xref: /qemu/target/riscv/kvm/kvm-cpu.c (revision 456a65546f6761aab09275ed7055d113608e3bb2)
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 
22 #include <linux/kvm.h>
23 
24 #include "qemu/timer.h"
25 #include "qapi/error.h"
26 #include "qemu/error-report.h"
27 #include "qemu/main-loop.h"
28 #include "qapi/visitor.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/kvm.h"
31 #include "sysemu/kvm_int.h"
32 #include "cpu.h"
33 #include "trace.h"
34 #include "hw/core/accel-cpu.h"
35 #include "hw/pci/pci.h"
36 #include "exec/memattrs.h"
37 #include "exec/address-spaces.h"
38 #include "hw/boards.h"
39 #include "hw/irq.h"
40 #include "hw/intc/riscv_imsic.h"
41 #include "qemu/log.h"
42 #include "hw/loader.h"
43 #include "kvm_riscv.h"
44 #include "sbi_ecall_interface.h"
45 #include "chardev/char-fe.h"
46 #include "migration/migration.h"
47 #include "sysemu/runstate.h"
48 #include "hw/riscv/numa.h"
49 
50 void riscv_kvm_aplic_request(void *opaque, int irq, int level)
51 {
52     kvm_set_irq(kvm_state, irq, !!level);
53 }
54 
55 static bool cap_has_mp_state;
56 
57 static uint64_t kvm_riscv_reg_id(CPURISCVState *env, uint64_t type,
58                                  uint64_t idx)
59 {
60     uint64_t id = KVM_REG_RISCV | type | idx;
61 
62     switch (riscv_cpu_mxl(env)) {
63     case MXL_RV32:
64         id |= KVM_REG_SIZE_U32;
65         break;
66     case MXL_RV64:
67         id |= KVM_REG_SIZE_U64;
68         break;
69     default:
70         g_assert_not_reached();
71     }
72     return id;
73 }
74 
75 #define RISCV_CORE_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, \
76                  KVM_REG_RISCV_CORE_REG(name))
77 
78 #define RISCV_CSR_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_CSR, \
79                  KVM_REG_RISCV_CSR_REG(name))
80 
81 #define RISCV_TIMER_REG(env, name)  kvm_riscv_reg_id(env, KVM_REG_RISCV_TIMER, \
82                  KVM_REG_RISCV_TIMER_REG(name))
83 
84 #define RISCV_FP_F_REG(env, idx)  kvm_riscv_reg_id(env, KVM_REG_RISCV_FP_F, idx)
85 
86 #define RISCV_FP_D_REG(env, idx)  kvm_riscv_reg_id(env, KVM_REG_RISCV_FP_D, idx)
87 
88 #define KVM_RISCV_GET_CSR(cs, env, csr, reg) \
89     do { \
90         int ret = kvm_get_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
91         if (ret) { \
92             return ret; \
93         } \
94     } while (0)
95 
96 #define KVM_RISCV_SET_CSR(cs, env, csr, reg) \
97     do { \
98         int ret = kvm_set_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
99         if (ret) { \
100             return ret; \
101         } \
102     } while (0)
103 
104 #define KVM_RISCV_GET_TIMER(cs, env, name, reg) \
105     do { \
106         int ret = kvm_get_one_reg(cs, RISCV_TIMER_REG(env, name), &reg); \
107         if (ret) { \
108             abort(); \
109         } \
110     } while (0)
111 
112 #define KVM_RISCV_SET_TIMER(cs, env, name, reg) \
113     do { \
114         int ret = kvm_set_one_reg(cs, RISCV_TIMER_REG(env, name), &reg); \
115         if (ret) { \
116             abort(); \
117         } \
118     } while (0)
119 
120 typedef struct KVMCPUConfig {
121     const char *name;
122     const char *description;
123     target_ulong offset;
124     int kvm_reg_id;
125     bool user_set;
126     bool supported;
127 } KVMCPUConfig;
128 
129 #define KVM_MISA_CFG(_bit, _reg_id) \
130     {.offset = _bit, .kvm_reg_id = _reg_id}
131 
132 /* KVM ISA extensions */
133 static KVMCPUConfig kvm_misa_ext_cfgs[] = {
134     KVM_MISA_CFG(RVA, KVM_RISCV_ISA_EXT_A),
135     KVM_MISA_CFG(RVC, KVM_RISCV_ISA_EXT_C),
136     KVM_MISA_CFG(RVD, KVM_RISCV_ISA_EXT_D),
137     KVM_MISA_CFG(RVF, KVM_RISCV_ISA_EXT_F),
138     KVM_MISA_CFG(RVH, KVM_RISCV_ISA_EXT_H),
139     KVM_MISA_CFG(RVI, KVM_RISCV_ISA_EXT_I),
140     KVM_MISA_CFG(RVM, KVM_RISCV_ISA_EXT_M),
141 };
142 
143 static void kvm_cpu_get_misa_ext_cfg(Object *obj, Visitor *v,
144                                      const char *name,
145                                      void *opaque, Error **errp)
146 {
147     KVMCPUConfig *misa_ext_cfg = opaque;
148     target_ulong misa_bit = misa_ext_cfg->offset;
149     RISCVCPU *cpu = RISCV_CPU(obj);
150     CPURISCVState *env = &cpu->env;
151     bool value = env->misa_ext_mask & misa_bit;
152 
153     visit_type_bool(v, name, &value, errp);
154 }
155 
156 static void kvm_cpu_set_misa_ext_cfg(Object *obj, Visitor *v,
157                                      const char *name,
158                                      void *opaque, Error **errp)
159 {
160     KVMCPUConfig *misa_ext_cfg = opaque;
161     target_ulong misa_bit = misa_ext_cfg->offset;
162     RISCVCPU *cpu = RISCV_CPU(obj);
163     CPURISCVState *env = &cpu->env;
164     bool value, host_bit;
165 
166     if (!visit_type_bool(v, name, &value, errp)) {
167         return;
168     }
169 
170     host_bit = env->misa_ext_mask & misa_bit;
171 
172     if (value == host_bit) {
173         return;
174     }
175 
176     if (!value) {
177         misa_ext_cfg->user_set = true;
178         return;
179     }
180 
181     /*
182      * Forbid users to enable extensions that aren't
183      * available in the hart.
184      */
185     error_setg(errp, "Enabling MISA bit '%s' is not allowed: it's not "
186                "enabled in the host", misa_ext_cfg->name);
187 }
188 
189 static void kvm_riscv_update_cpu_misa_ext(RISCVCPU *cpu, CPUState *cs)
190 {
191     CPURISCVState *env = &cpu->env;
192     uint64_t id, reg;
193     int i, ret;
194 
195     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
196         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
197         target_ulong misa_bit = misa_cfg->offset;
198 
199         if (!misa_cfg->user_set) {
200             continue;
201         }
202 
203         /* If we're here we're going to disable the MISA bit */
204         reg = 0;
205         id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
206                               misa_cfg->kvm_reg_id);
207         ret = kvm_set_one_reg(cs, id, &reg);
208         if (ret != 0) {
209             /*
210              * We're not checking for -EINVAL because if the bit is about
211              * to be disabled, it means that it was already enabled by
212              * KVM. We determined that by fetching the 'isa' register
213              * during init() time. Any error at this point is worth
214              * aborting.
215              */
216             error_report("Unable to set KVM reg %s, error %d",
217                          misa_cfg->name, ret);
218             exit(EXIT_FAILURE);
219         }
220         env->misa_ext &= ~misa_bit;
221     }
222 }
223 
224 #define KVM_EXT_CFG(_name, _prop, _reg_id) \
225     {.name = _name, .offset = CPU_CFG_OFFSET(_prop), \
226      .kvm_reg_id = _reg_id}
227 
228 static KVMCPUConfig kvm_multi_ext_cfgs[] = {
229     KVM_EXT_CFG("zicbom", ext_zicbom, KVM_RISCV_ISA_EXT_ZICBOM),
230     KVM_EXT_CFG("zicboz", ext_zicboz, KVM_RISCV_ISA_EXT_ZICBOZ),
231     KVM_EXT_CFG("zihintpause", ext_zihintpause, KVM_RISCV_ISA_EXT_ZIHINTPAUSE),
232     KVM_EXT_CFG("zbb", ext_zbb, KVM_RISCV_ISA_EXT_ZBB),
233     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
234     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
235     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
236     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
237 };
238 
239 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
240 {
241     return (void *)&cpu->cfg + kvmcfg->offset;
242 }
243 
244 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
245                             uint32_t val)
246 {
247     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
248 
249     *ext_enabled = val;
250 }
251 
252 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
253                                 KVMCPUConfig *multi_ext)
254 {
255     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
256 
257     return *ext_enabled;
258 }
259 
260 static void kvm_cpu_get_multi_ext_cfg(Object *obj, Visitor *v,
261                                       const char *name,
262                                       void *opaque, Error **errp)
263 {
264     KVMCPUConfig *multi_ext_cfg = opaque;
265     RISCVCPU *cpu = RISCV_CPU(obj);
266     bool value = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
267 
268     visit_type_bool(v, name, &value, errp);
269 }
270 
271 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
272                                       const char *name,
273                                       void *opaque, Error **errp)
274 {
275     KVMCPUConfig *multi_ext_cfg = opaque;
276     RISCVCPU *cpu = RISCV_CPU(obj);
277     bool value, host_val;
278 
279     if (!visit_type_bool(v, name, &value, errp)) {
280         return;
281     }
282 
283     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
284 
285     /*
286      * Ignore if the user is setting the same value
287      * as the host.
288      */
289     if (value == host_val) {
290         return;
291     }
292 
293     if (!multi_ext_cfg->supported) {
294         /*
295          * Error out if the user is trying to enable an
296          * extension that KVM doesn't support. Ignore
297          * option otherwise.
298          */
299         if (value) {
300             error_setg(errp, "KVM does not support disabling extension %s",
301                        multi_ext_cfg->name);
302         }
303 
304         return;
305     }
306 
307     multi_ext_cfg->user_set = true;
308     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
309 }
310 
311 static KVMCPUConfig kvm_cbom_blocksize = {
312     .name = "cbom_blocksize",
313     .offset = CPU_CFG_OFFSET(cbom_blocksize),
314     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
315 };
316 
317 static KVMCPUConfig kvm_cboz_blocksize = {
318     .name = "cboz_blocksize",
319     .offset = CPU_CFG_OFFSET(cboz_blocksize),
320     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
321 };
322 
323 static void kvm_cpu_set_cbomz_blksize(Object *obj, Visitor *v,
324                                       const char *name,
325                                       void *opaque, Error **errp)
326 {
327     KVMCPUConfig *cbomz_cfg = opaque;
328     RISCVCPU *cpu = RISCV_CPU(obj);
329     uint16_t value, *host_val;
330 
331     if (!visit_type_uint16(v, name, &value, errp)) {
332         return;
333     }
334 
335     host_val = kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
336 
337     if (value != *host_val) {
338         error_report("Unable to set %s to a different value than "
339                      "the host (%u)",
340                      cbomz_cfg->name, *host_val);
341         exit(EXIT_FAILURE);
342     }
343 
344     cbomz_cfg->user_set = true;
345 }
346 
347 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
348 {
349     CPURISCVState *env = &cpu->env;
350     uint64_t id, reg;
351     int i, ret;
352 
353     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
354         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
355 
356         if (!multi_ext_cfg->user_set) {
357             continue;
358         }
359 
360         id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
361                               multi_ext_cfg->kvm_reg_id);
362         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
363         ret = kvm_set_one_reg(cs, id, &reg);
364         if (ret != 0) {
365             error_report("Unable to %s extension %s in KVM, error %d",
366                          reg ? "enable" : "disable",
367                          multi_ext_cfg->name, ret);
368             exit(EXIT_FAILURE);
369         }
370     }
371 }
372 
373 static void cpu_get_cfg_unavailable(Object *obj, Visitor *v,
374                                     const char *name,
375                                     void *opaque, Error **errp)
376 {
377     bool value = false;
378 
379     visit_type_bool(v, name, &value, errp);
380 }
381 
382 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
383                                     const char *name,
384                                     void *opaque, Error **errp)
385 {
386     const char *propname = opaque;
387     bool value;
388 
389     if (!visit_type_bool(v, name, &value, errp)) {
390         return;
391     }
392 
393     if (value) {
394         error_setg(errp, "extension %s is not available with KVM",
395                    propname);
396     }
397 }
398 
399 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
400 {
401     /* Check if KVM created the property already */
402     if (object_property_find(obj, prop_name)) {
403         return;
404     }
405 
406     /*
407      * Set the default to disabled for every extension
408      * unknown to KVM and error out if the user attempts
409      * to enable any of them.
410      */
411     object_property_add(obj, prop_name, "bool",
412                         cpu_get_cfg_unavailable,
413                         cpu_set_cfg_unavailable,
414                         NULL, (void *)prop_name);
415 }
416 
417 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
418                                         const RISCVCPUMultiExtConfig *array)
419 {
420     const RISCVCPUMultiExtConfig *prop;
421 
422     g_assert(array);
423 
424     for (prop = array; prop && prop->name; prop++) {
425         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
426     }
427 }
428 
429 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
430 {
431     int i;
432 
433     riscv_add_satp_mode_properties(cpu_obj);
434 
435     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
436         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
437         int bit = misa_cfg->offset;
438 
439         misa_cfg->name = riscv_get_misa_ext_name(bit);
440         misa_cfg->description = riscv_get_misa_ext_description(bit);
441 
442         object_property_add(cpu_obj, misa_cfg->name, "bool",
443                             kvm_cpu_get_misa_ext_cfg,
444                             kvm_cpu_set_misa_ext_cfg,
445                             NULL, misa_cfg);
446         object_property_set_description(cpu_obj, misa_cfg->name,
447                                         misa_cfg->description);
448     }
449 
450     for (i = 0; misa_bits[i] != 0; i++) {
451         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
452         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
453     }
454 
455     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
456         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
457 
458         object_property_add(cpu_obj, multi_cfg->name, "bool",
459                             kvm_cpu_get_multi_ext_cfg,
460                             kvm_cpu_set_multi_ext_cfg,
461                             NULL, multi_cfg);
462     }
463 
464     object_property_add(cpu_obj, "cbom_blocksize", "uint16",
465                         NULL, kvm_cpu_set_cbomz_blksize,
466                         NULL, &kvm_cbom_blocksize);
467 
468     object_property_add(cpu_obj, "cboz_blocksize", "uint16",
469                         NULL, kvm_cpu_set_cbomz_blksize,
470                         NULL, &kvm_cboz_blocksize);
471 
472     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
473     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
474     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
475 }
476 
477 static int kvm_riscv_get_regs_core(CPUState *cs)
478 {
479     int ret = 0;
480     int i;
481     target_ulong reg;
482     CPURISCVState *env = &RISCV_CPU(cs)->env;
483 
484     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
485     if (ret) {
486         return ret;
487     }
488     env->pc = reg;
489 
490     for (i = 1; i < 32; i++) {
491         uint64_t id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, i);
492         ret = kvm_get_one_reg(cs, id, &reg);
493         if (ret) {
494             return ret;
495         }
496         env->gpr[i] = reg;
497     }
498 
499     return ret;
500 }
501 
502 static int kvm_riscv_put_regs_core(CPUState *cs)
503 {
504     int ret = 0;
505     int i;
506     target_ulong reg;
507     CPURISCVState *env = &RISCV_CPU(cs)->env;
508 
509     reg = env->pc;
510     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
511     if (ret) {
512         return ret;
513     }
514 
515     for (i = 1; i < 32; i++) {
516         uint64_t id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CORE, i);
517         reg = env->gpr[i];
518         ret = kvm_set_one_reg(cs, id, &reg);
519         if (ret) {
520             return ret;
521         }
522     }
523 
524     return ret;
525 }
526 
527 static int kvm_riscv_get_regs_csr(CPUState *cs)
528 {
529     int ret = 0;
530     CPURISCVState *env = &RISCV_CPU(cs)->env;
531 
532     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
533     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
534     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
535     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
536     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
537     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
538     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
539     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
540     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
541     return ret;
542 }
543 
544 static int kvm_riscv_put_regs_csr(CPUState *cs)
545 {
546     int ret = 0;
547     CPURISCVState *env = &RISCV_CPU(cs)->env;
548 
549     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
550     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
551     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
552     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
553     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
554     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
555     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
556     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
557     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
558 
559     return ret;
560 }
561 
562 static int kvm_riscv_get_regs_fp(CPUState *cs)
563 {
564     int ret = 0;
565     int i;
566     CPURISCVState *env = &RISCV_CPU(cs)->env;
567 
568     if (riscv_has_ext(env, RVD)) {
569         uint64_t reg;
570         for (i = 0; i < 32; i++) {
571             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(env, i), &reg);
572             if (ret) {
573                 return ret;
574             }
575             env->fpr[i] = reg;
576         }
577         return ret;
578     }
579 
580     if (riscv_has_ext(env, RVF)) {
581         uint32_t reg;
582         for (i = 0; i < 32; i++) {
583             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(env, i), &reg);
584             if (ret) {
585                 return ret;
586             }
587             env->fpr[i] = reg;
588         }
589         return ret;
590     }
591 
592     return ret;
593 }
594 
595 static int kvm_riscv_put_regs_fp(CPUState *cs)
596 {
597     int ret = 0;
598     int i;
599     CPURISCVState *env = &RISCV_CPU(cs)->env;
600 
601     if (riscv_has_ext(env, RVD)) {
602         uint64_t reg;
603         for (i = 0; i < 32; i++) {
604             reg = env->fpr[i];
605             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(env, i), &reg);
606             if (ret) {
607                 return ret;
608             }
609         }
610         return ret;
611     }
612 
613     if (riscv_has_ext(env, RVF)) {
614         uint32_t reg;
615         for (i = 0; i < 32; i++) {
616             reg = env->fpr[i];
617             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(env, i), &reg);
618             if (ret) {
619                 return ret;
620             }
621         }
622         return ret;
623     }
624 
625     return ret;
626 }
627 
628 static void kvm_riscv_get_regs_timer(CPUState *cs)
629 {
630     CPURISCVState *env = &RISCV_CPU(cs)->env;
631 
632     if (env->kvm_timer_dirty) {
633         return;
634     }
635 
636     KVM_RISCV_GET_TIMER(cs, env, time, env->kvm_timer_time);
637     KVM_RISCV_GET_TIMER(cs, env, compare, env->kvm_timer_compare);
638     KVM_RISCV_GET_TIMER(cs, env, state, env->kvm_timer_state);
639     KVM_RISCV_GET_TIMER(cs, env, frequency, env->kvm_timer_frequency);
640 
641     env->kvm_timer_dirty = true;
642 }
643 
644 static void kvm_riscv_put_regs_timer(CPUState *cs)
645 {
646     uint64_t reg;
647     CPURISCVState *env = &RISCV_CPU(cs)->env;
648 
649     if (!env->kvm_timer_dirty) {
650         return;
651     }
652 
653     KVM_RISCV_SET_TIMER(cs, env, time, env->kvm_timer_time);
654     KVM_RISCV_SET_TIMER(cs, env, compare, env->kvm_timer_compare);
655 
656     /*
657      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
658      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
659      * doesn't matter that adaping in QEMU now.
660      * TODO If KVM changes, adapt here.
661      */
662     if (env->kvm_timer_state) {
663         KVM_RISCV_SET_TIMER(cs, env, state, env->kvm_timer_state);
664     }
665 
666     /*
667      * For now, migration will not work between Hosts with different timer
668      * frequency. Therefore, we should check whether they are the same here
669      * during the migration.
670      */
671     if (migration_is_running(migrate_get_current()->state)) {
672         KVM_RISCV_GET_TIMER(cs, env, frequency, reg);
673         if (reg != env->kvm_timer_frequency) {
674             error_report("Dst Hosts timer frequency != Src Hosts");
675         }
676     }
677 
678     env->kvm_timer_dirty = false;
679 }
680 
681 typedef struct KVMScratchCPU {
682     int kvmfd;
683     int vmfd;
684     int cpufd;
685 } KVMScratchCPU;
686 
687 /*
688  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
689  * from target/arm/kvm.c.
690  */
691 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
692 {
693     int kvmfd = -1, vmfd = -1, cpufd = -1;
694 
695     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
696     if (kvmfd < 0) {
697         goto err;
698     }
699     do {
700         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
701     } while (vmfd == -1 && errno == EINTR);
702     if (vmfd < 0) {
703         goto err;
704     }
705     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
706     if (cpufd < 0) {
707         goto err;
708     }
709 
710     scratch->kvmfd =  kvmfd;
711     scratch->vmfd = vmfd;
712     scratch->cpufd = cpufd;
713 
714     return true;
715 
716  err:
717     if (cpufd >= 0) {
718         close(cpufd);
719     }
720     if (vmfd >= 0) {
721         close(vmfd);
722     }
723     if (kvmfd >= 0) {
724         close(kvmfd);
725     }
726 
727     return false;
728 }
729 
730 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
731 {
732     close(scratch->cpufd);
733     close(scratch->vmfd);
734     close(scratch->kvmfd);
735 }
736 
737 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
738 {
739     CPURISCVState *env = &cpu->env;
740     struct kvm_one_reg reg;
741     int ret;
742 
743     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
744                               KVM_REG_RISCV_CONFIG_REG(mvendorid));
745     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
746     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
747     if (ret != 0) {
748         error_report("Unable to retrieve mvendorid from host, error %d", ret);
749     }
750 
751     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
752                               KVM_REG_RISCV_CONFIG_REG(marchid));
753     reg.addr = (uint64_t)&cpu->cfg.marchid;
754     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
755     if (ret != 0) {
756         error_report("Unable to retrieve marchid from host, error %d", ret);
757     }
758 
759     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
760                               KVM_REG_RISCV_CONFIG_REG(mimpid));
761     reg.addr = (uint64_t)&cpu->cfg.mimpid;
762     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
763     if (ret != 0) {
764         error_report("Unable to retrieve mimpid from host, error %d", ret);
765     }
766 }
767 
768 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
769                                          KVMScratchCPU *kvmcpu)
770 {
771     CPURISCVState *env = &cpu->env;
772     struct kvm_one_reg reg;
773     int ret;
774 
775     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
776                               KVM_REG_RISCV_CONFIG_REG(isa));
777     reg.addr = (uint64_t)&env->misa_ext_mask;
778     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
779 
780     if (ret) {
781         error_report("Unable to fetch ISA register from KVM, "
782                      "error %d", ret);
783         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
784         exit(EXIT_FAILURE);
785     }
786 
787     env->misa_ext = env->misa_ext_mask;
788 }
789 
790 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
791                                          KVMCPUConfig *cbomz_cfg)
792 {
793     CPURISCVState *env = &cpu->env;
794     struct kvm_one_reg reg;
795     int ret;
796 
797     reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
798                               cbomz_cfg->kvm_reg_id);
799     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
800     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
801     if (ret != 0) {
802         error_report("Unable to read KVM reg %s, error %d",
803                      cbomz_cfg->name, ret);
804         exit(EXIT_FAILURE);
805     }
806 }
807 
808 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
809                                            KVMScratchCPU *kvmcpu)
810 {
811     CPURISCVState *env = &cpu->env;
812     uint64_t val;
813     int i, ret;
814 
815     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
816         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
817         struct kvm_one_reg reg;
818 
819         reg.id = kvm_riscv_reg_id(env, KVM_REG_RISCV_ISA_EXT,
820                                   multi_ext_cfg->kvm_reg_id);
821         reg.addr = (uint64_t)&val;
822         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
823         if (ret != 0) {
824             if (errno == EINVAL) {
825                 /* Silently default to 'false' if KVM does not support it. */
826                 multi_ext_cfg->supported = false;
827                 val = false;
828             } else {
829                 error_report("Unable to read ISA_EXT KVM register %s, "
830                              "error code: %s", multi_ext_cfg->name,
831                              strerrorname_np(errno));
832                 exit(EXIT_FAILURE);
833             }
834         } else {
835             multi_ext_cfg->supported = true;
836         }
837 
838         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
839     }
840 
841     if (cpu->cfg.ext_zicbom) {
842         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
843     }
844 
845     if (cpu->cfg.ext_zicboz) {
846         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
847     }
848 }
849 
850 static int uint64_cmp(const void *a, const void *b)
851 {
852     uint64_t val1 = *(const uint64_t *)a;
853     uint64_t val2 = *(const uint64_t *)b;
854 
855     if (val1 < val2) {
856         return -1;
857     }
858 
859     if (val1 > val2) {
860         return 1;
861     }
862 
863     return 0;
864 }
865 
866 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
867 {
868     KVMCPUConfig *multi_ext_cfg;
869     struct kvm_one_reg reg;
870     struct kvm_reg_list rl_struct;
871     struct kvm_reg_list *reglist;
872     uint64_t val, reg_id, *reg_search;
873     int i, ret;
874 
875     rl_struct.n = 0;
876     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
877 
878     /*
879      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
880      * (EINVAL). Use read_legacy() in this case.
881      */
882     if (errno == EINVAL) {
883         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
884     } else if (errno != E2BIG) {
885         /*
886          * E2BIG is an expected error message for the API since we
887          * don't know the number of registers. The right amount will
888          * be written in rl_struct.n.
889          *
890          * Error out if we get any other errno.
891          */
892         error_report("Error when accessing get-reg-list, code: %s",
893                      strerrorname_np(errno));
894         exit(EXIT_FAILURE);
895     }
896 
897     reglist = g_malloc(sizeof(struct kvm_reg_list) +
898                        rl_struct.n * sizeof(uint64_t));
899     reglist->n = rl_struct.n;
900     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
901     if (ret) {
902         error_report("Error when reading KVM_GET_REG_LIST, code %s ",
903                      strerrorname_np(errno));
904         exit(EXIT_FAILURE);
905     }
906 
907     /* sort reglist to use bsearch() */
908     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
909 
910     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
911         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
912         reg_id = kvm_riscv_reg_id(&cpu->env, KVM_REG_RISCV_ISA_EXT,
913                                   multi_ext_cfg->kvm_reg_id);
914         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
915                              sizeof(uint64_t), uint64_cmp);
916         if (!reg_search) {
917             continue;
918         }
919 
920         reg.id = reg_id;
921         reg.addr = (uint64_t)&val;
922         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
923         if (ret != 0) {
924             error_report("Unable to read ISA_EXT KVM register %s, "
925                          "error code: %s", multi_ext_cfg->name,
926                          strerrorname_np(errno));
927             exit(EXIT_FAILURE);
928         }
929 
930         multi_ext_cfg->supported = true;
931         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
932     }
933 
934     if (cpu->cfg.ext_zicbom) {
935         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
936     }
937 
938     if (cpu->cfg.ext_zicboz) {
939         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
940     }
941 }
942 
943 static void riscv_init_kvm_registers(Object *cpu_obj)
944 {
945     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
946     KVMScratchCPU kvmcpu;
947 
948     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
949         return;
950     }
951 
952     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
953     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
954     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
955 
956     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
957 }
958 
959 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
960     KVM_CAP_LAST_INFO
961 };
962 
963 int kvm_arch_get_registers(CPUState *cs)
964 {
965     int ret = 0;
966 
967     ret = kvm_riscv_get_regs_core(cs);
968     if (ret) {
969         return ret;
970     }
971 
972     ret = kvm_riscv_get_regs_csr(cs);
973     if (ret) {
974         return ret;
975     }
976 
977     ret = kvm_riscv_get_regs_fp(cs);
978     if (ret) {
979         return ret;
980     }
981 
982     return ret;
983 }
984 
985 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
986 {
987     if (cap_has_mp_state) {
988         struct kvm_mp_state mp_state = {
989             .mp_state = state
990         };
991 
992         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
993         if (ret) {
994             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
995                     __func__, ret, strerror(-ret));
996             return -1;
997         }
998     }
999 
1000     return 0;
1001 }
1002 
1003 int kvm_arch_put_registers(CPUState *cs, int level)
1004 {
1005     int ret = 0;
1006 
1007     ret = kvm_riscv_put_regs_core(cs);
1008     if (ret) {
1009         return ret;
1010     }
1011 
1012     ret = kvm_riscv_put_regs_csr(cs);
1013     if (ret) {
1014         return ret;
1015     }
1016 
1017     ret = kvm_riscv_put_regs_fp(cs);
1018     if (ret) {
1019         return ret;
1020     }
1021 
1022     if (KVM_PUT_RESET_STATE == level) {
1023         RISCVCPU *cpu = RISCV_CPU(cs);
1024         if (cs->cpu_index == 0) {
1025             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1026         } else {
1027             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1028         }
1029         if (ret) {
1030             return ret;
1031         }
1032     }
1033 
1034     return ret;
1035 }
1036 
1037 int kvm_arch_release_virq_post(int virq)
1038 {
1039     return 0;
1040 }
1041 
1042 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1043                              uint64_t address, uint32_t data, PCIDevice *dev)
1044 {
1045     return 0;
1046 }
1047 
1048 int kvm_arch_destroy_vcpu(CPUState *cs)
1049 {
1050     return 0;
1051 }
1052 
1053 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1054 {
1055     return cpu->cpu_index;
1056 }
1057 
1058 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1059                                       RunState state)
1060 {
1061     CPUState *cs = opaque;
1062 
1063     if (running) {
1064         kvm_riscv_put_regs_timer(cs);
1065     } else {
1066         kvm_riscv_get_regs_timer(cs);
1067     }
1068 }
1069 
1070 void kvm_arch_init_irq_routing(KVMState *s)
1071 {
1072 }
1073 
1074 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1075 {
1076     CPURISCVState *env = &cpu->env;
1077     target_ulong reg;
1078     uint64_t id;
1079     int ret;
1080 
1081     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1082                           KVM_REG_RISCV_CONFIG_REG(mvendorid));
1083     /*
1084      * cfg.mvendorid is an uint32 but a target_ulong will
1085      * be written. Assign it to a target_ulong var to avoid
1086      * writing pieces of other cpu->cfg fields in the reg.
1087      */
1088     reg = cpu->cfg.mvendorid;
1089     ret = kvm_set_one_reg(cs, id, &reg);
1090     if (ret != 0) {
1091         return ret;
1092     }
1093 
1094     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1095                           KVM_REG_RISCV_CONFIG_REG(marchid));
1096     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1097     if (ret != 0) {
1098         return ret;
1099     }
1100 
1101     id = kvm_riscv_reg_id(env, KVM_REG_RISCV_CONFIG,
1102                           KVM_REG_RISCV_CONFIG_REG(mimpid));
1103     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1104 
1105     return ret;
1106 }
1107 
1108 int kvm_arch_init_vcpu(CPUState *cs)
1109 {
1110     int ret = 0;
1111     RISCVCPU *cpu = RISCV_CPU(cs);
1112 
1113     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1114 
1115     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1116         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1117         if (ret != 0) {
1118             return ret;
1119         }
1120     }
1121 
1122     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1123     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1124 
1125     return ret;
1126 }
1127 
1128 int kvm_arch_msi_data_to_gsi(uint32_t data)
1129 {
1130     abort();
1131 }
1132 
1133 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1134                                 int vector, PCIDevice *dev)
1135 {
1136     return 0;
1137 }
1138 
1139 int kvm_arch_get_default_type(MachineState *ms)
1140 {
1141     return 0;
1142 }
1143 
1144 int kvm_arch_init(MachineState *ms, KVMState *s)
1145 {
1146     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1147     return 0;
1148 }
1149 
1150 int kvm_arch_irqchip_create(KVMState *s)
1151 {
1152     if (kvm_kernel_irqchip_split()) {
1153         error_report("-machine kernel_irqchip=split is not supported on RISC-V.");
1154         exit(1);
1155     }
1156 
1157     /*
1158      * We can create the VAIA using the newer device control API.
1159      */
1160     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1161 }
1162 
1163 int kvm_arch_process_async_events(CPUState *cs)
1164 {
1165     return 0;
1166 }
1167 
1168 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1169 {
1170 }
1171 
1172 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1173 {
1174     return MEMTXATTRS_UNSPECIFIED;
1175 }
1176 
1177 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1178 {
1179     return true;
1180 }
1181 
1182 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1183 {
1184     int ret = 0;
1185     unsigned char ch;
1186     switch (run->riscv_sbi.extension_id) {
1187     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1188         ch = run->riscv_sbi.args[0];
1189         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1190         break;
1191     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1192         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1193         if (ret == sizeof(ch)) {
1194             run->riscv_sbi.ret[0] = ch;
1195         } else {
1196             run->riscv_sbi.ret[0] = -1;
1197         }
1198         ret = 0;
1199         break;
1200     default:
1201         qemu_log_mask(LOG_UNIMP,
1202                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1203                       __func__, run->riscv_sbi.extension_id);
1204         ret = -1;
1205         break;
1206     }
1207     return ret;
1208 }
1209 
1210 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1211 {
1212     int ret = 0;
1213     switch (run->exit_reason) {
1214     case KVM_EXIT_RISCV_SBI:
1215         ret = kvm_riscv_handle_sbi(cs, run);
1216         break;
1217     default:
1218         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1219                       __func__, run->exit_reason);
1220         ret = -1;
1221         break;
1222     }
1223     return ret;
1224 }
1225 
1226 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1227 {
1228     CPURISCVState *env = &cpu->env;
1229     int i;
1230 
1231     if (!kvm_enabled()) {
1232         return;
1233     }
1234     for (i = 0; i < 32; i++) {
1235         env->gpr[i] = 0;
1236     }
1237     env->pc = cpu->env.kernel_addr;
1238     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1239     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1240     env->satp = 0;
1241     env->mie = 0;
1242     env->stvec = 0;
1243     env->sscratch = 0;
1244     env->sepc = 0;
1245     env->scause = 0;
1246     env->stval = 0;
1247     env->mip = 0;
1248 }
1249 
1250 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1251 {
1252     int ret;
1253     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1254 
1255     if (irq != IRQ_S_EXT) {
1256         perror("kvm riscv set irq != IRQ_S_EXT\n");
1257         abort();
1258     }
1259 
1260     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1261     if (ret < 0) {
1262         perror("Set irq failed");
1263         abort();
1264     }
1265 }
1266 
1267 bool kvm_arch_cpu_check_are_resettable(void)
1268 {
1269     return true;
1270 }
1271 
1272 static int aia_mode;
1273 
1274 static const char *kvm_aia_mode_str(uint64_t mode)
1275 {
1276     switch (mode) {
1277     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1278         return "emul";
1279     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1280         return "hwaccel";
1281     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1282     default:
1283         return "auto";
1284     };
1285 }
1286 
1287 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1288 {
1289     return g_strdup(kvm_aia_mode_str(aia_mode));
1290 }
1291 
1292 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1293 {
1294     if (!strcmp(val, "emul")) {
1295         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1296     } else if (!strcmp(val, "hwaccel")) {
1297         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1298     } else if (!strcmp(val, "auto")) {
1299         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1300     } else {
1301         error_setg(errp, "Invalid KVM AIA mode");
1302         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1303     }
1304 }
1305 
1306 void kvm_arch_accel_class_init(ObjectClass *oc)
1307 {
1308     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1309                                   riscv_set_kvm_aia);
1310     object_class_property_set_description(oc, "riscv-aia",
1311                                           "Set KVM AIA mode. Valid values are "
1312                                           "emul, hwaccel, and auto. Default "
1313                                           "is auto.");
1314     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1315                                     "auto");
1316 }
1317 
1318 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1319                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1320                           uint64_t aplic_base, uint64_t imsic_base,
1321                           uint64_t guest_num)
1322 {
1323     int ret, i;
1324     int aia_fd = -1;
1325     uint64_t default_aia_mode;
1326     uint64_t socket_count = riscv_socket_count(machine);
1327     uint64_t max_hart_per_socket = 0;
1328     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1329     uint64_t socket_bits, hart_bits, guest_bits;
1330 
1331     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1332 
1333     if (aia_fd < 0) {
1334         error_report("Unable to create in-kernel irqchip");
1335         exit(1);
1336     }
1337 
1338     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1339                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1340                             &default_aia_mode, false, NULL);
1341     if (ret < 0) {
1342         error_report("KVM AIA: failed to get current KVM AIA mode");
1343         exit(1);
1344     }
1345     qemu_log("KVM AIA: default mode is %s\n",
1346              kvm_aia_mode_str(default_aia_mode));
1347 
1348     if (default_aia_mode != aia_mode) {
1349         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1350                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1351                                 &aia_mode, true, NULL);
1352         if (ret < 0)
1353             warn_report("KVM AIA: failed to set KVM AIA mode");
1354         else
1355             qemu_log("KVM AIA: set current mode to %s\n",
1356                      kvm_aia_mode_str(aia_mode));
1357     }
1358 
1359     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1360                             KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1361                             &aia_irq_num, true, NULL);
1362     if (ret < 0) {
1363         error_report("KVM AIA: failed to set number of input irq lines");
1364         exit(1);
1365     }
1366 
1367     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1368                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1369                             &aia_msi_num, true, NULL);
1370     if (ret < 0) {
1371         error_report("KVM AIA: failed to set number of msi");
1372         exit(1);
1373     }
1374 
1375     socket_bits = find_last_bit(&socket_count, BITS_PER_LONG) + 1;
1376     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1377                             KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1378                             &socket_bits, true, NULL);
1379     if (ret < 0) {
1380         error_report("KVM AIA: failed to set group_bits");
1381         exit(1);
1382     }
1383 
1384     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1385                             KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1386                             &group_shift, true, NULL);
1387     if (ret < 0) {
1388         error_report("KVM AIA: failed to set group_shift");
1389         exit(1);
1390     }
1391 
1392     guest_bits = guest_num == 0 ? 0 :
1393                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1394     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1395                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1396                             &guest_bits, true, NULL);
1397     if (ret < 0) {
1398         error_report("KVM AIA: failed to set guest_bits");
1399         exit(1);
1400     }
1401 
1402     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1403                             KVM_DEV_RISCV_AIA_ADDR_APLIC,
1404                             &aplic_base, true, NULL);
1405     if (ret < 0) {
1406         error_report("KVM AIA: failed to set the base address of APLIC");
1407         exit(1);
1408     }
1409 
1410     for (socket = 0; socket < socket_count; socket++) {
1411         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1412         hart_count = riscv_socket_hart_count(machine, socket);
1413         base_hart = riscv_socket_first_hartid(machine, socket);
1414 
1415         if (max_hart_per_socket < hart_count) {
1416             max_hart_per_socket = hart_count;
1417         }
1418 
1419         for (i = 0; i < hart_count; i++) {
1420             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1421             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1422                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1423                                     &imsic_addr, true, NULL);
1424             if (ret < 0) {
1425                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1426                 exit(1);
1427             }
1428         }
1429     }
1430 
1431     hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1432     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1433                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1434                             &hart_bits, true, NULL);
1435     if (ret < 0) {
1436         error_report("KVM AIA: failed to set hart_bits");
1437         exit(1);
1438     }
1439 
1440     if (kvm_has_gsi_routing()) {
1441         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1442             /* KVM AIA only has one APLIC instance */
1443             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1444         }
1445         kvm_gsi_routing_allowed = true;
1446         kvm_irqchip_commit_routes(kvm_state);
1447     }
1448 
1449     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1450                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1451                             NULL, true, NULL);
1452     if (ret < 0) {
1453         error_report("KVM AIA: initialized fail");
1454         exit(1);
1455     }
1456 
1457     kvm_msi_via_irqfd_allowed = true;
1458 }
1459 
1460 static void kvm_cpu_instance_init(CPUState *cs)
1461 {
1462     Object *obj = OBJECT(RISCV_CPU(cs));
1463     DeviceState *dev = DEVICE(obj);
1464 
1465     riscv_init_kvm_registers(obj);
1466 
1467     kvm_riscv_add_cpu_user_properties(obj);
1468 
1469     for (Property *prop = riscv_cpu_options; prop && prop->name; prop++) {
1470         /* Check if we have a specific KVM handler for the option */
1471         if (object_property_find(obj, prop->name)) {
1472             continue;
1473         }
1474         qdev_property_add_static(dev, prop);
1475     }
1476 }
1477 
1478 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1479 {
1480     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1481 
1482     acc->cpu_instance_init = kvm_cpu_instance_init;
1483 }
1484 
1485 static const TypeInfo kvm_cpu_accel_type_info = {
1486     .name = ACCEL_CPU_NAME("kvm"),
1487 
1488     .parent = TYPE_ACCEL_CPU,
1489     .class_init = kvm_cpu_accel_class_init,
1490     .abstract = true,
1491 };
1492 static void kvm_cpu_accel_register_types(void)
1493 {
1494     type_register_static(&kvm_cpu_accel_type_info);
1495 }
1496 type_init(kvm_cpu_accel_register_types);
1497 
1498 static void riscv_host_cpu_init(Object *obj)
1499 {
1500     CPURISCVState *env = &RISCV_CPU(obj)->env;
1501 
1502 #if defined(TARGET_RISCV32)
1503     env->misa_mxl_max = env->misa_mxl = MXL_RV32;
1504 #elif defined(TARGET_RISCV64)
1505     env->misa_mxl_max = env->misa_mxl = MXL_RV64;
1506 #endif
1507 }
1508 
1509 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
1510     {
1511         .name = TYPE_RISCV_CPU_HOST,
1512         .parent = TYPE_RISCV_CPU,
1513         .instance_init = riscv_host_cpu_init,
1514     }
1515 };
1516 
1517 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
1518