->vmfd);
940     close(scratch->kvmfd);
941 }
942 
943 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
944 {
945     CPURISCVState *env = &cpu->env;
946     struct kvm_one_reg reg;
947     int ret;
948 
949     reg.id = RISCV_CONFIG_REG(env, mvendorid);
950     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
951     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
952     if (ret != 0) {
953         error_report("Unable to retrieve mvendorid from host, error %d", ret);
954     }
955 
956     reg.id = RISCV_CONFIG_REG(env, marchid);
957     reg.addr = (uint64_t)&cpu->cfg.marchid;
958     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
959     if (ret != 0) {
960         error_report("Unable to retrieve marchid from host, error %d", ret);
961     }
962 
963     reg.id = RISCV_CONFIG_REG(env, mimpid);
964     reg.addr = (uint64_t)&cpu->cfg.mimpid;
965     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
966     if (ret != 0) {
967         error_report("Unable to retrieve mimpid from host, error %d", ret);
968     }
969 }
970 
971 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
972                                          KVMScratchCPU *kvmcpu)
973 {
974     CPURISCVState *env = &cpu->env;
975     struct kvm_one_reg reg;
976     int ret;
977 
978     reg.id = RISCV_CONFIG_REG(env, isa);
979     reg.addr = (uint64_t)&env->misa_ext_mask;
980     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
981 
982     if (ret) {
983         error_report("Unable to fetch ISA register from KVM, "
984                      "error %d", ret);
985         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
986         exit(EXIT_FAILURE);
987     }
988 
989     env->misa_ext = env->misa_ext_mask;
990 }
991 
992 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
993                                          KVMCPUConfig *cbomz_cfg)
994 {
995     CPURISCVState *env = &cpu->env;
996     struct kvm_one_reg reg;
997     int ret;
998 
999     reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1000                                     cbomz_cfg->kvm_reg_id);
1001     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
1002     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1003     if (ret != 0) {
1004         error_report("Unable to read KVM reg %s, error %d",
1005                      cbomz_cfg->name, ret);
1006         exit(EXIT_FAILURE);
1007     }
1008 }
1009 
1010 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
1011                                            KVMScratchCPU *kvmcpu)
1012 {
1013     CPURISCVState *env = &cpu->env;
1014     uint64_t val;
1015     int i, ret;
1016 
1017     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1018         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1019         struct kvm_one_reg reg;
1020 
1021         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
1022                                         multi_ext_cfg->kvm_reg_id);
1023         reg.addr = (uint64_t)&val;
1024         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1025         if (ret != 0) {
1026             if (errno == EINVAL) {
1027                 /* Silently default to 'false' if KVM does not support it. */
1028                 multi_ext_cfg->supported = false;
1029                 val = false;
1030             } else {
1031                 error_report("Unable to read ISA_EXT KVM register %s: %s",
1032                              multi_ext_cfg->name, strerror(errno));
1033                 exit(EXIT_FAILURE);
1034             }
1035         } else {
1036             multi_ext_cfg->supported = true;
1037         }
1038 
1039         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1040     }
1041 
1042     if (cpu->cfg.ext_zicbom) {
1043         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1044     }
1045 
1046     if (cpu->cfg.ext_zicboz) {
1047         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1048     }
1049 }
1050 
1051 static int uint64_cmp(const void *a, const void *b)
1052 {
1053     uint64_t val1 = *(const uint64_t *)a;
1054     uint64_t val2 = *(const uint64_t *)b;
1055 
1056     if (val1 < val2) {
1057         return -1;
1058     }
1059 
1060     if (val1 > val2) {
1061         return 1;
1062     }
1063 
1064     return 0;
1065 }
1066 
1067 static void kvm_riscv_check_sbi_dbcn_support(RISCVCPU *cpu,
1068                                              KVMScratchCPU *kvmcpu,
1069                                              struct kvm_reg_list *reglist)
1070 {
1071     struct kvm_reg_list *reg_search;
1072 
1073     reg_search = bsearch(&kvm_sbi_dbcn.kvm_reg_id, reglist->reg, reglist->n,
1074                          sizeof(uint64_t), uint64_cmp);
1075 
1076     if (reg_search) {
1077         kvm_sbi_dbcn.supported = true;
1078     }
1079 }
1080 
1081 static void kvm_riscv_read_vlenb(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
1082                                  struct kvm_reg_list *reglist)
1083 {
1084     struct kvm_one_reg reg;
1085     struct kvm_reg_list *reg_search;
1086     uint64_t val;
1087     int ret;
1088 
1089     reg_search = bsearch(&kvm_v_vlenb.kvm_reg_id, reglist->reg, reglist->n,
1090                          sizeof(uint64_t), uint64_cmp);
1091 
1092     if (reg_search) {
1093         reg.id = kvm_v_vlenb.kvm_reg_id;
1094         reg.addr = (uint64_t)&val;
1095 
1096         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1097         if (ret != 0) {
1098             error_report("Unable to read vlenb register, error code: %d",
1099                          errno);
1100             exit(EXIT_FAILURE);
1101         }
1102 
1103         kvm_v_vlenb.supported = true;
1104         cpu->cfg.vlenb = val;
1105     }
1106 }
1107 
1108 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
1109 {
1110     KVMCPUConfig *multi_ext_cfg;
1111     struct kvm_one_reg reg;
1112     struct kvm_reg_list rl_struct;
1113     struct kvm_reg_list *reglist;
1114     uint64_t val, reg_id, *reg_search;
1115     int i, ret;
1116 
1117     rl_struct.n = 0;
1118     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
1119 
1120     /*
1121      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
1122      * (EINVAL). Use read_legacy() in this case.
1123      */
1124     if (errno == EINVAL) {
1125         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
1126     } else if (errno != E2BIG) {
1127         /*
1128          * E2BIG is an expected error message for the API since we
1129          * don't know the number of registers. The right amount will
1130          * be written in rl_struct.n.
1131          *
1132          * Error out if we get any other errno.
1133          */
1134         error_report("Error when accessing get-reg-list: %s",
1135                      strerror(errno));
1136         exit(EXIT_FAILURE);
1137     }
1138 
1139     reglist = g_malloc(sizeof(struct kvm_reg_list) +
1140                        rl_struct.n * sizeof(uint64_t));
1141     reglist->n = rl_struct.n;
1142     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
1143     if (ret) {
1144         error_report("Error when reading KVM_GET_REG_LIST: %s",
1145                      strerror(errno));
1146         exit(EXIT_FAILURE);
1147     }
1148 
1149     /* sort reglist to use bsearch() */
1150     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
1151 
1152     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1153         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1154         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
1155                                         multi_ext_cfg->kvm_reg_id);
1156         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
1157                              sizeof(uint64_t), uint64_cmp);
1158         if (!reg_search) {
1159             continue;
1160         }
1161 
1162         reg.id = reg_id;
1163         reg.addr = (uint64_t)&val;
1164         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1165         if (ret != 0) {
1166             error_report("Unable to read ISA_EXT KVM register %s: %s",
1167                          multi_ext_cfg->name, strerror(errno));
1168             exit(EXIT_FAILURE);
1169         }
1170 
1171         multi_ext_cfg->supported = true;
1172         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1173     }
1174 
1175     if (cpu->cfg.ext_zicbom) {
1176         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1177     }
1178 
1179     if (cpu->cfg.ext_zicboz) {
1180         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1181     }
1182 
1183     if (riscv_has_ext(&cpu->env, RVV)) {
1184         kvm_riscv_read_vlenb(cpu, kvmcpu, reglist);
1185     }
1186 
1187     kvm_riscv_check_sbi_dbcn_support(cpu, kvmcpu, reglist);
1188 }
1189 
1190 static void riscv_init_kvm_registers(Object *cpu_obj)
1191 {
1192     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
1193     KVMScratchCPU kvmcpu;
1194 
1195     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1196         return;
1197     }
1198 
1199     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
1200     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
1201     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
1202 
1203     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1204 }
1205 
1206 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
1207     KVM_CAP_LAST_INFO
1208 };
1209 
1210 int kvm_arch_get_registers(CPUState *cs, Error **errp)
1211 {
1212     int ret = 0;
1213 
1214     ret = kvm_riscv_get_regs_core(cs);
1215     if (ret) {
1216         return ret;
1217     }
1218 
1219     ret = kvm_riscv_get_regs_csr(cs);
1220     if (ret) {
1221         return ret;
1222     }
1223 
1224     ret = kvm_riscv_get_regs_fp(cs);
1225     if (ret) {
1226         return ret;
1227     }
1228 
1229     ret = kvm_riscv_get_regs_vector(cs);
1230     if (ret) {
1231         return ret;
1232     }
1233 
1234     return ret;
1235 }
1236 
1237 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1238 {
1239     if (cap_has_mp_state) {
1240         struct kvm_mp_state mp_state = {
1241             .mp_state = state
1242         };
1243 
1244         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1245         if (ret) {
1246             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1247                     __func__, ret, strerror(-ret));
1248             return -1;
1249         }
1250     }
1251 
1252     return 0;
1253 }
1254 
1255 int kvm_arch_put_registers(CPUState *cs, int level, Error **errp)
1256 {
1257     int ret = 0;
1258 
1259     ret = kvm_riscv_put_regs_core(cs);
1260     if (ret) {
1261         return ret;
1262     }
1263 
1264     ret = kvm_riscv_put_regs_csr(cs);
1265     if (ret) {
1266         return ret;
1267     }
1268 
1269     ret = kvm_riscv_put_regs_fp(cs);
1270     if (ret) {
1271         return ret;
1272     }
1273 
1274     ret = kvm_riscv_put_regs_vector(cs);
1275     if (ret) {
1276         return ret;
1277     }
1278 
1279     if (KVM_PUT_RESET_STATE == level) {
1280         RISCVCPU *cpu = RISCV_CPU(cs);
1281         if (cs->cpu_index == 0) {
1282             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1283         } else {
1284             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1285         }
1286         if (ret) {
1287             return ret;
1288         }
1289     }
1290 
1291     return ret;
1292 }
1293 
1294 int kvm_arch_release_virq_post(int virq)
1295 {
1296     return 0;
1297 }
1298 
1299 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1300                              uint64_t address, uint32_t data, PCIDevice *dev)
1301 {
1302     return 0;
1303 }
1304 
1305 int kvm_arch_destroy_vcpu(CPUState *cs)
1306 {
1307     return 0;
1308 }
1309 
1310 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1311 {
1312     return cpu->cpu_index;
1313 }
1314 
1315 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1316                                       RunState state)
1317 {
1318     CPUState *cs = opaque;
1319 
1320     if (running) {
1321         kvm_riscv_put_regs_timer(cs);
1322     } else {
1323         kvm_riscv_get_regs_timer(cs);
1324     }
1325 }
1326 
1327 void kvm_arch_init_irq_routing(KVMState *s)
1328 {
1329 }
1330 
1331 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1332 {
1333     CPURISCVState *env = &cpu->env;
1334     target_ulong reg;
1335     uint64_t id;
1336     int ret;
1337 
1338     id = RISCV_CONFIG_REG(env, mvendorid);
1339     /*
1340      * cfg.mvendorid is an uint32 but a target_ulong will
1341      * be written. Assign it to a target_ulong var to avoid
1342      * writing pieces of other cpu->cfg fields in the reg.
1343      */
1344     reg = cpu->cfg.mvendorid;
1345     ret = kvm_set_one_reg(cs, id, &reg);
1346     if (ret != 0) {
1347         return ret;
1348     }
1349 
1350     id = RISCV_CONFIG_REG(env, marchid);
1351     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1352     if (ret != 0) {
1353         return ret;
1354     }
1355 
1356     id = RISCV_CONFIG_REG(env, mimpid);
1357     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1358 
1359     return ret;
1360 }
1361 
1362 static int kvm_vcpu_enable_sbi_dbcn(RISCVCPU *cpu, CPUState *cs)
1363 {
1364     target_ulong reg = 1;
1365 
1366     if (!kvm_sbi_dbcn.supported) {
1367         return 0;
1368     }
1369 
1370     return kvm_set_one_reg(cs, kvm_sbi_dbcn.kvm_reg_id, &reg);
1371 }
1372 
1373 int kvm_arch_init_vcpu(CPUState *cs)
1374 {
1375     int ret = 0;
1376     RISCVCPU *cpu = RISCV_CPU(cs);
1377 
1378     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1379 
1380     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1381         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1382         if (ret != 0) {
1383             return ret;
1384         }
1385     }
1386 
1387     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1388     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1389 
1390     ret = kvm_vcpu_enable_sbi_dbcn(cpu, cs);
1391 
1392     return ret;
1393 }
1394 
1395 int kvm_arch_msi_data_to_gsi(uint32_t data)
1396 {
1397     abort();
1398 }
1399 
1400 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1401                                 int vector, PCIDevice *dev)
1402 {
1403     return 0;
1404 }
1405 
1406 int kvm_arch_get_default_type(MachineState *ms)
1407 {
1408     return 0;
1409 }
1410 
1411 int kvm_arch_init(MachineState *ms, KVMState *s)
1412 {
1413     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1414     return 0;
1415 }
1416 
1417 int kvm_arch_irqchip_create(KVMState *s)
1418 {
1419     /*
1420      * We can create the VAIA using the newer device control API.
1421      */
1422     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1423 }
1424 
1425 int kvm_arch_process_async_events(CPUState *cs)
1426 {
1427     return 0;
1428 }
1429 
1430 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1431 {
1432 }
1433 
1434 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1435 {
1436     return MEMTXATTRS_UNSPECIFIED;
1437 }
1438 
1439 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1440 {
1441     return true;
1442 }
1443 
1444 static void kvm_riscv_handle_sbi_dbcn(CPUState *cs, struct kvm_run *run)
1445 {
1446     g_autofree uint8_t *buf = NULL;
1447     RISCVCPU *cpu = RISCV_CPU(cs);
1448     target_ulong num_bytes;
1449     uint64_t addr;
1450     unsigned char ch;
1451     int ret;
1452 
1453     switch (run->riscv_sbi.function_id) {
1454     case SBI_EXT_DBCN_CONSOLE_READ:
1455     case SBI_EXT_DBCN_CONSOLE_WRITE:
1456         num_bytes = run->riscv_sbi.args[0];
1457 
1458         if (num_bytes == 0) {
1459             run->riscv_sbi.ret[0] = SBI_SUCCESS;
1460             run->riscv_sbi.ret[1] = 0;
1461             break;
1462         }
1463 
1464         addr = run->riscv_sbi.args[1];
1465 
1466         /*
1467          * Handle the case where a 32 bit CPU is running in a
1468          * 64 bit addressing env.
1469          */
1470         if (riscv_cpu_mxl(&cpu->env) == MXL_RV32) {
1471             addr |= (uint64_t)run->riscv_sbi.args[2] << 32;
1472         }
1473 
1474         buf = g_malloc0(num_bytes);
1475 
1476         if (run->riscv_sbi.function_id == SBI_EXT_DBCN_CONSOLE_READ) {
1477             ret = qemu_chr_fe_read_all(serial_hd(0)->be, buf, num_bytes);
1478             if (ret < 0) {
1479                 error_report("SBI_EXT_DBCN_CONSOLE_READ: error when "
1480                              "reading chardev");
1481                 exit(1);
1482             }
1483 
1484             cpu_physical_memory_write(addr, buf, ret);
1485         } else {
1486             cpu_physical_memory_read(addr, buf, num_bytes);
1487 
1488             ret = qemu_chr_fe_write_all(serial_hd(0)->be, buf, num_bytes);
1489             if (ret < 0) {
1490                 error_report("SBI_EXT_DBCN_CONSOLE_WRITE: error when "
1491                              "writing chardev");
1492                 exit(1);
1493             }
1494         }
1495 
1496         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1497         run->riscv_sbi.ret[1] = ret;
1498         break;
1499     case SBI_EXT_DBCN_CONSOLE_WRITE_BYTE:
1500         ch = run->riscv_sbi.args[0];
1501         ret = qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1502 
1503         if (ret < 0) {
1504             error_report("SBI_EXT_DBCN_CONSOLE_WRITE_BYTE: error when "
1505                          "writing chardev");
1506             exit(1);
1507         }
1508 
1509         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1510         run->riscv_sbi.ret[1] = 0;
1511         break;
1512     default:
1513         run->riscv_sbi.ret[0] = SBI_ERR_NOT_SUPPORTED;
1514     }
1515 }
1516 
1517 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1518 {
1519     int ret = 0;
1520     unsigned char ch;
1521     switch (run->riscv_sbi.extension_id) {
1522     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1523         ch = run->riscv_sbi.args[0];
1524         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1525         break;
1526     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1527         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1528         if (ret == sizeof(ch)) {
1529             run->riscv_sbi.ret[0] = ch;
1530         } else {
1531             run->riscv_sbi.ret[0] = -1;
1532         }
1533         ret = 0;
1534         break;
1535     case SBI_EXT_DBCN:
1536         kvm_riscv_handle_sbi_dbcn(cs, run);
1537         break;
1538     default:
1539         qemu_log_mask(LOG_UNIMP,
1540                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1541                       __func__, run->riscv_sbi.extension_id);
1542         ret = -1;
1543         break;
1544     }
1545     return ret;
1546 }
1547 
1548 static int kvm_riscv_handle_csr(CPUState *cs, struct kvm_run *run)
1549 {
1550     target_ulong csr_num = run->riscv_csr.csr_num;
1551     target_ulong new_value = run->riscv_csr.new_value;
1552     target_ulong write_mask = run->riscv_csr.write_mask;
1553     int ret = 0;
1554 
1555     switch (csr_num) {
1556     case CSR_SEED:
1557         run->riscv_csr.ret_value = riscv_new_csr_seed(new_value, write_mask);
1558         break;
1559     default:
1560         qemu_log_mask(LOG_UNIMP,
1561                       "%s: un-handled CSR EXIT for CSR %lx\n",
1562                       __func__, csr_num);
1563         ret = -1;
1564         break;
1565     }
1566 
1567     return ret;
1568 }
1569 
1570 static bool kvm_riscv_handle_debug(CPUState *cs)
1571 {
1572     RISCVCPU *cpu = RISCV_CPU(cs);
1573     CPURISCVState *env = &cpu->env;
1574 
1575     /* Ensure PC is synchronised */
1576     kvm_cpu_synchronize_state(cs);
1577 
1578     if (kvm_find_sw_breakpoint(cs, env->pc)) {
1579         return true;
1580     }
1581 
1582     return false;
1583 }
1584 
1585 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1586 {
1587     int ret = 0;
1588     switch (run->exit_reason) {
1589     case KVM_EXIT_RISCV_SBI:
1590         ret = kvm_riscv_handle_sbi(cs, run);
1591         break;
1592     case KVM_EXIT_RISCV_CSR:
1593         ret = kvm_riscv_handle_csr(cs, run);
1594         break;
1595     case KVM_EXIT_DEBUG:
1596         if (kvm_riscv_handle_debug(cs)) {
1597             ret = EXCP_DEBUG;
1598         }
1599         break;
1600     default:
1601         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1602                       __func__, run->exit_reason);
1603         ret = -1;
1604         break;
1605     }
1606     return ret;
1607 }
1608 
1609 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1610 {
1611     CPURISCVState *env = &cpu->env;
1612     int i;
1613 
1614     if (!kvm_enabled()) {
1615         return;
1616     }
1617     for (i = 0; i < 32; i++) {
1618         env->gpr[i] = 0;
1619     }
1620     env->pc = cpu->env.kernel_addr;
1621     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1622     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1623     env->satp = 0;
1624     env->mie = 0;
1625     env->stvec = 0;
1626     env->sscratch = 0;
1627     env->sepc = 0;
1628     env->scause = 0;
1629     env->stval = 0;
1630     env->mip = 0;
1631 }
1632 
1633 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1634 {
1635     int ret;
1636     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1637 
1638     if (irq != IRQ_S_EXT) {
1639         perror("kvm riscv set irq != IRQ_S_EXT\n");
1640         abort();
1641     }
1642 
1643     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1644     if (ret < 0) {
1645         perror("Set irq failed");
1646         abort();
1647     }
1648 }
1649 
1650 static int aia_mode;
1651 
1652 static const char *kvm_aia_mode_str(uint64_t mode)
1653 {
1654     switch (mode) {
1655     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1656         return "emul";
1657     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1658         return "hwaccel";
1659     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1660     default:
1661         return "auto";
1662     };
1663 }
1664 
1665 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1666 {
1667     return g_strdup(kvm_aia_mode_str(aia_mode));
1668 }
1669 
1670 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1671 {
1672     if (!strcmp(val, "emul")) {
1673         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1674     } else if (!strcmp(val, "hwaccel")) {
1675         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1676     } else if (!strcmp(val, "auto")) {
1677         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1678     } else {
1679         error_setg(errp, "Invalid KVM AIA mode");
1680         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1681     }
1682 }
1683 
1684 void kvm_arch_accel_class_init(ObjectClass *oc)
1685 {
1686     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1687                                   riscv_set_kvm_aia);
1688     object_class_property_set_description(oc, "riscv-aia",
1689         "Set KVM AIA mode. Valid values are 'emul', 'hwaccel' and 'auto'. "
1690         "Changing KVM AIA modes relies on host support. Defaults to 'auto' "
1691         "if the host supports it");
1692     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1693                                     "auto");
1694 }
1695 
1696 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1697                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1698                           uint64_t aplic_base, uint64_t imsic_base,
1699                           uint64_t guest_num)
1700 {
1701     int ret, i;
1702     int aia_fd = -1;
1703     uint64_t default_aia_mode;
1704     uint64_t socket_count = riscv_socket_count(machine);
1705     uint64_t max_hart_per_socket = 0;
1706     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1707     uint64_t socket_bits, hart_bits, guest_bits;
1708     uint64_t max_group_id;
1709 
1710     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1711 
1712     if (aia_fd < 0) {
1713         error_report("Unable to create in-kernel irqchip");
1714         exit(1);
1715     }
1716 
1717     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1718                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1719                             &default_aia_mode, false, NULL);
1720     if (ret < 0) {
1721         error_report("KVM AIA: failed to get current KVM AIA mode");
1722         exit(1);
1723     }
1724 
1725     if (default_aia_mode != aia_mode) {
1726         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1727                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1728                                 &aia_mode, true, NULL);
1729         if (ret < 0) {
1730             warn_report("KVM AIA: failed to set KVM AIA mode '%s', using "
1731                         "default host mode '%s'",
1732                         kvm_aia_mode_str(aia_mode),
1733                         kvm_aia_mode_str(default_aia_mode));
1734 
1735             /* failed to change AIA mode, use default */
1736             aia_mode = default_aia_mode;
1737         }
1738     }
1739 
1740     /*
1741      * Skip APLIC creation in KVM if we're running split mode.
1742      * This is done by leaving KVM_DEV_RISCV_AIA_CONFIG_SRCS
1743      * unset. We can also skip KVM_DEV_RISCV_AIA_ADDR_APLIC
1744      * since KVM won't be using it.
1745      */
1746     if (!kvm_kernel_irqchip_split()) {
1747         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1748                                 KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1749                                 &aia_irq_num, true, NULL);
1750         if (ret < 0) {
1751             error_report("KVM AIA: failed to set number of input irq lines");
1752             exit(1);
1753         }
1754 
1755         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1756                                 KVM_DEV_RISCV_AIA_ADDR_APLIC,
1757                                 &aplic_base, true, NULL);
1758         if (ret < 0) {
1759             error_report("KVM AIA: failed to set the base address of APLIC");
1760             exit(1);
1761         }
1762      }
1763 
1764     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1765                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1766                             &aia_msi_num, true, NULL);
1767     if (ret < 0) {
1768         error_report("KVM AIA: failed to set number of msi");
1769         exit(1);
1770     }
1771 
1772 
1773     if (socket_count > 1) {
1774         max_group_id = socket_count - 1;
1775         socket_bits = find_last_bit(&max_group_id, BITS_PER_LONG) + 1;
1776         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1777                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1778                                 &socket_bits, true, NULL);
1779         if (ret < 0) {
1780             error_report("KVM AIA: failed to set group_bits");
1781             exit(1);
1782         }
1783 
1784         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1785                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1786                                 &group_shift, true, NULL);
1787         if (ret < 0) {
1788             error_report("KVM AIA: failed to set group_shift");
1789             exit(1);
1790         }
1791     }
1792 
1793     guest_bits = guest_num == 0 ? 0 :
1794                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1795     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1796                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1797                             &guest_bits, true, NULL);
1798     if (ret < 0) {
1799         error_report("KVM AIA: failed to set guest_bits");
1800         exit(1);
1801     }
1802 
1803     for (socket = 0; socket < socket_count; socket++) {
1804         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1805         hart_count = riscv_socket_hart_count(machine, socket);
1806         base_hart = riscv_socket_first_hartid(machine, socket);
1807 
1808         if (max_hart_per_socket < hart_count) {
1809             max_hart_per_socket = hart_count;
1810         }
1811 
1812         for (i = 0; i < hart_count; i++) {
1813             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1814             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1815                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1816                                     &imsic_addr, true, NULL);
1817             if (ret < 0) {
1818                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1819                 exit(1);
1820             }
1821         }
1822     }
1823 
1824 
1825     if (max_hart_per_socket > 1) {
1826         max_hart_per_socket--;
1827         hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1828     } else {
1829         hart_bits = 0;
1830     }
1831 
1832     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1833                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1834                             &hart_bits, true, NULL);
1835     if (ret < 0) {
1836         error_report("KVM AIA: failed to set hart_bits");
1837         exit(1);
1838     }
1839 
1840     if (kvm_has_gsi_routing()) {
1841         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1842             /* KVM AIA only has one APLIC instance */
1843             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1844         }
1845         kvm_gsi_routing_allowed = true;
1846         kvm_irqchip_commit_routes(kvm_state);
1847     }
1848 
1849     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1850                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1851                             NULL, true, NULL);
1852     if (ret < 0) {
1853         error_report("KVM AIA: initialized fail");
1854         exit(1);
1855     }
1856 
1857     kvm_msi_via_irqfd_allowed = true;
1858 }
1859 
1860 static void kvm_cpu_instance_init(CPUState *cs)
1861 {
1862     Object *obj = OBJECT(RISCV_CPU(cs));
1863 
1864     riscv_init_kvm_registers(obj);
1865 
1866     kvm_riscv_add_cpu_user_properties(obj);
1867 }
1868 
1869 /*
1870  * We'll get here via the following path:
1871  *
1872  * riscv_cpu_realize()
1873  *   -> cpu_exec_realizefn()
1874  *      -> kvm_cpu_realize() (via accel_cpu_common_realize())
1875  */
1876 static bool kvm_cpu_realize(CPUState *cs, Error **errp)
1877 {
1878     RISCVCPU *cpu = RISCV_CPU(cs);
1879     int ret;
1880 
1881     if (riscv_has_ext(&cpu->env, RVV)) {
1882         ret = prctl(PR_RISCV_V_SET_CONTROL, PR_RISCV_V_VSTATE_CTRL_ON);
1883         if (ret) {
1884             error_setg(errp, "Error in prctl PR_RISCV_V_SET_CONTROL, code: %s",
1885                        strerrorname_np(errno));
1886             return false;
1887         }
1888     }
1889 
1890    return true;
1891 }
1892 
1893 void riscv_kvm_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
1894 {
1895     CPURISCVState *env = &cpu->env;
1896     KVMScratchCPU kvmcpu;
1897     struct kvm_one_reg reg;
1898     uint64_t val;
1899     int ret;
1900 
1901     /* short-circuit without spinning the scratch CPU */
1902     if (!cpu->cfg.ext_zicbom && !cpu->cfg.ext_zicboz &&
1903         !riscv_has_ext(env, RVV)) {
1904         return;
1905     }
1906 
1907     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1908         error_setg(errp, "Unable to create scratch KVM cpu");
1909         return;
1910     }
1911 
1912     if (cpu->cfg.ext_zicbom &&
1913         riscv_cpu_option_set(kvm_cbom_blocksize.name)) {
1914 
1915         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1916                                         kvm_cbom_blocksize.kvm_reg_id);
1917         reg.addr = (uint64_t)&val;
1918         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1919         if (ret != 0) {
1920             error_setg(errp, "Unable to read cbom_blocksize, error %d", errno);
1921             return;
1922         }
1923 
1924         if (cpu->cfg.cbom_blocksize != val) {
1925             error_setg(errp, "Unable to set cbom_blocksize to a different "
1926                        "value than the host (%lu)", val);
1927             return;
1928         }
1929     }
1930 
1931     if (cpu->cfg.ext_zicboz &&
1932         riscv_cpu_option_set(kvm_cboz_blocksize.name)) {
1933 
1934         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1935                                         kvm_cboz_blocksize.kvm_reg_id);
1936         reg.addr = (uint64_t)&val;
1937         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1938         if (ret != 0) {
1939             error_setg(errp, "Unable to read cboz_blocksize, error %d", errno);
1940             return;
1941         }
1942 
1943         if (cpu->cfg.cboz_blocksize != val) {
1944             error_setg(errp, "Unable to set cboz_blocksize to a different "
1945                        "value than the host (%lu)", val);
1946             return;
1947         }
1948     }
1949 
1950     /* Users are setting vlen, not vlenb */
1951     if (riscv_has_ext(env, RVV) && riscv_cpu_option_set("vlen")) {
1952         if (!kvm_v_vlenb.supported) {
1953             error_setg(errp, "Unable to set 'vlenb': register not supported");
1954             return;
1955         }
1956 
1957         reg.id = kvm_v_vlenb.kvm_reg_id;
1958         reg.addr = (uint64_t)&val;
1959         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1960         if (ret != 0) {
1961             error_setg(errp, "Unable to read vlenb register, error %d", errno);
1962             return;
1963         }
1964 
1965         if (cpu->cfg.vlenb != val) {
1966             error_setg(errp, "Unable to set 'vlen' to a different "
1967                        "value than the host (%lu)", val * 8);
1968             return;
1969         }
1970     }
1971 
1972     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1973 }
1974 
1975 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1976 {
1977     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1978 
1979     acc->cpu_instance_init = kvm_cpu_instance_init;
1980     acc->cpu_target_realize = kvm_cpu_realize;
1981 }
1982 
1983 static const TypeInfo kvm_cpu_accel_type_info = {
1984     .name = ACCEL_CPU_NAME("kvm"),
1985 
1986     .parent = TYPE_ACCEL_CPU,
1987     .class_init = kvm_cpu_accel_class_init,
1988     .abstract = true,
1989 };
1990 static void kvm_cpu_accel_register_types(void)
1991 {
1992     type_register_static(&kvm_cpu_accel_type_info);
1993 }
1994 type_init(kvm_cpu_accel_register_types);
1995 
1996 static void riscv_host_cpu_class_init(ObjectClass *c, void *data)
1997 {
1998     RISCVCPUClass *mcc = RISCV_CPU_CLASS(c);
1999 
2000 #if defined(TARGET_RISCV32)
2001     mcc->misa_mxl_max = MXL_RV32;
2002 #elif defined(TARGET_RISCV64)
2003     mcc->misa_mxl_max = MXL_RV64;
2004 #endif
2005 }
2006 
2007 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
2008     {
2009         .name = TYPE_RISCV_CPU_HOST,
2010         .parent = TYPE_RISCV_CPU,
2011         .class_init = riscv_host_cpu_class_init,
2012     }
2013 };
2014 
2015 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
2016 
2017 static const uint32_t ebreak_insn = 0x00100073;
2018 static const uint16_t c_ebreak_insn = 0x9002;
2019 
2020 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2021 {
2022     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 0)) {
2023         return -EINVAL;
2024     }
2025 
2026     if ((bp->saved_insn & 0x3) == 0x3) {
2027         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 0)
2028             || cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak_insn, 4, 1)) {
2029             return -EINVAL;
2030         }
2031     } else {
2032         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak_insn, 2, 1)) {
2033             return -EINVAL;
2034         }
2035     }
2036 
2037     return 0;
2038 }
2039 
2040 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2041 {
2042     uint32_t ebreak;
2043     uint16_t c_ebreak;
2044 
2045     if ((bp->saved_insn & 0x3) == 0x3) {
2046         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak, 4, 0) ||
2047             ebreak != ebreak_insn ||
2048             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 1)) {
2049             return -EINVAL;
2050         }
2051     } else {
2052         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak, 2, 0) ||
2053             c_ebreak != c_ebreak_insn ||
2054             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 1)) {
2055             return -EINVAL;
2056         }
2057     }
2058 
2059     return 0;
2060 }
2061 
2062 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type)
2063 {
2064     /* TODO; To be implemented later. */
2065     return -EINVAL;
2066 }
2067 
2068 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type)
2069 {
2070     /* TODO; To be implemented later. */
2071     return -EINVAL;
2072 }
2073 
2074 void kvm_arch_remove_all_hw_breakpoints(void)
2075 {
2076     /* TODO; To be implemented later. */
2077 }
2078 
2079 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
2080 {
2081     if (kvm_sw_breakpoints_active(cs)) {
2082         dbg->control |= KVM_GUESTDBG_ENABLE;
2083     }
2084 }
2085