xref: /openbmc/qemu/target/riscv/kvm/kvm-cpu.c (revision ee88df50)
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/migration.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 #define RISCV_CORE_REG(env, name) \
90     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, \
91                            KVM_REG_RISCV_CORE_REG(name))
92 
93 #define RISCV_CSR_REG(env, name) \
94     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CSR, \
95                            KVM_REG_RISCV_CSR_REG(name))
96 
97 #define RISCV_CONFIG_REG(env, name) \
98     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG, \
99                            KVM_REG_RISCV_CONFIG_REG(name))
100 
101 #define RISCV_TIMER_REG(name)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_TIMER, \
102                  KVM_REG_RISCV_TIMER_REG(name))
103 
104 #define RISCV_FP_F_REG(idx)  kvm_riscv_reg_id_u32(KVM_REG_RISCV_FP_F, idx)
105 
106 #define RISCV_FP_D_REG(idx)  kvm_riscv_reg_id_u64(KVM_REG_RISCV_FP_D, idx)
107 
108 #define RISCV_VECTOR_CSR_REG(env, name) \
109     kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_VECTOR, \
110                            KVM_REG_RISCV_VECTOR_CSR_REG(name))
111 
112 #define KVM_RISCV_GET_CSR(cs, env, csr, reg) \
113     do { \
114         int _ret = kvm_get_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
115         if (_ret) { \
116             return _ret; \
117         } \
118     } while (0)
119 
120 #define KVM_RISCV_SET_CSR(cs, env, csr, reg) \
121     do { \
122         int _ret = kvm_set_one_reg(cs, RISCV_CSR_REG(env, csr), &reg); \
123         if (_ret) { \
124             return _ret; \
125         } \
126     } while (0)
127 
128 #define KVM_RISCV_GET_TIMER(cs, name, reg) \
129     do { \
130         int ret = kvm_get_one_reg(cs, RISCV_TIMER_REG(name), &reg); \
131         if (ret) { \
132             abort(); \
133         } \
134     } while (0)
135 
136 #define KVM_RISCV_SET_TIMER(cs, name, reg) \
137     do { \
138         int ret = kvm_set_one_reg(cs, RISCV_TIMER_REG(name), &reg); \
139         if (ret) { \
140             abort(); \
141         } \
142     } while (0)
143 
144 typedef struct KVMCPUConfig {
145     const char *name;
146     const char *description;
147     target_ulong offset;
148     int kvm_reg_id;
149     bool user_set;
150     bool supported;
151 } KVMCPUConfig;
152 
153 #define KVM_MISA_CFG(_bit, _reg_id) \
154     {.offset = _bit, .kvm_reg_id = _reg_id}
155 
156 /* KVM ISA extensions */
157 static KVMCPUConfig kvm_misa_ext_cfgs[] = {
158     KVM_MISA_CFG(RVA, KVM_RISCV_ISA_EXT_A),
159     KVM_MISA_CFG(RVC, KVM_RISCV_ISA_EXT_C),
160     KVM_MISA_CFG(RVD, KVM_RISCV_ISA_EXT_D),
161     KVM_MISA_CFG(RVF, KVM_RISCV_ISA_EXT_F),
162     KVM_MISA_CFG(RVH, KVM_RISCV_ISA_EXT_H),
163     KVM_MISA_CFG(RVI, KVM_RISCV_ISA_EXT_I),
164     KVM_MISA_CFG(RVM, KVM_RISCV_ISA_EXT_M),
165     KVM_MISA_CFG(RVV, KVM_RISCV_ISA_EXT_V),
166 };
167 
168 static void kvm_cpu_get_misa_ext_cfg(Object *obj, Visitor *v,
169                                      const char *name,
170                                      void *opaque, Error **errp)
171 {
172     KVMCPUConfig *misa_ext_cfg = opaque;
173     target_ulong misa_bit = misa_ext_cfg->offset;
174     RISCVCPU *cpu = RISCV_CPU(obj);
175     CPURISCVState *env = &cpu->env;
176     bool value = env->misa_ext_mask & misa_bit;
177 
178     visit_type_bool(v, name, &value, errp);
179 }
180 
181 static void kvm_cpu_set_misa_ext_cfg(Object *obj, Visitor *v,
182                                      const char *name,
183                                      void *opaque, Error **errp)
184 {
185     KVMCPUConfig *misa_ext_cfg = opaque;
186     target_ulong misa_bit = misa_ext_cfg->offset;
187     RISCVCPU *cpu = RISCV_CPU(obj);
188     CPURISCVState *env = &cpu->env;
189     bool value, host_bit;
190 
191     if (!visit_type_bool(v, name, &value, errp)) {
192         return;
193     }
194 
195     host_bit = env->misa_ext_mask & misa_bit;
196 
197     if (value == host_bit) {
198         return;
199     }
200 
201     if (!value) {
202         misa_ext_cfg->user_set = true;
203         return;
204     }
205 
206     /*
207      * Forbid users to enable extensions that aren't
208      * available in the hart.
209      */
210     error_setg(errp, "Enabling MISA bit '%s' is not allowed: it's not "
211                "enabled in the host", misa_ext_cfg->name);
212 }
213 
214 static void kvm_riscv_update_cpu_misa_ext(RISCVCPU *cpu, CPUState *cs)
215 {
216     CPURISCVState *env = &cpu->env;
217     uint64_t id, reg;
218     int i, ret;
219 
220     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
221         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
222         target_ulong misa_bit = misa_cfg->offset;
223 
224         if (!misa_cfg->user_set) {
225             continue;
226         }
227 
228         /* If we're here we're going to disable the MISA bit */
229         reg = 0;
230         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
231                                     misa_cfg->kvm_reg_id);
232         ret = kvm_set_one_reg(cs, id, &reg);
233         if (ret != 0) {
234             /*
235              * We're not checking for -EINVAL because if the bit is about
236              * to be disabled, it means that it was already enabled by
237              * KVM. We determined that by fetching the 'isa' register
238              * during init() time. Any error at this point is worth
239              * aborting.
240              */
241             error_report("Unable to set KVM reg %s, error %d",
242                          misa_cfg->name, ret);
243             exit(EXIT_FAILURE);
244         }
245         env->misa_ext &= ~misa_bit;
246     }
247 }
248 
249 #define KVM_EXT_CFG(_name, _prop, _reg_id) \
250     {.name = _name, .offset = CPU_CFG_OFFSET(_prop), \
251      .kvm_reg_id = _reg_id}
252 
253 static KVMCPUConfig kvm_multi_ext_cfgs[] = {
254     KVM_EXT_CFG("zicbom", ext_zicbom, KVM_RISCV_ISA_EXT_ZICBOM),
255     KVM_EXT_CFG("zicboz", ext_zicboz, KVM_RISCV_ISA_EXT_ZICBOZ),
256     KVM_EXT_CFG("zicntr", ext_zicntr, KVM_RISCV_ISA_EXT_ZICNTR),
257     KVM_EXT_CFG("zicsr", ext_zicsr, KVM_RISCV_ISA_EXT_ZICSR),
258     KVM_EXT_CFG("zifencei", ext_zifencei, KVM_RISCV_ISA_EXT_ZIFENCEI),
259     KVM_EXT_CFG("zihintpause", ext_zihintpause, KVM_RISCV_ISA_EXT_ZIHINTPAUSE),
260     KVM_EXT_CFG("zihpm", ext_zihpm, KVM_RISCV_ISA_EXT_ZIHPM),
261     KVM_EXT_CFG("zba", ext_zba, KVM_RISCV_ISA_EXT_ZBA),
262     KVM_EXT_CFG("zbb", ext_zbb, KVM_RISCV_ISA_EXT_ZBB),
263     KVM_EXT_CFG("zbs", ext_zbs, KVM_RISCV_ISA_EXT_ZBS),
264     KVM_EXT_CFG("ssaia", ext_ssaia, KVM_RISCV_ISA_EXT_SSAIA),
265     KVM_EXT_CFG("sstc", ext_sstc, KVM_RISCV_ISA_EXT_SSTC),
266     KVM_EXT_CFG("svinval", ext_svinval, KVM_RISCV_ISA_EXT_SVINVAL),
267     KVM_EXT_CFG("svnapot", ext_svnapot, KVM_RISCV_ISA_EXT_SVNAPOT),
268     KVM_EXT_CFG("svpbmt", ext_svpbmt, KVM_RISCV_ISA_EXT_SVPBMT),
269 };
270 
271 static void *kvmconfig_get_cfg_addr(RISCVCPU *cpu, KVMCPUConfig *kvmcfg)
272 {
273     return (void *)&cpu->cfg + kvmcfg->offset;
274 }
275 
276 static void kvm_cpu_cfg_set(RISCVCPU *cpu, KVMCPUConfig *multi_ext,
277                             uint32_t val)
278 {
279     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
280 
281     *ext_enabled = val;
282 }
283 
284 static uint32_t kvm_cpu_cfg_get(RISCVCPU *cpu,
285                                 KVMCPUConfig *multi_ext)
286 {
287     bool *ext_enabled = kvmconfig_get_cfg_addr(cpu, multi_ext);
288 
289     return *ext_enabled;
290 }
291 
292 static void kvm_cpu_get_multi_ext_cfg(Object *obj, Visitor *v,
293                                       const char *name,
294                                       void *opaque, Error **errp)
295 {
296     KVMCPUConfig *multi_ext_cfg = opaque;
297     RISCVCPU *cpu = RISCV_CPU(obj);
298     bool value = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
299 
300     visit_type_bool(v, name, &value, errp);
301 }
302 
303 static void kvm_cpu_set_multi_ext_cfg(Object *obj, Visitor *v,
304                                       const char *name,
305                                       void *opaque, Error **errp)
306 {
307     KVMCPUConfig *multi_ext_cfg = opaque;
308     RISCVCPU *cpu = RISCV_CPU(obj);
309     bool value, host_val;
310 
311     if (!visit_type_bool(v, name, &value, errp)) {
312         return;
313     }
314 
315     host_val = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
316 
317     /*
318      * Ignore if the user is setting the same value
319      * as the host.
320      */
321     if (value == host_val) {
322         return;
323     }
324 
325     if (!multi_ext_cfg->supported) {
326         /*
327          * Error out if the user is trying to enable an
328          * extension that KVM doesn't support. Ignore
329          * option otherwise.
330          */
331         if (value) {
332             error_setg(errp, "KVM does not support disabling extension %s",
333                        multi_ext_cfg->name);
334         }
335 
336         return;
337     }
338 
339     multi_ext_cfg->user_set = true;
340     kvm_cpu_cfg_set(cpu, multi_ext_cfg, value);
341 }
342 
343 static KVMCPUConfig kvm_cbom_blocksize = {
344     .name = "cbom_blocksize",
345     .offset = CPU_CFG_OFFSET(cbom_blocksize),
346     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicbom_block_size)
347 };
348 
349 static KVMCPUConfig kvm_cboz_blocksize = {
350     .name = "cboz_blocksize",
351     .offset = CPU_CFG_OFFSET(cboz_blocksize),
352     .kvm_reg_id = KVM_REG_RISCV_CONFIG_REG(zicboz_block_size)
353 };
354 
355 static void kvm_cpu_set_cbomz_blksize(Object *obj, Visitor *v,
356                                       const char *name,
357                                       void *opaque, Error **errp)
358 {
359     KVMCPUConfig *cbomz_cfg = opaque;
360     RISCVCPU *cpu = RISCV_CPU(obj);
361     uint16_t value, *host_val;
362 
363     if (!visit_type_uint16(v, name, &value, errp)) {
364         return;
365     }
366 
367     host_val = kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
368 
369     if (value != *host_val) {
370         error_report("Unable to set %s to a different value than "
371                      "the host (%u)",
372                      cbomz_cfg->name, *host_val);
373         exit(EXIT_FAILURE);
374     }
375 
376     cbomz_cfg->user_set = true;
377 }
378 
379 static void kvm_riscv_update_cpu_cfg_isa_ext(RISCVCPU *cpu, CPUState *cs)
380 {
381     CPURISCVState *env = &cpu->env;
382     uint64_t id, reg;
383     int i, ret;
384 
385     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
386         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
387 
388         if (!multi_ext_cfg->user_set) {
389             continue;
390         }
391 
392         id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
393                                     multi_ext_cfg->kvm_reg_id);
394         reg = kvm_cpu_cfg_get(cpu, multi_ext_cfg);
395         ret = kvm_set_one_reg(cs, id, &reg);
396         if (ret != 0) {
397             error_report("Unable to %s extension %s in KVM, error %d",
398                          reg ? "enable" : "disable",
399                          multi_ext_cfg->name, ret);
400             exit(EXIT_FAILURE);
401         }
402     }
403 }
404 
405 static void cpu_get_cfg_unavailable(Object *obj, Visitor *v,
406                                     const char *name,
407                                     void *opaque, Error **errp)
408 {
409     bool value = false;
410 
411     visit_type_bool(v, name, &value, errp);
412 }
413 
414 static void cpu_set_cfg_unavailable(Object *obj, Visitor *v,
415                                     const char *name,
416                                     void *opaque, Error **errp)
417 {
418     const char *propname = opaque;
419     bool value;
420 
421     if (!visit_type_bool(v, name, &value, errp)) {
422         return;
423     }
424 
425     if (value) {
426         error_setg(errp, "'%s' is not available with KVM",
427                    propname);
428     }
429 }
430 
431 static void riscv_cpu_add_kvm_unavail_prop(Object *obj, const char *prop_name)
432 {
433     /* Check if KVM created the property already */
434     if (object_property_find(obj, prop_name)) {
435         return;
436     }
437 
438     /*
439      * Set the default to disabled for every extension
440      * unknown to KVM and error out if the user attempts
441      * to enable any of them.
442      */
443     object_property_add(obj, prop_name, "bool",
444                         cpu_get_cfg_unavailable,
445                         cpu_set_cfg_unavailable,
446                         NULL, (void *)prop_name);
447 }
448 
449 static void riscv_cpu_add_kvm_unavail_prop_array(Object *obj,
450                                         const RISCVCPUMultiExtConfig *array)
451 {
452     const RISCVCPUMultiExtConfig *prop;
453 
454     g_assert(array);
455 
456     for (prop = array; prop && prop->name; prop++) {
457         riscv_cpu_add_kvm_unavail_prop(obj, prop->name);
458     }
459 }
460 
461 static void kvm_riscv_add_cpu_user_properties(Object *cpu_obj)
462 {
463     int i;
464 
465     riscv_add_satp_mode_properties(cpu_obj);
466 
467     for (i = 0; i < ARRAY_SIZE(kvm_misa_ext_cfgs); i++) {
468         KVMCPUConfig *misa_cfg = &kvm_misa_ext_cfgs[i];
469         int bit = misa_cfg->offset;
470 
471         misa_cfg->name = riscv_get_misa_ext_name(bit);
472         misa_cfg->description = riscv_get_misa_ext_description(bit);
473 
474         object_property_add(cpu_obj, misa_cfg->name, "bool",
475                             kvm_cpu_get_misa_ext_cfg,
476                             kvm_cpu_set_misa_ext_cfg,
477                             NULL, misa_cfg);
478         object_property_set_description(cpu_obj, misa_cfg->name,
479                                         misa_cfg->description);
480     }
481 
482     for (i = 0; misa_bits[i] != 0; i++) {
483         const char *ext_name = riscv_get_misa_ext_name(misa_bits[i]);
484         riscv_cpu_add_kvm_unavail_prop(cpu_obj, ext_name);
485     }
486 
487     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
488         KVMCPUConfig *multi_cfg = &kvm_multi_ext_cfgs[i];
489 
490         object_property_add(cpu_obj, multi_cfg->name, "bool",
491                             kvm_cpu_get_multi_ext_cfg,
492                             kvm_cpu_set_multi_ext_cfg,
493                             NULL, multi_cfg);
494     }
495 
496     object_property_add(cpu_obj, "cbom_blocksize", "uint16",
497                         NULL, kvm_cpu_set_cbomz_blksize,
498                         NULL, &kvm_cbom_blocksize);
499 
500     object_property_add(cpu_obj, "cboz_blocksize", "uint16",
501                         NULL, kvm_cpu_set_cbomz_blksize,
502                         NULL, &kvm_cboz_blocksize);
503 
504     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
505     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
506     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
507 
508    /* We don't have the needed KVM support for profiles */
509     for (i = 0; riscv_profiles[i] != NULL; i++) {
510         riscv_cpu_add_kvm_unavail_prop(cpu_obj, riscv_profiles[i]->name);
511     }
512 }
513 
514 static int kvm_riscv_get_regs_core(CPUState *cs)
515 {
516     int ret = 0;
517     int i;
518     target_ulong reg;
519     CPURISCVState *env = &RISCV_CPU(cs)->env;
520 
521     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
522     if (ret) {
523         return ret;
524     }
525     env->pc = reg;
526 
527     for (i = 1; i < 32; i++) {
528         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
529         ret = kvm_get_one_reg(cs, id, &reg);
530         if (ret) {
531             return ret;
532         }
533         env->gpr[i] = reg;
534     }
535 
536     return ret;
537 }
538 
539 static int kvm_riscv_put_regs_core(CPUState *cs)
540 {
541     int ret = 0;
542     int i;
543     target_ulong reg;
544     CPURISCVState *env = &RISCV_CPU(cs)->env;
545 
546     reg = env->pc;
547     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
548     if (ret) {
549         return ret;
550     }
551 
552     for (i = 1; i < 32; i++) {
553         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
554         reg = env->gpr[i];
555         ret = kvm_set_one_reg(cs, id, &reg);
556         if (ret) {
557             return ret;
558         }
559     }
560 
561     return ret;
562 }
563 
564 static int kvm_riscv_get_regs_csr(CPUState *cs)
565 {
566     CPURISCVState *env = &RISCV_CPU(cs)->env;
567 
568     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
569     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
570     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
571     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
572     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
573     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
574     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
575     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
576     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
577 
578     return 0;
579 }
580 
581 static int kvm_riscv_put_regs_csr(CPUState *cs)
582 {
583     CPURISCVState *env = &RISCV_CPU(cs)->env;
584 
585     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
586     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
587     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
588     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
589     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
590     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
591     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
592     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
593     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
594 
595     return 0;
596 }
597 
598 static int kvm_riscv_get_regs_fp(CPUState *cs)
599 {
600     int ret = 0;
601     int i;
602     CPURISCVState *env = &RISCV_CPU(cs)->env;
603 
604     if (riscv_has_ext(env, RVD)) {
605         uint64_t reg;
606         for (i = 0; i < 32; i++) {
607             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
608             if (ret) {
609                 return ret;
610             }
611             env->fpr[i] = reg;
612         }
613         return ret;
614     }
615 
616     if (riscv_has_ext(env, RVF)) {
617         uint32_t reg;
618         for (i = 0; i < 32; i++) {
619             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
620             if (ret) {
621                 return ret;
622             }
623             env->fpr[i] = reg;
624         }
625         return ret;
626     }
627 
628     return ret;
629 }
630 
631 static int kvm_riscv_put_regs_fp(CPUState *cs)
632 {
633     int ret = 0;
634     int i;
635     CPURISCVState *env = &RISCV_CPU(cs)->env;
636 
637     if (riscv_has_ext(env, RVD)) {
638         uint64_t reg;
639         for (i = 0; i < 32; i++) {
640             reg = env->fpr[i];
641             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
642             if (ret) {
643                 return ret;
644             }
645         }
646         return ret;
647     }
648 
649     if (riscv_has_ext(env, RVF)) {
650         uint32_t reg;
651         for (i = 0; i < 32; i++) {
652             reg = env->fpr[i];
653             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
654             if (ret) {
655                 return ret;
656             }
657         }
658         return ret;
659     }
660 
661     return ret;
662 }
663 
664 static void kvm_riscv_get_regs_timer(CPUState *cs)
665 {
666     CPURISCVState *env = &RISCV_CPU(cs)->env;
667 
668     if (env->kvm_timer_dirty) {
669         return;
670     }
671 
672     KVM_RISCV_GET_TIMER(cs, time, env->kvm_timer_time);
673     KVM_RISCV_GET_TIMER(cs, compare, env->kvm_timer_compare);
674     KVM_RISCV_GET_TIMER(cs, state, env->kvm_timer_state);
675     KVM_RISCV_GET_TIMER(cs, frequency, env->kvm_timer_frequency);
676 
677     env->kvm_timer_dirty = true;
678 }
679 
680 static void kvm_riscv_put_regs_timer(CPUState *cs)
681 {
682     uint64_t reg;
683     CPURISCVState *env = &RISCV_CPU(cs)->env;
684 
685     if (!env->kvm_timer_dirty) {
686         return;
687     }
688 
689     KVM_RISCV_SET_TIMER(cs, time, env->kvm_timer_time);
690     KVM_RISCV_SET_TIMER(cs, compare, env->kvm_timer_compare);
691 
692     /*
693      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
694      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
695      * doesn't matter that adaping in QEMU now.
696      * TODO If KVM changes, adapt here.
697      */
698     if (env->kvm_timer_state) {
699         KVM_RISCV_SET_TIMER(cs, state, env->kvm_timer_state);
700     }
701 
702     /*
703      * For now, migration will not work between Hosts with different timer
704      * frequency. Therefore, we should check whether they are the same here
705      * during the migration.
706      */
707     if (migration_is_running(migrate_get_current()->state)) {
708         KVM_RISCV_GET_TIMER(cs, frequency, reg);
709         if (reg != env->kvm_timer_frequency) {
710             error_report("Dst Hosts timer frequency != Src Hosts");
711         }
712     }
713 
714     env->kvm_timer_dirty = false;
715 }
716 
717 static int kvm_riscv_get_regs_vector(CPUState *cs)
718 {
719     CPURISCVState *env = &RISCV_CPU(cs)->env;
720     target_ulong reg;
721     int ret = 0;
722 
723     if (!riscv_has_ext(env, RVV)) {
724         return 0;
725     }
726 
727     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
728     if (ret) {
729         return ret;
730     }
731     env->vstart = reg;
732 
733     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
734     if (ret) {
735         return ret;
736     }
737     env->vl = reg;
738 
739     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
740     if (ret) {
741         return ret;
742     }
743     env->vtype = reg;
744 
745     return 0;
746 }
747 
748 static int kvm_riscv_put_regs_vector(CPUState *cs)
749 {
750     CPURISCVState *env = &RISCV_CPU(cs)->env;
751     target_ulong reg;
752     int ret = 0;
753 
754     if (!riscv_has_ext(env, RVV)) {
755         return 0;
756     }
757 
758     reg = env->vstart;
759     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
760     if (ret) {
761         return ret;
762     }
763 
764     reg = env->vl;
765     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
766     if (ret) {
767         return ret;
768     }
769 
770     reg = env->vtype;
771     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
772 
773     return ret;
774 }
775 
776 typedef struct KVMScratchCPU {
777     int kvmfd;
778     int vmfd;
779     int cpufd;
780 } KVMScratchCPU;
781 
782 /*
783  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
784  * from target/arm/kvm.c.
785  */
786 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
787 {
788     int kvmfd = -1, vmfd = -1, cpufd = -1;
789 
790     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
791     if (kvmfd < 0) {
792         goto err;
793     }
794     do {
795         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
796     } while (vmfd == -1 && errno == EINTR);
797     if (vmfd < 0) {
798         goto err;
799     }
800     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
801     if (cpufd < 0) {
802         goto err;
803     }
804 
805     scratch->kvmfd =  kvmfd;
806     scratch->vmfd = vmfd;
807     scratch->cpufd = cpufd;
808 
809     return true;
810 
811  err:
812     if (cpufd >= 0) {
813         close(cpufd);
814     }
815     if (vmfd >= 0) {
816         close(vmfd);
817     }
818     if (kvmfd >= 0) {
819         close(kvmfd);
820     }
821 
822     return false;
823 }
824 
825 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
826 {
827     close(scratch->cpufd);
828     close(scratch->vmfd);
829     close(scratch->kvmfd);
830 }
831 
832 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
833 {
834     CPURISCVState *env = &cpu->env;
835     struct kvm_one_reg reg;
836     int ret;
837 
838     reg.id = RISCV_CONFIG_REG(env, mvendorid);
839     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
840     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
841     if (ret != 0) {
842         error_report("Unable to retrieve mvendorid from host, error %d", ret);
843     }
844 
845     reg.id = RISCV_CONFIG_REG(env, marchid);
846     reg.addr = (uint64_t)&cpu->cfg.marchid;
847     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
848     if (ret != 0) {
849         error_report("Unable to retrieve marchid from host, error %d", ret);
850     }
851 
852     reg.id = RISCV_CONFIG_REG(env, mimpid);
853     reg.addr = (uint64_t)&cpu->cfg.mimpid;
854     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
855     if (ret != 0) {
856         error_report("Unable to retrieve mimpid from host, error %d", ret);
857     }
858 }
859 
860 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
861                                          KVMScratchCPU *kvmcpu)
862 {
863     CPURISCVState *env = &cpu->env;
864     struct kvm_one_reg reg;
865     int ret;
866 
867     reg.id = RISCV_CONFIG_REG(env, isa);
868     reg.addr = (uint64_t)&env->misa_ext_mask;
869     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
870 
871     if (ret) {
872         error_report("Unable to fetch ISA register from KVM, "
873                      "error %d", ret);
874         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
875         exit(EXIT_FAILURE);
876     }
877 
878     env->misa_ext = env->misa_ext_mask;
879 }
880 
881 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
882                                          KVMCPUConfig *cbomz_cfg)
883 {
884     CPURISCVState *env = &cpu->env;
885     struct kvm_one_reg reg;
886     int ret;
887 
888     reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
889                                     cbomz_cfg->kvm_reg_id);
890     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
891     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
892     if (ret != 0) {
893         error_report("Unable to read KVM reg %s, error %d",
894                      cbomz_cfg->name, ret);
895         exit(EXIT_FAILURE);
896     }
897 }
898 
899 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
900                                            KVMScratchCPU *kvmcpu)
901 {
902     CPURISCVState *env = &cpu->env;
903     uint64_t val;
904     int i, ret;
905 
906     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
907         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
908         struct kvm_one_reg reg;
909 
910         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
911                                         multi_ext_cfg->kvm_reg_id);
912         reg.addr = (uint64_t)&val;
913         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
914         if (ret != 0) {
915             if (errno == EINVAL) {
916                 /* Silently default to 'false' if KVM does not support it. */
917                 multi_ext_cfg->supported = false;
918                 val = false;
919             } else {
920                 error_report("Unable to read ISA_EXT KVM register %s: %s",
921                              multi_ext_cfg->name, strerror(errno));
922                 exit(EXIT_FAILURE);
923             }
924         } else {
925             multi_ext_cfg->supported = true;
926         }
927 
928         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
929     }
930 
931     if (cpu->cfg.ext_zicbom) {
932         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
933     }
934 
935     if (cpu->cfg.ext_zicboz) {
936         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
937     }
938 }
939 
940 static int uint64_cmp(const void *a, const void *b)
941 {
942     uint64_t val1 = *(const uint64_t *)a;
943     uint64_t val2 = *(const uint64_t *)b;
944 
945     if (val1 < val2) {
946         return -1;
947     }
948 
949     if (val1 > val2) {
950         return 1;
951     }
952 
953     return 0;
954 }
955 
956 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
957 {
958     KVMCPUConfig *multi_ext_cfg;
959     struct kvm_one_reg reg;
960     struct kvm_reg_list rl_struct;
961     struct kvm_reg_list *reglist;
962     uint64_t val, reg_id, *reg_search;
963     int i, ret;
964 
965     rl_struct.n = 0;
966     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
967 
968     /*
969      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
970      * (EINVAL). Use read_legacy() in this case.
971      */
972     if (errno == EINVAL) {
973         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
974     } else if (errno != E2BIG) {
975         /*
976          * E2BIG is an expected error message for the API since we
977          * don't know the number of registers. The right amount will
978          * be written in rl_struct.n.
979          *
980          * Error out if we get any other errno.
981          */
982         error_report("Error when accessing get-reg-list: %s",
983                      strerror(errno));
984         exit(EXIT_FAILURE);
985     }
986 
987     reglist = g_malloc(sizeof(struct kvm_reg_list) +
988                        rl_struct.n * sizeof(uint64_t));
989     reglist->n = rl_struct.n;
990     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
991     if (ret) {
992         error_report("Error when reading KVM_GET_REG_LIST: %s",
993                      strerror(errno));
994         exit(EXIT_FAILURE);
995     }
996 
997     /* sort reglist to use bsearch() */
998     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
999 
1000     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
1001         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
1002         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
1003                                         multi_ext_cfg->kvm_reg_id);
1004         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
1005                              sizeof(uint64_t), uint64_cmp);
1006         if (!reg_search) {
1007             continue;
1008         }
1009 
1010         reg.id = reg_id;
1011         reg.addr = (uint64_t)&val;
1012         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1013         if (ret != 0) {
1014             error_report("Unable to read ISA_EXT KVM register %s: %s",
1015                          multi_ext_cfg->name, strerror(errno));
1016             exit(EXIT_FAILURE);
1017         }
1018 
1019         multi_ext_cfg->supported = true;
1020         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1021     }
1022 
1023     if (cpu->cfg.ext_zicbom) {
1024         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1025     }
1026 
1027     if (cpu->cfg.ext_zicboz) {
1028         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1029     }
1030 }
1031 
1032 static void riscv_init_kvm_registers(Object *cpu_obj)
1033 {
1034     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
1035     KVMScratchCPU kvmcpu;
1036 
1037     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1038         return;
1039     }
1040 
1041     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
1042     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
1043     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
1044 
1045     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1046 }
1047 
1048 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
1049     KVM_CAP_LAST_INFO
1050 };
1051 
1052 int kvm_arch_get_registers(CPUState *cs)
1053 {
1054     int ret = 0;
1055 
1056     ret = kvm_riscv_get_regs_core(cs);
1057     if (ret) {
1058         return ret;
1059     }
1060 
1061     ret = kvm_riscv_get_regs_csr(cs);
1062     if (ret) {
1063         return ret;
1064     }
1065 
1066     ret = kvm_riscv_get_regs_fp(cs);
1067     if (ret) {
1068         return ret;
1069     }
1070 
1071     ret = kvm_riscv_get_regs_vector(cs);
1072     if (ret) {
1073         return ret;
1074     }
1075 
1076     return ret;
1077 }
1078 
1079 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1080 {
1081     if (cap_has_mp_state) {
1082         struct kvm_mp_state mp_state = {
1083             .mp_state = state
1084         };
1085 
1086         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1087         if (ret) {
1088             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1089                     __func__, ret, strerror(-ret));
1090             return -1;
1091         }
1092     }
1093 
1094     return 0;
1095 }
1096 
1097 int kvm_arch_put_registers(CPUState *cs, int level)
1098 {
1099     int ret = 0;
1100 
1101     ret = kvm_riscv_put_regs_core(cs);
1102     if (ret) {
1103         return ret;
1104     }
1105 
1106     ret = kvm_riscv_put_regs_csr(cs);
1107     if (ret) {
1108         return ret;
1109     }
1110 
1111     ret = kvm_riscv_put_regs_fp(cs);
1112     if (ret) {
1113         return ret;
1114     }
1115 
1116     ret = kvm_riscv_put_regs_vector(cs);
1117     if (ret) {
1118         return ret;
1119     }
1120 
1121     if (KVM_PUT_RESET_STATE == level) {
1122         RISCVCPU *cpu = RISCV_CPU(cs);
1123         if (cs->cpu_index == 0) {
1124             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1125         } else {
1126             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1127         }
1128         if (ret) {
1129             return ret;
1130         }
1131     }
1132 
1133     return ret;
1134 }
1135 
1136 int kvm_arch_release_virq_post(int virq)
1137 {
1138     return 0;
1139 }
1140 
1141 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1142                              uint64_t address, uint32_t data, PCIDevice *dev)
1143 {
1144     return 0;
1145 }
1146 
1147 int kvm_arch_destroy_vcpu(CPUState *cs)
1148 {
1149     return 0;
1150 }
1151 
1152 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1153 {
1154     return cpu->cpu_index;
1155 }
1156 
1157 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1158                                       RunState state)
1159 {
1160     CPUState *cs = opaque;
1161 
1162     if (running) {
1163         kvm_riscv_put_regs_timer(cs);
1164     } else {
1165         kvm_riscv_get_regs_timer(cs);
1166     }
1167 }
1168 
1169 void kvm_arch_init_irq_routing(KVMState *s)
1170 {
1171 }
1172 
1173 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1174 {
1175     CPURISCVState *env = &cpu->env;
1176     target_ulong reg;
1177     uint64_t id;
1178     int ret;
1179 
1180     id = RISCV_CONFIG_REG(env, mvendorid);
1181     /*
1182      * cfg.mvendorid is an uint32 but a target_ulong will
1183      * be written. Assign it to a target_ulong var to avoid
1184      * writing pieces of other cpu->cfg fields in the reg.
1185      */
1186     reg = cpu->cfg.mvendorid;
1187     ret = kvm_set_one_reg(cs, id, &reg);
1188     if (ret != 0) {
1189         return ret;
1190     }
1191 
1192     id = RISCV_CONFIG_REG(env, marchid);
1193     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1194     if (ret != 0) {
1195         return ret;
1196     }
1197 
1198     id = RISCV_CONFIG_REG(env, mimpid);
1199     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1200 
1201     return ret;
1202 }
1203 
1204 int kvm_arch_init_vcpu(CPUState *cs)
1205 {
1206     int ret = 0;
1207     RISCVCPU *cpu = RISCV_CPU(cs);
1208 
1209     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1210 
1211     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1212         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1213         if (ret != 0) {
1214             return ret;
1215         }
1216     }
1217 
1218     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1219     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1220 
1221     return ret;
1222 }
1223 
1224 int kvm_arch_msi_data_to_gsi(uint32_t data)
1225 {
1226     abort();
1227 }
1228 
1229 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1230                                 int vector, PCIDevice *dev)
1231 {
1232     return 0;
1233 }
1234 
1235 int kvm_arch_get_default_type(MachineState *ms)
1236 {
1237     return 0;
1238 }
1239 
1240 int kvm_arch_init(MachineState *ms, KVMState *s)
1241 {
1242     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1243     return 0;
1244 }
1245 
1246 int kvm_arch_irqchip_create(KVMState *s)
1247 {
1248     if (kvm_kernel_irqchip_split()) {
1249         error_report("-machine kernel_irqchip=split is not supported on RISC-V.");
1250         exit(1);
1251     }
1252 
1253     /*
1254      * We can create the VAIA using the newer device control API.
1255      */
1256     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1257 }
1258 
1259 int kvm_arch_process_async_events(CPUState *cs)
1260 {
1261     return 0;
1262 }
1263 
1264 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1265 {
1266 }
1267 
1268 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1269 {
1270     return MEMTXATTRS_UNSPECIFIED;
1271 }
1272 
1273 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1274 {
1275     return true;
1276 }
1277 
1278 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1279 {
1280     int ret = 0;
1281     unsigned char ch;
1282     switch (run->riscv_sbi.extension_id) {
1283     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1284         ch = run->riscv_sbi.args[0];
1285         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1286         break;
1287     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1288         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1289         if (ret == sizeof(ch)) {
1290             run->riscv_sbi.ret[0] = ch;
1291         } else {
1292             run->riscv_sbi.ret[0] = -1;
1293         }
1294         ret = 0;
1295         break;
1296     default:
1297         qemu_log_mask(LOG_UNIMP,
1298                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1299                       __func__, run->riscv_sbi.extension_id);
1300         ret = -1;
1301         break;
1302     }
1303     return ret;
1304 }
1305 
1306 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1307 {
1308     int ret = 0;
1309     switch (run->exit_reason) {
1310     case KVM_EXIT_RISCV_SBI:
1311         ret = kvm_riscv_handle_sbi(cs, run);
1312         break;
1313     default:
1314         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1315                       __func__, run->exit_reason);
1316         ret = -1;
1317         break;
1318     }
1319     return ret;
1320 }
1321 
1322 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1323 {
1324     CPURISCVState *env = &cpu->env;
1325     int i;
1326 
1327     if (!kvm_enabled()) {
1328         return;
1329     }
1330     for (i = 0; i < 32; i++) {
1331         env->gpr[i] = 0;
1332     }
1333     env->pc = cpu->env.kernel_addr;
1334     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1335     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1336     env->satp = 0;
1337     env->mie = 0;
1338     env->stvec = 0;
1339     env->sscratch = 0;
1340     env->sepc = 0;
1341     env->scause = 0;
1342     env->stval = 0;
1343     env->mip = 0;
1344 }
1345 
1346 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1347 {
1348     int ret;
1349     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1350 
1351     if (irq != IRQ_S_EXT) {
1352         perror("kvm riscv set irq != IRQ_S_EXT\n");
1353         abort();
1354     }
1355 
1356     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1357     if (ret < 0) {
1358         perror("Set irq failed");
1359         abort();
1360     }
1361 }
1362 
1363 bool kvm_arch_cpu_check_are_resettable(void)
1364 {
1365     return true;
1366 }
1367 
1368 static int aia_mode;
1369 
1370 static const char *kvm_aia_mode_str(uint64_t mode)
1371 {
1372     switch (mode) {
1373     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1374         return "emul";
1375     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1376         return "hwaccel";
1377     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1378     default:
1379         return "auto";
1380     };
1381 }
1382 
1383 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1384 {
1385     return g_strdup(kvm_aia_mode_str(aia_mode));
1386 }
1387 
1388 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1389 {
1390     if (!strcmp(val, "emul")) {
1391         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1392     } else if (!strcmp(val, "hwaccel")) {
1393         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1394     } else if (!strcmp(val, "auto")) {
1395         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1396     } else {
1397         error_setg(errp, "Invalid KVM AIA mode");
1398         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1399     }
1400 }
1401 
1402 void kvm_arch_accel_class_init(ObjectClass *oc)
1403 {
1404     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1405                                   riscv_set_kvm_aia);
1406     object_class_property_set_description(oc, "riscv-aia",
1407                                           "Set KVM AIA mode. Valid values are "
1408                                           "emul, hwaccel, and auto. Default "
1409                                           "is auto.");
1410     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1411                                     "auto");
1412 }
1413 
1414 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1415                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1416                           uint64_t aplic_base, uint64_t imsic_base,
1417                           uint64_t guest_num)
1418 {
1419     int ret, i;
1420     int aia_fd = -1;
1421     uint64_t default_aia_mode;
1422     uint64_t socket_count = riscv_socket_count(machine);
1423     uint64_t max_hart_per_socket = 0;
1424     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1425     uint64_t socket_bits, hart_bits, guest_bits;
1426 
1427     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1428 
1429     if (aia_fd < 0) {
1430         error_report("Unable to create in-kernel irqchip");
1431         exit(1);
1432     }
1433 
1434     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1435                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1436                             &default_aia_mode, false, NULL);
1437     if (ret < 0) {
1438         error_report("KVM AIA: failed to get current KVM AIA mode");
1439         exit(1);
1440     }
1441     qemu_log("KVM AIA: default mode is %s\n",
1442              kvm_aia_mode_str(default_aia_mode));
1443 
1444     if (default_aia_mode != aia_mode) {
1445         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1446                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1447                                 &aia_mode, true, NULL);
1448         if (ret < 0)
1449             warn_report("KVM AIA: failed to set KVM AIA mode");
1450         else
1451             qemu_log("KVM AIA: set current mode to %s\n",
1452                      kvm_aia_mode_str(aia_mode));
1453     }
1454 
1455     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1456                             KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1457                             &aia_irq_num, true, NULL);
1458     if (ret < 0) {
1459         error_report("KVM AIA: failed to set number of input irq lines");
1460         exit(1);
1461     }
1462 
1463     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1464                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1465                             &aia_msi_num, true, NULL);
1466     if (ret < 0) {
1467         error_report("KVM AIA: failed to set number of msi");
1468         exit(1);
1469     }
1470 
1471 
1472     if (socket_count > 1) {
1473         socket_bits = find_last_bit(&socket_count, BITS_PER_LONG) + 1;
1474         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1475                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1476                                 &socket_bits, true, NULL);
1477         if (ret < 0) {
1478             error_report("KVM AIA: failed to set group_bits");
1479             exit(1);
1480         }
1481 
1482         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1483                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1484                                 &group_shift, true, NULL);
1485         if (ret < 0) {
1486             error_report("KVM AIA: failed to set group_shift");
1487             exit(1);
1488         }
1489     }
1490 
1491     guest_bits = guest_num == 0 ? 0 :
1492                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1493     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1494                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1495                             &guest_bits, true, NULL);
1496     if (ret < 0) {
1497         error_report("KVM AIA: failed to set guest_bits");
1498         exit(1);
1499     }
1500 
1501     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1502                             KVM_DEV_RISCV_AIA_ADDR_APLIC,
1503                             &aplic_base, true, NULL);
1504     if (ret < 0) {
1505         error_report("KVM AIA: failed to set the base address of APLIC");
1506         exit(1);
1507     }
1508 
1509     for (socket = 0; socket < socket_count; socket++) {
1510         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1511         hart_count = riscv_socket_hart_count(machine, socket);
1512         base_hart = riscv_socket_first_hartid(machine, socket);
1513 
1514         if (max_hart_per_socket < hart_count) {
1515             max_hart_per_socket = hart_count;
1516         }
1517 
1518         for (i = 0; i < hart_count; i++) {
1519             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1520             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1521                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1522                                     &imsic_addr, true, NULL);
1523             if (ret < 0) {
1524                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1525                 exit(1);
1526             }
1527         }
1528     }
1529 
1530     hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1531     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1532                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1533                             &hart_bits, true, NULL);
1534     if (ret < 0) {
1535         error_report("KVM AIA: failed to set hart_bits");
1536         exit(1);
1537     }
1538 
1539     if (kvm_has_gsi_routing()) {
1540         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1541             /* KVM AIA only has one APLIC instance */
1542             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1543         }
1544         kvm_gsi_routing_allowed = true;
1545         kvm_irqchip_commit_routes(kvm_state);
1546     }
1547 
1548     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1549                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1550                             NULL, true, NULL);
1551     if (ret < 0) {
1552         error_report("KVM AIA: initialized fail");
1553         exit(1);
1554     }
1555 
1556     kvm_msi_via_irqfd_allowed = true;
1557 }
1558 
1559 static void kvm_cpu_instance_init(CPUState *cs)
1560 {
1561     Object *obj = OBJECT(RISCV_CPU(cs));
1562     DeviceState *dev = DEVICE(obj);
1563 
1564     riscv_init_kvm_registers(obj);
1565 
1566     kvm_riscv_add_cpu_user_properties(obj);
1567 
1568     for (Property *prop = riscv_cpu_options; prop && prop->name; prop++) {
1569         /* Check if we have a specific KVM handler for the option */
1570         if (object_property_find(obj, prop->name)) {
1571             continue;
1572         }
1573         qdev_property_add_static(dev, prop);
1574     }
1575 }
1576 
1577 /*
1578  * We'll get here via the following path:
1579  *
1580  * riscv_cpu_realize()
1581  *   -> cpu_exec_realizefn()
1582  *      -> kvm_cpu_realize() (via accel_cpu_common_realize())
1583  */
1584 static bool kvm_cpu_realize(CPUState *cs, Error **errp)
1585 {
1586     RISCVCPU *cpu = RISCV_CPU(cs);
1587     int ret;
1588 
1589     if (riscv_has_ext(&cpu->env, RVV)) {
1590         ret = prctl(PR_RISCV_V_SET_CONTROL, PR_RISCV_V_VSTATE_CTRL_ON);
1591         if (ret) {
1592             error_setg(errp, "Error in prctl PR_RISCV_V_SET_CONTROL, code: %s",
1593                        strerrorname_np(errno));
1594             return false;
1595         }
1596     }
1597 
1598    return true;
1599 }
1600 
1601 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1602 {
1603     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1604 
1605     acc->cpu_instance_init = kvm_cpu_instance_init;
1606     acc->cpu_target_realize = kvm_cpu_realize;
1607 }
1608 
1609 static const TypeInfo kvm_cpu_accel_type_info = {
1610     .name = ACCEL_CPU_NAME("kvm"),
1611 
1612     .parent = TYPE_ACCEL_CPU,
1613     .class_init = kvm_cpu_accel_class_init,
1614     .abstract = true,
1615 };
1616 static void kvm_cpu_accel_register_types(void)
1617 {
1618     type_register_static(&kvm_cpu_accel_type_info);
1619 }
1620 type_init(kvm_cpu_accel_register_types);
1621 
1622 static void riscv_host_cpu_init(Object *obj)
1623 {
1624     CPURISCVState *env = &RISCV_CPU(obj)->env;
1625 
1626 #if defined(TARGET_RISCV32)
1627     env->misa_mxl_max = env->misa_mxl = MXL_RV32;
1628 #elif defined(TARGET_RISCV64)
1629     env->misa_mxl_max = env->misa_mxl = MXL_RV64;
1630 #endif
1631 }
1632 
1633 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
1634     {
1635         .name = TYPE_RISCV_CPU_HOST,
1636         .parent = TYPE_RISCV_CPU,
1637         .instance_init = riscv_host_cpu_init,
1638     }
1639 };
1640 
1641 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
1642