xref: /openbmc/qemu/target/riscv/cpu_helper.c (revision 43577499)
1 /*
2  * RISC-V CPU helpers for qemu.
3  *
4  * Copyright (c) 2016-2017 Sagar Karandikar, sagark@eecs.berkeley.edu
5  * Copyright (c) 2017-2018 SiFive, Inc.
6  *
7  * This program is free software; you can redistribute it and/or modify it
8  * under the terms and conditions of the GNU General Public License,
9  * version 2 or later, as published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope it will be useful, but WITHOUT
12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
14  * more details.
15  *
16  * You should have received a copy of the GNU General Public License along with
17  * this program.  If not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qemu/log.h"
22 #include "qemu/main-loop.h"
23 #include "cpu.h"
24 #include "exec/exec-all.h"
25 #include "tcg/tcg-op.h"
26 #include "trace.h"
27 #include "semihosting/common-semi.h"
28 
29 int riscv_cpu_mmu_index(CPURISCVState *env, bool ifetch)
30 {
31 #ifdef CONFIG_USER_ONLY
32     return 0;
33 #else
34     return env->priv;
35 #endif
36 }
37 
38 void cpu_get_tb_cpu_state(CPURISCVState *env, target_ulong *pc,
39                           target_ulong *cs_base, uint32_t *pflags)
40 {
41     CPUState *cs = env_cpu(env);
42     RISCVCPU *cpu = RISCV_CPU(cs);
43 
44     uint32_t flags = 0;
45 
46     *pc = env->xl == MXL_RV32 ? env->pc & UINT32_MAX : env->pc;
47     *cs_base = 0;
48 
49     if (riscv_has_ext(env, RVV) || cpu->cfg.ext_zve32f || cpu->cfg.ext_zve64f) {
50         /*
51          * If env->vl equals to VLMAX, we can use generic vector operation
52          * expanders (GVEC) to accerlate the vector operations.
53          * However, as LMUL could be a fractional number. The maximum
54          * vector size can be operated might be less than 8 bytes,
55          * which is not supported by GVEC. So we set vl_eq_vlmax flag to true
56          * only when maxsz >= 8 bytes.
57          */
58         uint32_t vlmax = vext_get_vlmax(env_archcpu(env), env->vtype);
59         uint32_t sew = FIELD_EX64(env->vtype, VTYPE, VSEW);
60         uint32_t maxsz = vlmax << sew;
61         bool vl_eq_vlmax = (env->vstart == 0) && (vlmax == env->vl) &&
62                            (maxsz >= 8);
63         flags = FIELD_DP32(flags, TB_FLAGS, VILL, env->vill);
64         flags = FIELD_DP32(flags, TB_FLAGS, SEW, sew);
65         flags = FIELD_DP32(flags, TB_FLAGS, LMUL,
66                     FIELD_EX64(env->vtype, VTYPE, VLMUL));
67         flags = FIELD_DP32(flags, TB_FLAGS, VL_EQ_VLMAX, vl_eq_vlmax);
68         flags = FIELD_DP32(flags, TB_FLAGS, VTA,
69                     FIELD_EX64(env->vtype, VTYPE, VTA));
70     } else {
71         flags = FIELD_DP32(flags, TB_FLAGS, VILL, 1);
72     }
73 
74 #ifdef CONFIG_USER_ONLY
75     flags |= TB_FLAGS_MSTATUS_FS;
76     flags |= TB_FLAGS_MSTATUS_VS;
77 #else
78     flags |= cpu_mmu_index(env, 0);
79     if (riscv_cpu_fp_enabled(env)) {
80         flags |= env->mstatus & MSTATUS_FS;
81     }
82 
83     if (riscv_cpu_vector_enabled(env)) {
84         flags |= env->mstatus & MSTATUS_VS;
85     }
86 
87     if (riscv_has_ext(env, RVH)) {
88         if (env->priv == PRV_M ||
89             (env->priv == PRV_S && !riscv_cpu_virt_enabled(env)) ||
90             (env->priv == PRV_U && !riscv_cpu_virt_enabled(env) &&
91                 get_field(env->hstatus, HSTATUS_HU))) {
92             flags = FIELD_DP32(flags, TB_FLAGS, HLSX, 1);
93         }
94 
95         flags = FIELD_DP32(flags, TB_FLAGS, MSTATUS_HS_FS,
96                            get_field(env->mstatus_hs, MSTATUS_FS));
97 
98         flags = FIELD_DP32(flags, TB_FLAGS, MSTATUS_HS_VS,
99                            get_field(env->mstatus_hs, MSTATUS_VS));
100     }
101 #endif
102 
103     flags = FIELD_DP32(flags, TB_FLAGS, XL, env->xl);
104     if (env->cur_pmmask < (env->xl == MXL_RV32 ? UINT32_MAX : UINT64_MAX)) {
105         flags = FIELD_DP32(flags, TB_FLAGS, PM_MASK_ENABLED, 1);
106     }
107     if (env->cur_pmbase != 0) {
108         flags = FIELD_DP32(flags, TB_FLAGS, PM_BASE_ENABLED, 1);
109     }
110 
111     *pflags = flags;
112 }
113 
114 void riscv_cpu_update_mask(CPURISCVState *env)
115 {
116     target_ulong mask = -1, base = 0;
117     /*
118      * TODO: Current RVJ spec does not specify
119      * how the extension interacts with XLEN.
120      */
121 #ifndef CONFIG_USER_ONLY
122     if (riscv_has_ext(env, RVJ)) {
123         switch (env->priv) {
124         case PRV_M:
125             if (env->mmte & M_PM_ENABLE) {
126                 mask = env->mpmmask;
127                 base = env->mpmbase;
128             }
129             break;
130         case PRV_S:
131             if (env->mmte & S_PM_ENABLE) {
132                 mask = env->spmmask;
133                 base = env->spmbase;
134             }
135             break;
136         case PRV_U:
137             if (env->mmte & U_PM_ENABLE) {
138                 mask = env->upmmask;
139                 base = env->upmbase;
140             }
141             break;
142         default:
143             g_assert_not_reached();
144         }
145     }
146 #endif
147     if (env->xl == MXL_RV32) {
148         env->cur_pmmask = mask & UINT32_MAX;
149         env->cur_pmbase = base & UINT32_MAX;
150     } else {
151         env->cur_pmmask = mask;
152         env->cur_pmbase = base;
153     }
154 }
155 
156 #ifndef CONFIG_USER_ONLY
157 
158 /*
159  * The HS-mode is allowed to configure priority only for the
160  * following VS-mode local interrupts:
161  *
162  * 0  (Reserved interrupt, reads as zero)
163  * 1  Supervisor software interrupt
164  * 4  (Reserved interrupt, reads as zero)
165  * 5  Supervisor timer interrupt
166  * 8  (Reserved interrupt, reads as zero)
167  * 13 (Reserved interrupt)
168  * 14 "
169  * 15 "
170  * 16 "
171  * 17 "
172  * 18 "
173  * 19 "
174  * 20 "
175  * 21 "
176  * 22 "
177  * 23 "
178  */
179 
180 static const int hviprio_index2irq[] = {
181     0, 1, 4, 5, 8, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 };
182 static const int hviprio_index2rdzero[] = {
183     1, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
184 
185 int riscv_cpu_hviprio_index2irq(int index, int *out_irq, int *out_rdzero)
186 {
187     if (index < 0 || ARRAY_SIZE(hviprio_index2irq) <= index) {
188         return -EINVAL;
189     }
190 
191     if (out_irq) {
192         *out_irq = hviprio_index2irq[index];
193     }
194 
195     if (out_rdzero) {
196         *out_rdzero = hviprio_index2rdzero[index];
197     }
198 
199     return 0;
200 }
201 
202 /*
203  * Default priorities of local interrupts are defined in the
204  * RISC-V Advanced Interrupt Architecture specification.
205  *
206  * ----------------------------------------------------------------
207  *  Default  |
208  *  Priority | Major Interrupt Numbers
209  * ----------------------------------------------------------------
210  *  Highest  | 47, 23, 46, 45, 22, 44,
211  *           | 43, 21, 42, 41, 20, 40
212  *           |
213  *           | 11 (0b),  3 (03),  7 (07)
214  *           |  9 (09),  1 (01),  5 (05)
215  *           | 12 (0c)
216  *           | 10 (0a),  2 (02),  6 (06)
217  *           |
218  *           | 39, 19, 38, 37, 18, 36,
219  *  Lowest   | 35, 17, 34, 33, 16, 32
220  * ----------------------------------------------------------------
221  */
222 static const uint8_t default_iprio[64] = {
223  /* Custom interrupts 48 to 63 */
224  [63] = IPRIO_MMAXIPRIO,
225  [62] = IPRIO_MMAXIPRIO,
226  [61] = IPRIO_MMAXIPRIO,
227  [60] = IPRIO_MMAXIPRIO,
228  [59] = IPRIO_MMAXIPRIO,
229  [58] = IPRIO_MMAXIPRIO,
230  [57] = IPRIO_MMAXIPRIO,
231  [56] = IPRIO_MMAXIPRIO,
232  [55] = IPRIO_MMAXIPRIO,
233  [54] = IPRIO_MMAXIPRIO,
234  [53] = IPRIO_MMAXIPRIO,
235  [52] = IPRIO_MMAXIPRIO,
236  [51] = IPRIO_MMAXIPRIO,
237  [50] = IPRIO_MMAXIPRIO,
238  [49] = IPRIO_MMAXIPRIO,
239  [48] = IPRIO_MMAXIPRIO,
240 
241  /* Custom interrupts 24 to 31 */
242  [31] = IPRIO_MMAXIPRIO,
243  [30] = IPRIO_MMAXIPRIO,
244  [29] = IPRIO_MMAXIPRIO,
245  [28] = IPRIO_MMAXIPRIO,
246  [27] = IPRIO_MMAXIPRIO,
247  [26] = IPRIO_MMAXIPRIO,
248  [25] = IPRIO_MMAXIPRIO,
249  [24] = IPRIO_MMAXIPRIO,
250 
251  [47] = IPRIO_DEFAULT_UPPER,
252  [23] = IPRIO_DEFAULT_UPPER + 1,
253  [46] = IPRIO_DEFAULT_UPPER + 2,
254  [45] = IPRIO_DEFAULT_UPPER + 3,
255  [22] = IPRIO_DEFAULT_UPPER + 4,
256  [44] = IPRIO_DEFAULT_UPPER + 5,
257 
258  [43] = IPRIO_DEFAULT_UPPER + 6,
259  [21] = IPRIO_DEFAULT_UPPER + 7,
260  [42] = IPRIO_DEFAULT_UPPER + 8,
261  [41] = IPRIO_DEFAULT_UPPER + 9,
262  [20] = IPRIO_DEFAULT_UPPER + 10,
263  [40] = IPRIO_DEFAULT_UPPER + 11,
264 
265  [11] = IPRIO_DEFAULT_M,
266  [3]  = IPRIO_DEFAULT_M + 1,
267  [7]  = IPRIO_DEFAULT_M + 2,
268 
269  [9]  = IPRIO_DEFAULT_S,
270  [1]  = IPRIO_DEFAULT_S + 1,
271  [5]  = IPRIO_DEFAULT_S + 2,
272 
273  [12] = IPRIO_DEFAULT_SGEXT,
274 
275  [10] = IPRIO_DEFAULT_VS,
276  [2]  = IPRIO_DEFAULT_VS + 1,
277  [6]  = IPRIO_DEFAULT_VS + 2,
278 
279  [39] = IPRIO_DEFAULT_LOWER,
280  [19] = IPRIO_DEFAULT_LOWER + 1,
281  [38] = IPRIO_DEFAULT_LOWER + 2,
282  [37] = IPRIO_DEFAULT_LOWER + 3,
283  [18] = IPRIO_DEFAULT_LOWER + 4,
284  [36] = IPRIO_DEFAULT_LOWER + 5,
285 
286  [35] = IPRIO_DEFAULT_LOWER + 6,
287  [17] = IPRIO_DEFAULT_LOWER + 7,
288  [34] = IPRIO_DEFAULT_LOWER + 8,
289  [33] = IPRIO_DEFAULT_LOWER + 9,
290  [16] = IPRIO_DEFAULT_LOWER + 10,
291  [32] = IPRIO_DEFAULT_LOWER + 11,
292 };
293 
294 uint8_t riscv_cpu_default_priority(int irq)
295 {
296     if (irq < 0 || irq > 63) {
297         return IPRIO_MMAXIPRIO;
298     }
299 
300     return default_iprio[irq] ? default_iprio[irq] : IPRIO_MMAXIPRIO;
301 };
302 
303 static int riscv_cpu_pending_to_irq(CPURISCVState *env,
304                                     int extirq, unsigned int extirq_def_prio,
305                                     uint64_t pending, uint8_t *iprio)
306 {
307     int irq, best_irq = RISCV_EXCP_NONE;
308     unsigned int prio, best_prio = UINT_MAX;
309 
310     if (!pending) {
311         return RISCV_EXCP_NONE;
312     }
313 
314     irq = ctz64(pending);
315     if (!riscv_feature(env, RISCV_FEATURE_AIA)) {
316         return irq;
317     }
318 
319     pending = pending >> irq;
320     while (pending) {
321         prio = iprio[irq];
322         if (!prio) {
323             if (irq == extirq) {
324                 prio = extirq_def_prio;
325             } else {
326                 prio = (riscv_cpu_default_priority(irq) < extirq_def_prio) ?
327                        1 : IPRIO_MMAXIPRIO;
328             }
329         }
330         if ((pending & 0x1) && (prio <= best_prio)) {
331             best_irq = irq;
332             best_prio = prio;
333         }
334         irq++;
335         pending = pending >> 1;
336     }
337 
338     return best_irq;
339 }
340 
341 uint64_t riscv_cpu_all_pending(CPURISCVState *env)
342 {
343     uint32_t gein = get_field(env->hstatus, HSTATUS_VGEIN);
344     uint64_t vsgein = (env->hgeip & (1ULL << gein)) ? MIP_VSEIP : 0;
345 
346     return (env->mip | vsgein) & env->mie;
347 }
348 
349 int riscv_cpu_mirq_pending(CPURISCVState *env)
350 {
351     uint64_t irqs = riscv_cpu_all_pending(env) & ~env->mideleg &
352                     ~(MIP_SGEIP | MIP_VSSIP | MIP_VSTIP | MIP_VSEIP);
353 
354     return riscv_cpu_pending_to_irq(env, IRQ_M_EXT, IPRIO_DEFAULT_M,
355                                     irqs, env->miprio);
356 }
357 
358 int riscv_cpu_sirq_pending(CPURISCVState *env)
359 {
360     uint64_t irqs = riscv_cpu_all_pending(env) & env->mideleg &
361                     ~(MIP_VSSIP | MIP_VSTIP | MIP_VSEIP);
362 
363     return riscv_cpu_pending_to_irq(env, IRQ_S_EXT, IPRIO_DEFAULT_S,
364                                     irqs, env->siprio);
365 }
366 
367 int riscv_cpu_vsirq_pending(CPURISCVState *env)
368 {
369     uint64_t irqs = riscv_cpu_all_pending(env) & env->mideleg &
370                     (MIP_VSSIP | MIP_VSTIP | MIP_VSEIP);
371 
372     return riscv_cpu_pending_to_irq(env, IRQ_S_EXT, IPRIO_DEFAULT_S,
373                                     irqs >> 1, env->hviprio);
374 }
375 
376 static int riscv_cpu_local_irq_pending(CPURISCVState *env)
377 {
378     int virq;
379     uint64_t irqs, pending, mie, hsie, vsie;
380 
381     /* Determine interrupt enable state of all privilege modes */
382     if (riscv_cpu_virt_enabled(env)) {
383         mie = 1;
384         hsie = 1;
385         vsie = (env->priv < PRV_S) ||
386                (env->priv == PRV_S && get_field(env->mstatus, MSTATUS_SIE));
387     } else {
388         mie = (env->priv < PRV_M) ||
389               (env->priv == PRV_M && get_field(env->mstatus, MSTATUS_MIE));
390         hsie = (env->priv < PRV_S) ||
391                (env->priv == PRV_S && get_field(env->mstatus, MSTATUS_SIE));
392         vsie = 0;
393     }
394 
395     /* Determine all pending interrupts */
396     pending = riscv_cpu_all_pending(env);
397 
398     /* Check M-mode interrupts */
399     irqs = pending & ~env->mideleg & -mie;
400     if (irqs) {
401         return riscv_cpu_pending_to_irq(env, IRQ_M_EXT, IPRIO_DEFAULT_M,
402                                         irqs, env->miprio);
403     }
404 
405     /* Check HS-mode interrupts */
406     irqs = pending & env->mideleg & ~env->hideleg & -hsie;
407     if (irqs) {
408         return riscv_cpu_pending_to_irq(env, IRQ_S_EXT, IPRIO_DEFAULT_S,
409                                         irqs, env->siprio);
410     }
411 
412     /* Check VS-mode interrupts */
413     irqs = pending & env->mideleg & env->hideleg & -vsie;
414     if (irqs) {
415         virq = riscv_cpu_pending_to_irq(env, IRQ_S_EXT, IPRIO_DEFAULT_S,
416                                         irqs >> 1, env->hviprio);
417         return (virq <= 0) ? virq : virq + 1;
418     }
419 
420     /* Indicate no pending interrupt */
421     return RISCV_EXCP_NONE;
422 }
423 
424 bool riscv_cpu_exec_interrupt(CPUState *cs, int interrupt_request)
425 {
426     if (interrupt_request & CPU_INTERRUPT_HARD) {
427         RISCVCPU *cpu = RISCV_CPU(cs);
428         CPURISCVState *env = &cpu->env;
429         int interruptno = riscv_cpu_local_irq_pending(env);
430         if (interruptno >= 0) {
431             cs->exception_index = RISCV_EXCP_INT_FLAG | interruptno;
432             riscv_cpu_do_interrupt(cs);
433             return true;
434         }
435     }
436     return false;
437 }
438 
439 /* Return true is floating point support is currently enabled */
440 bool riscv_cpu_fp_enabled(CPURISCVState *env)
441 {
442     if (env->mstatus & MSTATUS_FS) {
443         if (riscv_cpu_virt_enabled(env) && !(env->mstatus_hs & MSTATUS_FS)) {
444             return false;
445         }
446         return true;
447     }
448 
449     return false;
450 }
451 
452 /* Return true is vector support is currently enabled */
453 bool riscv_cpu_vector_enabled(CPURISCVState *env)
454 {
455     if (env->mstatus & MSTATUS_VS) {
456         if (riscv_cpu_virt_enabled(env) && !(env->mstatus_hs & MSTATUS_VS)) {
457             return false;
458         }
459         return true;
460     }
461 
462     return false;
463 }
464 
465 void riscv_cpu_swap_hypervisor_regs(CPURISCVState *env)
466 {
467     uint64_t mstatus_mask = MSTATUS_MXR | MSTATUS_SUM |
468                             MSTATUS_SPP | MSTATUS_SPIE | MSTATUS_SIE |
469                             MSTATUS64_UXL | MSTATUS_VS;
470 
471     if (riscv_has_ext(env, RVF)) {
472         mstatus_mask |= MSTATUS_FS;
473     }
474     bool current_virt = riscv_cpu_virt_enabled(env);
475 
476     g_assert(riscv_has_ext(env, RVH));
477 
478     if (current_virt) {
479         /* Current V=1 and we are about to change to V=0 */
480         env->vsstatus = env->mstatus & mstatus_mask;
481         env->mstatus &= ~mstatus_mask;
482         env->mstatus |= env->mstatus_hs;
483 
484         env->vstvec = env->stvec;
485         env->stvec = env->stvec_hs;
486 
487         env->vsscratch = env->sscratch;
488         env->sscratch = env->sscratch_hs;
489 
490         env->vsepc = env->sepc;
491         env->sepc = env->sepc_hs;
492 
493         env->vscause = env->scause;
494         env->scause = env->scause_hs;
495 
496         env->vstval = env->stval;
497         env->stval = env->stval_hs;
498 
499         env->vsatp = env->satp;
500         env->satp = env->satp_hs;
501     } else {
502         /* Current V=0 and we are about to change to V=1 */
503         env->mstatus_hs = env->mstatus & mstatus_mask;
504         env->mstatus &= ~mstatus_mask;
505         env->mstatus |= env->vsstatus;
506 
507         env->stvec_hs = env->stvec;
508         env->stvec = env->vstvec;
509 
510         env->sscratch_hs = env->sscratch;
511         env->sscratch = env->vsscratch;
512 
513         env->sepc_hs = env->sepc;
514         env->sepc = env->vsepc;
515 
516         env->scause_hs = env->scause;
517         env->scause = env->vscause;
518 
519         env->stval_hs = env->stval;
520         env->stval = env->vstval;
521 
522         env->satp_hs = env->satp;
523         env->satp = env->vsatp;
524     }
525 }
526 
527 target_ulong riscv_cpu_get_geilen(CPURISCVState *env)
528 {
529     if (!riscv_has_ext(env, RVH)) {
530         return 0;
531     }
532 
533     return env->geilen;
534 }
535 
536 void riscv_cpu_set_geilen(CPURISCVState *env, target_ulong geilen)
537 {
538     if (!riscv_has_ext(env, RVH)) {
539         return;
540     }
541 
542     if (geilen > (TARGET_LONG_BITS - 1)) {
543         return;
544     }
545 
546     env->geilen = geilen;
547 }
548 
549 bool riscv_cpu_virt_enabled(CPURISCVState *env)
550 {
551     if (!riscv_has_ext(env, RVH)) {
552         return false;
553     }
554 
555     return get_field(env->virt, VIRT_ONOFF);
556 }
557 
558 void riscv_cpu_set_virt_enabled(CPURISCVState *env, bool enable)
559 {
560     if (!riscv_has_ext(env, RVH)) {
561         return;
562     }
563 
564     /* Flush the TLB on all virt mode changes. */
565     if (get_field(env->virt, VIRT_ONOFF) != enable) {
566         tlb_flush(env_cpu(env));
567     }
568 
569     env->virt = set_field(env->virt, VIRT_ONOFF, enable);
570 
571     if (enable) {
572         /*
573          * The guest external interrupts from an interrupt controller are
574          * delivered only when the Guest/VM is running (i.e. V=1). This means
575          * any guest external interrupt which is triggered while the Guest/VM
576          * is not running (i.e. V=0) will be missed on QEMU resulting in guest
577          * with sluggish response to serial console input and other I/O events.
578          *
579          * To solve this, we check and inject interrupt after setting V=1.
580          */
581         riscv_cpu_update_mip(env_archcpu(env), 0, 0);
582     }
583 }
584 
585 bool riscv_cpu_two_stage_lookup(int mmu_idx)
586 {
587     return mmu_idx & TB_FLAGS_PRIV_HYP_ACCESS_MASK;
588 }
589 
590 int riscv_cpu_claim_interrupts(RISCVCPU *cpu, uint64_t interrupts)
591 {
592     CPURISCVState *env = &cpu->env;
593     if (env->miclaim & interrupts) {
594         return -1;
595     } else {
596         env->miclaim |= interrupts;
597         return 0;
598     }
599 }
600 
601 uint64_t riscv_cpu_update_mip(RISCVCPU *cpu, uint64_t mask, uint64_t value)
602 {
603     CPURISCVState *env = &cpu->env;
604     CPUState *cs = CPU(cpu);
605     uint64_t gein, vsgein = 0, old = env->mip;
606     bool locked = false;
607 
608     if (riscv_cpu_virt_enabled(env)) {
609         gein = get_field(env->hstatus, HSTATUS_VGEIN);
610         vsgein = (env->hgeip & (1ULL << gein)) ? MIP_VSEIP : 0;
611     }
612 
613     if (!qemu_mutex_iothread_locked()) {
614         locked = true;
615         qemu_mutex_lock_iothread();
616     }
617 
618     env->mip = (env->mip & ~mask) | (value & mask);
619 
620     if (env->mip | vsgein) {
621         cpu_interrupt(cs, CPU_INTERRUPT_HARD);
622     } else {
623         cpu_reset_interrupt(cs, CPU_INTERRUPT_HARD);
624     }
625 
626     if (locked) {
627         qemu_mutex_unlock_iothread();
628     }
629 
630     return old;
631 }
632 
633 void riscv_cpu_set_rdtime_fn(CPURISCVState *env, uint64_t (*fn)(void *),
634                              void *arg)
635 {
636     env->rdtime_fn = fn;
637     env->rdtime_fn_arg = arg;
638 }
639 
640 void riscv_cpu_set_aia_ireg_rmw_fn(CPURISCVState *env, uint32_t priv,
641                                    int (*rmw_fn)(void *arg,
642                                                  target_ulong reg,
643                                                  target_ulong *val,
644                                                  target_ulong new_val,
645                                                  target_ulong write_mask),
646                                    void *rmw_fn_arg)
647 {
648     if (priv <= PRV_M) {
649         env->aia_ireg_rmw_fn[priv] = rmw_fn;
650         env->aia_ireg_rmw_fn_arg[priv] = rmw_fn_arg;
651     }
652 }
653 
654 void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv)
655 {
656     if (newpriv > PRV_M) {
657         g_assert_not_reached();
658     }
659     if (newpriv == PRV_H) {
660         newpriv = PRV_U;
661     }
662     /* tlb_flush is unnecessary as mode is contained in mmu_idx */
663     env->priv = newpriv;
664     env->xl = cpu_recompute_xl(env);
665     riscv_cpu_update_mask(env);
666 
667     /*
668      * Clear the load reservation - otherwise a reservation placed in one
669      * context/process can be used by another, resulting in an SC succeeding
670      * incorrectly. Version 2.2 of the ISA specification explicitly requires
671      * this behaviour, while later revisions say that the kernel "should" use
672      * an SC instruction to force the yielding of a load reservation on a
673      * preemptive context switch. As a result, do both.
674      */
675     env->load_res = -1;
676 }
677 
678 /*
679  * get_physical_address_pmp - check PMP permission for this physical address
680  *
681  * Match the PMP region and check permission for this physical address and it's
682  * TLB page. Returns 0 if the permission checking was successful
683  *
684  * @env: CPURISCVState
685  * @prot: The returned protection attributes
686  * @tlb_size: TLB page size containing addr. It could be modified after PMP
687  *            permission checking. NULL if not set TLB page for addr.
688  * @addr: The physical address to be checked permission
689  * @access_type: The type of MMU access
690  * @mode: Indicates current privilege level.
691  */
692 static int get_physical_address_pmp(CPURISCVState *env, int *prot,
693                                     target_ulong *tlb_size, hwaddr addr,
694                                     int size, MMUAccessType access_type,
695                                     int mode)
696 {
697     pmp_priv_t pmp_priv;
698     target_ulong tlb_size_pmp = 0;
699 
700     if (!riscv_feature(env, RISCV_FEATURE_PMP)) {
701         *prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
702         return TRANSLATE_SUCCESS;
703     }
704 
705     if (!pmp_hart_has_privs(env, addr, size, 1 << access_type, &pmp_priv,
706                             mode)) {
707         *prot = 0;
708         return TRANSLATE_PMP_FAIL;
709     }
710 
711     *prot = pmp_priv_to_page_prot(pmp_priv);
712     if (tlb_size != NULL) {
713         if (pmp_is_range_in_tlb(env, addr & ~(*tlb_size - 1), &tlb_size_pmp)) {
714             *tlb_size = tlb_size_pmp;
715         }
716     }
717 
718     return TRANSLATE_SUCCESS;
719 }
720 
721 /* get_physical_address - get the physical address for this virtual address
722  *
723  * Do a page table walk to obtain the physical address corresponding to a
724  * virtual address. Returns 0 if the translation was successful
725  *
726  * Adapted from Spike's mmu_t::translate and mmu_t::walk
727  *
728  * @env: CPURISCVState
729  * @physical: This will be set to the calculated physical address
730  * @prot: The returned protection attributes
731  * @addr: The virtual address to be translated
732  * @fault_pte_addr: If not NULL, this will be set to fault pte address
733  *                  when a error occurs on pte address translation.
734  *                  This will already be shifted to match htval.
735  * @access_type: The type of MMU access
736  * @mmu_idx: Indicates current privilege level
737  * @first_stage: Are we in first stage translation?
738  *               Second stage is used for hypervisor guest translation
739  * @two_stage: Are we going to perform two stage translation
740  * @is_debug: Is this access from a debugger or the monitor?
741  */
742 static int get_physical_address(CPURISCVState *env, hwaddr *physical,
743                                 int *prot, target_ulong addr,
744                                 target_ulong *fault_pte_addr,
745                                 int access_type, int mmu_idx,
746                                 bool first_stage, bool two_stage,
747                                 bool is_debug)
748 {
749     /* NOTE: the env->pc value visible here will not be
750      * correct, but the value visible to the exception handler
751      * (riscv_cpu_do_interrupt) is correct */
752     MemTxResult res;
753     MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
754     int mode = mmu_idx & TB_FLAGS_PRIV_MMU_MASK;
755     bool use_background = false;
756     hwaddr ppn;
757     RISCVCPU *cpu = env_archcpu(env);
758     int napot_bits = 0;
759     target_ulong napot_mask;
760 
761     /*
762      * Check if we should use the background registers for the two
763      * stage translation. We don't need to check if we actually need
764      * two stage translation as that happened before this function
765      * was called. Background registers will be used if the guest has
766      * forced a two stage translation to be on (in HS or M mode).
767      */
768     if (!riscv_cpu_virt_enabled(env) && two_stage) {
769         use_background = true;
770     }
771 
772     /* MPRV does not affect the virtual-machine load/store
773        instructions, HLV, HLVX, and HSV. */
774     if (riscv_cpu_two_stage_lookup(mmu_idx)) {
775         mode = get_field(env->hstatus, HSTATUS_SPVP);
776     } else if (mode == PRV_M && access_type != MMU_INST_FETCH) {
777         if (get_field(env->mstatus, MSTATUS_MPRV)) {
778             mode = get_field(env->mstatus, MSTATUS_MPP);
779         }
780     }
781 
782     if (first_stage == false) {
783         /* We are in stage 2 translation, this is similar to stage 1. */
784         /* Stage 2 is always taken as U-mode */
785         mode = PRV_U;
786     }
787 
788     if (mode == PRV_M || !riscv_feature(env, RISCV_FEATURE_MMU)) {
789         *physical = addr;
790         *prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
791         return TRANSLATE_SUCCESS;
792     }
793 
794     *prot = 0;
795 
796     hwaddr base;
797     int levels, ptidxbits, ptesize, vm, sum, mxr, widened;
798 
799     if (first_stage == true) {
800         mxr = get_field(env->mstatus, MSTATUS_MXR);
801     } else {
802         mxr = get_field(env->vsstatus, MSTATUS_MXR);
803     }
804 
805     if (first_stage == true) {
806         if (use_background) {
807             if (riscv_cpu_mxl(env) == MXL_RV32) {
808                 base = (hwaddr)get_field(env->vsatp, SATP32_PPN) << PGSHIFT;
809                 vm = get_field(env->vsatp, SATP32_MODE);
810             } else {
811                 base = (hwaddr)get_field(env->vsatp, SATP64_PPN) << PGSHIFT;
812                 vm = get_field(env->vsatp, SATP64_MODE);
813             }
814         } else {
815             if (riscv_cpu_mxl(env) == MXL_RV32) {
816                 base = (hwaddr)get_field(env->satp, SATP32_PPN) << PGSHIFT;
817                 vm = get_field(env->satp, SATP32_MODE);
818             } else {
819                 base = (hwaddr)get_field(env->satp, SATP64_PPN) << PGSHIFT;
820                 vm = get_field(env->satp, SATP64_MODE);
821             }
822         }
823         widened = 0;
824     } else {
825         if (riscv_cpu_mxl(env) == MXL_RV32) {
826             base = (hwaddr)get_field(env->hgatp, SATP32_PPN) << PGSHIFT;
827             vm = get_field(env->hgatp, SATP32_MODE);
828         } else {
829             base = (hwaddr)get_field(env->hgatp, SATP64_PPN) << PGSHIFT;
830             vm = get_field(env->hgatp, SATP64_MODE);
831         }
832         widened = 2;
833     }
834     /* status.SUM will be ignored if execute on background */
835     sum = get_field(env->mstatus, MSTATUS_SUM) || use_background || is_debug;
836     switch (vm) {
837     case VM_1_10_SV32:
838       levels = 2; ptidxbits = 10; ptesize = 4; break;
839     case VM_1_10_SV39:
840       levels = 3; ptidxbits = 9; ptesize = 8; break;
841     case VM_1_10_SV48:
842       levels = 4; ptidxbits = 9; ptesize = 8; break;
843     case VM_1_10_SV57:
844       levels = 5; ptidxbits = 9; ptesize = 8; break;
845     case VM_1_10_MBARE:
846         *physical = addr;
847         *prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
848         return TRANSLATE_SUCCESS;
849     default:
850       g_assert_not_reached();
851     }
852 
853     CPUState *cs = env_cpu(env);
854     int va_bits = PGSHIFT + levels * ptidxbits + widened;
855     target_ulong mask, masked_msbs;
856 
857     if (TARGET_LONG_BITS > (va_bits - 1)) {
858         mask = (1L << (TARGET_LONG_BITS - (va_bits - 1))) - 1;
859     } else {
860         mask = 0;
861     }
862     masked_msbs = (addr >> (va_bits - 1)) & mask;
863 
864     if (masked_msbs != 0 && masked_msbs != mask) {
865         return TRANSLATE_FAIL;
866     }
867 
868     int ptshift = (levels - 1) * ptidxbits;
869     int i;
870 
871 #if !TCG_OVERSIZED_GUEST
872 restart:
873 #endif
874     for (i = 0; i < levels; i++, ptshift -= ptidxbits) {
875         target_ulong idx;
876         if (i == 0) {
877             idx = (addr >> (PGSHIFT + ptshift)) &
878                            ((1 << (ptidxbits + widened)) - 1);
879         } else {
880             idx = (addr >> (PGSHIFT + ptshift)) &
881                            ((1 << ptidxbits) - 1);
882         }
883 
884         /* check that physical address of PTE is legal */
885         hwaddr pte_addr;
886 
887         if (two_stage && first_stage) {
888             int vbase_prot;
889             hwaddr vbase;
890 
891             /* Do the second stage translation on the base PTE address. */
892             int vbase_ret = get_physical_address(env, &vbase, &vbase_prot,
893                                                  base, NULL, MMU_DATA_LOAD,
894                                                  mmu_idx, false, true,
895                                                  is_debug);
896 
897             if (vbase_ret != TRANSLATE_SUCCESS) {
898                 if (fault_pte_addr) {
899                     *fault_pte_addr = (base + idx * ptesize) >> 2;
900                 }
901                 return TRANSLATE_G_STAGE_FAIL;
902             }
903 
904             pte_addr = vbase + idx * ptesize;
905         } else {
906             pte_addr = base + idx * ptesize;
907         }
908 
909         int pmp_prot;
910         int pmp_ret = get_physical_address_pmp(env, &pmp_prot, NULL, pte_addr,
911                                                sizeof(target_ulong),
912                                                MMU_DATA_LOAD, PRV_S);
913         if (pmp_ret != TRANSLATE_SUCCESS) {
914             return TRANSLATE_PMP_FAIL;
915         }
916 
917         target_ulong pte;
918         if (riscv_cpu_mxl(env) == MXL_RV32) {
919             pte = address_space_ldl(cs->as, pte_addr, attrs, &res);
920         } else {
921             pte = address_space_ldq(cs->as, pte_addr, attrs, &res);
922         }
923 
924         if (res != MEMTX_OK) {
925             return TRANSLATE_FAIL;
926         }
927 
928         if (riscv_cpu_sxl(env) == MXL_RV32) {
929             ppn = pte >> PTE_PPN_SHIFT;
930         } else if (cpu->cfg.ext_svpbmt || cpu->cfg.ext_svnapot) {
931             ppn = (pte & (target_ulong)PTE_PPN_MASK) >> PTE_PPN_SHIFT;
932         } else {
933             ppn = pte >> PTE_PPN_SHIFT;
934             if ((pte & ~(target_ulong)PTE_PPN_MASK) >> PTE_PPN_SHIFT) {
935                 return TRANSLATE_FAIL;
936             }
937         }
938 
939         if (!(pte & PTE_V)) {
940             /* Invalid PTE */
941             return TRANSLATE_FAIL;
942         } else if (!cpu->cfg.ext_svpbmt && (pte & PTE_PBMT)) {
943             return TRANSLATE_FAIL;
944         } else if (!(pte & (PTE_R | PTE_W | PTE_X))) {
945             /* Inner PTE, continue walking */
946             if (pte & (PTE_D | PTE_A | PTE_U | PTE_ATTR)) {
947                 return TRANSLATE_FAIL;
948             }
949             base = ppn << PGSHIFT;
950         } else if ((pte & (PTE_R | PTE_W | PTE_X)) == PTE_W) {
951             /* Reserved leaf PTE flags: PTE_W */
952             return TRANSLATE_FAIL;
953         } else if ((pte & (PTE_R | PTE_W | PTE_X)) == (PTE_W | PTE_X)) {
954             /* Reserved leaf PTE flags: PTE_W + PTE_X */
955             return TRANSLATE_FAIL;
956         } else if ((pte & PTE_U) && ((mode != PRV_U) &&
957                    (!sum || access_type == MMU_INST_FETCH))) {
958             /* User PTE flags when not U mode and mstatus.SUM is not set,
959                or the access type is an instruction fetch */
960             return TRANSLATE_FAIL;
961         } else if (!(pte & PTE_U) && (mode != PRV_S)) {
962             /* Supervisor PTE flags when not S mode */
963             return TRANSLATE_FAIL;
964         } else if (ppn & ((1ULL << ptshift) - 1)) {
965             /* Misaligned PPN */
966             return TRANSLATE_FAIL;
967         } else if (access_type == MMU_DATA_LOAD && !((pte & PTE_R) ||
968                    ((pte & PTE_X) && mxr))) {
969             /* Read access check failed */
970             return TRANSLATE_FAIL;
971         } else if (access_type == MMU_DATA_STORE && !(pte & PTE_W)) {
972             /* Write access check failed */
973             return TRANSLATE_FAIL;
974         } else if (access_type == MMU_INST_FETCH && !(pte & PTE_X)) {
975             /* Fetch access check failed */
976             return TRANSLATE_FAIL;
977         } else {
978             /* if necessary, set accessed and dirty bits. */
979             target_ulong updated_pte = pte | PTE_A |
980                 (access_type == MMU_DATA_STORE ? PTE_D : 0);
981 
982             /* Page table updates need to be atomic with MTTCG enabled */
983             if (updated_pte != pte) {
984                 /*
985                  * - if accessed or dirty bits need updating, and the PTE is
986                  *   in RAM, then we do so atomically with a compare and swap.
987                  * - if the PTE is in IO space or ROM, then it can't be updated
988                  *   and we return TRANSLATE_FAIL.
989                  * - if the PTE changed by the time we went to update it, then
990                  *   it is no longer valid and we must re-walk the page table.
991                  */
992                 MemoryRegion *mr;
993                 hwaddr l = sizeof(target_ulong), addr1;
994                 mr = address_space_translate(cs->as, pte_addr,
995                     &addr1, &l, false, MEMTXATTRS_UNSPECIFIED);
996                 if (memory_region_is_ram(mr)) {
997                     target_ulong *pte_pa =
998                         qemu_map_ram_ptr(mr->ram_block, addr1);
999 #if TCG_OVERSIZED_GUEST
1000                     /* MTTCG is not enabled on oversized TCG guests so
1001                      * page table updates do not need to be atomic */
1002                     *pte_pa = pte = updated_pte;
1003 #else
1004                     target_ulong old_pte =
1005                         qatomic_cmpxchg(pte_pa, pte, updated_pte);
1006                     if (old_pte != pte) {
1007                         goto restart;
1008                     } else {
1009                         pte = updated_pte;
1010                     }
1011 #endif
1012                 } else {
1013                     /* misconfigured PTE in ROM (AD bits are not preset) or
1014                      * PTE is in IO space and can't be updated atomically */
1015                     return TRANSLATE_FAIL;
1016                 }
1017             }
1018 
1019             /* for superpage mappings, make a fake leaf PTE for the TLB's
1020                benefit. */
1021             target_ulong vpn = addr >> PGSHIFT;
1022 
1023             if (cpu->cfg.ext_svnapot && (pte & PTE_N)) {
1024                 napot_bits = ctzl(ppn) + 1;
1025                 if ((i != (levels - 1)) || (napot_bits != 4)) {
1026                     return TRANSLATE_FAIL;
1027                 }
1028             }
1029 
1030             napot_mask = (1 << napot_bits) - 1;
1031             *physical = (((ppn & ~napot_mask) | (vpn & napot_mask) |
1032                           (vpn & (((target_ulong)1 << ptshift) - 1))
1033                          ) << PGSHIFT) | (addr & ~TARGET_PAGE_MASK);
1034 
1035             /* set permissions on the TLB entry */
1036             if ((pte & PTE_R) || ((pte & PTE_X) && mxr)) {
1037                 *prot |= PAGE_READ;
1038             }
1039             if ((pte & PTE_X)) {
1040                 *prot |= PAGE_EXEC;
1041             }
1042             /* add write permission on stores or if the page is already dirty,
1043                so that we TLB miss on later writes to update the dirty bit */
1044             if ((pte & PTE_W) &&
1045                     (access_type == MMU_DATA_STORE || (pte & PTE_D))) {
1046                 *prot |= PAGE_WRITE;
1047             }
1048             return TRANSLATE_SUCCESS;
1049         }
1050     }
1051     return TRANSLATE_FAIL;
1052 }
1053 
1054 static void raise_mmu_exception(CPURISCVState *env, target_ulong address,
1055                                 MMUAccessType access_type, bool pmp_violation,
1056                                 bool first_stage, bool two_stage)
1057 {
1058     CPUState *cs = env_cpu(env);
1059     int page_fault_exceptions, vm;
1060     uint64_t stap_mode;
1061 
1062     if (riscv_cpu_mxl(env) == MXL_RV32) {
1063         stap_mode = SATP32_MODE;
1064     } else {
1065         stap_mode = SATP64_MODE;
1066     }
1067 
1068     if (first_stage) {
1069         vm = get_field(env->satp, stap_mode);
1070     } else {
1071         vm = get_field(env->hgatp, stap_mode);
1072     }
1073 
1074     page_fault_exceptions = vm != VM_1_10_MBARE && !pmp_violation;
1075 
1076     switch (access_type) {
1077     case MMU_INST_FETCH:
1078         if (riscv_cpu_virt_enabled(env) && !first_stage) {
1079             cs->exception_index = RISCV_EXCP_INST_GUEST_PAGE_FAULT;
1080         } else {
1081             cs->exception_index = page_fault_exceptions ?
1082                 RISCV_EXCP_INST_PAGE_FAULT : RISCV_EXCP_INST_ACCESS_FAULT;
1083         }
1084         break;
1085     case MMU_DATA_LOAD:
1086         if (two_stage && !first_stage) {
1087             cs->exception_index = RISCV_EXCP_LOAD_GUEST_ACCESS_FAULT;
1088         } else {
1089             cs->exception_index = page_fault_exceptions ?
1090                 RISCV_EXCP_LOAD_PAGE_FAULT : RISCV_EXCP_LOAD_ACCESS_FAULT;
1091         }
1092         break;
1093     case MMU_DATA_STORE:
1094         if (two_stage && !first_stage) {
1095             cs->exception_index = RISCV_EXCP_STORE_GUEST_AMO_ACCESS_FAULT;
1096         } else {
1097             cs->exception_index = page_fault_exceptions ?
1098                 RISCV_EXCP_STORE_PAGE_FAULT : RISCV_EXCP_STORE_AMO_ACCESS_FAULT;
1099         }
1100         break;
1101     default:
1102         g_assert_not_reached();
1103     }
1104     env->badaddr = address;
1105     env->two_stage_lookup = two_stage;
1106 }
1107 
1108 hwaddr riscv_cpu_get_phys_page_debug(CPUState *cs, vaddr addr)
1109 {
1110     RISCVCPU *cpu = RISCV_CPU(cs);
1111     CPURISCVState *env = &cpu->env;
1112     hwaddr phys_addr;
1113     int prot;
1114     int mmu_idx = cpu_mmu_index(&cpu->env, false);
1115 
1116     if (get_physical_address(env, &phys_addr, &prot, addr, NULL, 0, mmu_idx,
1117                              true, riscv_cpu_virt_enabled(env), true)) {
1118         return -1;
1119     }
1120 
1121     if (riscv_cpu_virt_enabled(env)) {
1122         if (get_physical_address(env, &phys_addr, &prot, phys_addr, NULL,
1123                                  0, mmu_idx, false, true, true)) {
1124             return -1;
1125         }
1126     }
1127 
1128     return phys_addr & TARGET_PAGE_MASK;
1129 }
1130 
1131 void riscv_cpu_do_transaction_failed(CPUState *cs, hwaddr physaddr,
1132                                      vaddr addr, unsigned size,
1133                                      MMUAccessType access_type,
1134                                      int mmu_idx, MemTxAttrs attrs,
1135                                      MemTxResult response, uintptr_t retaddr)
1136 {
1137     RISCVCPU *cpu = RISCV_CPU(cs);
1138     CPURISCVState *env = &cpu->env;
1139 
1140     if (access_type == MMU_DATA_STORE) {
1141         cs->exception_index = RISCV_EXCP_STORE_AMO_ACCESS_FAULT;
1142     } else if (access_type == MMU_DATA_LOAD) {
1143         cs->exception_index = RISCV_EXCP_LOAD_ACCESS_FAULT;
1144     } else {
1145         cs->exception_index = RISCV_EXCP_INST_ACCESS_FAULT;
1146     }
1147 
1148     env->badaddr = addr;
1149     env->two_stage_lookup = riscv_cpu_virt_enabled(env) ||
1150                             riscv_cpu_two_stage_lookup(mmu_idx);
1151     cpu_loop_exit_restore(cs, retaddr);
1152 }
1153 
1154 void riscv_cpu_do_unaligned_access(CPUState *cs, vaddr addr,
1155                                    MMUAccessType access_type, int mmu_idx,
1156                                    uintptr_t retaddr)
1157 {
1158     RISCVCPU *cpu = RISCV_CPU(cs);
1159     CPURISCVState *env = &cpu->env;
1160     switch (access_type) {
1161     case MMU_INST_FETCH:
1162         cs->exception_index = RISCV_EXCP_INST_ADDR_MIS;
1163         break;
1164     case MMU_DATA_LOAD:
1165         cs->exception_index = RISCV_EXCP_LOAD_ADDR_MIS;
1166         break;
1167     case MMU_DATA_STORE:
1168         cs->exception_index = RISCV_EXCP_STORE_AMO_ADDR_MIS;
1169         break;
1170     default:
1171         g_assert_not_reached();
1172     }
1173     env->badaddr = addr;
1174     env->two_stage_lookup = riscv_cpu_virt_enabled(env) ||
1175                             riscv_cpu_two_stage_lookup(mmu_idx);
1176     cpu_loop_exit_restore(cs, retaddr);
1177 }
1178 
1179 bool riscv_cpu_tlb_fill(CPUState *cs, vaddr address, int size,
1180                         MMUAccessType access_type, int mmu_idx,
1181                         bool probe, uintptr_t retaddr)
1182 {
1183     RISCVCPU *cpu = RISCV_CPU(cs);
1184     CPURISCVState *env = &cpu->env;
1185     vaddr im_address;
1186     hwaddr pa = 0;
1187     int prot, prot2, prot_pmp;
1188     bool pmp_violation = false;
1189     bool first_stage_error = true;
1190     bool two_stage_lookup = false;
1191     int ret = TRANSLATE_FAIL;
1192     int mode = mmu_idx;
1193     /* default TLB page size */
1194     target_ulong tlb_size = TARGET_PAGE_SIZE;
1195 
1196     env->guest_phys_fault_addr = 0;
1197 
1198     qemu_log_mask(CPU_LOG_MMU, "%s ad %" VADDR_PRIx " rw %d mmu_idx %d\n",
1199                   __func__, address, access_type, mmu_idx);
1200 
1201     /* MPRV does not affect the virtual-machine load/store
1202        instructions, HLV, HLVX, and HSV. */
1203     if (riscv_cpu_two_stage_lookup(mmu_idx)) {
1204         mode = get_field(env->hstatus, HSTATUS_SPVP);
1205     } else if (mode == PRV_M && access_type != MMU_INST_FETCH &&
1206                get_field(env->mstatus, MSTATUS_MPRV)) {
1207         mode = get_field(env->mstatus, MSTATUS_MPP);
1208         if (riscv_has_ext(env, RVH) && get_field(env->mstatus, MSTATUS_MPV)) {
1209             two_stage_lookup = true;
1210         }
1211     }
1212 
1213     if (riscv_cpu_virt_enabled(env) ||
1214         ((riscv_cpu_two_stage_lookup(mmu_idx) || two_stage_lookup) &&
1215          access_type != MMU_INST_FETCH)) {
1216         /* Two stage lookup */
1217         ret = get_physical_address(env, &pa, &prot, address,
1218                                    &env->guest_phys_fault_addr, access_type,
1219                                    mmu_idx, true, true, false);
1220 
1221         /*
1222          * A G-stage exception may be triggered during two state lookup.
1223          * And the env->guest_phys_fault_addr has already been set in
1224          * get_physical_address().
1225          */
1226         if (ret == TRANSLATE_G_STAGE_FAIL) {
1227             first_stage_error = false;
1228             access_type = MMU_DATA_LOAD;
1229         }
1230 
1231         qemu_log_mask(CPU_LOG_MMU,
1232                       "%s 1st-stage address=%" VADDR_PRIx " ret %d physical "
1233                       TARGET_FMT_plx " prot %d\n",
1234                       __func__, address, ret, pa, prot);
1235 
1236         if (ret == TRANSLATE_SUCCESS) {
1237             /* Second stage lookup */
1238             im_address = pa;
1239 
1240             ret = get_physical_address(env, &pa, &prot2, im_address, NULL,
1241                                        access_type, mmu_idx, false, true,
1242                                        false);
1243 
1244             qemu_log_mask(CPU_LOG_MMU,
1245                     "%s 2nd-stage address=%" VADDR_PRIx " ret %d physical "
1246                     TARGET_FMT_plx " prot %d\n",
1247                     __func__, im_address, ret, pa, prot2);
1248 
1249             prot &= prot2;
1250 
1251             if (ret == TRANSLATE_SUCCESS) {
1252                 ret = get_physical_address_pmp(env, &prot_pmp, &tlb_size, pa,
1253                                                size, access_type, mode);
1254 
1255                 qemu_log_mask(CPU_LOG_MMU,
1256                               "%s PMP address=" TARGET_FMT_plx " ret %d prot"
1257                               " %d tlb_size " TARGET_FMT_lu "\n",
1258                               __func__, pa, ret, prot_pmp, tlb_size);
1259 
1260                 prot &= prot_pmp;
1261             }
1262 
1263             if (ret != TRANSLATE_SUCCESS) {
1264                 /*
1265                  * Guest physical address translation failed, this is a HS
1266                  * level exception
1267                  */
1268                 first_stage_error = false;
1269                 env->guest_phys_fault_addr = (im_address |
1270                                               (address &
1271                                                (TARGET_PAGE_SIZE - 1))) >> 2;
1272             }
1273         }
1274     } else {
1275         /* Single stage lookup */
1276         ret = get_physical_address(env, &pa, &prot, address, NULL,
1277                                    access_type, mmu_idx, true, false, false);
1278 
1279         qemu_log_mask(CPU_LOG_MMU,
1280                       "%s address=%" VADDR_PRIx " ret %d physical "
1281                       TARGET_FMT_plx " prot %d\n",
1282                       __func__, address, ret, pa, prot);
1283 
1284         if (ret == TRANSLATE_SUCCESS) {
1285             ret = get_physical_address_pmp(env, &prot_pmp, &tlb_size, pa,
1286                                            size, access_type, mode);
1287 
1288             qemu_log_mask(CPU_LOG_MMU,
1289                           "%s PMP address=" TARGET_FMT_plx " ret %d prot"
1290                           " %d tlb_size " TARGET_FMT_lu "\n",
1291                           __func__, pa, ret, prot_pmp, tlb_size);
1292 
1293             prot &= prot_pmp;
1294         }
1295     }
1296 
1297     if (ret == TRANSLATE_PMP_FAIL) {
1298         pmp_violation = true;
1299     }
1300 
1301     if (ret == TRANSLATE_SUCCESS) {
1302         tlb_set_page(cs, address & ~(tlb_size - 1), pa & ~(tlb_size - 1),
1303                      prot, mmu_idx, tlb_size);
1304         return true;
1305     } else if (probe) {
1306         return false;
1307     } else {
1308         raise_mmu_exception(env, address, access_type, pmp_violation,
1309                             first_stage_error,
1310                             riscv_cpu_virt_enabled(env) ||
1311                                 riscv_cpu_two_stage_lookup(mmu_idx));
1312         cpu_loop_exit_restore(cs, retaddr);
1313     }
1314 
1315     return true;
1316 }
1317 #endif /* !CONFIG_USER_ONLY */
1318 
1319 /*
1320  * Handle Traps
1321  *
1322  * Adapted from Spike's processor_t::take_trap.
1323  *
1324  */
1325 void riscv_cpu_do_interrupt(CPUState *cs)
1326 {
1327 #if !defined(CONFIG_USER_ONLY)
1328 
1329     RISCVCPU *cpu = RISCV_CPU(cs);
1330     CPURISCVState *env = &cpu->env;
1331     bool write_gva = false;
1332     uint64_t s;
1333 
1334     /* cs->exception is 32-bits wide unlike mcause which is XLEN-bits wide
1335      * so we mask off the MSB and separate into trap type and cause.
1336      */
1337     bool async = !!(cs->exception_index & RISCV_EXCP_INT_FLAG);
1338     target_ulong cause = cs->exception_index & RISCV_EXCP_INT_MASK;
1339     uint64_t deleg = async ? env->mideleg : env->medeleg;
1340     target_ulong tval = 0;
1341     target_ulong htval = 0;
1342     target_ulong mtval2 = 0;
1343 
1344     if  (cause == RISCV_EXCP_SEMIHOST) {
1345         if (env->priv >= PRV_S) {
1346             do_common_semihosting(cs);
1347             env->pc += 4;
1348             return;
1349         }
1350         cause = RISCV_EXCP_BREAKPOINT;
1351     }
1352 
1353     if (!async) {
1354         /* set tval to badaddr for traps with address information */
1355         switch (cause) {
1356         case RISCV_EXCP_INST_GUEST_PAGE_FAULT:
1357         case RISCV_EXCP_LOAD_GUEST_ACCESS_FAULT:
1358         case RISCV_EXCP_STORE_GUEST_AMO_ACCESS_FAULT:
1359         case RISCV_EXCP_INST_ADDR_MIS:
1360         case RISCV_EXCP_INST_ACCESS_FAULT:
1361         case RISCV_EXCP_LOAD_ADDR_MIS:
1362         case RISCV_EXCP_STORE_AMO_ADDR_MIS:
1363         case RISCV_EXCP_LOAD_ACCESS_FAULT:
1364         case RISCV_EXCP_STORE_AMO_ACCESS_FAULT:
1365         case RISCV_EXCP_INST_PAGE_FAULT:
1366         case RISCV_EXCP_LOAD_PAGE_FAULT:
1367         case RISCV_EXCP_STORE_PAGE_FAULT:
1368             write_gva = env->two_stage_lookup;
1369             tval = env->badaddr;
1370             break;
1371         case RISCV_EXCP_ILLEGAL_INST:
1372         case RISCV_EXCP_VIRT_INSTRUCTION_FAULT:
1373             tval = env->bins;
1374             break;
1375         default:
1376             break;
1377         }
1378         /* ecall is dispatched as one cause so translate based on mode */
1379         if (cause == RISCV_EXCP_U_ECALL) {
1380             assert(env->priv <= 3);
1381 
1382             if (env->priv == PRV_M) {
1383                 cause = RISCV_EXCP_M_ECALL;
1384             } else if (env->priv == PRV_S && riscv_cpu_virt_enabled(env)) {
1385                 cause = RISCV_EXCP_VS_ECALL;
1386             } else if (env->priv == PRV_S && !riscv_cpu_virt_enabled(env)) {
1387                 cause = RISCV_EXCP_S_ECALL;
1388             } else if (env->priv == PRV_U) {
1389                 cause = RISCV_EXCP_U_ECALL;
1390             }
1391         }
1392     }
1393 
1394     trace_riscv_trap(env->mhartid, async, cause, env->pc, tval,
1395                      riscv_cpu_get_trap_name(cause, async));
1396 
1397     qemu_log_mask(CPU_LOG_INT,
1398                   "%s: hart:"TARGET_FMT_ld", async:%d, cause:"TARGET_FMT_lx", "
1399                   "epc:0x"TARGET_FMT_lx", tval:0x"TARGET_FMT_lx", desc=%s\n",
1400                   __func__, env->mhartid, async, cause, env->pc, tval,
1401                   riscv_cpu_get_trap_name(cause, async));
1402 
1403     if (env->priv <= PRV_S &&
1404             cause < TARGET_LONG_BITS && ((deleg >> cause) & 1)) {
1405         /* handle the trap in S-mode */
1406         if (riscv_has_ext(env, RVH)) {
1407             uint64_t hdeleg = async ? env->hideleg : env->hedeleg;
1408 
1409             if (riscv_cpu_virt_enabled(env) && ((hdeleg >> cause) & 1)) {
1410                 /* Trap to VS mode */
1411                 /*
1412                  * See if we need to adjust cause. Yes if its VS mode interrupt
1413                  * no if hypervisor has delegated one of hs mode's interrupt
1414                  */
1415                 if (cause == IRQ_VS_TIMER || cause == IRQ_VS_SOFT ||
1416                     cause == IRQ_VS_EXT) {
1417                     cause = cause - 1;
1418                 }
1419                 write_gva = false;
1420             } else if (riscv_cpu_virt_enabled(env)) {
1421                 /* Trap into HS mode, from virt */
1422                 riscv_cpu_swap_hypervisor_regs(env);
1423                 env->hstatus = set_field(env->hstatus, HSTATUS_SPVP,
1424                                          env->priv);
1425                 env->hstatus = set_field(env->hstatus, HSTATUS_SPV,
1426                                          riscv_cpu_virt_enabled(env));
1427 
1428 
1429                 htval = env->guest_phys_fault_addr;
1430 
1431                 riscv_cpu_set_virt_enabled(env, 0);
1432             } else {
1433                 /* Trap into HS mode */
1434                 env->hstatus = set_field(env->hstatus, HSTATUS_SPV, false);
1435                 htval = env->guest_phys_fault_addr;
1436             }
1437             env->hstatus = set_field(env->hstatus, HSTATUS_GVA, write_gva);
1438         }
1439 
1440         s = env->mstatus;
1441         s = set_field(s, MSTATUS_SPIE, get_field(s, MSTATUS_SIE));
1442         s = set_field(s, MSTATUS_SPP, env->priv);
1443         s = set_field(s, MSTATUS_SIE, 0);
1444         env->mstatus = s;
1445         env->scause = cause | ((target_ulong)async << (TARGET_LONG_BITS - 1));
1446         env->sepc = env->pc;
1447         env->stval = tval;
1448         env->htval = htval;
1449         env->pc = (env->stvec >> 2 << 2) +
1450             ((async && (env->stvec & 3) == 1) ? cause * 4 : 0);
1451         riscv_cpu_set_mode(env, PRV_S);
1452     } else {
1453         /* handle the trap in M-mode */
1454         if (riscv_has_ext(env, RVH)) {
1455             if (riscv_cpu_virt_enabled(env)) {
1456                 riscv_cpu_swap_hypervisor_regs(env);
1457             }
1458             env->mstatus = set_field(env->mstatus, MSTATUS_MPV,
1459                                      riscv_cpu_virt_enabled(env));
1460             if (riscv_cpu_virt_enabled(env) && tval) {
1461                 env->mstatus = set_field(env->mstatus, MSTATUS_GVA, 1);
1462             }
1463 
1464             mtval2 = env->guest_phys_fault_addr;
1465 
1466             /* Trapping to M mode, virt is disabled */
1467             riscv_cpu_set_virt_enabled(env, 0);
1468         }
1469 
1470         s = env->mstatus;
1471         s = set_field(s, MSTATUS_MPIE, get_field(s, MSTATUS_MIE));
1472         s = set_field(s, MSTATUS_MPP, env->priv);
1473         s = set_field(s, MSTATUS_MIE, 0);
1474         env->mstatus = s;
1475         env->mcause = cause | ~(((target_ulong)-1) >> async);
1476         env->mepc = env->pc;
1477         env->mtval = tval;
1478         env->mtval2 = mtval2;
1479         env->pc = (env->mtvec >> 2 << 2) +
1480             ((async && (env->mtvec & 3) == 1) ? cause * 4 : 0);
1481         riscv_cpu_set_mode(env, PRV_M);
1482     }
1483 
1484     /* NOTE: it is not necessary to yield load reservations here. It is only
1485      * necessary for an SC from "another hart" to cause a load reservation
1486      * to be yielded. Refer to the memory consistency model section of the
1487      * RISC-V ISA Specification.
1488      */
1489 
1490     env->two_stage_lookup = false;
1491 #endif
1492     cs->exception_index = RISCV_EXCP_NONE; /* mark handled to qemu */
1493 }
1494