xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision 906af6de9462c5192547cca0beac2c134659a437)
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 "accel/accel-cpu-target.h"
36 #include "hw/pci/pci.h"
37 #include "exec/memattrs.h"
38 #include "system/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 void kvm_riscv_reset_regs_csr(CPURISCVState *env)
651 {
652     env->mstatus = 0;
653     env->mie = 0;
654     env->stvec = 0;
655     env->sscratch = 0;
656     env->sepc = 0;
657     env->scause = 0;
658     env->stval = 0;
659     env->mip = 0;
660     env->satp = 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                                              struct kvm_reg_list *reglist)
1082 {
1083     struct kvm_reg_list *reg_search;
1084 
1085     reg_search = bsearch(&kvm_sbi_dbcn.kvm_reg_id, reglist->reg, reglist->n,
1086                          sizeof(uint64_t), uint64_cmp);
1087 
1088     if (reg_search) {
1089         kvm_sbi_dbcn.supported = true;
1090     }
1091 }
1092 
1093 static void kvm_riscv_read_vlenb(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
1094                                  struct kvm_reg_list *reglist)
1095 {
1096     struct kvm_one_reg reg;
1097     struct kvm_reg_list *reg_search;
1098     uint64_t val;
1099     int ret;
1100 
1101     reg_search = bsearch(&kvm_v_vlenb.kvm_reg_id, reglist->reg, reglist->n,
1102                          sizeof(uint64_t), uint64_cmp);
1103 
1104     if (reg_search) {
1105         reg.id = kvm_v_vlenb.kvm_reg_id;
1106         reg.addr = (uint64_t)&val;
1107 
1108         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1109         if (ret != 0) {
1110             error_report("Unable to read vlenb register, error code: %d",
1111                          errno);
1112             exit(EXIT_FAILURE);
1113         }
1114 
1115         kvm_v_vlenb.supported = true;
1116         cpu->cfg.vlenb = val;
1117     }
1118 }
1119 
1120 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
1121 {
1122     g_autofree struct kvm_reg_list *reglist = NULL;
1123     KVMCPUConfig *multi_ext_cfg;
1124     struct kvm_one_reg reg;
1125     struct kvm_reg_list rl_struct;
1126     uint64_t val, reg_id, *reg_search;
1127     int i, ret;
1128 
1129     rl_struct.n = 0;
1130     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
1131 
1132     /*
1133      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
1134      * (EINVAL). Use read_legacy() in this case.
1135      */
1136     if (errno == EINVAL) {
1137         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
1138     } else if (errno != E2BIG) {
1139         /*
1140          * E2BIG is an expected error message for the API since we
1141          * don't know the number of registers. The right amount will
1142          * be written in rl_struct.n.
1143          *
1144          * Error out if we get any other errno.
1145          */
1146         error_report("Error when accessing get-reg-list: %s",
1147                      strerror(errno));
1148         exit(EXIT_FAILURE);
1149     }
1150 
1151     reglist = g_malloc(sizeof(struct kvm_reg_list) +
1152                        rl_struct.n * sizeof(uint64_t));
1153     reglist->n = rl_struct.n;
1154     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
1155     if (ret) {
1156         error_report("Error when reading KVM_GET_REG_LIST: %s",
1157                      strerror(errno));
1158         exit(EXIT_FAILURE);
1159     }
1160 
1161     /* sort reglist to use bsearch() */
1162     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
1163 
1164     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1165         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1166         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
1167                                         multi_ext_cfg->kvm_reg_id);
1168         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
1169                              sizeof(uint64_t), uint64_cmp);
1170         if (!reg_search) {
1171             continue;
1172         }
1173 
1174         reg.id = reg_id;
1175         reg.addr = (uint64_t)&val;
1176         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1177         if (ret != 0) {
1178             error_report("Unable to read ISA_EXT KVM register %s: %s",
1179                          multi_ext_cfg->name, strerror(errno));
1180             exit(EXIT_FAILURE);
1181         }
1182 
1183         multi_ext_cfg->supported = true;
1184         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1185     }
1186 
1187     if (cpu->cfg.ext_zicbom) {
1188         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1189     }
1190 
1191     if (cpu->cfg.ext_zicboz) {
1192         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1193     }
1194 
1195     if (riscv_has_ext(&cpu->env, RVV)) {
1196         kvm_riscv_read_vlenb(cpu, kvmcpu, reglist);
1197     }
1198 
1199     kvm_riscv_check_sbi_dbcn_support(cpu, reglist);
1200 }
1201 
1202 static void riscv_init_kvm_registers(Object *cpu_obj)
1203 {
1204     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
1205     KVMScratchCPU kvmcpu;
1206 
1207     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1208         return;
1209     }
1210 
1211     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
1212     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
1213     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
1214 
1215     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1216 }
1217 
1218 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
1219     KVM_CAP_LAST_INFO
1220 };
1221 
1222 int kvm_arch_get_registers(CPUState *cs, Error **errp)
1223 {
1224     int ret = 0;
1225 
1226     ret = kvm_riscv_get_regs_core(cs);
1227     if (ret) {
1228         return ret;
1229     }
1230 
1231     ret = kvm_riscv_get_regs_csr(cs);
1232     if (ret) {
1233         return ret;
1234     }
1235 
1236     ret = kvm_riscv_get_regs_fp(cs);
1237     if (ret) {
1238         return ret;
1239     }
1240 
1241     ret = kvm_riscv_get_regs_vector(cs);
1242     if (ret) {
1243         return ret;
1244     }
1245 
1246     return ret;
1247 }
1248 
1249 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1250 {
1251     if (cap_has_mp_state) {
1252         struct kvm_mp_state mp_state = {
1253             .mp_state = state
1254         };
1255 
1256         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1257         if (ret) {
1258             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1259                     __func__, ret, strerror(-ret));
1260             return -1;
1261         }
1262     }
1263 
1264     return 0;
1265 }
1266 
1267 int kvm_arch_put_registers(CPUState *cs, int level, Error **errp)
1268 {
1269     int ret = 0;
1270 
1271     ret = kvm_riscv_put_regs_core(cs);
1272     if (ret) {
1273         return ret;
1274     }
1275 
1276     ret = kvm_riscv_put_regs_csr(cs);
1277     if (ret) {
1278         return ret;
1279     }
1280 
1281     ret = kvm_riscv_put_regs_fp(cs);
1282     if (ret) {
1283         return ret;
1284     }
1285 
1286     ret = kvm_riscv_put_regs_vector(cs);
1287     if (ret) {
1288         return ret;
1289     }
1290 
1291     if (KVM_PUT_RESET_STATE == level) {
1292         RISCVCPU *cpu = RISCV_CPU(cs);
1293         if (cs->cpu_index == 0) {
1294             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1295         } else {
1296             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1297         }
1298         if (ret) {
1299             return ret;
1300         }
1301     }
1302 
1303     return ret;
1304 }
1305 
1306 int kvm_arch_release_virq_post(int virq)
1307 {
1308     return 0;
1309 }
1310 
1311 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1312                              uint64_t address, uint32_t data, PCIDevice *dev)
1313 {
1314     return 0;
1315 }
1316 
1317 int kvm_arch_destroy_vcpu(CPUState *cs)
1318 {
1319     return 0;
1320 }
1321 
1322 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1323 {
1324     return cpu->cpu_index;
1325 }
1326 
1327 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1328                                       RunState state)
1329 {
1330     CPUState *cs = opaque;
1331 
1332     if (running) {
1333         kvm_riscv_put_regs_timer(cs);
1334     } else {
1335         kvm_riscv_get_regs_timer(cs);
1336     }
1337 }
1338 
1339 void kvm_arch_init_irq_routing(KVMState *s)
1340 {
1341 }
1342 
1343 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1344 {
1345     CPURISCVState *env = &cpu->env;
1346     target_ulong reg;
1347     uint64_t id;
1348     int ret;
1349 
1350     id = RISCV_CONFIG_REG(env, mvendorid);
1351     /*
1352      * cfg.mvendorid is an uint32 but a target_ulong will
1353      * be written. Assign it to a target_ulong var to avoid
1354      * writing pieces of other cpu->cfg fields in the reg.
1355      */
1356     reg = cpu->cfg.mvendorid;
1357     ret = kvm_set_one_reg(cs, id, &reg);
1358     if (ret != 0) {
1359         return ret;
1360     }
1361 
1362     id = RISCV_CONFIG_REG(env, marchid);
1363     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1364     if (ret != 0) {
1365         return ret;
1366     }
1367 
1368     id = RISCV_CONFIG_REG(env, mimpid);
1369     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1370 
1371     return ret;
1372 }
1373 
1374 static int kvm_vcpu_enable_sbi_dbcn(RISCVCPU *cpu, CPUState *cs)
1375 {
1376     target_ulong reg = 1;
1377 
1378     if (!kvm_sbi_dbcn.supported) {
1379         return 0;
1380     }
1381 
1382     return kvm_set_one_reg(cs, kvm_sbi_dbcn.kvm_reg_id, &reg);
1383 }
1384 
1385 int kvm_arch_init_vcpu(CPUState *cs)
1386 {
1387     int ret = 0;
1388     RISCVCPU *cpu = RISCV_CPU(cs);
1389 
1390     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1391 
1392     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1393         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1394         if (ret != 0) {
1395             return ret;
1396         }
1397     }
1398 
1399     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1400     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1401 
1402     ret = kvm_vcpu_enable_sbi_dbcn(cpu, cs);
1403 
1404     return ret;
1405 }
1406 
1407 int kvm_arch_msi_data_to_gsi(uint32_t data)
1408 {
1409     abort();
1410 }
1411 
1412 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1413                                 int vector, PCIDevice *dev)
1414 {
1415     return 0;
1416 }
1417 
1418 int kvm_arch_get_default_type(MachineState *ms)
1419 {
1420     return 0;
1421 }
1422 
1423 int kvm_arch_init(MachineState *ms, KVMState *s)
1424 {
1425     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1426     return 0;
1427 }
1428 
1429 int kvm_arch_irqchip_create(KVMState *s)
1430 {
1431     /*
1432      * We can create the VAIA using the newer device control API.
1433      */
1434     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1435 }
1436 
1437 int kvm_arch_process_async_events(CPUState *cs)
1438 {
1439     return 0;
1440 }
1441 
1442 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1443 {
1444 }
1445 
1446 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1447 {
1448     return MEMTXATTRS_UNSPECIFIED;
1449 }
1450 
1451 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1452 {
1453     return true;
1454 }
1455 
1456 static void kvm_riscv_handle_sbi_dbcn(CPUState *cs, struct kvm_run *run)
1457 {
1458     g_autofree uint8_t *buf = NULL;
1459     RISCVCPU *cpu = RISCV_CPU(cs);
1460     target_ulong num_bytes;
1461     uint64_t addr;
1462     unsigned char ch;
1463     int ret;
1464 
1465     switch (run->riscv_sbi.function_id) {
1466     case SBI_EXT_DBCN_CONSOLE_READ:
1467     case SBI_EXT_DBCN_CONSOLE_WRITE:
1468         num_bytes = run->riscv_sbi.args[0];
1469 
1470         if (num_bytes == 0) {
1471             run->riscv_sbi.ret[0] = SBI_SUCCESS;
1472             run->riscv_sbi.ret[1] = 0;
1473             break;
1474         }
1475 
1476         addr = run->riscv_sbi.args[1];
1477 
1478         /*
1479          * Handle the case where a 32 bit CPU is running in a
1480          * 64 bit addressing env.
1481          */
1482         if (riscv_cpu_mxl(&cpu->env) == MXL_RV32) {
1483             addr |= (uint64_t)run->riscv_sbi.args[2] << 32;
1484         }
1485 
1486         buf = g_malloc0(num_bytes);
1487 
1488         if (run->riscv_sbi.function_id == SBI_EXT_DBCN_CONSOLE_READ) {
1489             ret = qemu_chr_fe_read_all(serial_hd(0)->be, buf, num_bytes);
1490             if (ret < 0) {
1491                 error_report("SBI_EXT_DBCN_CONSOLE_READ: error when "
1492                              "reading chardev");
1493                 exit(1);
1494             }
1495 
1496             cpu_physical_memory_write(addr, buf, ret);
1497         } else {
1498             cpu_physical_memory_read(addr, buf, num_bytes);
1499 
1500             ret = qemu_chr_fe_write_all(serial_hd(0)->be, buf, num_bytes);
1501             if (ret < 0) {
1502                 error_report("SBI_EXT_DBCN_CONSOLE_WRITE: error when "
1503                              "writing chardev");
1504                 exit(1);
1505             }
1506         }
1507 
1508         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1509         run->riscv_sbi.ret[1] = ret;
1510         break;
1511     case SBI_EXT_DBCN_CONSOLE_WRITE_BYTE:
1512         ch = run->riscv_sbi.args[0];
1513         ret = qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1514 
1515         if (ret < 0) {
1516             error_report("SBI_EXT_DBCN_CONSOLE_WRITE_BYTE: error when "
1517                          "writing chardev");
1518             exit(1);
1519         }
1520 
1521         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1522         run->riscv_sbi.ret[1] = 0;
1523         break;
1524     default:
1525         run->riscv_sbi.ret[0] = SBI_ERR_NOT_SUPPORTED;
1526     }
1527 }
1528 
1529 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1530 {
1531     int ret = 0;
1532     unsigned char ch;
1533     switch (run->riscv_sbi.extension_id) {
1534     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1535         ch = run->riscv_sbi.args[0];
1536         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1537         break;
1538     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1539         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1540         if (ret == sizeof(ch)) {
1541             run->riscv_sbi.ret[0] = ch;
1542         } else {
1543             run->riscv_sbi.ret[0] = -1;
1544         }
1545         ret = 0;
1546         break;
1547     case SBI_EXT_DBCN:
1548         kvm_riscv_handle_sbi_dbcn(cs, run);
1549         break;
1550     default:
1551         qemu_log_mask(LOG_UNIMP,
1552                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1553                       __func__, run->riscv_sbi.extension_id);
1554         ret = -1;
1555         break;
1556     }
1557     return ret;
1558 }
1559 
1560 static int kvm_riscv_handle_csr(CPUState *cs, struct kvm_run *run)
1561 {
1562     target_ulong csr_num = run->riscv_csr.csr_num;
1563     target_ulong new_value = run->riscv_csr.new_value;
1564     target_ulong write_mask = run->riscv_csr.write_mask;
1565     int ret = 0;
1566 
1567     switch (csr_num) {
1568     case CSR_SEED:
1569         run->riscv_csr.ret_value = riscv_new_csr_seed(new_value, write_mask);
1570         break;
1571     default:
1572         qemu_log_mask(LOG_UNIMP,
1573                       "%s: un-handled CSR EXIT for CSR %lx\n",
1574                       __func__, csr_num);
1575         ret = -1;
1576         break;
1577     }
1578 
1579     return ret;
1580 }
1581 
1582 static bool kvm_riscv_handle_debug(CPUState *cs)
1583 {
1584     RISCVCPU *cpu = RISCV_CPU(cs);
1585     CPURISCVState *env = &cpu->env;
1586 
1587     /* Ensure PC is synchronised */
1588     kvm_cpu_synchronize_state(cs);
1589 
1590     if (kvm_find_sw_breakpoint(cs, env->pc)) {
1591         return true;
1592     }
1593 
1594     return false;
1595 }
1596 
1597 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1598 {
1599     int ret = 0;
1600     switch (run->exit_reason) {
1601     case KVM_EXIT_RISCV_SBI:
1602         ret = kvm_riscv_handle_sbi(cs, run);
1603         break;
1604     case KVM_EXIT_RISCV_CSR:
1605         ret = kvm_riscv_handle_csr(cs, run);
1606         break;
1607     case KVM_EXIT_DEBUG:
1608         if (kvm_riscv_handle_debug(cs)) {
1609             ret = EXCP_DEBUG;
1610         }
1611         break;
1612     default:
1613         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1614                       __func__, run->exit_reason);
1615         ret = -1;
1616         break;
1617     }
1618     return ret;
1619 }
1620 
1621 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1622 {
1623     CPURISCVState *env = &cpu->env;
1624     int i;
1625 
1626     for (i = 0; i < 32; i++) {
1627         env->gpr[i] = 0;
1628     }
1629     env->pc = cpu->env.kernel_addr;
1630     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1631     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1632 
1633     kvm_riscv_reset_regs_csr(env);
1634 }
1635 
1636 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1637 {
1638     int ret;
1639     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1640 
1641     if (irq != IRQ_S_EXT) {
1642         perror("kvm riscv set irq != IRQ_S_EXT\n");
1643         abort();
1644     }
1645 
1646     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1647     if (ret < 0) {
1648         perror("Set irq failed");
1649         abort();
1650     }
1651 }
1652 
1653 static int aia_mode;
1654 
1655 static const char *kvm_aia_mode_str(uint64_t mode)
1656 {
1657     switch (mode) {
1658     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1659         return "emul";
1660     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1661         return "hwaccel";
1662     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1663     default:
1664         return "auto";
1665     };
1666 }
1667 
1668 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1669 {
1670     return g_strdup(kvm_aia_mode_str(aia_mode));
1671 }
1672 
1673 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1674 {
1675     if (!strcmp(val, "emul")) {
1676         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1677     } else if (!strcmp(val, "hwaccel")) {
1678         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1679     } else if (!strcmp(val, "auto")) {
1680         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1681     } else {
1682         error_setg(errp, "Invalid KVM AIA mode");
1683         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1684     }
1685 }
1686 
1687 void kvm_arch_accel_class_init(ObjectClass *oc)
1688 {
1689     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1690                                   riscv_set_kvm_aia);
1691     object_class_property_set_description(oc, "riscv-aia",
1692         "Set KVM AIA mode. Valid values are 'emul', 'hwaccel' and 'auto'. "
1693         "Changing KVM AIA modes relies on host support. Defaults to 'auto' "
1694         "if the host supports it");
1695     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1696                                     "auto");
1697 }
1698 
1699 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1700                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1701                           uint64_t aplic_base, uint64_t imsic_base,
1702                           uint64_t guest_num)
1703 {
1704     int ret, i;
1705     int aia_fd = -1;
1706     uint64_t default_aia_mode;
1707     uint64_t socket_count = riscv_socket_count(machine);
1708     uint64_t max_hart_per_socket = 0;
1709     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1710     uint64_t socket_bits, hart_bits, guest_bits;
1711     uint64_t max_group_id;
1712 
1713     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1714 
1715     if (aia_fd < 0) {
1716         error_report("Unable to create in-kernel irqchip");
1717         exit(1);
1718     }
1719 
1720     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1721                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1722                             &default_aia_mode, false, NULL);
1723     if (ret < 0) {
1724         error_report("KVM AIA: failed to get current KVM AIA mode");
1725         exit(1);
1726     }
1727 
1728     if (default_aia_mode != aia_mode) {
1729         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1730                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1731                                 &aia_mode, true, NULL);
1732         if (ret < 0) {
1733             warn_report("KVM AIA: failed to set KVM AIA mode '%s', using "
1734                         "default host mode '%s'",
1735                         kvm_aia_mode_str(aia_mode),
1736                         kvm_aia_mode_str(default_aia_mode));
1737 
1738             /* failed to change AIA mode, use default */
1739             aia_mode = default_aia_mode;
1740         }
1741     }
1742 
1743     /*
1744      * Skip APLIC creation in KVM if we're running split mode.
1745      * This is done by leaving KVM_DEV_RISCV_AIA_CONFIG_SRCS
1746      * unset. We can also skip KVM_DEV_RISCV_AIA_ADDR_APLIC
1747      * since KVM won't be using it.
1748      */
1749     if (!kvm_kernel_irqchip_split()) {
1750         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1751                                 KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1752                                 &aia_irq_num, true, NULL);
1753         if (ret < 0) {
1754             error_report("KVM AIA: failed to set number of input irq lines");
1755             exit(1);
1756         }
1757 
1758         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1759                                 KVM_DEV_RISCV_AIA_ADDR_APLIC,
1760                                 &aplic_base, true, NULL);
1761         if (ret < 0) {
1762             error_report("KVM AIA: failed to set the base address of APLIC");
1763             exit(1);
1764         }
1765      }
1766 
1767     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1768                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1769                             &aia_msi_num, true, NULL);
1770     if (ret < 0) {
1771         error_report("KVM AIA: failed to set number of msi");
1772         exit(1);
1773     }
1774 
1775 
1776     if (socket_count > 1) {
1777         max_group_id = socket_count - 1;
1778         socket_bits = find_last_bit(&max_group_id, BITS_PER_LONG) + 1;
1779         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1780                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1781                                 &socket_bits, true, NULL);
1782         if (ret < 0) {
1783             error_report("KVM AIA: failed to set group_bits");
1784             exit(1);
1785         }
1786 
1787         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1788                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1789                                 &group_shift, true, NULL);
1790         if (ret < 0) {
1791             error_report("KVM AIA: failed to set group_shift");
1792             exit(1);
1793         }
1794     }
1795 
1796     guest_bits = guest_num == 0 ? 0 :
1797                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1798     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1799                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1800                             &guest_bits, true, NULL);
1801     if (ret < 0) {
1802         error_report("KVM AIA: failed to set guest_bits");
1803         exit(1);
1804     }
1805 
1806     for (socket = 0; socket < socket_count; socket++) {
1807         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1808         hart_count = riscv_socket_hart_count(machine, socket);
1809         base_hart = riscv_socket_first_hartid(machine, socket);
1810 
1811         if (max_hart_per_socket < hart_count) {
1812             max_hart_per_socket = hart_count;
1813         }
1814 
1815         for (i = 0; i < hart_count; i++) {
1816             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1817             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1818                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1819                                     &imsic_addr, true, NULL);
1820             if (ret < 0) {
1821                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1822                 exit(1);
1823             }
1824         }
1825     }
1826 
1827 
1828     if (max_hart_per_socket > 1) {
1829         max_hart_per_socket--;
1830         hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1831     } else {
1832         hart_bits = 0;
1833     }
1834 
1835     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1836                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1837                             &hart_bits, true, NULL);
1838     if (ret < 0) {
1839         error_report("KVM AIA: failed to set hart_bits");
1840         exit(1);
1841     }
1842 
1843     if (kvm_has_gsi_routing()) {
1844         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1845             /* KVM AIA only has one APLIC instance */
1846             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1847         }
1848         kvm_gsi_routing_allowed = true;
1849         kvm_irqchip_commit_routes(kvm_state);
1850     }
1851 
1852     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1853                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1854                             NULL, true, NULL);
1855     if (ret < 0) {
1856         error_report("KVM AIA: initialized fail");
1857         exit(1);
1858     }
1859 
1860     kvm_msi_via_irqfd_allowed = true;
1861 }
1862 
1863 static void kvm_cpu_instance_init(CPUState *cs)
1864 {
1865     Object *obj = OBJECT(RISCV_CPU(cs));
1866 
1867     riscv_init_kvm_registers(obj);
1868 
1869     kvm_riscv_add_cpu_user_properties(obj);
1870 }
1871 
1872 /*
1873  * We'll get here via the following path:
1874  *
1875  * riscv_cpu_realize()
1876  *   -> cpu_exec_realizefn()
1877  *      -> kvm_cpu_realize() (via accel_cpu_common_realize())
1878  */
1879 static bool kvm_cpu_realize(CPUState *cs, Error **errp)
1880 {
1881     RISCVCPU *cpu = RISCV_CPU(cs);
1882     int ret;
1883 
1884     if (riscv_has_ext(&cpu->env, RVV)) {
1885         ret = prctl(PR_RISCV_V_SET_CONTROL, PR_RISCV_V_VSTATE_CTRL_ON);
1886         if (ret) {
1887             error_setg(errp, "Error in prctl PR_RISCV_V_SET_CONTROL, code: %s",
1888                        strerrorname_np(errno));
1889             return false;
1890         }
1891     }
1892 
1893    return true;
1894 }
1895 
1896 void riscv_kvm_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
1897 {
1898     CPURISCVState *env = &cpu->env;
1899     KVMScratchCPU kvmcpu;
1900     struct kvm_one_reg reg;
1901     uint64_t val;
1902     int ret;
1903 
1904     /* short-circuit without spinning the scratch CPU */
1905     if (!cpu->cfg.ext_zicbom && !cpu->cfg.ext_zicboz &&
1906         !riscv_has_ext(env, RVV)) {
1907         return;
1908     }
1909 
1910     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1911         error_setg(errp, "Unable to create scratch KVM cpu");
1912         return;
1913     }
1914 
1915     if (cpu->cfg.ext_zicbom &&
1916         riscv_cpu_option_set(kvm_cbom_blocksize.name)) {
1917 
1918         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1919                                         kvm_cbom_blocksize.kvm_reg_id);
1920         reg.addr = (uint64_t)&val;
1921         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1922         if (ret != 0) {
1923             error_setg(errp, "Unable to read cbom_blocksize, error %d", errno);
1924             return;
1925         }
1926 
1927         if (cpu->cfg.cbom_blocksize != val) {
1928             error_setg(errp, "Unable to set cbom_blocksize to a different "
1929                        "value than the host (%lu)", val);
1930             return;
1931         }
1932     }
1933 
1934     if (cpu->cfg.ext_zicboz &&
1935         riscv_cpu_option_set(kvm_cboz_blocksize.name)) {
1936 
1937         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1938                                         kvm_cboz_blocksize.kvm_reg_id);
1939         reg.addr = (uint64_t)&val;
1940         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1941         if (ret != 0) {
1942             error_setg(errp, "Unable to read cboz_blocksize, error %d", errno);
1943             return;
1944         }
1945 
1946         if (cpu->cfg.cboz_blocksize != val) {
1947             error_setg(errp, "Unable to set cboz_blocksize to a different "
1948                        "value than the host (%lu)", val);
1949             return;
1950         }
1951     }
1952 
1953     /* Users are setting vlen, not vlenb */
1954     if (riscv_has_ext(env, RVV) && riscv_cpu_option_set("vlen")) {
1955         if (!kvm_v_vlenb.supported) {
1956             error_setg(errp, "Unable to set 'vlenb': register not supported");
1957             return;
1958         }
1959 
1960         reg.id = kvm_v_vlenb.kvm_reg_id;
1961         reg.addr = (uint64_t)&val;
1962         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1963         if (ret != 0) {
1964             error_setg(errp, "Unable to read vlenb register, error %d", errno);
1965             return;
1966         }
1967 
1968         if (cpu->cfg.vlenb != val) {
1969             error_setg(errp, "Unable to set 'vlen' to a different "
1970                        "value than the host (%lu)", val * 8);
1971             return;
1972         }
1973     }
1974 
1975     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1976 }
1977 
1978 static void kvm_cpu_accel_class_init(ObjectClass *oc, const void *data)
1979 {
1980     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1981 
1982     acc->cpu_instance_init = kvm_cpu_instance_init;
1983     acc->cpu_target_realize = kvm_cpu_realize;
1984 }
1985 
1986 static const TypeInfo kvm_cpu_accel_type_info = {
1987     .name = ACCEL_CPU_NAME("kvm"),
1988 
1989     .parent = TYPE_ACCEL_CPU,
1990     .class_init = kvm_cpu_accel_class_init,
1991     .abstract = true,
1992 };
1993 static void kvm_cpu_accel_register_types(void)
1994 {
1995     type_register_static(&kvm_cpu_accel_type_info);
1996 }
1997 type_init(kvm_cpu_accel_register_types);
1998 
1999 static void riscv_host_cpu_class_init(ObjectClass *c, const void *data)
2000 {
2001     RISCVCPUClass *mcc = RISCV_CPU_CLASS(c);
2002 
2003 #if defined(TARGET_RISCV32)
2004     mcc->misa_mxl_max = MXL_RV32;
2005 #elif defined(TARGET_RISCV64)
2006     mcc->misa_mxl_max = MXL_RV64;
2007 #endif
2008 }
2009 
2010 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
2011     {
2012         .name = TYPE_RISCV_CPU_HOST,
2013         .parent = TYPE_RISCV_CPU,
2014         .class_init = riscv_host_cpu_class_init,
2015     }
2016 };
2017 
2018 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
2019 
2020 static const uint32_t ebreak_insn = 0x00100073;
2021 static const uint16_t c_ebreak_insn = 0x9002;
2022 
2023 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2024 {
2025     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 0)) {
2026         return -EINVAL;
2027     }
2028 
2029     if ((bp->saved_insn & 0x3) == 0x3) {
2030         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 0)
2031             || cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak_insn, 4, 1)) {
2032             return -EINVAL;
2033         }
2034     } else {
2035         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak_insn, 2, 1)) {
2036             return -EINVAL;
2037         }
2038     }
2039 
2040     return 0;
2041 }
2042 
2043 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2044 {
2045     uint32_t ebreak;
2046     uint16_t c_ebreak;
2047 
2048     if ((bp->saved_insn & 0x3) == 0x3) {
2049         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak, 4, 0) ||
2050             ebreak != ebreak_insn ||
2051             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 1)) {
2052             return -EINVAL;
2053         }
2054     } else {
2055         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak, 2, 0) ||
2056             c_ebreak != c_ebreak_insn ||
2057             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 1)) {
2058             return -EINVAL;
2059         }
2060     }
2061 
2062     return 0;
2063 }
2064 
2065 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type)
2066 {
2067     /* TODO; To be implemented later. */
2068     return -EINVAL;
2069 }
2070 
2071 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type)
2072 {
2073     /* TODO; To be implemented later. */
2074     return -EINVAL;
2075 }
2076 
2077 void kvm_arch_remove_all_hw_breakpoints(void)
2078 {
2079     /* TODO; To be implemented later. */
2080 }
2081 
2082 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
2083 {
2084     if (kvm_sw_breakpoints_active(cs)) {
2085         dbg->control |= KVM_GUESTDBG_ENABLE;
2086     }
2087 }
2088