xref: /openbmc/qemu/target/riscv/kvm/kvm-cpu.c (revision 646b5378)
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 "sysemu/sysemu.h"
31 #include "sysemu/kvm.h"
32 #include "sysemu/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 "sysemu/runstate.h"
49 #include "hw/riscv/numa.h"
50 
51 #define PR_RISCV_V_SET_CONTROL            69
52 #define PR_RISCV_V_VSTATE_CTRL_ON          2
53 
54 void riscv_kvm_aplic_request(void *opaque, int irq, int level)
55 {
56     kvm_set_irq(kvm_state, irq, !!level);
57 }
58 
59 static bool cap_has_mp_state;
60 
61 static uint64_t kvm_riscv_reg_id_ulong(CPURISCVState *env, uint64_t type,
62                                  uint64_t idx)
63 {
64     uint64_t id = KVM_REG_RISCV | type | idx;
65 
66     switch (riscv_cpu_mxl(env)) {
67     case MXL_RV32:
68         id |= KVM_REG_SIZE_U32;
69         break;
70     case MXL_RV64:
71         id |= KVM_REG_SIZE_U64;
72         break;
73     default:
74         g_assert_not_reached();
75     }
76     return id;
77 }
78 
79 static uint64_t kvm_riscv_reg_id_u32(uint64_t type, uint64_t idx)
80 {
81     return KVM_REG_RISCV | KVM_REG_SIZE_U32 | type | idx;
82 }
83 
84 static uint64_t kvm_riscv_reg_id_u64(uint64_t type, uint64_t idx)
85 {
86     return KVM_REG_RISCV | KVM_REG_SIZE_U64 | type | idx;
87 }
88 
89 static uint64_t kvm_encode_reg_size_id(uint64_t id, size_t size_b)
90 {
91     uint64_t size_ctz = __builtin_ctz(size_b);
92 
93     return id | (size_ctz << KVM_REG_SIZE_SHIFT);
94 }
95 
96 static uint64_t kvm_riscv_vector_reg_id(RISCVCPU *cpu,
97                                         uint64_t idx)
98 {
99     uint64_t id;
100     size_t size_b;
101 
102     g_assert(idx < 32);
103 
104     id = KVM_REG_RISCV | KVM_REG_RISCV_VECTOR | KVM_REG_RISCV_VECTOR_REG(idx);
105     size_b = cpu->cfg.vlenb;
106 
107     return kvm_encode_reg_size_id(id, size_b);
108 }
109 
110 #define RISCV_CORE_REG(env, name) \
111     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, \
112                            KVM_REG_RISCV_CORE_REG(name))
113 
114 #define RISCV_CSR_REG(env, name) \
115     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CSR, \
116                            KVM_REG_RISCV_CSR_REG(name))
117 
118 #define RISCV_CONFIG_REG(env, name) \
119     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG, \
120                            KVM_REG_RISCV_CONFIG_REG(name))
121 
122 #define RISCV_TIMER_REG(name)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_TIMER, \
123                  KVM_REG_RISCV_TIMER_REG(name))
124 
125 #define RISCV_FP_F_REG(idx)  kvm_riscv_reg_id_u32(KVM_REG_RISCV_FP_F, idx)
126 
127 #define RISCV_FP_D_REG(idx)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_FP_D, idx)
128 
129 #define RISCV_VECTOR_CSR_REG(env, name) \
130     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_VECTOR, \
131                            KVM_REG_RISCV_VECTOR_CSR_REG(name))
132 
133 #define KVM_RISCV_GET_CSR(cs, env, csr, reg) \
134     do { \
135         int _ret = kvm_get_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
136         if (_ret) { \
137             return _ret; \
138         } \
139     } while (0)
140 
141 #define KVM_RISCV_SET_CSR(cs, env, csr, reg) \
142     do { \
143         int _ret = kvm_set_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
144         if (_ret) { \
145             return _ret; \
146         } \
147     } while (0)
148 
149 #define KVM_RISCV_GET_TIMER(cs, name, reg) \
150     do { \
151         int ret = kvm_get_one_reg(cs, RISCV_TIMER_REG(name), &reg); \
152         if (ret) { \
153             abort(); \
154         } \
155     } while (0)
156 
157 #define KVM_RISCV_SET_TIMER(cs, name, reg) \
158     do { \
159         int ret = kvm_set_one_reg(cs, RISCV_TIMER_REG(name), &reg); \
160         if (ret) { \
161             abort(); \
162         } \
163     } while (0)
164 
165 typedef struct KVMCPUConfig {
166     const char *name;
167     const char *description;
168     target_ulong offset;
169     uint64_t kvm_reg_id;
170     bool user_set;
171     bool supported;
172 } KVMCPUConfig;
173 
174 #define KVM_MISA_CFG(_bit, _reg_id) \
175     {.offset = _bit, .kvm_reg_id = _reg_id}
176 
177 /* KVM ISA extensions */
178 static KVMCPUConfig kvm_misa_ext_cfgs[] = {
179     KVM_MISA_CFG(RVA, KVM_RISCV_ISA_EXT_A),
180     KVM_MISA_CFG(RVC, KVM_RISCV_ISA_EXT_C),
181     KVM_MISA_CFG(RVD, KVM_RISCV_ISA_EXT_D),
182     KVM_MISA_CFG(RVF, KVM_RISCV_ISA_EXT_F),
183     KVM_MISA_CFG(RVH, KVM_RISCV_ISA_EXT_H),
184     KVM_MISA_CFG(RVI, KVM_RISCV_ISA_EXT_I),
185     KVM_MISA_CFG(RVM, KVM_RISCV_ISA_EXT_M),
186     KVM_MISA_CFG(RVV, KVM_RISCV_ISA_EXT_V),
187 };
188 
189 static void kvm_cpu_get_misa_ext_cfg(Object *obj, Visitor *v,
190                                      const char *name,
191                                      void *opaque, Error **errp)
192 {
193     KVMCPUConfig *misa_ext_cfg = opaque;
194     target_ulong misa_bit = misa_ext_cfg->offset;
195     RISCVCPU *cpu = RISCV_CPU(obj);
196     CPURISCVState *env = &cpu->env;
197     bool value = env->misa_ext_mask & misa_bit;
198 
199     visit_type_bool(v, name, &value, errp);
200 }
201 
202 static void kvm_cpu_set_misa_ext_cfg(Object *obj, Visitor *v,
203                                      const char *name,
204                                      void *opaque, Error **errp)
205 {
206     KVMCPUConfig *misa_ext_cfg = opaque;
207     target_ulong misa_bit = misa_ext_cfg->offset;
208     RISCVCPU *cpu = RISCV_CPU(obj);
209     CPURISCVState *env = &cpu->env;
210     bool value, host_bit;
211 
212     if (!visit_type_bool(v, name, &value, errp)) {
213         return;
214     }
215 
216     host_bit = env->misa_ext_mask & misa_bit;
217 
218     if (value == host_bit) {
219         return;
220     }
221 
222     if (!value) {
223         misa_ext_cfg->user_set = true;
224         return;
225     }
226 
227     /*
228      * Forbid users to enable extensions that aren't
229      * available in the hart.
230      */
231     error_setg(errp, "Enabling MISA bit '%s' is not allowed: it's not "
232                "enabled in the host", misa_ext_cfg->name);
233 }
234 
235 static void kvm_riscv_update_cpu_misa_ext(RISCVCPU *cpu, CPUState *cs)
236 {
237     CPURISCVState *env = &cpu->env;
238     uint64_t id, reg;
239     int i, ret;
240 
241     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
242         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
243         target_ulong misa_bit = misa_cfg->offset;
244 
245         if (!misa_cfg->user_set) {
246             continue;
247         }
248 
249         /* If we're here we're going to disable the MISA bit */
250         reg = 0;
251         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
252                                     misa_cfg->kvm_reg_id);
253         ret = kvm_set_one_reg(cs, id, &reg);
254         if (ret != 0) {
255             /*
256              * We're not checking for -EINVAL because if the bit is about
257              * to be disabled, it means that it was already enabled by
258              * KVM. We determined that by fetching the 'isa' register
259              * during init() time. Any error at this point is worth
260              * aborting.
261              */
262             error_report("Unable to set KVM reg %s, error %d",
263                          misa_cfg->name, ret);
264             exit(EXIT_FAILURE);
265         }
266         env->misa_ext &= ~misa_bit;
267     }
268 }
269 
270 #define KVM_EXT_CFG(_name, _prop, _reg_id) \
271     {.name = _name, .offset = CPU_CFG_OFFSET(_prop), \
272      .kvm_reg_id = _reg_id}
273 
274 static KVMCPUConfig kvm_multi_ext_cfgs[] = {
275     KVM_EXT_CFG("zicbom", ext_zicbom, KVM_RISCV_ISA_EXT_ZICBOM),
276     KVM_EXT_CFG("zicboz", ext_zicboz, KVM_RISCV_ISA_EXT_ZICBOZ),
277     KVM_EXT_CFG("zicntr", ext_zicntr, KVM_RISCV_ISA_EXT_ZICNTR),
278     KVM_EXT_CFG("zicond", ext_zicond, KVM_RISCV_ISA_EXT_ZICOND),
279     KVM_EXT_CFG("zicsr", ext_zicsr, KVM_RISCV_ISA_EXT_ZICSR),
280     KVM_EXT_CFG("zifencei", ext_zifencei, KVM_RISCV_ISA_EXT_ZIFENCEI),
281     KVM_EXT_CFG("zihintntl", ext_zihintntl, KVM_RISCV_ISA_EXT_ZIHINTNTL),
282     KVM_EXT_CFG("zihintpause", ext_zihintpause, KVM_RISCV_ISA_EXT_ZIHINTPAUSE),
283     KVM_EXT_CFG("zihpm", ext_zihpm, KVM_RISCV_ISA_EXT_ZIHPM),
284     KVM_EXT_CFG("zacas", ext_zacas, KVM_RISCV_ISA_EXT_ZACAS),
285     KVM_EXT_CFG("zfa", ext_zfa, KVM_RISCV_ISA_EXT_ZFA),
286     KVM_EXT_CFG("zfh", ext_zfh, KVM_RISCV_ISA_EXT_ZFH),
287     KVM_EXT_CFG("zfhmin", ext_zfhmin, KVM_RISCV_ISA_EXT_ZFHMIN),
288     KVM_EXT_CFG("zba", ext_zba, KVM_RISCV_ISA_EXT_ZBA),
289     KVM_EXT_CFG("zbb", ext_zbb, KVM_RISCV_ISA_EXT_ZBB),
290     KVM_EXT_CFG("zbc", ext_zbc, KVM_RISCV_ISA_EXT_ZBC),
291     KVM_EXT_CFG("zbkb", ext_zbkb, KVM_RISCV_ISA_EXT_ZBKB),
292     KVM_EXT_CFG("zbkc", ext_zbkc, KVM_RISCV_ISA_EXT_ZBKC),
293     KVM_EXT_CFG("zbkx", ext_zbkx, KVM_RISCV_ISA_EXT_ZBKX),
294     KVM_EXT_CFG("zbs", ext_zbs, KVM_RISCV_ISA_EXT_ZBS),
295     KVM_EXT_CFG("zknd", ext_zknd, KVM_RISCV_ISA_EXT_ZKND),
296     KVM_EXT_CFG("zkne", ext_zkne, KVM_RISCV_ISA_EXT_ZKNE),
297     KVM_EXT_CFG("zknh", ext_zknh, KVM_RISCV_ISA_EXT_ZKNH),
298     KVM_EXT_CFG("zkr", ext_zkr, KVM_RISCV_ISA_EXT_ZKR),
299     KVM_EXT_CFG("zksed", ext_zksed, KVM_RISCV_ISA_EXT_ZKSED),
300     KVM_EXT_CFG("zksh", ext_zksh, KVM_RISCV_ISA_EXT_ZKSH),
301     KVM_EXT_CFG("zkt", ext_zkt, KVM_RISCV_ISA_EXT_ZKT),
302     KVM_EXT_CFG("ztso", ext_ztso, KVM_RISCV_ISA_EXT_ZTSO),
303     KVM_EXT_CFG("zvbb", ext_zvbb, KVM_RISCV_ISA_EXT_ZVBB),
304     KVM_EXT_CFG("zvbc", ext_zvbc, KVM_RISCV_ISA_EXT_ZVBC),
305     KVM_EXT_CFG("zvfh", ext_zvfh, KVM_RISCV_ISA_EXT_ZVFH),
306     KVM_EXT_CFG("zvfhmin", ext_zvfhmin, KVM_RISCV_ISA_EXT_ZVFHMIN),
307     KVM_EXT_CFG("zvkb", ext_zvkb, KVM_RISCV_ISA_EXT_ZVKB),
308     KVM_EXT_CFG("zvkg", ext_zvkg, KVM_RISCV_ISA_EXT_ZVKG),
309     KVM_EXT_CFG("zvkned", ext_zvkned, KVM_RISCV_ISA_EXT_ZVKNED),
310     KVM_EXT_CFG("zvknha", ext_zvknha, KVM_RISCV_ISA_EXT_ZVKNHA),
311     KVM_EXT_CFG("zvknhb", ext_zvknhb, KVM_RISCV_ISA_EXT_ZVKNHB),
312     KVM_EXT_CFG("zvksed", ext_zvksed, KVM_RISCV_ISA_EXT_ZVKSED),
313     KVM_EXT_CFG("zvksh", ext_zvksh, KVM_RISCV_ISA_EXT_ZVKSH),
314     KVM_EXT_CFG("zvkt", ext_zvkt, KVM_RISCV_ISA_EXT_ZVKT),
315     KVM_EXT_CFG("smstateen", ext_smstateen, KVM_RISCV_ISA_EXT_SMSTATEEN),
316     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
317     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
318     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
319     KVM_EXT_CFG("svnapot", ext_svnapot, KVM_RISCV_ISA_EXT_SVNAPOT),
320     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
321 };
322 
323 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
324 {
325     return (void *)&cpu->cfg + kvmcfg->offset;
326 }
327 
328 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
329                             uint32_t val)
330 {
331     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
332 
333     *ext_enabled = val;
334 }
335 
336 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
337                                 KVMCPUConfig *multi_ext)
338 {
339     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
340 
341     return *ext_enabled;
342 }
343 
344 static void kvm_cpu_get_multi_ext_cfg(Object *obj, Visitor *v,
345                                       const char *name,
346                                       void *opaque, Error **errp)
347 {
348     KVMCPUConfig *multi_ext_cfg = opaque;
349     RISCVCPU *cpu = RISCV_CPU(obj);
350     bool value = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
351 
352     visit_type_bool(v, name, &value, errp);
353 }
354 
355 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
356                                       const char *name,
357                                       void *opaque, Error **errp)
358 {
359     KVMCPUConfig *multi_ext_cfg = opaque;
360     RISCVCPU *cpu = RISCV_CPU(obj);
361     bool value, host_val;
362 
363     if (!visit_type_bool(v, name, &value, errp)) {
364         return;
365     }
366 
367     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
368 
369     /*
370      * Ignore if the user is setting the same value
371      * as the host.
372      */
373     if (value == host_val) {
374         return;
375     }
376 
377     if (!multi_ext_cfg->supported) {
378         /*
379          * Error out if the user is trying to enable an
380          * extension that KVM doesn't support. Ignore
381          * option otherwise.
382          */
383         if (value) {
384             error_setg(errp, "KVM does not support disabling extension %s",
385                        multi_ext_cfg->name);
386         }
387 
388         return;
389     }
390 
391     multi_ext_cfg->user_set = true;
392     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
393 }
394 
395 static KVMCPUConfig kvm_cbom_blocksize = {
396     .name = "cbom_blocksize",
397     .offset = CPU_CFG_OFFSET(cbom_blocksize),
398     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
399 };
400 
401 static KVMCPUConfig kvm_cboz_blocksize = {
402     .name = "cboz_blocksize",
403     .offset = CPU_CFG_OFFSET(cboz_blocksize),
404     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
405 };
406 
407 static KVMCPUConfig kvm_v_vlenb = {
408     .name = "vlenb",
409     .offset = CPU_CFG_OFFSET(vlenb),
410     .kvm_reg_id =  KVM_REG_RISCV | KVM_REG_SIZE_U64 | KVM_REG_RISCV_VECTOR |
411                    KVM_REG_RISCV_VECTOR_CSR_REG(vlenb)
412 };
413 
414 static KVMCPUConfig kvm_sbi_dbcn = {
415     .name = "sbi_dbcn",
416     .kvm_reg_id = KVM_REG_RISCV | KVM_REG_SIZE_U64 |
417                   KVM_REG_RISCV_SBI_EXT | KVM_RISCV_SBI_EXT_DBCN
418 };
419 
420 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
421 {
422     CPURISCVState *env = &cpu->env;
423     uint64_t id, reg;
424     int i, ret;
425 
426     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
427         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
428 
429         if (!multi_ext_cfg->user_set) {
430             continue;
431         }
432 
433         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
434                                     multi_ext_cfg->kvm_reg_id);
435         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
436         ret = kvm_set_one_reg(cs, id, &reg);
437         if (ret != 0) {
438             if (!reg && ret == -EINVAL) {
439                 warn_report("KVM cannot disable extension %s",
440                             multi_ext_cfg->name);
441             } else {
442                 error_report("Unable to enable extension %s in KVM, error %d",
443                              multi_ext_cfg->name, ret);
444                 exit(EXIT_FAILURE);
445             }
446         }
447     }
448 }
449 
450 static void cpu_get_cfg_unavailable(Object *obj, Visitor *v,
451                                     const char *name,
452                                     void *opaque, Error **errp)
453 {
454     bool value = false;
455 
456     visit_type_bool(v, name, &value, errp);
457 }
458 
459 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
460                                     const char *name,
461                                     void *opaque, Error **errp)
462 {
463     const char *propname = opaque;
464     bool value;
465 
466     if (!visit_type_bool(v, name, &value, errp)) {
467         return;
468     }
469 
470     if (value) {
471         error_setg(errp, "'%s' is not available with KVM",
472                    propname);
473     }
474 }
475 
476 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
477 {
478     /* Check if KVM created the property already */
479     if (object_property_find(obj, prop_name)) {
480         return;
481     }
482 
483     /*
484      * Set the default to disabled for every extension
485      * unknown to KVM and error out if the user attempts
486      * to enable any of them.
487      */
488     object_property_add(obj, prop_name, "bool",
489                         cpu_get_cfg_unavailable,
490                         cpu_set_cfg_unavailable,
491                         NULL, (void *)prop_name);
492 }
493 
494 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
495                                         const RISCVCPUMultiExtConfig *array)
496 {
497     const RISCVCPUMultiExtConfig *prop;
498 
499     g_assert(array);
500 
501     for (prop = array; prop && prop->name; prop++) {
502         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
503     }
504 }
505 
506 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
507 {
508     int i;
509 
510     riscv_add_satp_mode_properties(cpu_obj);
511 
512     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
513         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
514         int bit = misa_cfg->offset;
515 
516         misa_cfg->name = riscv_get_misa_ext_name(bit);
517         misa_cfg->description = riscv_get_misa_ext_description(bit);
518 
519         object_property_add(cpu_obj, misa_cfg->name, "bool",
520                             kvm_cpu_get_misa_ext_cfg,
521                             kvm_cpu_set_misa_ext_cfg,
522                             NULL, misa_cfg);
523         object_property_set_description(cpu_obj, misa_cfg->name,
524                                         misa_cfg->description);
525     }
526 
527     for (i = 0; misa_bits[i] != 0; i++) {
528         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
529         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
530     }
531 
532     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
533         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
534 
535         object_property_add(cpu_obj, multi_cfg->name, "bool",
536                             kvm_cpu_get_multi_ext_cfg,
537                             kvm_cpu_set_multi_ext_cfg,
538                             NULL, multi_cfg);
539     }
540 
541     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
542     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
543     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
544 
545    /* We don't have the needed KVM support for profiles */
546     for (i = 0; riscv_profiles[i] != NULL; i++) {
547         riscv_cpu_add_kvm_unavail_prop(cpu_obj, riscv_profiles[i]->name);
548     }
549 }
550 
551 static int kvm_riscv_get_regs_core(CPUState *cs)
552 {
553     int ret = 0;
554     int i;
555     target_ulong reg;
556     CPURISCVState *env = &RISCV_CPU(cs)->env;
557 
558     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
559     if (ret) {
560         return ret;
561     }
562     env->pc = reg;
563 
564     for (i = 1; i < 32; i++) {
565         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
566         ret = kvm_get_one_reg(cs, id, &reg);
567         if (ret) {
568             return ret;
569         }
570         env->gpr[i] = reg;
571     }
572 
573     return ret;
574 }
575 
576 static int kvm_riscv_put_regs_core(CPUState *cs)
577 {
578     int ret = 0;
579     int i;
580     target_ulong reg;
581     CPURISCVState *env = &RISCV_CPU(cs)->env;
582 
583     reg = env->pc;
584     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
585     if (ret) {
586         return ret;
587     }
588 
589     for (i = 1; i < 32; i++) {
590         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
591         reg = env->gpr[i];
592         ret = kvm_set_one_reg(cs, id, &reg);
593         if (ret) {
594             return ret;
595         }
596     }
597 
598     return ret;
599 }
600 
601 static int kvm_riscv_get_regs_csr(CPUState *cs)
602 {
603     CPURISCVState *env = &RISCV_CPU(cs)->env;
604 
605     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
606     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
607     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
608     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
609     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
610     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
611     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
612     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
613     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
614 
615     return 0;
616 }
617 
618 static int kvm_riscv_put_regs_csr(CPUState *cs)
619 {
620     CPURISCVState *env = &RISCV_CPU(cs)->env;
621 
622     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
623     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
624     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
625     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
626     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
627     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
628     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
629     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
630     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
631 
632     return 0;
633 }
634 
635 static int kvm_riscv_get_regs_fp(CPUState *cs)
636 {
637     int ret = 0;
638     int i;
639     CPURISCVState *env = &RISCV_CPU(cs)->env;
640 
641     if (riscv_has_ext(env, RVD)) {
642         uint64_t reg;
643         for (i = 0; i < 32; i++) {
644             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
645             if (ret) {
646                 return ret;
647             }
648             env->fpr[i] = reg;
649         }
650         return ret;
651     }
652 
653     if (riscv_has_ext(env, RVF)) {
654         uint32_t reg;
655         for (i = 0; i < 32; i++) {
656             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
657             if (ret) {
658                 return ret;
659             }
660             env->fpr[i] = reg;
661         }
662         return ret;
663     }
664 
665     return ret;
666 }
667 
668 static int kvm_riscv_put_regs_fp(CPUState *cs)
669 {
670     int ret = 0;
671     int i;
672     CPURISCVState *env = &RISCV_CPU(cs)->env;
673 
674     if (riscv_has_ext(env, RVD)) {
675         uint64_t reg;
676         for (i = 0; i < 32; i++) {
677             reg = env->fpr[i];
678             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
679             if (ret) {
680                 return ret;
681             }
682         }
683         return ret;
684     }
685 
686     if (riscv_has_ext(env, RVF)) {
687         uint32_t reg;
688         for (i = 0; i < 32; i++) {
689             reg = env->fpr[i];
690             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
691             if (ret) {
692                 return ret;
693             }
694         }
695         return ret;
696     }
697 
698     return ret;
699 }
700 
701 static void kvm_riscv_get_regs_timer(CPUState *cs)
702 {
703     CPURISCVState *env = &RISCV_CPU(cs)->env;
704 
705     if (env->kvm_timer_dirty) {
706         return;
707     }
708 
709     KVM_RISCV_GET_TIMER(cs, time, env->kvm_timer_time);
710     KVM_RISCV_GET_TIMER(cs, compare, env->kvm_timer_compare);
711     KVM_RISCV_GET_TIMER(cs, state, env->kvm_timer_state);
712     KVM_RISCV_GET_TIMER(cs, frequency, env->kvm_timer_frequency);
713 
714     env->kvm_timer_dirty = true;
715 }
716 
717 static void kvm_riscv_put_regs_timer(CPUState *cs)
718 {
719     uint64_t reg;
720     CPURISCVState *env = &RISCV_CPU(cs)->env;
721 
722     if (!env->kvm_timer_dirty) {
723         return;
724     }
725 
726     KVM_RISCV_SET_TIMER(cs, time, env->kvm_timer_time);
727     KVM_RISCV_SET_TIMER(cs, compare, env->kvm_timer_compare);
728 
729     /*
730      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
731      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
732      * doesn't matter that adaping in QEMU now.
733      * TODO If KVM changes, adapt here.
734      */
735     if (env->kvm_timer_state) {
736         KVM_RISCV_SET_TIMER(cs, state, env->kvm_timer_state);
737     }
738 
739     /*
740      * For now, migration will not work between Hosts with different timer
741      * frequency. Therefore, we should check whether they are the same here
742      * during the migration.
743      */
744     if (migration_is_running()) {
745         KVM_RISCV_GET_TIMER(cs, frequency, reg);
746         if (reg != env->kvm_timer_frequency) {
747             error_report("Dst Hosts timer frequency != Src Hosts");
748         }
749     }
750 
751     env->kvm_timer_dirty = false;
752 }
753 
754 uint64_t kvm_riscv_get_timebase_frequency(CPUState *cs)
755 {
756     uint64_t reg;
757 
758     KVM_RISCV_GET_TIMER(cs, frequency, reg);
759 
760     return reg;
761 }
762 
763 static int kvm_riscv_get_regs_vector(CPUState *cs)
764 {
765     RISCVCPU *cpu = RISCV_CPU(cs);
766     CPURISCVState *env = &cpu->env;
767     target_ulong reg;
768     uint64_t vreg_id;
769     int vreg_idx, ret = 0;
770 
771     if (!riscv_has_ext(env, RVV)) {
772         return 0;
773     }
774 
775     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
776     if (ret) {
777         return ret;
778     }
779     env->vstart = reg;
780 
781     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
782     if (ret) {
783         return ret;
784     }
785     env->vl = reg;
786 
787     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
788     if (ret) {
789         return ret;
790     }
791     env->vtype = reg;
792 
793     if (kvm_v_vlenb.supported) {
794         ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
795         if (ret) {
796             return ret;
797         }
798         cpu->cfg.vlenb = reg;
799 
800         for (int i = 0; i < 32; i++) {
801             /*
802              * vreg[] is statically allocated using RV_VLEN_MAX.
803              * Use it instead of vlenb to calculate vreg_idx for
804              * simplicity.
805              */
806             vreg_idx = i * RV_VLEN_MAX / 64;
807             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
808 
809             ret = kvm_get_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
810             if (ret) {
811                 return ret;
812             }
813         }
814     }
815 
816     return 0;
817 }
818 
819 static int kvm_riscv_put_regs_vector(CPUState *cs)
820 {
821     RISCVCPU *cpu = RISCV_CPU(cs);
822     CPURISCVState *env = &cpu->env;
823     target_ulong reg;
824     uint64_t vreg_id;
825     int vreg_idx, ret = 0;
826 
827     if (!riscv_has_ext(env, RVV)) {
828         return 0;
829     }
830 
831     reg = env->vstart;
832     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
833     if (ret) {
834         return ret;
835     }
836 
837     reg = env->vl;
838     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
839     if (ret) {
840         return ret;
841     }
842 
843     reg = env->vtype;
844     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
845     if (ret) {
846         return ret;
847     }
848 
849     if (kvm_v_vlenb.supported) {
850         reg = cpu->cfg.vlenb;
851         ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vlenb), &reg);
852 
853         for (int i = 0; i < 32; i++) {
854             /*
855              * vreg[] is statically allocated using RV_VLEN_MAX.
856              * Use it instead of vlenb to calculate vreg_idx for
857              * simplicity.
858              */
859             vreg_idx = i * RV_VLEN_MAX / 64;
860             vreg_id = kvm_riscv_vector_reg_id(cpu, i);
861 
862             ret = kvm_set_one_reg(cs, vreg_id, &env->vreg[vreg_idx]);
863             if (ret) {
864                 return ret;
865             }
866         }
867     }
868 
869     return ret;
870 }
871 
872 typedef struct KVMScratchCPU {
873     int kvmfd;
874     int vmfd;
875     int cpufd;
876 } KVMScratchCPU;
877 
878 /*
879  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
880  * from target/arm/kvm.c.
881  */
882 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
883 {
884     int kvmfd = -1, vmfd = -1, cpufd = -1;
885 
886     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
887     if (kvmfd < 0) {
888         goto err;
889     }
890     do {
891         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
892     } while (vmfd == -1 && errno == EINTR);
893     if (vmfd < 0) {
894         goto err;
895     }
896     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
897     if (cpufd < 0) {
898         goto err;
899     }
900 
901     scratch->kvmfd =  kvmfd;
902     scratch->vmfd = vmfd;
903     scratch->cpufd = cpufd;
904 
905     return true;
906 
907  err:
908     if (cpufd >= 0) {
909         close(cpufd);
910     }
911     if (vmfd >= 0) {
912         close(vmfd);
913     }
914     if (kvmfd >= 0) {
915         close(kvmfd);
916     }
917 
918     return false;
919 }
920 
921 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
922 {
923     close(scratch->cpufd);
924     close(scratch->vmfd);
925     close(scratch->kvmfd);
926 }
927 
928 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
929 {
930     CPURISCVState *env = &cpu->env;
931     struct kvm_one_reg reg;
932     int ret;
933 
934     reg.id = RISCV_CONFIG_REG(env, mvendorid);
935     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
936     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
937     if (ret != 0) {
938         error_report("Unable to retrieve mvendorid from host, error %d", ret);
939     }
940 
941     reg.id = RISCV_CONFIG_REG(env, marchid);
942     reg.addr = (uint64_t)&cpu->cfg.marchid;
943     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
944     if (ret != 0) {
945         error_report("Unable to retrieve marchid from host, error %d", ret);
946     }
947 
948     reg.id = RISCV_CONFIG_REG(env, mimpid);
949     reg.addr = (uint64_t)&cpu->cfg.mimpid;
950     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
951     if (ret != 0) {
952         error_report("Unable to retrieve mimpid from host, error %d", ret);
953     }
954 }
955 
956 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
957                                          KVMScratchCPU *kvmcpu)
958 {
959     CPURISCVState *env = &cpu->env;
960     struct kvm_one_reg reg;
961     int ret;
962 
963     reg.id = RISCV_CONFIG_REG(env, isa);
964     reg.addr = (uint64_t)&env->misa_ext_mask;
965     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
966 
967     if (ret) {
968         error_report("Unable to fetch ISA register from KVM, "
969                      "error %d", ret);
970         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
971         exit(EXIT_FAILURE);
972     }
973 
974     env->misa_ext = env->misa_ext_mask;
975 }
976 
977 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
978                                          KVMCPUConfig *cbomz_cfg)
979 {
980     CPURISCVState *env = &cpu->env;
981     struct kvm_one_reg reg;
982     int ret;
983 
984     reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
985                                     cbomz_cfg->kvm_reg_id);
986     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
987     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
988     if (ret != 0) {
989         error_report("Unable to read KVM reg %s, error %d",
990                      cbomz_cfg->name, ret);
991         exit(EXIT_FAILURE);
992     }
993 }
994 
995 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
996                                            KVMScratchCPU *kvmcpu)
997 {
998     CPURISCVState *env = &cpu->env;
999     uint64_t val;
1000     int i, ret;
1001 
1002     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1003         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1004         struct kvm_one_reg reg;
1005 
1006         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
1007                                         multi_ext_cfg->kvm_reg_id);
1008         reg.addr = (uint64_t)&val;
1009         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1010         if (ret != 0) {
1011             if (errno == EINVAL) {
1012                 /* Silently default to 'false' if KVM does not support it. */
1013                 multi_ext_cfg->supported = false;
1014                 val = false;
1015             } else {
1016                 error_report("Unable to read ISA_EXT KVM register %s: %s",
1017                              multi_ext_cfg->name, strerror(errno));
1018                 exit(EXIT_FAILURE);
1019             }
1020         } else {
1021             multi_ext_cfg->supported = true;
1022         }
1023 
1024         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1025     }
1026 
1027     if (cpu->cfg.ext_zicbom) {
1028         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1029     }
1030 
1031     if (cpu->cfg.ext_zicboz) {
1032         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1033     }
1034 }
1035 
1036 static int uint64_cmp(const void *a, const void *b)
1037 {
1038     uint64_t val1 = *(const uint64_t *)a;
1039     uint64_t val2 = *(const uint64_t *)b;
1040 
1041     if (val1 < val2) {
1042         return -1;
1043     }
1044 
1045     if (val1 > val2) {
1046         return 1;
1047     }
1048 
1049     return 0;
1050 }
1051 
1052 static void kvm_riscv_check_sbi_dbcn_support(RISCVCPU *cpu,
1053                                              KVMScratchCPU *kvmcpu,
1054                                              struct kvm_reg_list *reglist)
1055 {
1056     struct kvm_reg_list *reg_search;
1057 
1058     reg_search = bsearch(&kvm_sbi_dbcn.kvm_reg_id, reglist->reg, reglist->n,
1059                          sizeof(uint64_t), uint64_cmp);
1060 
1061     if (reg_search) {
1062         kvm_sbi_dbcn.supported = true;
1063     }
1064 }
1065 
1066 static void kvm_riscv_read_vlenb(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
1067                                  struct kvm_reg_list *reglist)
1068 {
1069     struct kvm_one_reg reg;
1070     struct kvm_reg_list *reg_search;
1071     uint64_t val;
1072     int ret;
1073 
1074     reg_search = bsearch(&kvm_v_vlenb.kvm_reg_id, reglist->reg, reglist->n,
1075                          sizeof(uint64_t), uint64_cmp);
1076 
1077     if (reg_search) {
1078         reg.id = kvm_v_vlenb.kvm_reg_id;
1079         reg.addr = (uint64_t)&val;
1080 
1081         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1082         if (ret != 0) {
1083             error_report("Unable to read vlenb register, error code: %d",
1084                          errno);
1085             exit(EXIT_FAILURE);
1086         }
1087 
1088         kvm_v_vlenb.supported = true;
1089         cpu->cfg.vlenb = val;
1090     }
1091 }
1092 
1093 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
1094 {
1095     KVMCPUConfig *multi_ext_cfg;
1096     struct kvm_one_reg reg;
1097     struct kvm_reg_list rl_struct;
1098     struct kvm_reg_list *reglist;
1099     uint64_t val, reg_id, *reg_search;
1100     int i, ret;
1101 
1102     rl_struct.n = 0;
1103     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
1104 
1105     /*
1106      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
1107      * (EINVAL). Use read_legacy() in this case.
1108      */
1109     if (errno == EINVAL) {
1110         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
1111     } else if (errno != E2BIG) {
1112         /*
1113          * E2BIG is an expected error message for the API since we
1114          * don't know the number of registers. The right amount will
1115          * be written in rl_struct.n.
1116          *
1117          * Error out if we get any other errno.
1118          */
1119         error_report("Error when accessing get-reg-list: %s",
1120                      strerror(errno));
1121         exit(EXIT_FAILURE);
1122     }
1123 
1124     reglist = g_malloc(sizeof(struct kvm_reg_list) +
1125                        rl_struct.n * sizeof(uint64_t));
1126     reglist->n = rl_struct.n;
1127     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
1128     if (ret) {
1129         error_report("Error when reading KVM_GET_REG_LIST: %s",
1130                      strerror(errno));
1131         exit(EXIT_FAILURE);
1132     }
1133 
1134     /* sort reglist to use bsearch() */
1135     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
1136 
1137     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1138         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1139         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
1140                                         multi_ext_cfg->kvm_reg_id);
1141         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
1142                              sizeof(uint64_t), uint64_cmp);
1143         if (!reg_search) {
1144             continue;
1145         }
1146 
1147         reg.id = reg_id;
1148         reg.addr = (uint64_t)&val;
1149         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1150         if (ret != 0) {
1151             error_report("Unable to read ISA_EXT KVM register %s: %s",
1152                          multi_ext_cfg->name, strerror(errno));
1153             exit(EXIT_FAILURE);
1154         }
1155 
1156         multi_ext_cfg->supported = true;
1157         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1158     }
1159 
1160     if (cpu->cfg.ext_zicbom) {
1161         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1162     }
1163 
1164     if (cpu->cfg.ext_zicboz) {
1165         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1166     }
1167 
1168     if (riscv_has_ext(&cpu->env, RVV)) {
1169         kvm_riscv_read_vlenb(cpu, kvmcpu, reglist);
1170     }
1171 
1172     kvm_riscv_check_sbi_dbcn_support(cpu, kvmcpu, reglist);
1173 }
1174 
1175 static void riscv_init_kvm_registers(Object *cpu_obj)
1176 {
1177     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
1178     KVMScratchCPU kvmcpu;
1179 
1180     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1181         return;
1182     }
1183 
1184     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
1185     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
1186     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
1187 
1188     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1189 }
1190 
1191 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
1192     KVM_CAP_LAST_INFO
1193 };
1194 
1195 int kvm_arch_get_registers(CPUState *cs, Error **errp)
1196 {
1197     int ret = 0;
1198 
1199     ret = kvm_riscv_get_regs_core(cs);
1200     if (ret) {
1201         return ret;
1202     }
1203 
1204     ret = kvm_riscv_get_regs_csr(cs);
1205     if (ret) {
1206         return ret;
1207     }
1208 
1209     ret = kvm_riscv_get_regs_fp(cs);
1210     if (ret) {
1211         return ret;
1212     }
1213 
1214     ret = kvm_riscv_get_regs_vector(cs);
1215     if (ret) {
1216         return ret;
1217     }
1218 
1219     return ret;
1220 }
1221 
1222 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1223 {
1224     if (cap_has_mp_state) {
1225         struct kvm_mp_state mp_state = {
1226             .mp_state = state
1227         };
1228 
1229         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1230         if (ret) {
1231             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1232                     __func__, ret, strerror(-ret));
1233             return -1;
1234         }
1235     }
1236 
1237     return 0;
1238 }
1239 
1240 int kvm_arch_put_registers(CPUState *cs, int level, Error **errp)
1241 {
1242     int ret = 0;
1243 
1244     ret = kvm_riscv_put_regs_core(cs);
1245     if (ret) {
1246         return ret;
1247     }
1248 
1249     ret = kvm_riscv_put_regs_csr(cs);
1250     if (ret) {
1251         return ret;
1252     }
1253 
1254     ret = kvm_riscv_put_regs_fp(cs);
1255     if (ret) {
1256         return ret;
1257     }
1258 
1259     ret = kvm_riscv_put_regs_vector(cs);
1260     if (ret) {
1261         return ret;
1262     }
1263 
1264     if (KVM_PUT_RESET_STATE == level) {
1265         RISCVCPU *cpu = RISCV_CPU(cs);
1266         if (cs->cpu_index == 0) {
1267             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1268         } else {
1269             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1270         }
1271         if (ret) {
1272             return ret;
1273         }
1274     }
1275 
1276     return ret;
1277 }
1278 
1279 int kvm_arch_release_virq_post(int virq)
1280 {
1281     return 0;
1282 }
1283 
1284 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1285                              uint64_t address, uint32_t data, PCIDevice *dev)
1286 {
1287     return 0;
1288 }
1289 
1290 int kvm_arch_destroy_vcpu(CPUState *cs)
1291 {
1292     return 0;
1293 }
1294 
1295 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1296 {
1297     return cpu->cpu_index;
1298 }
1299 
1300 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1301                                       RunState state)
1302 {
1303     CPUState *cs = opaque;
1304 
1305     if (running) {
1306         kvm_riscv_put_regs_timer(cs);
1307     } else {
1308         kvm_riscv_get_regs_timer(cs);
1309     }
1310 }
1311 
1312 void kvm_arch_init_irq_routing(KVMState *s)
1313 {
1314 }
1315 
1316 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1317 {
1318     CPURISCVState *env = &cpu->env;
1319     target_ulong reg;
1320     uint64_t id;
1321     int ret;
1322 
1323     id = RISCV_CONFIG_REG(env, mvendorid);
1324     /*
1325      * cfg.mvendorid is an uint32 but a target_ulong will
1326      * be written. Assign it to a target_ulong var to avoid
1327      * writing pieces of other cpu->cfg fields in the reg.
1328      */
1329     reg = cpu->cfg.mvendorid;
1330     ret = kvm_set_one_reg(cs, id, &reg);
1331     if (ret != 0) {
1332         return ret;
1333     }
1334 
1335     id = RISCV_CONFIG_REG(env, marchid);
1336     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1337     if (ret != 0) {
1338         return ret;
1339     }
1340 
1341     id = RISCV_CONFIG_REG(env, mimpid);
1342     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1343 
1344     return ret;
1345 }
1346 
1347 static int kvm_vcpu_enable_sbi_dbcn(RISCVCPU *cpu, CPUState *cs)
1348 {
1349     target_ulong reg = 1;
1350 
1351     if (!kvm_sbi_dbcn.supported) {
1352         return 0;
1353     }
1354 
1355     return kvm_set_one_reg(cs, kvm_sbi_dbcn.kvm_reg_id, &reg);
1356 }
1357 
1358 int kvm_arch_init_vcpu(CPUState *cs)
1359 {
1360     int ret = 0;
1361     RISCVCPU *cpu = RISCV_CPU(cs);
1362 
1363     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1364 
1365     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1366         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1367         if (ret != 0) {
1368             return ret;
1369         }
1370     }
1371 
1372     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1373     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1374 
1375     ret = kvm_vcpu_enable_sbi_dbcn(cpu, cs);
1376 
1377     return ret;
1378 }
1379 
1380 int kvm_arch_msi_data_to_gsi(uint32_t data)
1381 {
1382     abort();
1383 }
1384 
1385 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1386                                 int vector, PCIDevice *dev)
1387 {
1388     return 0;
1389 }
1390 
1391 int kvm_arch_get_default_type(MachineState *ms)
1392 {
1393     return 0;
1394 }
1395 
1396 int kvm_arch_init(MachineState *ms, KVMState *s)
1397 {
1398     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1399     return 0;
1400 }
1401 
1402 int kvm_arch_irqchip_create(KVMState *s)
1403 {
1404     if (kvm_kernel_irqchip_split()) {
1405         error_report("-machine kernel_irqchip=split is not supported on RISC-V.");
1406         exit(1);
1407     }
1408 
1409     /*
1410      * We can create the VAIA using the newer device control API.
1411      */
1412     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1413 }
1414 
1415 int kvm_arch_process_async_events(CPUState *cs)
1416 {
1417     return 0;
1418 }
1419 
1420 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1421 {
1422 }
1423 
1424 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1425 {
1426     return MEMTXATTRS_UNSPECIFIED;
1427 }
1428 
1429 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1430 {
1431     return true;
1432 }
1433 
1434 static void kvm_riscv_handle_sbi_dbcn(CPUState *cs, struct kvm_run *run)
1435 {
1436     g_autofree uint8_t *buf = NULL;
1437     RISCVCPU *cpu = RISCV_CPU(cs);
1438     target_ulong num_bytes;
1439     uint64_t addr;
1440     unsigned char ch;
1441     int ret;
1442 
1443     switch (run->riscv_sbi.function_id) {
1444     case SBI_EXT_DBCN_CONSOLE_READ:
1445     case SBI_EXT_DBCN_CONSOLE_WRITE:
1446         num_bytes = run->riscv_sbi.args[0];
1447 
1448         if (num_bytes == 0) {
1449             run->riscv_sbi.ret[0] = SBI_SUCCESS;
1450             run->riscv_sbi.ret[1] = 0;
1451             break;
1452         }
1453 
1454         addr = run->riscv_sbi.args[1];
1455 
1456         /*
1457          * Handle the case where a 32 bit CPU is running in a
1458          * 64 bit addressing env.
1459          */
1460         if (riscv_cpu_mxl(&cpu->env) == MXL_RV32) {
1461             addr |= (uint64_t)run->riscv_sbi.args[2] << 32;
1462         }
1463 
1464         buf = g_malloc0(num_bytes);
1465 
1466         if (run->riscv_sbi.function_id == SBI_EXT_DBCN_CONSOLE_READ) {
1467             ret = qemu_chr_fe_read_all(serial_hd(0)->be, buf, num_bytes);
1468             if (ret < 0) {
1469                 error_report("SBI_EXT_DBCN_CONSOLE_READ: error when "
1470                              "reading chardev");
1471                 exit(1);
1472             }
1473 
1474             cpu_physical_memory_write(addr, buf, ret);
1475         } else {
1476             cpu_physical_memory_read(addr, buf, num_bytes);
1477 
1478             ret = qemu_chr_fe_write_all(serial_hd(0)->be, buf, num_bytes);
1479             if (ret < 0) {
1480                 error_report("SBI_EXT_DBCN_CONSOLE_WRITE: error when "
1481                              "writing chardev");
1482                 exit(1);
1483             }
1484         }
1485 
1486         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1487         run->riscv_sbi.ret[1] = ret;
1488         break;
1489     case SBI_EXT_DBCN_CONSOLE_WRITE_BYTE:
1490         ch = run->riscv_sbi.args[0];
1491         ret = qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1492 
1493         if (ret < 0) {
1494             error_report("SBI_EXT_DBCN_CONSOLE_WRITE_BYTE: error when "
1495                          "writing chardev");
1496             exit(1);
1497         }
1498 
1499         run->riscv_sbi.ret[0] = SBI_SUCCESS;
1500         run->riscv_sbi.ret[1] = 0;
1501         break;
1502     default:
1503         run->riscv_sbi.ret[0] = SBI_ERR_NOT_SUPPORTED;
1504     }
1505 }
1506 
1507 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1508 {
1509     int ret = 0;
1510     unsigned char ch;
1511     switch (run->riscv_sbi.extension_id) {
1512     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1513         ch = run->riscv_sbi.args[0];
1514         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1515         break;
1516     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1517         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1518         if (ret == sizeof(ch)) {
1519             run->riscv_sbi.ret[0] = ch;
1520         } else {
1521             run->riscv_sbi.ret[0] = -1;
1522         }
1523         ret = 0;
1524         break;
1525     case SBI_EXT_DBCN:
1526         kvm_riscv_handle_sbi_dbcn(cs, run);
1527         break;
1528     default:
1529         qemu_log_mask(LOG_UNIMP,
1530                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1531                       __func__, run->riscv_sbi.extension_id);
1532         ret = -1;
1533         break;
1534     }
1535     return ret;
1536 }
1537 
1538 static int kvm_riscv_handle_csr(CPUState *cs, struct kvm_run *run)
1539 {
1540     target_ulong csr_num = run->riscv_csr.csr_num;
1541     target_ulong new_value = run->riscv_csr.new_value;
1542     target_ulong write_mask = run->riscv_csr.write_mask;
1543     int ret = 0;
1544 
1545     switch (csr_num) {
1546     case CSR_SEED:
1547         run->riscv_csr.ret_value = riscv_new_csr_seed(new_value, write_mask);
1548         break;
1549     default:
1550         qemu_log_mask(LOG_UNIMP,
1551                       "%s: un-handled CSR EXIT for CSR %lx\n",
1552                       __func__, csr_num);
1553         ret = -1;
1554         break;
1555     }
1556 
1557     return ret;
1558 }
1559 
1560 static bool kvm_riscv_handle_debug(CPUState *cs)
1561 {
1562     RISCVCPU *cpu = RISCV_CPU(cs);
1563     CPURISCVState *env = &cpu->env;
1564 
1565     /* Ensure PC is synchronised */
1566     kvm_cpu_synchronize_state(cs);
1567 
1568     if (kvm_find_sw_breakpoint(cs, env->pc)) {
1569         return true;
1570     }
1571 
1572     return false;
1573 }
1574 
1575 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1576 {
1577     int ret = 0;
1578     switch (run->exit_reason) {
1579     case KVM_EXIT_RISCV_SBI:
1580         ret = kvm_riscv_handle_sbi(cs, run);
1581         break;
1582     case KVM_EXIT_RISCV_CSR:
1583         ret = kvm_riscv_handle_csr(cs, run);
1584         break;
1585     case KVM_EXIT_DEBUG:
1586         if (kvm_riscv_handle_debug(cs)) {
1587             ret = EXCP_DEBUG;
1588         }
1589         break;
1590     default:
1591         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1592                       __func__, run->exit_reason);
1593         ret = -1;
1594         break;
1595     }
1596     return ret;
1597 }
1598 
1599 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1600 {
1601     CPURISCVState *env = &cpu->env;
1602     int i;
1603 
1604     if (!kvm_enabled()) {
1605         return;
1606     }
1607     for (i = 0; i < 32; i++) {
1608         env->gpr[i] = 0;
1609     }
1610     env->pc = cpu->env.kernel_addr;
1611     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1612     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1613     env->satp = 0;
1614     env->mie = 0;
1615     env->stvec = 0;
1616     env->sscratch = 0;
1617     env->sepc = 0;
1618     env->scause = 0;
1619     env->stval = 0;
1620     env->mip = 0;
1621 }
1622 
1623 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1624 {
1625     int ret;
1626     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1627 
1628     if (irq != IRQ_S_EXT) {
1629         perror("kvm riscv set irq != IRQ_S_EXT\n");
1630         abort();
1631     }
1632 
1633     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1634     if (ret < 0) {
1635         perror("Set irq failed");
1636         abort();
1637     }
1638 }
1639 
1640 static int aia_mode;
1641 
1642 static const char *kvm_aia_mode_str(uint64_t mode)
1643 {
1644     switch (mode) {
1645     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1646         return "emul";
1647     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1648         return "hwaccel";
1649     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1650     default:
1651         return "auto";
1652     };
1653 }
1654 
1655 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1656 {
1657     return g_strdup(kvm_aia_mode_str(aia_mode));
1658 }
1659 
1660 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1661 {
1662     if (!strcmp(val, "emul")) {
1663         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1664     } else if (!strcmp(val, "hwaccel")) {
1665         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1666     } else if (!strcmp(val, "auto")) {
1667         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1668     } else {
1669         error_setg(errp, "Invalid KVM AIA mode");
1670         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1671     }
1672 }
1673 
1674 void kvm_arch_accel_class_init(ObjectClass *oc)
1675 {
1676     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1677                                   riscv_set_kvm_aia);
1678     object_class_property_set_description(oc, "riscv-aia",
1679         "Set KVM AIA mode. Valid values are 'emul', 'hwaccel' and 'auto'. "
1680         "Changing KVM AIA modes relies on host support. Defaults to 'auto' "
1681         "if the host supports it");
1682     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1683                                     "auto");
1684 }
1685 
1686 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1687                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1688                           uint64_t aplic_base, uint64_t imsic_base,
1689                           uint64_t guest_num)
1690 {
1691     int ret, i;
1692     int aia_fd = -1;
1693     uint64_t default_aia_mode;
1694     uint64_t socket_count = riscv_socket_count(machine);
1695     uint64_t max_hart_per_socket = 0;
1696     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1697     uint64_t socket_bits, hart_bits, guest_bits;
1698     uint64_t max_group_id;
1699 
1700     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1701 
1702     if (aia_fd < 0) {
1703         error_report("Unable to create in-kernel irqchip");
1704         exit(1);
1705     }
1706 
1707     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1708                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1709                             &default_aia_mode, false, NULL);
1710     if (ret < 0) {
1711         error_report("KVM AIA: failed to get current KVM AIA mode");
1712         exit(1);
1713     }
1714 
1715     if (default_aia_mode != aia_mode) {
1716         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1717                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1718                                 &aia_mode, true, NULL);
1719         if (ret < 0) {
1720             warn_report("KVM AIA: failed to set KVM AIA mode '%s', using "
1721                         "default host mode '%s'",
1722                         kvm_aia_mode_str(aia_mode),
1723                         kvm_aia_mode_str(default_aia_mode));
1724 
1725             /* failed to change AIA mode, use default */
1726             aia_mode = default_aia_mode;
1727         }
1728     }
1729 
1730     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1731                             KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1732                             &aia_irq_num, true, NULL);
1733     if (ret < 0) {
1734         error_report("KVM AIA: failed to set number of input irq lines");
1735         exit(1);
1736     }
1737 
1738     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1739                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1740                             &aia_msi_num, true, NULL);
1741     if (ret < 0) {
1742         error_report("KVM AIA: failed to set number of msi");
1743         exit(1);
1744     }
1745 
1746 
1747     if (socket_count > 1) {
1748         max_group_id = socket_count - 1;
1749         socket_bits = find_last_bit(&max_group_id, BITS_PER_LONG) + 1;
1750         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1751                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1752                                 &socket_bits, true, NULL);
1753         if (ret < 0) {
1754             error_report("KVM AIA: failed to set group_bits");
1755             exit(1);
1756         }
1757 
1758         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1759                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1760                                 &group_shift, true, NULL);
1761         if (ret < 0) {
1762             error_report("KVM AIA: failed to set group_shift");
1763             exit(1);
1764         }
1765     }
1766 
1767     guest_bits = guest_num == 0 ? 0 :
1768                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1769     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1770                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1771                             &guest_bits, true, NULL);
1772     if (ret < 0) {
1773         error_report("KVM AIA: failed to set guest_bits");
1774         exit(1);
1775     }
1776 
1777     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1778                             KVM_DEV_RISCV_AIA_ADDR_APLIC,
1779                             &aplic_base, true, NULL);
1780     if (ret < 0) {
1781         error_report("KVM AIA: failed to set the base address of APLIC");
1782         exit(1);
1783     }
1784 
1785     for (socket = 0; socket < socket_count; socket++) {
1786         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1787         hart_count = riscv_socket_hart_count(machine, socket);
1788         base_hart = riscv_socket_first_hartid(machine, socket);
1789 
1790         if (max_hart_per_socket < hart_count) {
1791             max_hart_per_socket = hart_count;
1792         }
1793 
1794         for (i = 0; i < hart_count; i++) {
1795             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1796             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1797                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1798                                     &imsic_addr, true, NULL);
1799             if (ret < 0) {
1800                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1801                 exit(1);
1802             }
1803         }
1804     }
1805 
1806 
1807     if (max_hart_per_socket > 1) {
1808         max_hart_per_socket--;
1809         hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1810     } else {
1811         hart_bits = 0;
1812     }
1813 
1814     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1815                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1816                             &hart_bits, true, NULL);
1817     if (ret < 0) {
1818         error_report("KVM AIA: failed to set hart_bits");
1819         exit(1);
1820     }
1821 
1822     if (kvm_has_gsi_routing()) {
1823         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1824             /* KVM AIA only has one APLIC instance */
1825             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1826         }
1827         kvm_gsi_routing_allowed = true;
1828         kvm_irqchip_commit_routes(kvm_state);
1829     }
1830 
1831     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1832                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1833                             NULL, true, NULL);
1834     if (ret < 0) {
1835         error_report("KVM AIA: initialized fail");
1836         exit(1);
1837     }
1838 
1839     kvm_msi_via_irqfd_allowed = true;
1840 }
1841 
1842 static void kvm_cpu_instance_init(CPUState *cs)
1843 {
1844     Object *obj = OBJECT(RISCV_CPU(cs));
1845 
1846     riscv_init_kvm_registers(obj);
1847 
1848     kvm_riscv_add_cpu_user_properties(obj);
1849 }
1850 
1851 /*
1852  * We'll get here via the following path:
1853  *
1854  * riscv_cpu_realize()
1855  *   -> cpu_exec_realizefn()
1856  *      -> kvm_cpu_realize() (via accel_cpu_common_realize())
1857  */
1858 static bool kvm_cpu_realize(CPUState *cs, Error **errp)
1859 {
1860     RISCVCPU *cpu = RISCV_CPU(cs);
1861     int ret;
1862 
1863     if (riscv_has_ext(&cpu->env, RVV)) {
1864         ret = prctl(PR_RISCV_V_SET_CONTROL, PR_RISCV_V_VSTATE_CTRL_ON);
1865         if (ret) {
1866             error_setg(errp, "Error in prctl PR_RISCV_V_SET_CONTROL, code: %s",
1867                        strerrorname_np(errno));
1868             return false;
1869         }
1870     }
1871 
1872    return true;
1873 }
1874 
1875 void riscv_kvm_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
1876 {
1877     CPURISCVState *env = &cpu->env;
1878     KVMScratchCPU kvmcpu;
1879     struct kvm_one_reg reg;
1880     uint64_t val;
1881     int ret;
1882 
1883     /* short-circuit without spinning the scratch CPU */
1884     if (!cpu->cfg.ext_zicbom && !cpu->cfg.ext_zicboz &&
1885         !riscv_has_ext(env, RVV)) {
1886         return;
1887     }
1888 
1889     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1890         error_setg(errp, "Unable to create scratch KVM cpu");
1891         return;
1892     }
1893 
1894     if (cpu->cfg.ext_zicbom &&
1895         riscv_cpu_option_set(kvm_cbom_blocksize.name)) {
1896 
1897         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1898                                         kvm_cbom_blocksize.kvm_reg_id);
1899         reg.addr = (uint64_t)&val;
1900         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1901         if (ret != 0) {
1902             error_setg(errp, "Unable to read cbom_blocksize, error %d", errno);
1903             return;
1904         }
1905 
1906         if (cpu->cfg.cbom_blocksize != val) {
1907             error_setg(errp, "Unable to set cbom_blocksize to a different "
1908                        "value than the host (%lu)", val);
1909             return;
1910         }
1911     }
1912 
1913     if (cpu->cfg.ext_zicboz &&
1914         riscv_cpu_option_set(kvm_cboz_blocksize.name)) {
1915 
1916         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1917                                         kvm_cboz_blocksize.kvm_reg_id);
1918         reg.addr = (uint64_t)&val;
1919         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1920         if (ret != 0) {
1921             error_setg(errp, "Unable to read cboz_blocksize, error %d", errno);
1922             return;
1923         }
1924 
1925         if (cpu->cfg.cboz_blocksize != val) {
1926             error_setg(errp, "Unable to set cboz_blocksize to a different "
1927                        "value than the host (%lu)", val);
1928             return;
1929         }
1930     }
1931 
1932     /* Users are setting vlen, not vlenb */
1933     if (riscv_has_ext(env, RVV) && riscv_cpu_option_set("vlen")) {
1934         if (!kvm_v_vlenb.supported) {
1935             error_setg(errp, "Unable to set 'vlenb': register not supported");
1936             return;
1937         }
1938 
1939         reg.id = kvm_v_vlenb.kvm_reg_id;
1940         reg.addr = (uint64_t)&val;
1941         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1942         if (ret != 0) {
1943             error_setg(errp, "Unable to read vlenb register, error %d", errno);
1944             return;
1945         }
1946 
1947         if (cpu->cfg.vlenb != val) {
1948             error_setg(errp, "Unable to set 'vlen' to a different "
1949                        "value than the host (%lu)", val * 8);
1950             return;
1951         }
1952     }
1953 
1954     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1955 }
1956 
1957 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1958 {
1959     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1960 
1961     acc->cpu_instance_init = kvm_cpu_instance_init;
1962     acc->cpu_target_realize = kvm_cpu_realize;
1963 }
1964 
1965 static const TypeInfo kvm_cpu_accel_type_info = {
1966     .name = ACCEL_CPU_NAME("kvm"),
1967 
1968     .parent = TYPE_ACCEL_CPU,
1969     .class_init = kvm_cpu_accel_class_init,
1970     .abstract = true,
1971 };
1972 static void kvm_cpu_accel_register_types(void)
1973 {
1974     type_register_static(&kvm_cpu_accel_type_info);
1975 }
1976 type_init(kvm_cpu_accel_register_types);
1977 
1978 static void riscv_host_cpu_class_init(ObjectClass *c, void *data)
1979 {
1980     RISCVCPUClass *mcc = RISCV_CPU_CLASS(c);
1981 
1982 #if defined(TARGET_RISCV32)
1983     mcc->misa_mxl_max = MXL_RV32;
1984 #elif defined(TARGET_RISCV64)
1985     mcc->misa_mxl_max = MXL_RV64;
1986 #endif
1987 }
1988 
1989 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
1990     {
1991         .name = TYPE_RISCV_CPU_HOST,
1992         .parent = TYPE_RISCV_CPU,
1993         .class_init = riscv_host_cpu_class_init,
1994     }
1995 };
1996 
1997 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
1998 
1999 static const uint32_t ebreak_insn = 0x00100073;
2000 static const uint16_t c_ebreak_insn = 0x9002;
2001 
2002 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2003 {
2004     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 0)) {
2005         return -EINVAL;
2006     }
2007 
2008     if ((bp->saved_insn & 0x3) == 0x3) {
2009         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 0)
2010             || cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak_insn, 4, 1)) {
2011             return -EINVAL;
2012         }
2013     } else {
2014         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak_insn, 2, 1)) {
2015             return -EINVAL;
2016         }
2017     }
2018 
2019     return 0;
2020 }
2021 
2022 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
2023 {
2024     uint32_t ebreak;
2025     uint16_t c_ebreak;
2026 
2027     if ((bp->saved_insn & 0x3) == 0x3) {
2028         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&ebreak, 4, 0) ||
2029             ebreak != ebreak_insn ||
2030             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 1)) {
2031             return -EINVAL;
2032         }
2033     } else {
2034         if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&c_ebreak, 2, 0) ||
2035             c_ebreak != c_ebreak_insn ||
2036             cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 2, 1)) {
2037             return -EINVAL;
2038         }
2039     }
2040 
2041     return 0;
2042 }
2043 
2044 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type)
2045 {
2046     /* TODO; To be implemented later. */
2047     return -EINVAL;
2048 }
2049 
2050 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type)
2051 {
2052     /* TODO; To be implemented later. */
2053     return -EINVAL;
2054 }
2055 
2056 void kvm_arch_remove_all_hw_breakpoints(void)
2057 {
2058     /* TODO; To be implemented later. */
2059 }
2060 
2061 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
2062 {
2063     if (kvm_sw_breakpoints_active(cs)) {
2064         dbg->control |= KVM_GUESTDBG_ENABLE;
2065     }
2066 }
2067