xref: /openbmc/qemu/target/riscv/kvm/kvm-cpu.c (revision b84efa39)
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, "cboz_blocksize", "uint16",
497                         NULL, kvm_cpu_set_cbomz_blksize,
498                         NULL, &kvm_cboz_blocksize);
499 
500     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_extensions);
501     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_vendor_exts);
502     riscv_cpu_add_kvm_unavail_prop_array(cpu_obj, riscv_cpu_experimental_exts);
503 
504    /* We don't have the needed KVM support for profiles */
505     for (i = 0; riscv_profiles[i] != NULL; i++) {
506         riscv_cpu_add_kvm_unavail_prop(cpu_obj, riscv_profiles[i]->name);
507     }
508 }
509 
510 static int kvm_riscv_get_regs_core(CPUState *cs)
511 {
512     int ret = 0;
513     int i;
514     target_ulong reg;
515     CPURISCVState *env = &RISCV_CPU(cs)->env;
516 
517     ret = kvm_get_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
518     if (ret) {
519         return ret;
520     }
521     env->pc = reg;
522 
523     for (i = 1; i < 32; i++) {
524         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
525         ret = kvm_get_one_reg(cs, id, &reg);
526         if (ret) {
527             return ret;
528         }
529         env->gpr[i] = reg;
530     }
531 
532     return ret;
533 }
534 
535 static int kvm_riscv_put_regs_core(CPUState *cs)
536 {
537     int ret = 0;
538     int i;
539     target_ulong reg;
540     CPURISCVState *env = &RISCV_CPU(cs)->env;
541 
542     reg = env->pc;
543     ret = kvm_set_one_reg(cs, RISCV_CORE_REG(env, regs.pc), &reg);
544     if (ret) {
545         return ret;
546     }
547 
548     for (i = 1; i < 32; i++) {
549         uint64_t id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CORE, i);
550         reg = env->gpr[i];
551         ret = kvm_set_one_reg(cs, id, &reg);
552         if (ret) {
553             return ret;
554         }
555     }
556 
557     return ret;
558 }
559 
560 static int kvm_riscv_get_regs_csr(CPUState *cs)
561 {
562     CPURISCVState *env = &RISCV_CPU(cs)->env;
563 
564     KVM_RISCV_GET_CSR(cs, env, sstatus, env->mstatus);
565     KVM_RISCV_GET_CSR(cs, env, sie, env->mie);
566     KVM_RISCV_GET_CSR(cs, env, stvec, env->stvec);
567     KVM_RISCV_GET_CSR(cs, env, sscratch, env->sscratch);
568     KVM_RISCV_GET_CSR(cs, env, sepc, env->sepc);
569     KVM_RISCV_GET_CSR(cs, env, scause, env->scause);
570     KVM_RISCV_GET_CSR(cs, env, stval, env->stval);
571     KVM_RISCV_GET_CSR(cs, env, sip, env->mip);
572     KVM_RISCV_GET_CSR(cs, env, satp, env->satp);
573 
574     return 0;
575 }
576 
577 static int kvm_riscv_put_regs_csr(CPUState *cs)
578 {
579     CPURISCVState *env = &RISCV_CPU(cs)->env;
580 
581     KVM_RISCV_SET_CSR(cs, env, sstatus, env->mstatus);
582     KVM_RISCV_SET_CSR(cs, env, sie, env->mie);
583     KVM_RISCV_SET_CSR(cs, env, stvec, env->stvec);
584     KVM_RISCV_SET_CSR(cs, env, sscratch, env->sscratch);
585     KVM_RISCV_SET_CSR(cs, env, sepc, env->sepc);
586     KVM_RISCV_SET_CSR(cs, env, scause, env->scause);
587     KVM_RISCV_SET_CSR(cs, env, stval, env->stval);
588     KVM_RISCV_SET_CSR(cs, env, sip, env->mip);
589     KVM_RISCV_SET_CSR(cs, env, satp, env->satp);
590 
591     return 0;
592 }
593 
594 static int kvm_riscv_get_regs_fp(CPUState *cs)
595 {
596     int ret = 0;
597     int i;
598     CPURISCVState *env = &RISCV_CPU(cs)->env;
599 
600     if (riscv_has_ext(env, RVD)) {
601         uint64_t reg;
602         for (i = 0; i < 32; i++) {
603             ret = kvm_get_one_reg(cs, RISCV_FP_D_REG(i), &reg);
604             if (ret) {
605                 return ret;
606             }
607             env->fpr[i] = reg;
608         }
609         return ret;
610     }
611 
612     if (riscv_has_ext(env, RVF)) {
613         uint32_t reg;
614         for (i = 0; i < 32; i++) {
615             ret = kvm_get_one_reg(cs, RISCV_FP_F_REG(i), &reg);
616             if (ret) {
617                 return ret;
618             }
619             env->fpr[i] = reg;
620         }
621         return ret;
622     }
623 
624     return ret;
625 }
626 
627 static int kvm_riscv_put_regs_fp(CPUState *cs)
628 {
629     int ret = 0;
630     int i;
631     CPURISCVState *env = &RISCV_CPU(cs)->env;
632 
633     if (riscv_has_ext(env, RVD)) {
634         uint64_t reg;
635         for (i = 0; i < 32; i++) {
636             reg = env->fpr[i];
637             ret = kvm_set_one_reg(cs, RISCV_FP_D_REG(i), &reg);
638             if (ret) {
639                 return ret;
640             }
641         }
642         return ret;
643     }
644 
645     if (riscv_has_ext(env, RVF)) {
646         uint32_t reg;
647         for (i = 0; i < 32; i++) {
648             reg = env->fpr[i];
649             ret = kvm_set_one_reg(cs, RISCV_FP_F_REG(i), &reg);
650             if (ret) {
651                 return ret;
652             }
653         }
654         return ret;
655     }
656 
657     return ret;
658 }
659 
660 static void kvm_riscv_get_regs_timer(CPUState *cs)
661 {
662     CPURISCVState *env = &RISCV_CPU(cs)->env;
663 
664     if (env->kvm_timer_dirty) {
665         return;
666     }
667 
668     KVM_RISCV_GET_TIMER(cs, time, env->kvm_timer_time);
669     KVM_RISCV_GET_TIMER(cs, compare, env->kvm_timer_compare);
670     KVM_RISCV_GET_TIMER(cs, state, env->kvm_timer_state);
671     KVM_RISCV_GET_TIMER(cs, frequency, env->kvm_timer_frequency);
672 
673     env->kvm_timer_dirty = true;
674 }
675 
676 static void kvm_riscv_put_regs_timer(CPUState *cs)
677 {
678     uint64_t reg;
679     CPURISCVState *env = &RISCV_CPU(cs)->env;
680 
681     if (!env->kvm_timer_dirty) {
682         return;
683     }
684 
685     KVM_RISCV_SET_TIMER(cs, time, env->kvm_timer_time);
686     KVM_RISCV_SET_TIMER(cs, compare, env->kvm_timer_compare);
687 
688     /*
689      * To set register of RISCV_TIMER_REG(state) will occur a error from KVM
690      * on env->kvm_timer_state == 0, It's better to adapt in KVM, but it
691      * doesn't matter that adaping in QEMU now.
692      * TODO If KVM changes, adapt here.
693      */
694     if (env->kvm_timer_state) {
695         KVM_RISCV_SET_TIMER(cs, state, env->kvm_timer_state);
696     }
697 
698     /*
699      * For now, migration will not work between Hosts with different timer
700      * frequency. Therefore, we should check whether they are the same here
701      * during the migration.
702      */
703     if (migration_is_running(migrate_get_current()->state)) {
704         KVM_RISCV_GET_TIMER(cs, frequency, reg);
705         if (reg != env->kvm_timer_frequency) {
706             error_report("Dst Hosts timer frequency != Src Hosts");
707         }
708     }
709 
710     env->kvm_timer_dirty = false;
711 }
712 
713 static int kvm_riscv_get_regs_vector(CPUState *cs)
714 {
715     CPURISCVState *env = &RISCV_CPU(cs)->env;
716     target_ulong reg;
717     int ret = 0;
718 
719     if (!riscv_has_ext(env, RVV)) {
720         return 0;
721     }
722 
723     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
724     if (ret) {
725         return ret;
726     }
727     env->vstart = reg;
728 
729     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
730     if (ret) {
731         return ret;
732     }
733     env->vl = reg;
734 
735     ret = kvm_get_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
736     if (ret) {
737         return ret;
738     }
739     env->vtype = reg;
740 
741     return 0;
742 }
743 
744 static int kvm_riscv_put_regs_vector(CPUState *cs)
745 {
746     CPURISCVState *env = &RISCV_CPU(cs)->env;
747     target_ulong reg;
748     int ret = 0;
749 
750     if (!riscv_has_ext(env, RVV)) {
751         return 0;
752     }
753 
754     reg = env->vstart;
755     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vstart), &reg);
756     if (ret) {
757         return ret;
758     }
759 
760     reg = env->vl;
761     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vl), &reg);
762     if (ret) {
763         return ret;
764     }
765 
766     reg = env->vtype;
767     ret = kvm_set_one_reg(cs, RISCV_VECTOR_CSR_REG(env, vtype), &reg);
768 
769     return ret;
770 }
771 
772 typedef struct KVMScratchCPU {
773     int kvmfd;
774     int vmfd;
775     int cpufd;
776 } KVMScratchCPU;
777 
778 /*
779  * Heavily inspired by kvm_arm_create_scratch_host_vcpu()
780  * from target/arm/kvm.c.
781  */
782 static bool kvm_riscv_create_scratch_vcpu(KVMScratchCPU *scratch)
783 {
784     int kvmfd = -1, vmfd = -1, cpufd = -1;
785 
786     kvmfd = qemu_open_old("/dev/kvm", O_RDWR);
787     if (kvmfd < 0) {
788         goto err;
789     }
790     do {
791         vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
792     } while (vmfd == -1 && errno == EINTR);
793     if (vmfd < 0) {
794         goto err;
795     }
796     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
797     if (cpufd < 0) {
798         goto err;
799     }
800 
801     scratch->kvmfd =  kvmfd;
802     scratch->vmfd = vmfd;
803     scratch->cpufd = cpufd;
804 
805     return true;
806 
807  err:
808     if (cpufd >= 0) {
809         close(cpufd);
810     }
811     if (vmfd >= 0) {
812         close(vmfd);
813     }
814     if (kvmfd >= 0) {
815         close(kvmfd);
816     }
817 
818     return false;
819 }
820 
821 static void kvm_riscv_destroy_scratch_vcpu(KVMScratchCPU *scratch)
822 {
823     close(scratch->cpufd);
824     close(scratch->vmfd);
825     close(scratch->kvmfd);
826 }
827 
828 static void kvm_riscv_init_machine_ids(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
829 {
830     CPURISCVState *env = &cpu->env;
831     struct kvm_one_reg reg;
832     int ret;
833 
834     reg.id = RISCV_CONFIG_REG(env, mvendorid);
835     reg.addr = (uint64_t)&cpu->cfg.mvendorid;
836     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
837     if (ret != 0) {
838         error_report("Unable to retrieve mvendorid from host, error %d", ret);
839     }
840 
841     reg.id = RISCV_CONFIG_REG(env, marchid);
842     reg.addr = (uint64_t)&cpu->cfg.marchid;
843     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
844     if (ret != 0) {
845         error_report("Unable to retrieve marchid from host, error %d", ret);
846     }
847 
848     reg.id = RISCV_CONFIG_REG(env, mimpid);
849     reg.addr = (uint64_t)&cpu->cfg.mimpid;
850     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
851     if (ret != 0) {
852         error_report("Unable to retrieve mimpid from host, error %d", ret);
853     }
854 }
855 
856 static void kvm_riscv_init_misa_ext_mask(RISCVCPU *cpu,
857                                          KVMScratchCPU *kvmcpu)
858 {
859     CPURISCVState *env = &cpu->env;
860     struct kvm_one_reg reg;
861     int ret;
862 
863     reg.id = RISCV_CONFIG_REG(env, isa);
864     reg.addr = (uint64_t)&env->misa_ext_mask;
865     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
866 
867     if (ret) {
868         error_report("Unable to fetch ISA register from KVM, "
869                      "error %d", ret);
870         kvm_riscv_destroy_scratch_vcpu(kvmcpu);
871         exit(EXIT_FAILURE);
872     }
873 
874     env->misa_ext = env->misa_ext_mask;
875 }
876 
877 static void kvm_riscv_read_cbomz_blksize(RISCVCPU *cpu, KVMScratchCPU *kvmcpu,
878                                          KVMCPUConfig *cbomz_cfg)
879 {
880     CPURISCVState *env = &cpu->env;
881     struct kvm_one_reg reg;
882     int ret;
883 
884     reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
885                                     cbomz_cfg->kvm_reg_id);
886     reg.addr = (uint64_t)kvmconfig_get_cfg_addr(cpu, cbomz_cfg);
887     ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
888     if (ret != 0) {
889         error_report("Unable to read KVM reg %s, error %d",
890                      cbomz_cfg->name, ret);
891         exit(EXIT_FAILURE);
892     }
893 }
894 
895 static void kvm_riscv_read_multiext_legacy(RISCVCPU *cpu,
896                                            KVMScratchCPU *kvmcpu)
897 {
898     CPURISCVState *env = &cpu->env;
899     uint64_t val;
900     int i, ret;
901 
902     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
903         KVMCPUConfig *multi_ext_cfg = &kvm_multi_ext_cfgs[i];
904         struct kvm_one_reg reg;
905 
906         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_ISA_EXT,
907                                         multi_ext_cfg->kvm_reg_id);
908         reg.addr = (uint64_t)&val;
909         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
910         if (ret != 0) {
911             if (errno == EINVAL) {
912                 /* Silently default to 'false' if KVM does not support it. */
913                 multi_ext_cfg->supported = false;
914                 val = false;
915             } else {
916                 error_report("Unable to read ISA_EXT KVM register %s: %s",
917                              multi_ext_cfg->name, strerror(errno));
918                 exit(EXIT_FAILURE);
919             }
920         } else {
921             multi_ext_cfg->supported = true;
922         }
923 
924         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
925     }
926 
927     if (cpu->cfg.ext_zicbom) {
928         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
929     }
930 
931     if (cpu->cfg.ext_zicboz) {
932         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
933     }
934 }
935 
936 static int uint64_cmp(const void *a, const void *b)
937 {
938     uint64_t val1 = *(const uint64_t *)a;
939     uint64_t val2 = *(const uint64_t *)b;
940 
941     if (val1 < val2) {
942         return -1;
943     }
944 
945     if (val1 > val2) {
946         return 1;
947     }
948 
949     return 0;
950 }
951 
952 static void kvm_riscv_init_multiext_cfg(RISCVCPU *cpu, KVMScratchCPU *kvmcpu)
953 {
954     KVMCPUConfig *multi_ext_cfg;
955     struct kvm_one_reg reg;
956     struct kvm_reg_list rl_struct;
957     struct kvm_reg_list *reglist;
958     uint64_t val, reg_id, *reg_search;
959     int i, ret;
960 
961     rl_struct.n = 0;
962     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, &rl_struct);
963 
964     /*
965      * If KVM_GET_REG_LIST isn't supported we'll get errno 22
966      * (EINVAL). Use read_legacy() in this case.
967      */
968     if (errno == EINVAL) {
969         return kvm_riscv_read_multiext_legacy(cpu, kvmcpu);
970     } else if (errno != E2BIG) {
971         /*
972          * E2BIG is an expected error message for the API since we
973          * don't know the number of registers. The right amount will
974          * be written in rl_struct.n.
975          *
976          * Error out if we get any other errno.
977          */
978         error_report("Error when accessing get-reg-list: %s",
979                      strerror(errno));
980         exit(EXIT_FAILURE);
981     }
982 
983     reglist = g_malloc(sizeof(struct kvm_reg_list) +
984                        rl_struct.n * sizeof(uint64_t));
985     reglist->n = rl_struct.n;
986     ret = ioctl(kvmcpu->cpufd, KVM_GET_REG_LIST, reglist);
987     if (ret) {
988         error_report("Error when reading KVM_GET_REG_LIST: %s",
989                      strerror(errno));
990         exit(EXIT_FAILURE);
991     }
992 
993     /* sort reglist to use bsearch() */
994     qsort(&reglist->reg, reglist->n, sizeof(uint64_t), uint64_cmp);
995 
996     for (i = 0; i < ARRAY_SIZE(kvm_multi_ext_cfgs); i++) {
997         multi_ext_cfg = &kvm_multi_ext_cfgs[i];
998         reg_id = kvm_riscv_reg_id_ulong(&cpu->env, KVM_REG_RISCV_ISA_EXT,
999                                         multi_ext_cfg->kvm_reg_id);
1000         reg_search = bsearch(&reg_id, reglist->reg, reglist->n,
1001                              sizeof(uint64_t), uint64_cmp);
1002         if (!reg_search) {
1003             continue;
1004         }
1005 
1006         reg.id = reg_id;
1007         reg.addr = (uint64_t)&val;
1008         ret = ioctl(kvmcpu->cpufd, KVM_GET_ONE_REG, &reg);
1009         if (ret != 0) {
1010             error_report("Unable to read ISA_EXT KVM register %s: %s",
1011                          multi_ext_cfg->name, strerror(errno));
1012             exit(EXIT_FAILURE);
1013         }
1014 
1015         multi_ext_cfg->supported = true;
1016         kvm_cpu_cfg_set(cpu, multi_ext_cfg, val);
1017     }
1018 
1019     if (cpu->cfg.ext_zicbom) {
1020         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cbom_blocksize);
1021     }
1022 
1023     if (cpu->cfg.ext_zicboz) {
1024         kvm_riscv_read_cbomz_blksize(cpu, kvmcpu, &kvm_cboz_blocksize);
1025     }
1026 }
1027 
1028 static void riscv_init_kvm_registers(Object *cpu_obj)
1029 {
1030     RISCVCPU *cpu = RISCV_CPU(cpu_obj);
1031     KVMScratchCPU kvmcpu;
1032 
1033     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1034         return;
1035     }
1036 
1037     kvm_riscv_init_machine_ids(cpu, &kvmcpu);
1038     kvm_riscv_init_misa_ext_mask(cpu, &kvmcpu);
1039     kvm_riscv_init_multiext_cfg(cpu, &kvmcpu);
1040 
1041     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1042 }
1043 
1044 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
1045     KVM_CAP_LAST_INFO
1046 };
1047 
1048 int kvm_arch_get_registers(CPUState *cs)
1049 {
1050     int ret = 0;
1051 
1052     ret = kvm_riscv_get_regs_core(cs);
1053     if (ret) {
1054         return ret;
1055     }
1056 
1057     ret = kvm_riscv_get_regs_csr(cs);
1058     if (ret) {
1059         return ret;
1060     }
1061 
1062     ret = kvm_riscv_get_regs_fp(cs);
1063     if (ret) {
1064         return ret;
1065     }
1066 
1067     ret = kvm_riscv_get_regs_vector(cs);
1068     if (ret) {
1069         return ret;
1070     }
1071 
1072     return ret;
1073 }
1074 
1075 int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
1076 {
1077     if (cap_has_mp_state) {
1078         struct kvm_mp_state mp_state = {
1079             .mp_state = state
1080         };
1081 
1082         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1083         if (ret) {
1084             fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
1085                     __func__, ret, strerror(-ret));
1086             return -1;
1087         }
1088     }
1089 
1090     return 0;
1091 }
1092 
1093 int kvm_arch_put_registers(CPUState *cs, int level)
1094 {
1095     int ret = 0;
1096 
1097     ret = kvm_riscv_put_regs_core(cs);
1098     if (ret) {
1099         return ret;
1100     }
1101 
1102     ret = kvm_riscv_put_regs_csr(cs);
1103     if (ret) {
1104         return ret;
1105     }
1106 
1107     ret = kvm_riscv_put_regs_fp(cs);
1108     if (ret) {
1109         return ret;
1110     }
1111 
1112     ret = kvm_riscv_put_regs_vector(cs);
1113     if (ret) {
1114         return ret;
1115     }
1116 
1117     if (KVM_PUT_RESET_STATE == level) {
1118         RISCVCPU *cpu = RISCV_CPU(cs);
1119         if (cs->cpu_index == 0) {
1120             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
1121         } else {
1122             ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
1123         }
1124         if (ret) {
1125             return ret;
1126         }
1127     }
1128 
1129     return ret;
1130 }
1131 
1132 int kvm_arch_release_virq_post(int virq)
1133 {
1134     return 0;
1135 }
1136 
1137 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1138                              uint64_t address, uint32_t data, PCIDevice *dev)
1139 {
1140     return 0;
1141 }
1142 
1143 int kvm_arch_destroy_vcpu(CPUState *cs)
1144 {
1145     return 0;
1146 }
1147 
1148 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
1149 {
1150     return cpu->cpu_index;
1151 }
1152 
1153 static void kvm_riscv_vm_state_change(void *opaque, bool running,
1154                                       RunState state)
1155 {
1156     CPUState *cs = opaque;
1157 
1158     if (running) {
1159         kvm_riscv_put_regs_timer(cs);
1160     } else {
1161         kvm_riscv_get_regs_timer(cs);
1162     }
1163 }
1164 
1165 void kvm_arch_init_irq_routing(KVMState *s)
1166 {
1167 }
1168 
1169 static int kvm_vcpu_set_machine_ids(RISCVCPU *cpu, CPUState *cs)
1170 {
1171     CPURISCVState *env = &cpu->env;
1172     target_ulong reg;
1173     uint64_t id;
1174     int ret;
1175 
1176     id = RISCV_CONFIG_REG(env, mvendorid);
1177     /*
1178      * cfg.mvendorid is an uint32 but a target_ulong will
1179      * be written. Assign it to a target_ulong var to avoid
1180      * writing pieces of other cpu->cfg fields in the reg.
1181      */
1182     reg = cpu->cfg.mvendorid;
1183     ret = kvm_set_one_reg(cs, id, &reg);
1184     if (ret != 0) {
1185         return ret;
1186     }
1187 
1188     id = RISCV_CONFIG_REG(env, marchid);
1189     ret = kvm_set_one_reg(cs, id, &cpu->cfg.marchid);
1190     if (ret != 0) {
1191         return ret;
1192     }
1193 
1194     id = RISCV_CONFIG_REG(env, mimpid);
1195     ret = kvm_set_one_reg(cs, id, &cpu->cfg.mimpid);
1196 
1197     return ret;
1198 }
1199 
1200 int kvm_arch_init_vcpu(CPUState *cs)
1201 {
1202     int ret = 0;
1203     RISCVCPU *cpu = RISCV_CPU(cs);
1204 
1205     qemu_add_vm_change_state_handler(kvm_riscv_vm_state_change, cs);
1206 
1207     if (!object_dynamic_cast(OBJECT(cpu), TYPE_RISCV_CPU_HOST)) {
1208         ret = kvm_vcpu_set_machine_ids(cpu, cs);
1209         if (ret != 0) {
1210             return ret;
1211         }
1212     }
1213 
1214     kvm_riscv_update_cpu_misa_ext(cpu, cs);
1215     kvm_riscv_update_cpu_cfg_isa_ext(cpu, cs);
1216 
1217     return ret;
1218 }
1219 
1220 int kvm_arch_msi_data_to_gsi(uint32_t data)
1221 {
1222     abort();
1223 }
1224 
1225 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1226                                 int vector, PCIDevice *dev)
1227 {
1228     return 0;
1229 }
1230 
1231 int kvm_arch_get_default_type(MachineState *ms)
1232 {
1233     return 0;
1234 }
1235 
1236 int kvm_arch_init(MachineState *ms, KVMState *s)
1237 {
1238     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
1239     return 0;
1240 }
1241 
1242 int kvm_arch_irqchip_create(KVMState *s)
1243 {
1244     if (kvm_kernel_irqchip_split()) {
1245         error_report("-machine kernel_irqchip=split is not supported on RISC-V.");
1246         exit(1);
1247     }
1248 
1249     /*
1250      * We can create the VAIA using the newer device control API.
1251      */
1252     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
1253 }
1254 
1255 int kvm_arch_process_async_events(CPUState *cs)
1256 {
1257     return 0;
1258 }
1259 
1260 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1261 {
1262 }
1263 
1264 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1265 {
1266     return MEMTXATTRS_UNSPECIFIED;
1267 }
1268 
1269 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
1270 {
1271     return true;
1272 }
1273 
1274 static int kvm_riscv_handle_sbi(CPUState *cs, struct kvm_run *run)
1275 {
1276     int ret = 0;
1277     unsigned char ch;
1278     switch (run->riscv_sbi.extension_id) {
1279     case SBI_EXT_0_1_CONSOLE_PUTCHAR:
1280         ch = run->riscv_sbi.args[0];
1281         qemu_chr_fe_write(serial_hd(0)->be, &ch, sizeof(ch));
1282         break;
1283     case SBI_EXT_0_1_CONSOLE_GETCHAR:
1284         ret = qemu_chr_fe_read_all(serial_hd(0)->be, &ch, sizeof(ch));
1285         if (ret == sizeof(ch)) {
1286             run->riscv_sbi.ret[0] = ch;
1287         } else {
1288             run->riscv_sbi.ret[0] = -1;
1289         }
1290         ret = 0;
1291         break;
1292     default:
1293         qemu_log_mask(LOG_UNIMP,
1294                       "%s: un-handled SBI EXIT, specific reasons is %lu\n",
1295                       __func__, run->riscv_sbi.extension_id);
1296         ret = -1;
1297         break;
1298     }
1299     return ret;
1300 }
1301 
1302 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1303 {
1304     int ret = 0;
1305     switch (run->exit_reason) {
1306     case KVM_EXIT_RISCV_SBI:
1307         ret = kvm_riscv_handle_sbi(cs, run);
1308         break;
1309     default:
1310         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
1311                       __func__, run->exit_reason);
1312         ret = -1;
1313         break;
1314     }
1315     return ret;
1316 }
1317 
1318 void kvm_riscv_reset_vcpu(RISCVCPU *cpu)
1319 {
1320     CPURISCVState *env = &cpu->env;
1321     int i;
1322 
1323     if (!kvm_enabled()) {
1324         return;
1325     }
1326     for (i = 0; i < 32; i++) {
1327         env->gpr[i] = 0;
1328     }
1329     env->pc = cpu->env.kernel_addr;
1330     env->gpr[10] = kvm_arch_vcpu_id(CPU(cpu)); /* a0 */
1331     env->gpr[11] = cpu->env.fdt_addr;          /* a1 */
1332     env->satp = 0;
1333     env->mie = 0;
1334     env->stvec = 0;
1335     env->sscratch = 0;
1336     env->sepc = 0;
1337     env->scause = 0;
1338     env->stval = 0;
1339     env->mip = 0;
1340 }
1341 
1342 void kvm_riscv_set_irq(RISCVCPU *cpu, int irq, int level)
1343 {
1344     int ret;
1345     unsigned virq = level ? KVM_INTERRUPT_SET : KVM_INTERRUPT_UNSET;
1346 
1347     if (irq != IRQ_S_EXT) {
1348         perror("kvm riscv set irq != IRQ_S_EXT\n");
1349         abort();
1350     }
1351 
1352     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1353     if (ret < 0) {
1354         perror("Set irq failed");
1355         abort();
1356     }
1357 }
1358 
1359 bool kvm_arch_cpu_check_are_resettable(void)
1360 {
1361     return true;
1362 }
1363 
1364 static int aia_mode;
1365 
1366 static const char *kvm_aia_mode_str(uint64_t mode)
1367 {
1368     switch (mode) {
1369     case KVM_DEV_RISCV_AIA_MODE_EMUL:
1370         return "emul";
1371     case KVM_DEV_RISCV_AIA_MODE_HWACCEL:
1372         return "hwaccel";
1373     case KVM_DEV_RISCV_AIA_MODE_AUTO:
1374     default:
1375         return "auto";
1376     };
1377 }
1378 
1379 static char *riscv_get_kvm_aia(Object *obj, Error **errp)
1380 {
1381     return g_strdup(kvm_aia_mode_str(aia_mode));
1382 }
1383 
1384 static void riscv_set_kvm_aia(Object *obj, const char *val, Error **errp)
1385 {
1386     if (!strcmp(val, "emul")) {
1387         aia_mode = KVM_DEV_RISCV_AIA_MODE_EMUL;
1388     } else if (!strcmp(val, "hwaccel")) {
1389         aia_mode = KVM_DEV_RISCV_AIA_MODE_HWACCEL;
1390     } else if (!strcmp(val, "auto")) {
1391         aia_mode = KVM_DEV_RISCV_AIA_MODE_AUTO;
1392     } else {
1393         error_setg(errp, "Invalid KVM AIA mode");
1394         error_append_hint(errp, "Valid values are emul, hwaccel, and auto.\n");
1395     }
1396 }
1397 
1398 void kvm_arch_accel_class_init(ObjectClass *oc)
1399 {
1400     object_class_property_add_str(oc, "riscv-aia", riscv_get_kvm_aia,
1401                                   riscv_set_kvm_aia);
1402     object_class_property_set_description(oc, "riscv-aia",
1403                                           "Set KVM AIA mode. Valid values are "
1404                                           "emul, hwaccel, and auto. Default "
1405                                           "is auto.");
1406     object_property_set_default_str(object_class_property_find(oc, "riscv-aia"),
1407                                     "auto");
1408 }
1409 
1410 void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
1411                           uint64_t aia_irq_num, uint64_t aia_msi_num,
1412                           uint64_t aplic_base, uint64_t imsic_base,
1413                           uint64_t guest_num)
1414 {
1415     int ret, i;
1416     int aia_fd = -1;
1417     uint64_t default_aia_mode;
1418     uint64_t socket_count = riscv_socket_count(machine);
1419     uint64_t max_hart_per_socket = 0;
1420     uint64_t socket, base_hart, hart_count, socket_imsic_base, imsic_addr;
1421     uint64_t socket_bits, hart_bits, guest_bits;
1422 
1423     aia_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_RISCV_AIA, false);
1424 
1425     if (aia_fd < 0) {
1426         error_report("Unable to create in-kernel irqchip");
1427         exit(1);
1428     }
1429 
1430     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1431                             KVM_DEV_RISCV_AIA_CONFIG_MODE,
1432                             &default_aia_mode, false, NULL);
1433     if (ret < 0) {
1434         error_report("KVM AIA: failed to get current KVM AIA mode");
1435         exit(1);
1436     }
1437     qemu_log("KVM AIA: default mode is %s\n",
1438              kvm_aia_mode_str(default_aia_mode));
1439 
1440     if (default_aia_mode != aia_mode) {
1441         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1442                                 KVM_DEV_RISCV_AIA_CONFIG_MODE,
1443                                 &aia_mode, true, NULL);
1444         if (ret < 0)
1445             warn_report("KVM AIA: failed to set KVM AIA mode");
1446         else
1447             qemu_log("KVM AIA: set current mode to %s\n",
1448                      kvm_aia_mode_str(aia_mode));
1449     }
1450 
1451     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1452                             KVM_DEV_RISCV_AIA_CONFIG_SRCS,
1453                             &aia_irq_num, true, NULL);
1454     if (ret < 0) {
1455         error_report("KVM AIA: failed to set number of input irq lines");
1456         exit(1);
1457     }
1458 
1459     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1460                             KVM_DEV_RISCV_AIA_CONFIG_IDS,
1461                             &aia_msi_num, true, NULL);
1462     if (ret < 0) {
1463         error_report("KVM AIA: failed to set number of msi");
1464         exit(1);
1465     }
1466 
1467 
1468     if (socket_count > 1) {
1469         socket_bits = find_last_bit(&socket_count, BITS_PER_LONG) + 1;
1470         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1471                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_BITS,
1472                                 &socket_bits, true, NULL);
1473         if (ret < 0) {
1474             error_report("KVM AIA: failed to set group_bits");
1475             exit(1);
1476         }
1477 
1478         ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1479                                 KVM_DEV_RISCV_AIA_CONFIG_GROUP_SHIFT,
1480                                 &group_shift, true, NULL);
1481         if (ret < 0) {
1482             error_report("KVM AIA: failed to set group_shift");
1483             exit(1);
1484         }
1485     }
1486 
1487     guest_bits = guest_num == 0 ? 0 :
1488                  find_last_bit(&guest_num, BITS_PER_LONG) + 1;
1489     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1490                             KVM_DEV_RISCV_AIA_CONFIG_GUEST_BITS,
1491                             &guest_bits, true, NULL);
1492     if (ret < 0) {
1493         error_report("KVM AIA: failed to set guest_bits");
1494         exit(1);
1495     }
1496 
1497     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1498                             KVM_DEV_RISCV_AIA_ADDR_APLIC,
1499                             &aplic_base, true, NULL);
1500     if (ret < 0) {
1501         error_report("KVM AIA: failed to set the base address of APLIC");
1502         exit(1);
1503     }
1504 
1505     for (socket = 0; socket < socket_count; socket++) {
1506         socket_imsic_base = imsic_base + socket * (1U << group_shift);
1507         hart_count = riscv_socket_hart_count(machine, socket);
1508         base_hart = riscv_socket_first_hartid(machine, socket);
1509 
1510         if (max_hart_per_socket < hart_count) {
1511             max_hart_per_socket = hart_count;
1512         }
1513 
1514         for (i = 0; i < hart_count; i++) {
1515             imsic_addr = socket_imsic_base + i * IMSIC_HART_SIZE(guest_bits);
1516             ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_ADDR,
1517                                     KVM_DEV_RISCV_AIA_ADDR_IMSIC(i + base_hart),
1518                                     &imsic_addr, true, NULL);
1519             if (ret < 0) {
1520                 error_report("KVM AIA: failed to set the IMSIC address for hart %d", i);
1521                 exit(1);
1522             }
1523         }
1524     }
1525 
1526     hart_bits = find_last_bit(&max_hart_per_socket, BITS_PER_LONG) + 1;
1527     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CONFIG,
1528                             KVM_DEV_RISCV_AIA_CONFIG_HART_BITS,
1529                             &hart_bits, true, NULL);
1530     if (ret < 0) {
1531         error_report("KVM AIA: failed to set hart_bits");
1532         exit(1);
1533     }
1534 
1535     if (kvm_has_gsi_routing()) {
1536         for (uint64_t idx = 0; idx < aia_irq_num + 1; ++idx) {
1537             /* KVM AIA only has one APLIC instance */
1538             kvm_irqchip_add_irq_route(kvm_state, idx, 0, idx);
1539         }
1540         kvm_gsi_routing_allowed = true;
1541         kvm_irqchip_commit_routes(kvm_state);
1542     }
1543 
1544     ret = kvm_device_access(aia_fd, KVM_DEV_RISCV_AIA_GRP_CTRL,
1545                             KVM_DEV_RISCV_AIA_CTRL_INIT,
1546                             NULL, true, NULL);
1547     if (ret < 0) {
1548         error_report("KVM AIA: initialized fail");
1549         exit(1);
1550     }
1551 
1552     kvm_msi_via_irqfd_allowed = true;
1553 }
1554 
1555 static void kvm_cpu_instance_init(CPUState *cs)
1556 {
1557     Object *obj = OBJECT(RISCV_CPU(cs));
1558     DeviceState *dev = DEVICE(obj);
1559 
1560     riscv_init_kvm_registers(obj);
1561 
1562     kvm_riscv_add_cpu_user_properties(obj);
1563 
1564     for (Property *prop = riscv_cpu_options; prop && prop->name; prop++) {
1565         /* Check if we have a specific KVM handler for the option */
1566         if (object_property_find(obj, prop->name)) {
1567             continue;
1568         }
1569         qdev_property_add_static(dev, prop);
1570     }
1571 }
1572 
1573 /*
1574  * We'll get here via the following path:
1575  *
1576  * riscv_cpu_realize()
1577  *   -> cpu_exec_realizefn()
1578  *      -> kvm_cpu_realize() (via accel_cpu_common_realize())
1579  */
1580 static bool kvm_cpu_realize(CPUState *cs, Error **errp)
1581 {
1582     RISCVCPU *cpu = RISCV_CPU(cs);
1583     int ret;
1584 
1585     if (riscv_has_ext(&cpu->env, RVV)) {
1586         ret = prctl(PR_RISCV_V_SET_CONTROL, PR_RISCV_V_VSTATE_CTRL_ON);
1587         if (ret) {
1588             error_setg(errp, "Error in prctl PR_RISCV_V_SET_CONTROL, code: %s",
1589                        strerrorname_np(errno));
1590             return false;
1591         }
1592     }
1593 
1594    return true;
1595 }
1596 
1597 void riscv_kvm_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
1598 {
1599     CPURISCVState *env = &cpu->env;
1600     KVMScratchCPU kvmcpu;
1601     struct kvm_one_reg reg;
1602     uint64_t val;
1603     int ret;
1604 
1605     /* short-circuit without spinning the scratch CPU */
1606     if (!cpu->cfg.ext_zicbom && !cpu->cfg.ext_zicboz) {
1607         return;
1608     }
1609 
1610     if (!kvm_riscv_create_scratch_vcpu(&kvmcpu)) {
1611         error_setg(errp, "Unable to create scratch KVM cpu");
1612         return;
1613     }
1614 
1615     if (cpu->cfg.ext_zicbom &&
1616         riscv_cpu_option_set(kvm_cbom_blocksize.name)) {
1617 
1618         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1619                                         kvm_cbom_blocksize.kvm_reg_id);
1620         reg.addr = (uint64_t)&val;
1621         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1622         if (ret != 0) {
1623             error_setg(errp, "Unable to read cbom_blocksize, error %d", errno);
1624             return;
1625         }
1626 
1627         if (cpu->cfg.cbom_blocksize != val) {
1628             error_setg(errp, "Unable to set cbom_blocksize to a different "
1629                        "value than the host (%lu)", val);
1630             return;
1631         }
1632     }
1633 
1634     if (cpu->cfg.ext_zicboz &&
1635         riscv_cpu_option_set(kvm_cboz_blocksize.name)) {
1636 
1637         reg.id = kvm_riscv_reg_id_ulong(env, KVM_REG_RISCV_CONFIG,
1638                                         kvm_cboz_blocksize.kvm_reg_id);
1639         reg.addr = (uint64_t)&val;
1640         ret = ioctl(kvmcpu.cpufd, KVM_GET_ONE_REG, &reg);
1641         if (ret != 0) {
1642             error_setg(errp, "Unable to read cboz_blocksize, error %d", errno);
1643             return;
1644         }
1645 
1646         if (cpu->cfg.cboz_blocksize != val) {
1647             error_setg(errp, "Unable to set cboz_blocksize to a different "
1648                        "value than the host (%lu)", val);
1649             return;
1650         }
1651     }
1652 
1653     kvm_riscv_destroy_scratch_vcpu(&kvmcpu);
1654 }
1655 
1656 static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
1657 {
1658     AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
1659 
1660     acc->cpu_instance_init = kvm_cpu_instance_init;
1661     acc->cpu_target_realize = kvm_cpu_realize;
1662 }
1663 
1664 static const TypeInfo kvm_cpu_accel_type_info = {
1665     .name = ACCEL_CPU_NAME("kvm"),
1666 
1667     .parent = TYPE_ACCEL_CPU,
1668     .class_init = kvm_cpu_accel_class_init,
1669     .abstract = true,
1670 };
1671 static void kvm_cpu_accel_register_types(void)
1672 {
1673     type_register_static(&kvm_cpu_accel_type_info);
1674 }
1675 type_init(kvm_cpu_accel_register_types);
1676 
1677 static void riscv_host_cpu_init(Object *obj)
1678 {
1679     CPURISCVState *env = &RISCV_CPU(obj)->env;
1680 
1681 #if defined(TARGET_RISCV32)
1682     env->misa_mxl_max = env->misa_mxl = MXL_RV32;
1683 #elif defined(TARGET_RISCV64)
1684     env->misa_mxl_max = env->misa_mxl = MXL_RV64;
1685 #endif
1686 }
1687 
1688 static const TypeInfo riscv_kvm_cpu_type_infos[] = {
1689     {
1690         .name = TYPE_RISCV_CPU_HOST,
1691         .parent = TYPE_RISCV_CPU,
1692         .instance_init = riscv_host_cpu_init,
1693     }
1694 };
1695 
1696 DEFINE_TYPES(riscv_kvm_cpu_type_infos)
1697