xref: /openbmc/qemu/target/riscv/cpu_helper.c (revision 48eaeb56)
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 static RISCVMXL cpu_get_xl(CPURISCVState *env)
39 {
40 #if defined(TARGET_RISCV32)
41     return MXL_RV32;
42 #elif defined(CONFIG_USER_ONLY)
43     return MXL_RV64;
44 #else
45     RISCVMXL xl = riscv_cpu_mxl(env);
46 
47     /*
48      * When emulating a 32-bit-only cpu, use RV32.
49      * When emulating a 64-bit cpu, and MXL has been reduced to RV32,
50      * MSTATUSH doesn't have UXL/SXL, therefore XLEN cannot be widened
51      * back to RV64 for lower privs.
52      */
53     if (xl != MXL_RV32) {
54         switch (env->priv) {
55         case PRV_M:
56             break;
57         case PRV_U:
58             xl = get_field(env->mstatus, MSTATUS64_UXL);
59             break;
60         default: /* PRV_S | PRV_H */
61             xl = get_field(env->mstatus, MSTATUS64_SXL);
62             break;
63         }
64     }
65     return xl;
66 #endif
67 }
68 
69 void cpu_get_tb_cpu_state(CPURISCVState *env, target_ulong *pc,
70                           target_ulong *cs_base, uint32_t *pflags)
71 {
72     uint32_t flags = 0;
73 
74     *pc = env->pc;
75     *cs_base = 0;
76 
77     if (riscv_has_ext(env, RVV)) {
78         /*
79          * If env->vl equals to VLMAX, we can use generic vector operation
80          * expanders (GVEC) to accerlate the vector operations.
81          * However, as LMUL could be a fractional number. The maximum
82          * vector size can be operated might be less than 8 bytes,
83          * which is not supported by GVEC. So we set vl_eq_vlmax flag to true
84          * only when maxsz >= 8 bytes.
85          */
86         uint32_t vlmax = vext_get_vlmax(env_archcpu(env), env->vtype);
87         uint32_t sew = FIELD_EX64(env->vtype, VTYPE, VSEW);
88         uint32_t maxsz = vlmax << sew;
89         bool vl_eq_vlmax = (env->vstart == 0) && (vlmax == env->vl) &&
90                            (maxsz >= 8);
91         flags = FIELD_DP32(flags, TB_FLAGS, VILL,
92                     FIELD_EX64(env->vtype, VTYPE, VILL));
93         flags = FIELD_DP32(flags, TB_FLAGS, SEW, sew);
94         flags = FIELD_DP32(flags, TB_FLAGS, LMUL,
95                     FIELD_EX64(env->vtype, VTYPE, VLMUL));
96         flags = FIELD_DP32(flags, TB_FLAGS, VL_EQ_VLMAX, vl_eq_vlmax);
97     } else {
98         flags = FIELD_DP32(flags, TB_FLAGS, VILL, 1);
99     }
100 
101 #ifdef CONFIG_USER_ONLY
102     flags |= TB_FLAGS_MSTATUS_FS;
103     flags |= TB_FLAGS_MSTATUS_VS;
104 #else
105     flags |= cpu_mmu_index(env, 0);
106     if (riscv_cpu_fp_enabled(env)) {
107         flags |= env->mstatus & MSTATUS_FS;
108     }
109 
110     if (riscv_cpu_vector_enabled(env)) {
111         flags |= env->mstatus & MSTATUS_VS;
112     }
113 
114     if (riscv_has_ext(env, RVH)) {
115         if (env->priv == PRV_M ||
116             (env->priv == PRV_S && !riscv_cpu_virt_enabled(env)) ||
117             (env->priv == PRV_U && !riscv_cpu_virt_enabled(env) &&
118                 get_field(env->hstatus, HSTATUS_HU))) {
119             flags = FIELD_DP32(flags, TB_FLAGS, HLSX, 1);
120         }
121 
122         flags = FIELD_DP32(flags, TB_FLAGS, MSTATUS_HS_FS,
123                            get_field(env->mstatus_hs, MSTATUS_FS));
124 
125         flags = FIELD_DP32(flags, TB_FLAGS, MSTATUS_HS_VS,
126                            get_field(env->mstatus_hs, MSTATUS_VS));
127     }
128     if (riscv_has_ext(env, RVJ)) {
129         int priv = flags & TB_FLAGS_PRIV_MMU_MASK;
130         bool pm_enabled = false;
131         switch (priv) {
132         case PRV_U:
133             pm_enabled = env->mmte & U_PM_ENABLE;
134             break;
135         case PRV_S:
136             pm_enabled = env->mmte & S_PM_ENABLE;
137             break;
138         case PRV_M:
139             pm_enabled = env->mmte & M_PM_ENABLE;
140             break;
141         default:
142             g_assert_not_reached();
143         }
144         flags = FIELD_DP32(flags, TB_FLAGS, PM_ENABLED, pm_enabled);
145     }
146 #endif
147 
148     flags = FIELD_DP32(flags, TB_FLAGS, XL, cpu_get_xl(env));
149 
150     *pflags = flags;
151 }
152 
153 #ifndef CONFIG_USER_ONLY
154 static int riscv_cpu_local_irq_pending(CPURISCVState *env)
155 {
156     target_ulong virt_enabled = riscv_cpu_virt_enabled(env);
157 
158     target_ulong mstatus_mie = get_field(env->mstatus, MSTATUS_MIE);
159     target_ulong mstatus_sie = get_field(env->mstatus, MSTATUS_SIE);
160 
161     target_ulong pending = env->mip & env->mie;
162 
163     target_ulong mie    = env->priv < PRV_M ||
164                           (env->priv == PRV_M && mstatus_mie);
165     target_ulong sie    = env->priv < PRV_S ||
166                           (env->priv == PRV_S && mstatus_sie);
167     target_ulong hsie   = virt_enabled || sie;
168     target_ulong vsie   = virt_enabled && sie;
169 
170     target_ulong irqs =
171             (pending & ~env->mideleg & -mie) |
172             (pending &  env->mideleg & ~env->hideleg & -hsie) |
173             (pending &  env->mideleg &  env->hideleg & -vsie);
174 
175     if (irqs) {
176         return ctz64(irqs); /* since non-zero */
177     } else {
178         return RISCV_EXCP_NONE; /* indicates no pending interrupt */
179     }
180 }
181 
182 bool riscv_cpu_exec_interrupt(CPUState *cs, int interrupt_request)
183 {
184     if (interrupt_request & CPU_INTERRUPT_HARD) {
185         RISCVCPU *cpu = RISCV_CPU(cs);
186         CPURISCVState *env = &cpu->env;
187         int interruptno = riscv_cpu_local_irq_pending(env);
188         if (interruptno >= 0) {
189             cs->exception_index = RISCV_EXCP_INT_FLAG | interruptno;
190             riscv_cpu_do_interrupt(cs);
191             return true;
192         }
193     }
194     return false;
195 }
196 
197 /* Return true is floating point support is currently enabled */
198 bool riscv_cpu_fp_enabled(CPURISCVState *env)
199 {
200     if (env->mstatus & MSTATUS_FS) {
201         if (riscv_cpu_virt_enabled(env) && !(env->mstatus_hs & MSTATUS_FS)) {
202             return false;
203         }
204         return true;
205     }
206 
207     return false;
208 }
209 
210 /* Return true is vector support is currently enabled */
211 bool riscv_cpu_vector_enabled(CPURISCVState *env)
212 {
213     if (env->mstatus & MSTATUS_VS) {
214         if (riscv_cpu_virt_enabled(env) && !(env->mstatus_hs & MSTATUS_VS)) {
215             return false;
216         }
217         return true;
218     }
219 
220     return false;
221 }
222 
223 void riscv_cpu_swap_hypervisor_regs(CPURISCVState *env)
224 {
225     uint64_t mstatus_mask = MSTATUS_MXR | MSTATUS_SUM | MSTATUS_FS |
226                             MSTATUS_SPP | MSTATUS_SPIE | MSTATUS_SIE |
227                             MSTATUS64_UXL | MSTATUS_VS;
228     bool current_virt = riscv_cpu_virt_enabled(env);
229 
230     g_assert(riscv_has_ext(env, RVH));
231 
232     if (current_virt) {
233         /* Current V=1 and we are about to change to V=0 */
234         env->vsstatus = env->mstatus & mstatus_mask;
235         env->mstatus &= ~mstatus_mask;
236         env->mstatus |= env->mstatus_hs;
237 
238         env->vstvec = env->stvec;
239         env->stvec = env->stvec_hs;
240 
241         env->vsscratch = env->sscratch;
242         env->sscratch = env->sscratch_hs;
243 
244         env->vsepc = env->sepc;
245         env->sepc = env->sepc_hs;
246 
247         env->vscause = env->scause;
248         env->scause = env->scause_hs;
249 
250         env->vstval = env->stval;
251         env->stval = env->stval_hs;
252 
253         env->vsatp = env->satp;
254         env->satp = env->satp_hs;
255     } else {
256         /* Current V=0 and we are about to change to V=1 */
257         env->mstatus_hs = env->mstatus & mstatus_mask;
258         env->mstatus &= ~mstatus_mask;
259         env->mstatus |= env->vsstatus;
260 
261         env->stvec_hs = env->stvec;
262         env->stvec = env->vstvec;
263 
264         env->sscratch_hs = env->sscratch;
265         env->sscratch = env->vsscratch;
266 
267         env->sepc_hs = env->sepc;
268         env->sepc = env->vsepc;
269 
270         env->scause_hs = env->scause;
271         env->scause = env->vscause;
272 
273         env->stval_hs = env->stval;
274         env->stval = env->vstval;
275 
276         env->satp_hs = env->satp;
277         env->satp = env->vsatp;
278     }
279 }
280 
281 bool riscv_cpu_virt_enabled(CPURISCVState *env)
282 {
283     if (!riscv_has_ext(env, RVH)) {
284         return false;
285     }
286 
287     return get_field(env->virt, VIRT_ONOFF);
288 }
289 
290 void riscv_cpu_set_virt_enabled(CPURISCVState *env, bool enable)
291 {
292     if (!riscv_has_ext(env, RVH)) {
293         return;
294     }
295 
296     /* Flush the TLB on all virt mode changes. */
297     if (get_field(env->virt, VIRT_ONOFF) != enable) {
298         tlb_flush(env_cpu(env));
299     }
300 
301     env->virt = set_field(env->virt, VIRT_ONOFF, enable);
302 }
303 
304 bool riscv_cpu_two_stage_lookup(int mmu_idx)
305 {
306     return mmu_idx & TB_FLAGS_PRIV_HYP_ACCESS_MASK;
307 }
308 
309 int riscv_cpu_claim_interrupts(RISCVCPU *cpu, uint32_t interrupts)
310 {
311     CPURISCVState *env = &cpu->env;
312     if (env->miclaim & interrupts) {
313         return -1;
314     } else {
315         env->miclaim |= interrupts;
316         return 0;
317     }
318 }
319 
320 uint32_t riscv_cpu_update_mip(RISCVCPU *cpu, uint32_t mask, uint32_t value)
321 {
322     CPURISCVState *env = &cpu->env;
323     CPUState *cs = CPU(cpu);
324     uint32_t old = env->mip;
325     bool locked = false;
326 
327     if (!qemu_mutex_iothread_locked()) {
328         locked = true;
329         qemu_mutex_lock_iothread();
330     }
331 
332     env->mip = (env->mip & ~mask) | (value & mask);
333 
334     if (env->mip) {
335         cpu_interrupt(cs, CPU_INTERRUPT_HARD);
336     } else {
337         cpu_reset_interrupt(cs, CPU_INTERRUPT_HARD);
338     }
339 
340     if (locked) {
341         qemu_mutex_unlock_iothread();
342     }
343 
344     return old;
345 }
346 
347 void riscv_cpu_set_rdtime_fn(CPURISCVState *env, uint64_t (*fn)(uint32_t),
348                              uint32_t arg)
349 {
350     env->rdtime_fn = fn;
351     env->rdtime_fn_arg = arg;
352 }
353 
354 void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv)
355 {
356     if (newpriv > PRV_M) {
357         g_assert_not_reached();
358     }
359     if (newpriv == PRV_H) {
360         newpriv = PRV_U;
361     }
362     /* tlb_flush is unnecessary as mode is contained in mmu_idx */
363     env->priv = newpriv;
364 
365     /*
366      * Clear the load reservation - otherwise a reservation placed in one
367      * context/process can be used by another, resulting in an SC succeeding
368      * incorrectly. Version 2.2 of the ISA specification explicitly requires
369      * this behaviour, while later revisions say that the kernel "should" use
370      * an SC instruction to force the yielding of a load reservation on a
371      * preemptive context switch. As a result, do both.
372      */
373     env->load_res = -1;
374 }
375 
376 /*
377  * get_physical_address_pmp - check PMP permission for this physical address
378  *
379  * Match the PMP region and check permission for this physical address and it's
380  * TLB page. Returns 0 if the permission checking was successful
381  *
382  * @env: CPURISCVState
383  * @prot: The returned protection attributes
384  * @tlb_size: TLB page size containing addr. It could be modified after PMP
385  *            permission checking. NULL if not set TLB page for addr.
386  * @addr: The physical address to be checked permission
387  * @access_type: The type of MMU access
388  * @mode: Indicates current privilege level.
389  */
390 static int get_physical_address_pmp(CPURISCVState *env, int *prot,
391                                     target_ulong *tlb_size, hwaddr addr,
392                                     int size, MMUAccessType access_type,
393                                     int mode)
394 {
395     pmp_priv_t pmp_priv;
396     target_ulong tlb_size_pmp = 0;
397 
398     if (!riscv_feature(env, RISCV_FEATURE_PMP)) {
399         *prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
400         return TRANSLATE_SUCCESS;
401     }
402 
403     if (!pmp_hart_has_privs(env, addr, size, 1 << access_type, &pmp_priv,
404                             mode)) {
405         *prot = 0;
406         return TRANSLATE_PMP_FAIL;
407     }
408 
409     *prot = pmp_priv_to_page_prot(pmp_priv);
410     if (tlb_size != NULL) {
411         if (pmp_is_range_in_tlb(env, addr & ~(*tlb_size - 1), &tlb_size_pmp)) {
412             *tlb_size = tlb_size_pmp;
413         }
414     }
415 
416     return TRANSLATE_SUCCESS;
417 }
418 
419 /* get_physical_address - get the physical address for this virtual address
420  *
421  * Do a page table walk to obtain the physical address corresponding to a
422  * virtual address. Returns 0 if the translation was successful
423  *
424  * Adapted from Spike's mmu_t::translate and mmu_t::walk
425  *
426  * @env: CPURISCVState
427  * @physical: This will be set to the calculated physical address
428  * @prot: The returned protection attributes
429  * @addr: The virtual address to be translated
430  * @fault_pte_addr: If not NULL, this will be set to fault pte address
431  *                  when a error occurs on pte address translation.
432  *                  This will already be shifted to match htval.
433  * @access_type: The type of MMU access
434  * @mmu_idx: Indicates current privilege level
435  * @first_stage: Are we in first stage translation?
436  *               Second stage is used for hypervisor guest translation
437  * @two_stage: Are we going to perform two stage translation
438  * @is_debug: Is this access from a debugger or the monitor?
439  */
440 static int get_physical_address(CPURISCVState *env, hwaddr *physical,
441                                 int *prot, target_ulong addr,
442                                 target_ulong *fault_pte_addr,
443                                 int access_type, int mmu_idx,
444                                 bool first_stage, bool two_stage,
445                                 bool is_debug)
446 {
447     /* NOTE: the env->pc value visible here will not be
448      * correct, but the value visible to the exception handler
449      * (riscv_cpu_do_interrupt) is correct */
450     MemTxResult res;
451     MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
452     int mode = mmu_idx & TB_FLAGS_PRIV_MMU_MASK;
453     bool use_background = false;
454 
455     /*
456      * Check if we should use the background registers for the two
457      * stage translation. We don't need to check if we actually need
458      * two stage translation as that happened before this function
459      * was called. Background registers will be used if the guest has
460      * forced a two stage translation to be on (in HS or M mode).
461      */
462     if (!riscv_cpu_virt_enabled(env) && two_stage) {
463         use_background = true;
464     }
465 
466     /* MPRV does not affect the virtual-machine load/store
467        instructions, HLV, HLVX, and HSV. */
468     if (riscv_cpu_two_stage_lookup(mmu_idx)) {
469         mode = get_field(env->hstatus, HSTATUS_SPVP);
470     } else if (mode == PRV_M && access_type != MMU_INST_FETCH) {
471         if (get_field(env->mstatus, MSTATUS_MPRV)) {
472             mode = get_field(env->mstatus, MSTATUS_MPP);
473         }
474     }
475 
476     if (first_stage == false) {
477         /* We are in stage 2 translation, this is similar to stage 1. */
478         /* Stage 2 is always taken as U-mode */
479         mode = PRV_U;
480     }
481 
482     if (mode == PRV_M || !riscv_feature(env, RISCV_FEATURE_MMU)) {
483         *physical = addr;
484         *prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
485         return TRANSLATE_SUCCESS;
486     }
487 
488     *prot = 0;
489 
490     hwaddr base;
491     int levels, ptidxbits, ptesize, vm, sum, mxr, widened;
492 
493     if (first_stage == true) {
494         mxr = get_field(env->mstatus, MSTATUS_MXR);
495     } else {
496         mxr = get_field(env->vsstatus, MSTATUS_MXR);
497     }
498 
499     if (first_stage == true) {
500         if (use_background) {
501             if (riscv_cpu_mxl(env) == MXL_RV32) {
502                 base = (hwaddr)get_field(env->vsatp, SATP32_PPN) << PGSHIFT;
503                 vm = get_field(env->vsatp, SATP32_MODE);
504             } else {
505                 base = (hwaddr)get_field(env->vsatp, SATP64_PPN) << PGSHIFT;
506                 vm = get_field(env->vsatp, SATP64_MODE);
507             }
508         } else {
509             if (riscv_cpu_mxl(env) == MXL_RV32) {
510                 base = (hwaddr)get_field(env->satp, SATP32_PPN) << PGSHIFT;
511                 vm = get_field(env->satp, SATP32_MODE);
512             } else {
513                 base = (hwaddr)get_field(env->satp, SATP64_PPN) << PGSHIFT;
514                 vm = get_field(env->satp, SATP64_MODE);
515             }
516         }
517         widened = 0;
518     } else {
519         if (riscv_cpu_mxl(env) == MXL_RV32) {
520             base = (hwaddr)get_field(env->hgatp, SATP32_PPN) << PGSHIFT;
521             vm = get_field(env->hgatp, SATP32_MODE);
522         } else {
523             base = (hwaddr)get_field(env->hgatp, SATP64_PPN) << PGSHIFT;
524             vm = get_field(env->hgatp, SATP64_MODE);
525         }
526         widened = 2;
527     }
528     /* status.SUM will be ignored if execute on background */
529     sum = get_field(env->mstatus, MSTATUS_SUM) || use_background || is_debug;
530     switch (vm) {
531     case VM_1_10_SV32:
532       levels = 2; ptidxbits = 10; ptesize = 4; break;
533     case VM_1_10_SV39:
534       levels = 3; ptidxbits = 9; ptesize = 8; break;
535     case VM_1_10_SV48:
536       levels = 4; ptidxbits = 9; ptesize = 8; break;
537     case VM_1_10_SV57:
538       levels = 5; ptidxbits = 9; ptesize = 8; break;
539     case VM_1_10_MBARE:
540         *physical = addr;
541         *prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
542         return TRANSLATE_SUCCESS;
543     default:
544       g_assert_not_reached();
545     }
546 
547     CPUState *cs = env_cpu(env);
548     int va_bits = PGSHIFT + levels * ptidxbits + widened;
549     target_ulong mask, masked_msbs;
550 
551     if (TARGET_LONG_BITS > (va_bits - 1)) {
552         mask = (1L << (TARGET_LONG_BITS - (va_bits - 1))) - 1;
553     } else {
554         mask = 0;
555     }
556     masked_msbs = (addr >> (va_bits - 1)) & mask;
557 
558     if (masked_msbs != 0 && masked_msbs != mask) {
559         return TRANSLATE_FAIL;
560     }
561 
562     int ptshift = (levels - 1) * ptidxbits;
563     int i;
564 
565 #if !TCG_OVERSIZED_GUEST
566 restart:
567 #endif
568     for (i = 0; i < levels; i++, ptshift -= ptidxbits) {
569         target_ulong idx;
570         if (i == 0) {
571             idx = (addr >> (PGSHIFT + ptshift)) &
572                            ((1 << (ptidxbits + widened)) - 1);
573         } else {
574             idx = (addr >> (PGSHIFT + ptshift)) &
575                            ((1 << ptidxbits) - 1);
576         }
577 
578         /* check that physical address of PTE is legal */
579         hwaddr pte_addr;
580 
581         if (two_stage && first_stage) {
582             int vbase_prot;
583             hwaddr vbase;
584 
585             /* Do the second stage translation on the base PTE address. */
586             int vbase_ret = get_physical_address(env, &vbase, &vbase_prot,
587                                                  base, NULL, MMU_DATA_LOAD,
588                                                  mmu_idx, false, true,
589                                                  is_debug);
590 
591             if (vbase_ret != TRANSLATE_SUCCESS) {
592                 if (fault_pte_addr) {
593                     *fault_pte_addr = (base + idx * ptesize) >> 2;
594                 }
595                 return TRANSLATE_G_STAGE_FAIL;
596             }
597 
598             pte_addr = vbase + idx * ptesize;
599         } else {
600             pte_addr = base + idx * ptesize;
601         }
602 
603         int pmp_prot;
604         int pmp_ret = get_physical_address_pmp(env, &pmp_prot, NULL, pte_addr,
605                                                sizeof(target_ulong),
606                                                MMU_DATA_LOAD, PRV_S);
607         if (pmp_ret != TRANSLATE_SUCCESS) {
608             return TRANSLATE_PMP_FAIL;
609         }
610 
611         target_ulong pte;
612         if (riscv_cpu_mxl(env) == MXL_RV32) {
613             pte = address_space_ldl(cs->as, pte_addr, attrs, &res);
614         } else {
615             pte = address_space_ldq(cs->as, pte_addr, attrs, &res);
616         }
617 
618         if (res != MEMTX_OK) {
619             return TRANSLATE_FAIL;
620         }
621 
622         hwaddr ppn = pte >> PTE_PPN_SHIFT;
623 
624         if (!(pte & PTE_V)) {
625             /* Invalid PTE */
626             return TRANSLATE_FAIL;
627         } else if (!(pte & (PTE_R | PTE_W | PTE_X))) {
628             /* Inner PTE, continue walking */
629             base = ppn << PGSHIFT;
630         } else if ((pte & (PTE_R | PTE_W | PTE_X)) == PTE_W) {
631             /* Reserved leaf PTE flags: PTE_W */
632             return TRANSLATE_FAIL;
633         } else if ((pte & (PTE_R | PTE_W | PTE_X)) == (PTE_W | PTE_X)) {
634             /* Reserved leaf PTE flags: PTE_W + PTE_X */
635             return TRANSLATE_FAIL;
636         } else if ((pte & PTE_U) && ((mode != PRV_U) &&
637                    (!sum || access_type == MMU_INST_FETCH))) {
638             /* User PTE flags when not U mode and mstatus.SUM is not set,
639                or the access type is an instruction fetch */
640             return TRANSLATE_FAIL;
641         } else if (!(pte & PTE_U) && (mode != PRV_S)) {
642             /* Supervisor PTE flags when not S mode */
643             return TRANSLATE_FAIL;
644         } else if (ppn & ((1ULL << ptshift) - 1)) {
645             /* Misaligned PPN */
646             return TRANSLATE_FAIL;
647         } else if (access_type == MMU_DATA_LOAD && !((pte & PTE_R) ||
648                    ((pte & PTE_X) && mxr))) {
649             /* Read access check failed */
650             return TRANSLATE_FAIL;
651         } else if (access_type == MMU_DATA_STORE && !(pte & PTE_W)) {
652             /* Write access check failed */
653             return TRANSLATE_FAIL;
654         } else if (access_type == MMU_INST_FETCH && !(pte & PTE_X)) {
655             /* Fetch access check failed */
656             return TRANSLATE_FAIL;
657         } else {
658             /* if necessary, set accessed and dirty bits. */
659             target_ulong updated_pte = pte | PTE_A |
660                 (access_type == MMU_DATA_STORE ? PTE_D : 0);
661 
662             /* Page table updates need to be atomic with MTTCG enabled */
663             if (updated_pte != pte) {
664                 /*
665                  * - if accessed or dirty bits need updating, and the PTE is
666                  *   in RAM, then we do so atomically with a compare and swap.
667                  * - if the PTE is in IO space or ROM, then it can't be updated
668                  *   and we return TRANSLATE_FAIL.
669                  * - if the PTE changed by the time we went to update it, then
670                  *   it is no longer valid and we must re-walk the page table.
671                  */
672                 MemoryRegion *mr;
673                 hwaddr l = sizeof(target_ulong), addr1;
674                 mr = address_space_translate(cs->as, pte_addr,
675                     &addr1, &l, false, MEMTXATTRS_UNSPECIFIED);
676                 if (memory_region_is_ram(mr)) {
677                     target_ulong *pte_pa =
678                         qemu_map_ram_ptr(mr->ram_block, addr1);
679 #if TCG_OVERSIZED_GUEST
680                     /* MTTCG is not enabled on oversized TCG guests so
681                      * page table updates do not need to be atomic */
682                     *pte_pa = pte = updated_pte;
683 #else
684                     target_ulong old_pte =
685                         qatomic_cmpxchg(pte_pa, pte, updated_pte);
686                     if (old_pte != pte) {
687                         goto restart;
688                     } else {
689                         pte = updated_pte;
690                     }
691 #endif
692                 } else {
693                     /* misconfigured PTE in ROM (AD bits are not preset) or
694                      * PTE is in IO space and can't be updated atomically */
695                     return TRANSLATE_FAIL;
696                 }
697             }
698 
699             /* for superpage mappings, make a fake leaf PTE for the TLB's
700                benefit. */
701             target_ulong vpn = addr >> PGSHIFT;
702             *physical = ((ppn | (vpn & ((1L << ptshift) - 1))) << PGSHIFT) |
703                         (addr & ~TARGET_PAGE_MASK);
704 
705             /* set permissions on the TLB entry */
706             if ((pte & PTE_R) || ((pte & PTE_X) && mxr)) {
707                 *prot |= PAGE_READ;
708             }
709             if ((pte & PTE_X)) {
710                 *prot |= PAGE_EXEC;
711             }
712             /* add write permission on stores or if the page is already dirty,
713                so that we TLB miss on later writes to update the dirty bit */
714             if ((pte & PTE_W) &&
715                     (access_type == MMU_DATA_STORE || (pte & PTE_D))) {
716                 *prot |= PAGE_WRITE;
717             }
718             return TRANSLATE_SUCCESS;
719         }
720     }
721     return TRANSLATE_FAIL;
722 }
723 
724 static void raise_mmu_exception(CPURISCVState *env, target_ulong address,
725                                 MMUAccessType access_type, bool pmp_violation,
726                                 bool first_stage, bool two_stage)
727 {
728     CPUState *cs = env_cpu(env);
729     int page_fault_exceptions, vm;
730     uint64_t stap_mode;
731 
732     if (riscv_cpu_mxl(env) == MXL_RV32) {
733         stap_mode = SATP32_MODE;
734     } else {
735         stap_mode = SATP64_MODE;
736     }
737 
738     if (first_stage) {
739         vm = get_field(env->satp, stap_mode);
740     } else {
741         vm = get_field(env->hgatp, stap_mode);
742     }
743 
744     page_fault_exceptions = vm != VM_1_10_MBARE && !pmp_violation;
745 
746     switch (access_type) {
747     case MMU_INST_FETCH:
748         if (riscv_cpu_virt_enabled(env) && !first_stage) {
749             cs->exception_index = RISCV_EXCP_INST_GUEST_PAGE_FAULT;
750         } else {
751             cs->exception_index = page_fault_exceptions ?
752                 RISCV_EXCP_INST_PAGE_FAULT : RISCV_EXCP_INST_ACCESS_FAULT;
753         }
754         break;
755     case MMU_DATA_LOAD:
756         if (two_stage && !first_stage) {
757             cs->exception_index = RISCV_EXCP_LOAD_GUEST_ACCESS_FAULT;
758         } else {
759             cs->exception_index = page_fault_exceptions ?
760                 RISCV_EXCP_LOAD_PAGE_FAULT : RISCV_EXCP_LOAD_ACCESS_FAULT;
761         }
762         break;
763     case MMU_DATA_STORE:
764         if (two_stage && !first_stage) {
765             cs->exception_index = RISCV_EXCP_STORE_GUEST_AMO_ACCESS_FAULT;
766         } else {
767             cs->exception_index = page_fault_exceptions ?
768                 RISCV_EXCP_STORE_PAGE_FAULT : RISCV_EXCP_STORE_AMO_ACCESS_FAULT;
769         }
770         break;
771     default:
772         g_assert_not_reached();
773     }
774     env->badaddr = address;
775     env->two_stage_lookup = two_stage;
776 }
777 
778 hwaddr riscv_cpu_get_phys_page_debug(CPUState *cs, vaddr addr)
779 {
780     RISCVCPU *cpu = RISCV_CPU(cs);
781     CPURISCVState *env = &cpu->env;
782     hwaddr phys_addr;
783     int prot;
784     int mmu_idx = cpu_mmu_index(&cpu->env, false);
785 
786     if (get_physical_address(env, &phys_addr, &prot, addr, NULL, 0, mmu_idx,
787                              true, riscv_cpu_virt_enabled(env), true)) {
788         return -1;
789     }
790 
791     if (riscv_cpu_virt_enabled(env)) {
792         if (get_physical_address(env, &phys_addr, &prot, phys_addr, NULL,
793                                  0, mmu_idx, false, true, true)) {
794             return -1;
795         }
796     }
797 
798     return phys_addr & TARGET_PAGE_MASK;
799 }
800 
801 void riscv_cpu_do_transaction_failed(CPUState *cs, hwaddr physaddr,
802                                      vaddr addr, unsigned size,
803                                      MMUAccessType access_type,
804                                      int mmu_idx, MemTxAttrs attrs,
805                                      MemTxResult response, uintptr_t retaddr)
806 {
807     RISCVCPU *cpu = RISCV_CPU(cs);
808     CPURISCVState *env = &cpu->env;
809 
810     if (access_type == MMU_DATA_STORE) {
811         cs->exception_index = RISCV_EXCP_STORE_AMO_ACCESS_FAULT;
812     } else if (access_type == MMU_DATA_LOAD) {
813         cs->exception_index = RISCV_EXCP_LOAD_ACCESS_FAULT;
814     } else {
815         cs->exception_index = RISCV_EXCP_INST_ACCESS_FAULT;
816     }
817 
818     env->badaddr = addr;
819     env->two_stage_lookup = riscv_cpu_virt_enabled(env) ||
820                             riscv_cpu_two_stage_lookup(mmu_idx);
821     riscv_raise_exception(&cpu->env, cs->exception_index, retaddr);
822 }
823 
824 void riscv_cpu_do_unaligned_access(CPUState *cs, vaddr addr,
825                                    MMUAccessType access_type, int mmu_idx,
826                                    uintptr_t retaddr)
827 {
828     RISCVCPU *cpu = RISCV_CPU(cs);
829     CPURISCVState *env = &cpu->env;
830     switch (access_type) {
831     case MMU_INST_FETCH:
832         cs->exception_index = RISCV_EXCP_INST_ADDR_MIS;
833         break;
834     case MMU_DATA_LOAD:
835         cs->exception_index = RISCV_EXCP_LOAD_ADDR_MIS;
836         break;
837     case MMU_DATA_STORE:
838         cs->exception_index = RISCV_EXCP_STORE_AMO_ADDR_MIS;
839         break;
840     default:
841         g_assert_not_reached();
842     }
843     env->badaddr = addr;
844     env->two_stage_lookup = riscv_cpu_virt_enabled(env) ||
845                             riscv_cpu_two_stage_lookup(mmu_idx);
846     riscv_raise_exception(env, cs->exception_index, retaddr);
847 }
848 
849 bool riscv_cpu_tlb_fill(CPUState *cs, vaddr address, int size,
850                         MMUAccessType access_type, int mmu_idx,
851                         bool probe, uintptr_t retaddr)
852 {
853     RISCVCPU *cpu = RISCV_CPU(cs);
854     CPURISCVState *env = &cpu->env;
855     vaddr im_address;
856     hwaddr pa = 0;
857     int prot, prot2, prot_pmp;
858     bool pmp_violation = false;
859     bool first_stage_error = true;
860     bool two_stage_lookup = false;
861     int ret = TRANSLATE_FAIL;
862     int mode = mmu_idx;
863     /* default TLB page size */
864     target_ulong tlb_size = TARGET_PAGE_SIZE;
865 
866     env->guest_phys_fault_addr = 0;
867 
868     qemu_log_mask(CPU_LOG_MMU, "%s ad %" VADDR_PRIx " rw %d mmu_idx %d\n",
869                   __func__, address, access_type, mmu_idx);
870 
871     /* MPRV does not affect the virtual-machine load/store
872        instructions, HLV, HLVX, and HSV. */
873     if (riscv_cpu_two_stage_lookup(mmu_idx)) {
874         mode = get_field(env->hstatus, HSTATUS_SPVP);
875     } else if (mode == PRV_M && access_type != MMU_INST_FETCH &&
876                get_field(env->mstatus, MSTATUS_MPRV)) {
877         mode = get_field(env->mstatus, MSTATUS_MPP);
878         if (riscv_has_ext(env, RVH) && get_field(env->mstatus, MSTATUS_MPV)) {
879             two_stage_lookup = true;
880         }
881     }
882 
883     if (riscv_cpu_virt_enabled(env) ||
884         ((riscv_cpu_two_stage_lookup(mmu_idx) || two_stage_lookup) &&
885          access_type != MMU_INST_FETCH)) {
886         /* Two stage lookup */
887         ret = get_physical_address(env, &pa, &prot, address,
888                                    &env->guest_phys_fault_addr, access_type,
889                                    mmu_idx, true, true, false);
890 
891         /*
892          * A G-stage exception may be triggered during two state lookup.
893          * And the env->guest_phys_fault_addr has already been set in
894          * get_physical_address().
895          */
896         if (ret == TRANSLATE_G_STAGE_FAIL) {
897             first_stage_error = false;
898             access_type = MMU_DATA_LOAD;
899         }
900 
901         qemu_log_mask(CPU_LOG_MMU,
902                       "%s 1st-stage address=%" VADDR_PRIx " ret %d physical "
903                       TARGET_FMT_plx " prot %d\n",
904                       __func__, address, ret, pa, prot);
905 
906         if (ret == TRANSLATE_SUCCESS) {
907             /* Second stage lookup */
908             im_address = pa;
909 
910             ret = get_physical_address(env, &pa, &prot2, im_address, NULL,
911                                        access_type, mmu_idx, false, true,
912                                        false);
913 
914             qemu_log_mask(CPU_LOG_MMU,
915                     "%s 2nd-stage address=%" VADDR_PRIx " ret %d physical "
916                     TARGET_FMT_plx " prot %d\n",
917                     __func__, im_address, ret, pa, prot2);
918 
919             prot &= prot2;
920 
921             if (ret == TRANSLATE_SUCCESS) {
922                 ret = get_physical_address_pmp(env, &prot_pmp, &tlb_size, pa,
923                                                size, access_type, mode);
924 
925                 qemu_log_mask(CPU_LOG_MMU,
926                               "%s PMP address=" TARGET_FMT_plx " ret %d prot"
927                               " %d tlb_size " TARGET_FMT_lu "\n",
928                               __func__, pa, ret, prot_pmp, tlb_size);
929 
930                 prot &= prot_pmp;
931             }
932 
933             if (ret != TRANSLATE_SUCCESS) {
934                 /*
935                  * Guest physical address translation failed, this is a HS
936                  * level exception
937                  */
938                 first_stage_error = false;
939                 env->guest_phys_fault_addr = (im_address |
940                                               (address &
941                                                (TARGET_PAGE_SIZE - 1))) >> 2;
942             }
943         }
944     } else {
945         /* Single stage lookup */
946         ret = get_physical_address(env, &pa, &prot, address, NULL,
947                                    access_type, mmu_idx, true, false, false);
948 
949         qemu_log_mask(CPU_LOG_MMU,
950                       "%s address=%" VADDR_PRIx " ret %d physical "
951                       TARGET_FMT_plx " prot %d\n",
952                       __func__, address, ret, pa, prot);
953 
954         if (ret == TRANSLATE_SUCCESS) {
955             ret = get_physical_address_pmp(env, &prot_pmp, &tlb_size, pa,
956                                            size, access_type, mode);
957 
958             qemu_log_mask(CPU_LOG_MMU,
959                           "%s PMP address=" TARGET_FMT_plx " ret %d prot"
960                           " %d tlb_size " TARGET_FMT_lu "\n",
961                           __func__, pa, ret, prot_pmp, tlb_size);
962 
963             prot &= prot_pmp;
964         }
965     }
966 
967     if (ret == TRANSLATE_PMP_FAIL) {
968         pmp_violation = true;
969     }
970 
971     if (ret == TRANSLATE_SUCCESS) {
972         tlb_set_page(cs, address & ~(tlb_size - 1), pa & ~(tlb_size - 1),
973                      prot, mmu_idx, tlb_size);
974         return true;
975     } else if (probe) {
976         return false;
977     } else {
978         raise_mmu_exception(env, address, access_type, pmp_violation,
979                             first_stage_error,
980                             riscv_cpu_virt_enabled(env) ||
981                                 riscv_cpu_two_stage_lookup(mmu_idx));
982         riscv_raise_exception(env, cs->exception_index, retaddr);
983     }
984 
985     return true;
986 }
987 #endif /* !CONFIG_USER_ONLY */
988 
989 /*
990  * Handle Traps
991  *
992  * Adapted from Spike's processor_t::take_trap.
993  *
994  */
995 void riscv_cpu_do_interrupt(CPUState *cs)
996 {
997 #if !defined(CONFIG_USER_ONLY)
998 
999     RISCVCPU *cpu = RISCV_CPU(cs);
1000     CPURISCVState *env = &cpu->env;
1001     bool write_gva = false;
1002     uint64_t s;
1003 
1004     /* cs->exception is 32-bits wide unlike mcause which is XLEN-bits wide
1005      * so we mask off the MSB and separate into trap type and cause.
1006      */
1007     bool async = !!(cs->exception_index & RISCV_EXCP_INT_FLAG);
1008     target_ulong cause = cs->exception_index & RISCV_EXCP_INT_MASK;
1009     target_ulong deleg = async ? env->mideleg : env->medeleg;
1010     target_ulong tval = 0;
1011     target_ulong htval = 0;
1012     target_ulong mtval2 = 0;
1013 
1014     if  (cause == RISCV_EXCP_SEMIHOST) {
1015         if (env->priv >= PRV_S) {
1016             env->gpr[xA0] = do_common_semihosting(cs);
1017             env->pc += 4;
1018             return;
1019         }
1020         cause = RISCV_EXCP_BREAKPOINT;
1021     }
1022 
1023     if (!async) {
1024         /* set tval to badaddr for traps with address information */
1025         switch (cause) {
1026         case RISCV_EXCP_INST_GUEST_PAGE_FAULT:
1027         case RISCV_EXCP_LOAD_GUEST_ACCESS_FAULT:
1028         case RISCV_EXCP_STORE_GUEST_AMO_ACCESS_FAULT:
1029         case RISCV_EXCP_INST_ADDR_MIS:
1030         case RISCV_EXCP_INST_ACCESS_FAULT:
1031         case RISCV_EXCP_LOAD_ADDR_MIS:
1032         case RISCV_EXCP_STORE_AMO_ADDR_MIS:
1033         case RISCV_EXCP_LOAD_ACCESS_FAULT:
1034         case RISCV_EXCP_STORE_AMO_ACCESS_FAULT:
1035         case RISCV_EXCP_INST_PAGE_FAULT:
1036         case RISCV_EXCP_LOAD_PAGE_FAULT:
1037         case RISCV_EXCP_STORE_PAGE_FAULT:
1038             write_gva = true;
1039             tval = env->badaddr;
1040             break;
1041         case RISCV_EXCP_ILLEGAL_INST:
1042             tval = env->bins;
1043             break;
1044         default:
1045             break;
1046         }
1047         /* ecall is dispatched as one cause so translate based on mode */
1048         if (cause == RISCV_EXCP_U_ECALL) {
1049             assert(env->priv <= 3);
1050 
1051             if (env->priv == PRV_M) {
1052                 cause = RISCV_EXCP_M_ECALL;
1053             } else if (env->priv == PRV_S && riscv_cpu_virt_enabled(env)) {
1054                 cause = RISCV_EXCP_VS_ECALL;
1055             } else if (env->priv == PRV_S && !riscv_cpu_virt_enabled(env)) {
1056                 cause = RISCV_EXCP_S_ECALL;
1057             } else if (env->priv == PRV_U) {
1058                 cause = RISCV_EXCP_U_ECALL;
1059             }
1060         }
1061     }
1062 
1063     trace_riscv_trap(env->mhartid, async, cause, env->pc, tval,
1064                      riscv_cpu_get_trap_name(cause, async));
1065 
1066     qemu_log_mask(CPU_LOG_INT,
1067                   "%s: hart:"TARGET_FMT_ld", async:%d, cause:"TARGET_FMT_lx", "
1068                   "epc:0x"TARGET_FMT_lx", tval:0x"TARGET_FMT_lx", desc=%s\n",
1069                   __func__, env->mhartid, async, cause, env->pc, tval,
1070                   riscv_cpu_get_trap_name(cause, async));
1071 
1072     if (env->priv <= PRV_S &&
1073             cause < TARGET_LONG_BITS && ((deleg >> cause) & 1)) {
1074         /* handle the trap in S-mode */
1075         if (riscv_has_ext(env, RVH)) {
1076             target_ulong hdeleg = async ? env->hideleg : env->hedeleg;
1077 
1078             if (riscv_cpu_virt_enabled(env) && ((hdeleg >> cause) & 1)) {
1079                 /* Trap to VS mode */
1080                 /*
1081                  * See if we need to adjust cause. Yes if its VS mode interrupt
1082                  * no if hypervisor has delegated one of hs mode's interrupt
1083                  */
1084                 if (cause == IRQ_VS_TIMER || cause == IRQ_VS_SOFT ||
1085                     cause == IRQ_VS_EXT) {
1086                     cause = cause - 1;
1087                 }
1088                 write_gva = false;
1089             } else if (riscv_cpu_virt_enabled(env)) {
1090                 /* Trap into HS mode, from virt */
1091                 riscv_cpu_swap_hypervisor_regs(env);
1092                 env->hstatus = set_field(env->hstatus, HSTATUS_SPVP,
1093                                          env->priv);
1094                 env->hstatus = set_field(env->hstatus, HSTATUS_SPV,
1095                                          riscv_cpu_virt_enabled(env));
1096 
1097 
1098                 htval = env->guest_phys_fault_addr;
1099 
1100                 riscv_cpu_set_virt_enabled(env, 0);
1101             } else {
1102                 /* Trap into HS mode */
1103                 env->hstatus = set_field(env->hstatus, HSTATUS_SPV, false);
1104                 htval = env->guest_phys_fault_addr;
1105                 write_gva = false;
1106             }
1107             env->hstatus = set_field(env->hstatus, HSTATUS_GVA, write_gva);
1108         }
1109 
1110         s = env->mstatus;
1111         s = set_field(s, MSTATUS_SPIE, get_field(s, MSTATUS_SIE));
1112         s = set_field(s, MSTATUS_SPP, env->priv);
1113         s = set_field(s, MSTATUS_SIE, 0);
1114         env->mstatus = s;
1115         env->scause = cause | ((target_ulong)async << (TARGET_LONG_BITS - 1));
1116         env->sepc = env->pc;
1117         env->stval = tval;
1118         env->htval = htval;
1119         env->pc = (env->stvec >> 2 << 2) +
1120             ((async && (env->stvec & 3) == 1) ? cause * 4 : 0);
1121         riscv_cpu_set_mode(env, PRV_S);
1122     } else {
1123         /* handle the trap in M-mode */
1124         if (riscv_has_ext(env, RVH)) {
1125             if (riscv_cpu_virt_enabled(env)) {
1126                 riscv_cpu_swap_hypervisor_regs(env);
1127             }
1128             env->mstatus = set_field(env->mstatus, MSTATUS_MPV,
1129                                      riscv_cpu_virt_enabled(env));
1130             if (riscv_cpu_virt_enabled(env) && tval) {
1131                 env->mstatus = set_field(env->mstatus, MSTATUS_GVA, 1);
1132             }
1133 
1134             mtval2 = env->guest_phys_fault_addr;
1135 
1136             /* Trapping to M mode, virt is disabled */
1137             riscv_cpu_set_virt_enabled(env, 0);
1138         }
1139 
1140         s = env->mstatus;
1141         s = set_field(s, MSTATUS_MPIE, get_field(s, MSTATUS_MIE));
1142         s = set_field(s, MSTATUS_MPP, env->priv);
1143         s = set_field(s, MSTATUS_MIE, 0);
1144         env->mstatus = s;
1145         env->mcause = cause | ~(((target_ulong)-1) >> async);
1146         env->mepc = env->pc;
1147         env->mtval = tval;
1148         env->mtval2 = mtval2;
1149         env->pc = (env->mtvec >> 2 << 2) +
1150             ((async && (env->mtvec & 3) == 1) ? cause * 4 : 0);
1151         riscv_cpu_set_mode(env, PRV_M);
1152     }
1153 
1154     /* NOTE: it is not necessary to yield load reservations here. It is only
1155      * necessary for an SC from "another hart" to cause a load reservation
1156      * to be yielded. Refer to the memory consistency model section of the
1157      * RISC-V ISA Specification.
1158      */
1159 
1160     env->two_stage_lookup = false;
1161 #endif
1162     cs->exception_index = RISCV_EXCP_NONE; /* mark handled to qemu */
1163 }
1164