xref: /openbmc/qemu/target/ppc/mmu-hash64.c (revision f58f084e71c6aeee066a30fc87422820e94a6cfc)
1 /*
2  *  PowerPC MMU, TLB, SLB and BAT emulation helpers for QEMU.
3  *
4  *  Copyright (c) 2003-2007 Jocelyn Mayer
5  *  Copyright (c) 2013 David Gibson, IBM Corporation
6  *
7  * This library is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2 of the License, or (at your option) any later version.
11  *
12  * This library is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
19  */
20 #include "qemu/osdep.h"
21 #include "qemu/units.h"
22 #include "cpu.h"
23 #include "exec/exec-all.h"
24 #include "exec/helper-proto.h"
25 #include "qemu/error-report.h"
26 #include "qemu/qemu-print.h"
27 #include "sysemu/hw_accel.h"
28 #include "kvm_ppc.h"
29 #include "mmu-hash64.h"
30 #include "exec/log.h"
31 #include "hw/hw.h"
32 #include "mmu-book3s-v3.h"
33 
34 /* #define DEBUG_SLB */
35 
36 #ifdef DEBUG_SLB
37 #  define LOG_SLB(...) qemu_log_mask(CPU_LOG_MMU, __VA_ARGS__)
38 #else
39 #  define LOG_SLB(...) do { } while (0)
40 #endif
41 
42 /*
43  * SLB handling
44  */
45 
46 static ppc_slb_t *slb_lookup(PowerPCCPU *cpu, target_ulong eaddr)
47 {
48     CPUPPCState *env = &cpu->env;
49     uint64_t esid_256M, esid_1T;
50     int n;
51 
52     LOG_SLB("%s: eaddr " TARGET_FMT_lx "\n", __func__, eaddr);
53 
54     esid_256M = (eaddr & SEGMENT_MASK_256M) | SLB_ESID_V;
55     esid_1T = (eaddr & SEGMENT_MASK_1T) | SLB_ESID_V;
56 
57     for (n = 0; n < cpu->hash64_opts->slb_size; n++) {
58         ppc_slb_t *slb = &env->slb[n];
59 
60         LOG_SLB("%s: slot %d %016" PRIx64 " %016"
61                     PRIx64 "\n", __func__, n, slb->esid, slb->vsid);
62         /*
63          * We check for 1T matches on all MMUs here - if the MMU
64          * doesn't have 1T segment support, we will have prevented 1T
65          * entries from being inserted in the slbmte code.
66          */
67         if (((slb->esid == esid_256M) &&
68              ((slb->vsid & SLB_VSID_B) == SLB_VSID_B_256M))
69             || ((slb->esid == esid_1T) &&
70                 ((slb->vsid & SLB_VSID_B) == SLB_VSID_B_1T))) {
71             return slb;
72         }
73     }
74 
75     return NULL;
76 }
77 
78 void dump_slb(PowerPCCPU *cpu)
79 {
80     CPUPPCState *env = &cpu->env;
81     int i;
82     uint64_t slbe, slbv;
83 
84     cpu_synchronize_state(CPU(cpu));
85 
86     qemu_printf("SLB\tESID\t\t\tVSID\n");
87     for (i = 0; i < cpu->hash64_opts->slb_size; i++) {
88         slbe = env->slb[i].esid;
89         slbv = env->slb[i].vsid;
90         if (slbe == 0 && slbv == 0) {
91             continue;
92         }
93         qemu_printf("%d\t0x%016" PRIx64 "\t0x%016" PRIx64 "\n",
94                     i, slbe, slbv);
95     }
96 }
97 
98 void helper_slbia(CPUPPCState *env)
99 {
100     PowerPCCPU *cpu = env_archcpu(env);
101     int n;
102 
103     /* XXX: Warning: slbia never invalidates the first segment */
104     for (n = 1; n < cpu->hash64_opts->slb_size; n++) {
105         ppc_slb_t *slb = &env->slb[n];
106 
107         if (slb->esid & SLB_ESID_V) {
108             slb->esid &= ~SLB_ESID_V;
109             /*
110              * XXX: given the fact that segment size is 256 MB or 1TB,
111              *      and we still don't have a tlb_flush_mask(env, n, mask)
112              *      in QEMU, we just invalidate all TLBs
113              */
114             env->tlb_need_flush |= TLB_NEED_LOCAL_FLUSH;
115         }
116     }
117 }
118 
119 static void __helper_slbie(CPUPPCState *env, target_ulong addr,
120                            target_ulong global)
121 {
122     PowerPCCPU *cpu = env_archcpu(env);
123     ppc_slb_t *slb;
124 
125     slb = slb_lookup(cpu, addr);
126     if (!slb) {
127         return;
128     }
129 
130     if (slb->esid & SLB_ESID_V) {
131         slb->esid &= ~SLB_ESID_V;
132 
133         /*
134          * XXX: given the fact that segment size is 256 MB or 1TB,
135          *      and we still don't have a tlb_flush_mask(env, n, mask)
136          *      in QEMU, we just invalidate all TLBs
137          */
138         env->tlb_need_flush |=
139             (global == false ? TLB_NEED_LOCAL_FLUSH : TLB_NEED_GLOBAL_FLUSH);
140     }
141 }
142 
143 void helper_slbie(CPUPPCState *env, target_ulong addr)
144 {
145     __helper_slbie(env, addr, false);
146 }
147 
148 void helper_slbieg(CPUPPCState *env, target_ulong addr)
149 {
150     __helper_slbie(env, addr, true);
151 }
152 
153 int ppc_store_slb(PowerPCCPU *cpu, target_ulong slot,
154                   target_ulong esid, target_ulong vsid)
155 {
156     CPUPPCState *env = &cpu->env;
157     ppc_slb_t *slb = &env->slb[slot];
158     const PPCHash64SegmentPageSizes *sps = NULL;
159     int i;
160 
161     if (slot >= cpu->hash64_opts->slb_size) {
162         return -1; /* Bad slot number */
163     }
164     if (esid & ~(SLB_ESID_ESID | SLB_ESID_V)) {
165         return -1; /* Reserved bits set */
166     }
167     if (vsid & (SLB_VSID_B & ~SLB_VSID_B_1T)) {
168         return -1; /* Bad segment size */
169     }
170     if ((vsid & SLB_VSID_B) && !(ppc_hash64_has(cpu, PPC_HASH64_1TSEG))) {
171         return -1; /* 1T segment on MMU that doesn't support it */
172     }
173 
174     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
175         const PPCHash64SegmentPageSizes *sps1 = &cpu->hash64_opts->sps[i];
176 
177         if (!sps1->page_shift) {
178             break;
179         }
180 
181         if ((vsid & SLB_VSID_LLP_MASK) == sps1->slb_enc) {
182             sps = sps1;
183             break;
184         }
185     }
186 
187     if (!sps) {
188         error_report("Bad page size encoding in SLB store: slot "TARGET_FMT_lu
189                      " esid 0x"TARGET_FMT_lx" vsid 0x"TARGET_FMT_lx,
190                      slot, esid, vsid);
191         return -1;
192     }
193 
194     slb->esid = esid;
195     slb->vsid = vsid;
196     slb->sps = sps;
197 
198     LOG_SLB("%s: " TARGET_FMT_lu " " TARGET_FMT_lx " - " TARGET_FMT_lx
199             " => %016" PRIx64 " %016" PRIx64 "\n", __func__, slot, esid, vsid,
200             slb->esid, slb->vsid);
201 
202     return 0;
203 }
204 
205 static int ppc_load_slb_esid(PowerPCCPU *cpu, target_ulong rb,
206                              target_ulong *rt)
207 {
208     CPUPPCState *env = &cpu->env;
209     int slot = rb & 0xfff;
210     ppc_slb_t *slb = &env->slb[slot];
211 
212     if (slot >= cpu->hash64_opts->slb_size) {
213         return -1;
214     }
215 
216     *rt = slb->esid;
217     return 0;
218 }
219 
220 static int ppc_load_slb_vsid(PowerPCCPU *cpu, target_ulong rb,
221                              target_ulong *rt)
222 {
223     CPUPPCState *env = &cpu->env;
224     int slot = rb & 0xfff;
225     ppc_slb_t *slb = &env->slb[slot];
226 
227     if (slot >= cpu->hash64_opts->slb_size) {
228         return -1;
229     }
230 
231     *rt = slb->vsid;
232     return 0;
233 }
234 
235 static int ppc_find_slb_vsid(PowerPCCPU *cpu, target_ulong rb,
236                              target_ulong *rt)
237 {
238     CPUPPCState *env = &cpu->env;
239     ppc_slb_t *slb;
240 
241     if (!msr_is_64bit(env, env->msr)) {
242         rb &= 0xffffffff;
243     }
244     slb = slb_lookup(cpu, rb);
245     if (slb == NULL) {
246         *rt = (target_ulong)-1ul;
247     } else {
248         *rt = slb->vsid;
249     }
250     return 0;
251 }
252 
253 void helper_store_slb(CPUPPCState *env, target_ulong rb, target_ulong rs)
254 {
255     PowerPCCPU *cpu = env_archcpu(env);
256 
257     if (ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs) < 0) {
258         raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
259                                POWERPC_EXCP_INVAL, GETPC());
260     }
261 }
262 
263 target_ulong helper_load_slb_esid(CPUPPCState *env, target_ulong rb)
264 {
265     PowerPCCPU *cpu = env_archcpu(env);
266     target_ulong rt = 0;
267 
268     if (ppc_load_slb_esid(cpu, rb, &rt) < 0) {
269         raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
270                                POWERPC_EXCP_INVAL, GETPC());
271     }
272     return rt;
273 }
274 
275 target_ulong helper_find_slb_vsid(CPUPPCState *env, target_ulong rb)
276 {
277     PowerPCCPU *cpu = env_archcpu(env);
278     target_ulong rt = 0;
279 
280     if (ppc_find_slb_vsid(cpu, rb, &rt) < 0) {
281         raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
282                                POWERPC_EXCP_INVAL, GETPC());
283     }
284     return rt;
285 }
286 
287 target_ulong helper_load_slb_vsid(CPUPPCState *env, target_ulong rb)
288 {
289     PowerPCCPU *cpu = env_archcpu(env);
290     target_ulong rt = 0;
291 
292     if (ppc_load_slb_vsid(cpu, rb, &rt) < 0) {
293         raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
294                                POWERPC_EXCP_INVAL, GETPC());
295     }
296     return rt;
297 }
298 
299 /* Check No-Execute or Guarded Storage */
300 static inline int ppc_hash64_pte_noexec_guard(PowerPCCPU *cpu,
301                                               ppc_hash_pte64_t pte)
302 {
303     /* Exec permissions CANNOT take away read or write permissions */
304     return (pte.pte1 & HPTE64_R_N) || (pte.pte1 & HPTE64_R_G) ?
305             PAGE_READ | PAGE_WRITE : PAGE_READ | PAGE_WRITE | PAGE_EXEC;
306 }
307 
308 /* Check Basic Storage Protection */
309 static int ppc_hash64_pte_prot(PowerPCCPU *cpu,
310                                ppc_slb_t *slb, ppc_hash_pte64_t pte)
311 {
312     CPUPPCState *env = &cpu->env;
313     unsigned pp, key;
314     /*
315      * Some pp bit combinations have undefined behaviour, so default
316      * to no access in those cases
317      */
318     int prot = 0;
319 
320     key = !!(msr_pr ? (slb->vsid & SLB_VSID_KP)
321              : (slb->vsid & SLB_VSID_KS));
322     pp = (pte.pte1 & HPTE64_R_PP) | ((pte.pte1 & HPTE64_R_PP0) >> 61);
323 
324     if (key == 0) {
325         switch (pp) {
326         case 0x0:
327         case 0x1:
328         case 0x2:
329             prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
330             break;
331 
332         case 0x3:
333         case 0x6:
334             prot = PAGE_READ | PAGE_EXEC;
335             break;
336         }
337     } else {
338         switch (pp) {
339         case 0x0:
340         case 0x6:
341             break;
342 
343         case 0x1:
344         case 0x3:
345             prot = PAGE_READ | PAGE_EXEC;
346             break;
347 
348         case 0x2:
349             prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
350             break;
351         }
352     }
353 
354     return prot;
355 }
356 
357 /* Check the instruction access permissions specified in the IAMR */
358 static int ppc_hash64_iamr_prot(PowerPCCPU *cpu, int key)
359 {
360     CPUPPCState *env = &cpu->env;
361     int iamr_bits = (env->spr[SPR_IAMR] >> 2 * (31 - key)) & 0x3;
362 
363     /*
364      * An instruction fetch is permitted if the IAMR bit is 0.
365      * If the bit is set, return PAGE_READ | PAGE_WRITE because this bit
366      * can only take away EXEC permissions not READ or WRITE permissions.
367      * If bit is cleared return PAGE_READ | PAGE_WRITE | PAGE_EXEC since
368      * EXEC permissions are allowed.
369      */
370     return (iamr_bits & 0x1) ? PAGE_READ | PAGE_WRITE :
371                                PAGE_READ | PAGE_WRITE | PAGE_EXEC;
372 }
373 
374 static int ppc_hash64_amr_prot(PowerPCCPU *cpu, ppc_hash_pte64_t pte)
375 {
376     CPUPPCState *env = &cpu->env;
377     int key, amrbits;
378     int prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
379 
380     /* Only recent MMUs implement Virtual Page Class Key Protection */
381     if (!ppc_hash64_has(cpu, PPC_HASH64_AMR)) {
382         return prot;
383     }
384 
385     key = HPTE64_R_KEY(pte.pte1);
386     amrbits = (env->spr[SPR_AMR] >> 2 * (31 - key)) & 0x3;
387 
388     /* fprintf(stderr, "AMR protection: key=%d AMR=0x%" PRIx64 "\n", key, */
389     /*         env->spr[SPR_AMR]); */
390 
391     /*
392      * A store is permitted if the AMR bit is 0. Remove write
393      * protection if it is set.
394      */
395     if (amrbits & 0x2) {
396         prot &= ~PAGE_WRITE;
397     }
398     /*
399      * A load is permitted if the AMR bit is 0. Remove read
400      * protection if it is set.
401      */
402     if (amrbits & 0x1) {
403         prot &= ~PAGE_READ;
404     }
405 
406     switch (env->mmu_model) {
407     /*
408      * MMU version 2.07 and later support IAMR
409      * Check if the IAMR allows the instruction access - it will return
410      * PAGE_EXEC if it doesn't (and thus that bit will be cleared) or 0
411      * if it does (and prot will be unchanged indicating execution support).
412      */
413     case POWERPC_MMU_2_07:
414     case POWERPC_MMU_3_00:
415         prot &= ppc_hash64_iamr_prot(cpu, key);
416         break;
417     default:
418         break;
419     }
420 
421     return prot;
422 }
423 
424 const ppc_hash_pte64_t *ppc_hash64_map_hptes(PowerPCCPU *cpu,
425                                              hwaddr ptex, int n)
426 {
427     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
428     hwaddr base;
429     hwaddr plen = n * HASH_PTE_SIZE_64;
430     const ppc_hash_pte64_t *hptes;
431 
432     if (cpu->vhyp) {
433         PPCVirtualHypervisorClass *vhc =
434             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
435         return vhc->map_hptes(cpu->vhyp, ptex, n);
436     }
437     base = ppc_hash64_hpt_base(cpu);
438 
439     if (!base) {
440         return NULL;
441     }
442 
443     hptes = address_space_map(CPU(cpu)->as, base + pte_offset, &plen, false,
444                               MEMTXATTRS_UNSPECIFIED);
445     if (plen < (n * HASH_PTE_SIZE_64)) {
446         hw_error("%s: Unable to map all requested HPTEs\n", __func__);
447     }
448     return hptes;
449 }
450 
451 void ppc_hash64_unmap_hptes(PowerPCCPU *cpu, const ppc_hash_pte64_t *hptes,
452                             hwaddr ptex, int n)
453 {
454     if (cpu->vhyp) {
455         PPCVirtualHypervisorClass *vhc =
456             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
457         vhc->unmap_hptes(cpu->vhyp, hptes, ptex, n);
458         return;
459     }
460 
461     address_space_unmap(CPU(cpu)->as, (void *)hptes, n * HASH_PTE_SIZE_64,
462                         false, n * HASH_PTE_SIZE_64);
463 }
464 
465 static unsigned hpte_page_shift(const PPCHash64SegmentPageSizes *sps,
466                                 uint64_t pte0, uint64_t pte1)
467 {
468     int i;
469 
470     if (!(pte0 & HPTE64_V_LARGE)) {
471         if (sps->page_shift != 12) {
472             /* 4kiB page in a non 4kiB segment */
473             return 0;
474         }
475         /* Normal 4kiB page */
476         return 12;
477     }
478 
479     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
480         const PPCHash64PageSize *ps = &sps->enc[i];
481         uint64_t mask;
482 
483         if (!ps->page_shift) {
484             break;
485         }
486 
487         if (ps->page_shift == 12) {
488             /* L bit is set so this can't be a 4kiB page */
489             continue;
490         }
491 
492         mask = ((1ULL << ps->page_shift) - 1) & HPTE64_R_RPN;
493 
494         if ((pte1 & mask) == ((uint64_t)ps->pte_enc << HPTE64_R_RPN_SHIFT)) {
495             return ps->page_shift;
496         }
497     }
498 
499     return 0; /* Bad page size encoding */
500 }
501 
502 static void ppc64_v3_new_to_old_hpte(target_ulong *pte0, target_ulong *pte1)
503 {
504     /* Insert B into pte0 */
505     *pte0 = (*pte0 & HPTE64_V_COMMON_BITS) |
506             ((*pte1 & HPTE64_R_3_0_SSIZE_MASK) <<
507              (HPTE64_V_SSIZE_SHIFT - HPTE64_R_3_0_SSIZE_SHIFT));
508 
509     /* Remove B from pte1 */
510     *pte1 = *pte1 & ~HPTE64_R_3_0_SSIZE_MASK;
511 }
512 
513 
514 static hwaddr ppc_hash64_pteg_search(PowerPCCPU *cpu, hwaddr hash,
515                                      const PPCHash64SegmentPageSizes *sps,
516                                      target_ulong ptem,
517                                      ppc_hash_pte64_t *pte, unsigned *pshift)
518 {
519     int i;
520     const ppc_hash_pte64_t *pteg;
521     target_ulong pte0, pte1;
522     target_ulong ptex;
523 
524     ptex = (hash & ppc_hash64_hpt_mask(cpu)) * HPTES_PER_GROUP;
525     pteg = ppc_hash64_map_hptes(cpu, ptex, HPTES_PER_GROUP);
526     if (!pteg) {
527         return -1;
528     }
529     for (i = 0; i < HPTES_PER_GROUP; i++) {
530         pte0 = ppc_hash64_hpte0(cpu, pteg, i);
531         /*
532          * pte0 contains the valid bit and must be read before pte1,
533          * otherwise we might see an old pte1 with a new valid bit and
534          * thus an inconsistent hpte value
535          */
536         smp_rmb();
537         pte1 = ppc_hash64_hpte1(cpu, pteg, i);
538 
539         /* Convert format if necessary */
540         if (cpu->env.mmu_model == POWERPC_MMU_3_00 && !cpu->vhyp) {
541             ppc64_v3_new_to_old_hpte(&pte0, &pte1);
542         }
543 
544         /* This compares V, B, H (secondary) and the AVPN */
545         if (HPTE64_V_COMPARE(pte0, ptem)) {
546             *pshift = hpte_page_shift(sps, pte0, pte1);
547             /*
548              * If there is no match, ignore the PTE, it could simply
549              * be for a different segment size encoding and the
550              * architecture specifies we should not match. Linux will
551              * potentially leave behind PTEs for the wrong base page
552              * size when demoting segments.
553              */
554             if (*pshift == 0) {
555                 continue;
556             }
557             /*
558              * We don't do anything with pshift yet as qemu TLB only
559              * deals with 4K pages anyway
560              */
561             pte->pte0 = pte0;
562             pte->pte1 = pte1;
563             ppc_hash64_unmap_hptes(cpu, pteg, ptex, HPTES_PER_GROUP);
564             return ptex + i;
565         }
566     }
567     ppc_hash64_unmap_hptes(cpu, pteg, ptex, HPTES_PER_GROUP);
568     /*
569      * We didn't find a valid entry.
570      */
571     return -1;
572 }
573 
574 static hwaddr ppc_hash64_htab_lookup(PowerPCCPU *cpu,
575                                      ppc_slb_t *slb, target_ulong eaddr,
576                                      ppc_hash_pte64_t *pte, unsigned *pshift)
577 {
578     CPUPPCState *env = &cpu->env;
579     hwaddr hash, ptex;
580     uint64_t vsid, epnmask, epn, ptem;
581     const PPCHash64SegmentPageSizes *sps = slb->sps;
582 
583     /*
584      * The SLB store path should prevent any bad page size encodings
585      * getting in there, so:
586      */
587     assert(sps);
588 
589     /* If ISL is set in LPCR we need to clamp the page size to 4K */
590     if (env->spr[SPR_LPCR] & LPCR_ISL) {
591         /* We assume that when using TCG, 4k is first entry of SPS */
592         sps = &cpu->hash64_opts->sps[0];
593         assert(sps->page_shift == 12);
594     }
595 
596     epnmask = ~((1ULL << sps->page_shift) - 1);
597 
598     if (slb->vsid & SLB_VSID_B) {
599         /* 1TB segment */
600         vsid = (slb->vsid & SLB_VSID_VSID) >> SLB_VSID_SHIFT_1T;
601         epn = (eaddr & ~SEGMENT_MASK_1T) & epnmask;
602         hash = vsid ^ (vsid << 25) ^ (epn >> sps->page_shift);
603     } else {
604         /* 256M segment */
605         vsid = (slb->vsid & SLB_VSID_VSID) >> SLB_VSID_SHIFT;
606         epn = (eaddr & ~SEGMENT_MASK_256M) & epnmask;
607         hash = vsid ^ (epn >> sps->page_shift);
608     }
609     ptem = (slb->vsid & SLB_VSID_PTEM) | ((epn >> 16) & HPTE64_V_AVPN);
610     ptem |= HPTE64_V_VALID;
611 
612     /* Page address translation */
613     qemu_log_mask(CPU_LOG_MMU,
614             "htab_base " TARGET_FMT_plx " htab_mask " TARGET_FMT_plx
615             " hash " TARGET_FMT_plx "\n",
616             ppc_hash64_hpt_base(cpu), ppc_hash64_hpt_mask(cpu), hash);
617 
618     /* Primary PTEG lookup */
619     qemu_log_mask(CPU_LOG_MMU,
620             "0 htab=" TARGET_FMT_plx "/" TARGET_FMT_plx
621             " vsid=" TARGET_FMT_lx " ptem=" TARGET_FMT_lx
622             " hash=" TARGET_FMT_plx "\n",
623             ppc_hash64_hpt_base(cpu), ppc_hash64_hpt_mask(cpu),
624             vsid, ptem,  hash);
625     ptex = ppc_hash64_pteg_search(cpu, hash, sps, ptem, pte, pshift);
626 
627     if (ptex == -1) {
628         /* Secondary PTEG lookup */
629         ptem |= HPTE64_V_SECONDARY;
630         qemu_log_mask(CPU_LOG_MMU,
631                 "1 htab=" TARGET_FMT_plx "/" TARGET_FMT_plx
632                 " vsid=" TARGET_FMT_lx " api=" TARGET_FMT_lx
633                 " hash=" TARGET_FMT_plx "\n", ppc_hash64_hpt_base(cpu),
634                 ppc_hash64_hpt_mask(cpu), vsid, ptem, ~hash);
635 
636         ptex = ppc_hash64_pteg_search(cpu, ~hash, sps, ptem, pte, pshift);
637     }
638 
639     return ptex;
640 }
641 
642 unsigned ppc_hash64_hpte_page_shift_noslb(PowerPCCPU *cpu,
643                                           uint64_t pte0, uint64_t pte1)
644 {
645     int i;
646 
647     if (!(pte0 & HPTE64_V_LARGE)) {
648         return 12;
649     }
650 
651     /*
652      * The encodings in env->sps need to be carefully chosen so that
653      * this gives an unambiguous result.
654      */
655     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
656         const PPCHash64SegmentPageSizes *sps = &cpu->hash64_opts->sps[i];
657         unsigned shift;
658 
659         if (!sps->page_shift) {
660             break;
661         }
662 
663         shift = hpte_page_shift(sps, pte0, pte1);
664         if (shift) {
665             return shift;
666         }
667     }
668 
669     return 0;
670 }
671 
672 static bool ppc_hash64_use_vrma(CPUPPCState *env)
673 {
674     switch (env->mmu_model) {
675     case POWERPC_MMU_3_00:
676         /*
677          * ISAv3.0 (POWER9) always uses VRMA, the VPM0 field and RMOR
678          * register no longer exist
679          */
680         return true;
681 
682     default:
683         return !!(env->spr[SPR_LPCR] & LPCR_VPM0);
684     }
685 }
686 
687 static void ppc_hash64_set_isi(CPUState *cs, uint64_t error_code)
688 {
689     CPUPPCState *env = &POWERPC_CPU(cs)->env;
690     bool vpm;
691 
692     if (msr_ir) {
693         vpm = !!(env->spr[SPR_LPCR] & LPCR_VPM1);
694     } else {
695         vpm = ppc_hash64_use_vrma(env);
696     }
697     if (vpm && !msr_hv) {
698         cs->exception_index = POWERPC_EXCP_HISI;
699     } else {
700         cs->exception_index = POWERPC_EXCP_ISI;
701     }
702     env->error_code = error_code;
703 }
704 
705 static void ppc_hash64_set_dsi(CPUState *cs, uint64_t dar, uint64_t dsisr)
706 {
707     CPUPPCState *env = &POWERPC_CPU(cs)->env;
708     bool vpm;
709 
710     if (msr_dr) {
711         vpm = !!(env->spr[SPR_LPCR] & LPCR_VPM1);
712     } else {
713         vpm = ppc_hash64_use_vrma(env);
714     }
715     if (vpm && !msr_hv) {
716         cs->exception_index = POWERPC_EXCP_HDSI;
717         env->spr[SPR_HDAR] = dar;
718         env->spr[SPR_HDSISR] = dsisr;
719     } else {
720         cs->exception_index = POWERPC_EXCP_DSI;
721         env->spr[SPR_DAR] = dar;
722         env->spr[SPR_DSISR] = dsisr;
723    }
724     env->error_code = 0;
725 }
726 
727 
728 static void ppc_hash64_set_r(PowerPCCPU *cpu, hwaddr ptex, uint64_t pte1)
729 {
730     hwaddr base, offset = ptex * HASH_PTE_SIZE_64 + 16;
731 
732     if (cpu->vhyp) {
733         PPCVirtualHypervisorClass *vhc =
734             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
735         vhc->hpte_set_r(cpu->vhyp, ptex, pte1);
736         return;
737     }
738     base = ppc_hash64_hpt_base(cpu);
739 
740 
741     /* The HW performs a non-atomic byte update */
742     stb_phys(CPU(cpu)->as, base + offset, ((pte1 >> 8) & 0xff) | 0x01);
743 }
744 
745 static void ppc_hash64_set_c(PowerPCCPU *cpu, hwaddr ptex, uint64_t pte1)
746 {
747     hwaddr base, offset = ptex * HASH_PTE_SIZE_64 + 15;
748 
749     if (cpu->vhyp) {
750         PPCVirtualHypervisorClass *vhc =
751             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
752         vhc->hpte_set_c(cpu->vhyp, ptex, pte1);
753         return;
754     }
755     base = ppc_hash64_hpt_base(cpu);
756 
757     /* The HW performs a non-atomic byte update */
758     stb_phys(CPU(cpu)->as, base + offset, (pte1 & 0xff) | 0x80);
759 }
760 
761 static target_ulong rmls_limit(PowerPCCPU *cpu)
762 {
763     CPUPPCState *env = &cpu->env;
764     /*
765      * In theory the meanings of RMLS values are implementation
766      * dependent.  In practice, this seems to have been the set from
767      * POWER4+..POWER8, and RMLS is no longer supported in POWER9.
768      *
769      * Unsupported values mean the OS has shot itself in the
770      * foot. Return a 0-sized RMA in this case, which we expect
771      * to trigger an immediate DSI or ISI
772      */
773     static const target_ulong rma_sizes[16] = {
774         [0] = 256 * GiB,
775         [1] = 16 * GiB,
776         [2] = 1 * GiB,
777         [3] = 64 * MiB,
778         [4] = 256 * MiB,
779         [7] = 128 * MiB,
780         [8] = 32 * MiB,
781     };
782     target_ulong rmls = (env->spr[SPR_LPCR] & LPCR_RMLS) >> LPCR_RMLS_SHIFT;
783 
784     return rma_sizes[rmls];
785 }
786 
787 static int build_vrma_slbe(PowerPCCPU *cpu, ppc_slb_t *slb)
788 {
789     CPUPPCState *env = &cpu->env;
790     target_ulong lpcr = env->spr[SPR_LPCR];
791     uint32_t vrmasd = (lpcr & LPCR_VRMASD) >> LPCR_VRMASD_SHIFT;
792     target_ulong vsid = SLB_VSID_VRMA | ((vrmasd << 4) & SLB_VSID_LLP_MASK);
793     int i;
794 
795     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
796         const PPCHash64SegmentPageSizes *sps = &cpu->hash64_opts->sps[i];
797 
798         if (!sps->page_shift) {
799             break;
800         }
801 
802         if ((vsid & SLB_VSID_LLP_MASK) == sps->slb_enc) {
803             slb->esid = SLB_ESID_V;
804             slb->vsid = vsid;
805             slb->sps = sps;
806             return 0;
807         }
808     }
809 
810     error_report("Bad page size encoding in LPCR[VRMASD]; LPCR=0x"
811                  TARGET_FMT_lx"\n", lpcr);
812 
813     return -1;
814 }
815 
816 int ppc_hash64_handle_mmu_fault(PowerPCCPU *cpu, vaddr eaddr,
817                                 int rwx, int mmu_idx)
818 {
819     CPUState *cs = CPU(cpu);
820     CPUPPCState *env = &cpu->env;
821     ppc_slb_t vrma_slbe;
822     ppc_slb_t *slb;
823     unsigned apshift;
824     hwaddr ptex;
825     ppc_hash_pte64_t pte;
826     int exec_prot, pp_prot, amr_prot, prot;
827     const int need_prot[] = {PAGE_READ, PAGE_WRITE, PAGE_EXEC};
828     hwaddr raddr;
829 
830     assert((rwx == 0) || (rwx == 1) || (rwx == 2));
831 
832     /*
833      * Note on LPCR usage: 970 uses HID4, but our special variant of
834      * store_spr copies relevant fields into env->spr[SPR_LPCR].
835      * Similarily we filter unimplemented bits when storing into LPCR
836      * depending on the MMU version. This code can thus just use the
837      * LPCR "as-is".
838      */
839 
840     /* 1. Handle real mode accesses */
841     if (((rwx == 2) && (msr_ir == 0)) || ((rwx != 2) && (msr_dr == 0))) {
842         /*
843          * Translation is supposedly "off", but in real mode the top 4
844          * effective address bits are (mostly) ignored
845          */
846         raddr = eaddr & 0x0FFFFFFFFFFFFFFFULL;
847 
848         if (cpu->vhyp) {
849             /*
850              * In virtual hypervisor mode, there's nothing to do:
851              *   EA == GPA == qemu guest address
852              */
853         } else if (msr_hv || !env->has_hv_mode) {
854             /* In HV mode, add HRMOR if top EA bit is clear */
855             if (!(eaddr >> 63)) {
856                 raddr |= env->spr[SPR_HRMOR];
857             }
858         } else if (ppc_hash64_use_vrma(env)) {
859             /* Emulated VRMA mode */
860             slb = &vrma_slbe;
861             if (build_vrma_slbe(cpu, slb) != 0) {
862                 /* Invalid VRMA setup, machine check */
863                 cs->exception_index = POWERPC_EXCP_MCHECK;
864                 env->error_code = 0;
865                 return 1;
866             }
867 
868             goto skip_slb_search;
869         } else {
870             target_ulong limit = rmls_limit(cpu);
871 
872             /* Emulated old-style RMO mode, bounds check against RMLS */
873             if (raddr >= limit) {
874                 if (rwx == 2) {
875                     ppc_hash64_set_isi(cs, SRR1_PROTFAULT);
876                 } else {
877                     int dsisr = DSISR_PROTFAULT;
878                     if (rwx == 1) {
879                         dsisr |= DSISR_ISSTORE;
880                     }
881                     ppc_hash64_set_dsi(cs, eaddr, dsisr);
882                 }
883                 return 1;
884             }
885 
886             raddr |= env->spr[SPR_RMOR];
887         }
888         tlb_set_page(cs, eaddr & TARGET_PAGE_MASK, raddr & TARGET_PAGE_MASK,
889                      PAGE_READ | PAGE_WRITE | PAGE_EXEC, mmu_idx,
890                      TARGET_PAGE_SIZE);
891         return 0;
892     }
893 
894     /* 2. Translation is on, so look up the SLB */
895     slb = slb_lookup(cpu, eaddr);
896     if (!slb) {
897         /* No entry found, check if in-memory segment tables are in use */
898         if (ppc64_use_proc_tbl(cpu)) {
899             /* TODO - Unsupported */
900             error_report("Segment Table Support Unimplemented");
901             exit(1);
902         }
903         /* Segment still not found, generate the appropriate interrupt */
904         if (rwx == 2) {
905             cs->exception_index = POWERPC_EXCP_ISEG;
906             env->error_code = 0;
907         } else {
908             cs->exception_index = POWERPC_EXCP_DSEG;
909             env->error_code = 0;
910             env->spr[SPR_DAR] = eaddr;
911         }
912         return 1;
913     }
914 
915 skip_slb_search:
916 
917     /* 3. Check for segment level no-execute violation */
918     if ((rwx == 2) && (slb->vsid & SLB_VSID_N)) {
919         ppc_hash64_set_isi(cs, SRR1_NOEXEC_GUARD);
920         return 1;
921     }
922 
923     /* 4. Locate the PTE in the hash table */
924     ptex = ppc_hash64_htab_lookup(cpu, slb, eaddr, &pte, &apshift);
925     if (ptex == -1) {
926         if (rwx == 2) {
927             ppc_hash64_set_isi(cs, SRR1_NOPTE);
928         } else {
929             int dsisr = DSISR_NOPTE;
930             if (rwx == 1) {
931                 dsisr |= DSISR_ISSTORE;
932             }
933             ppc_hash64_set_dsi(cs, eaddr, dsisr);
934         }
935         return 1;
936     }
937     qemu_log_mask(CPU_LOG_MMU,
938                   "found PTE at index %08" HWADDR_PRIx "\n", ptex);
939 
940     /* 5. Check access permissions */
941 
942     exec_prot = ppc_hash64_pte_noexec_guard(cpu, pte);
943     pp_prot = ppc_hash64_pte_prot(cpu, slb, pte);
944     amr_prot = ppc_hash64_amr_prot(cpu, pte);
945     prot = exec_prot & pp_prot & amr_prot;
946 
947     if ((need_prot[rwx] & ~prot) != 0) {
948         /* Access right violation */
949         qemu_log_mask(CPU_LOG_MMU, "PTE access rejected\n");
950         if (rwx == 2) {
951             int srr1 = 0;
952             if (PAGE_EXEC & ~exec_prot) {
953                 srr1 |= SRR1_NOEXEC_GUARD; /* Access violates noexec or guard */
954             } else if (PAGE_EXEC & ~pp_prot) {
955                 srr1 |= SRR1_PROTFAULT; /* Access violates access authority */
956             }
957             if (PAGE_EXEC & ~amr_prot) {
958                 srr1 |= SRR1_IAMR; /* Access violates virt pg class key prot */
959             }
960             ppc_hash64_set_isi(cs, srr1);
961         } else {
962             int dsisr = 0;
963             if (need_prot[rwx] & ~pp_prot) {
964                 dsisr |= DSISR_PROTFAULT;
965             }
966             if (rwx == 1) {
967                 dsisr |= DSISR_ISSTORE;
968             }
969             if (need_prot[rwx] & ~amr_prot) {
970                 dsisr |= DSISR_AMR;
971             }
972             ppc_hash64_set_dsi(cs, eaddr, dsisr);
973         }
974         return 1;
975     }
976 
977     qemu_log_mask(CPU_LOG_MMU, "PTE access granted !\n");
978 
979     /* 6. Update PTE referenced and changed bits if necessary */
980 
981     if (!(pte.pte1 & HPTE64_R_R)) {
982         ppc_hash64_set_r(cpu, ptex, pte.pte1);
983     }
984     if (!(pte.pte1 & HPTE64_R_C)) {
985         if (rwx == 1) {
986             ppc_hash64_set_c(cpu, ptex, pte.pte1);
987         } else {
988             /*
989              * Treat the page as read-only for now, so that a later write
990              * will pass through this function again to set the C bit
991              */
992             prot &= ~PAGE_WRITE;
993         }
994     }
995 
996     /* 7. Determine the real address from the PTE */
997 
998     raddr = deposit64(pte.pte1 & HPTE64_R_RPN, 0, apshift, eaddr);
999 
1000     tlb_set_page(cs, eaddr & TARGET_PAGE_MASK, raddr & TARGET_PAGE_MASK,
1001                  prot, mmu_idx, 1ULL << apshift);
1002 
1003     return 0;
1004 }
1005 
1006 hwaddr ppc_hash64_get_phys_page_debug(PowerPCCPU *cpu, target_ulong addr)
1007 {
1008     CPUPPCState *env = &cpu->env;
1009     ppc_slb_t vrma_slbe;
1010     ppc_slb_t *slb;
1011     hwaddr ptex, raddr;
1012     ppc_hash_pte64_t pte;
1013     unsigned apshift;
1014 
1015     /* Handle real mode */
1016     if (msr_dr == 0) {
1017         /* In real mode the top 4 effective address bits are ignored */
1018         raddr = addr & 0x0FFFFFFFFFFFFFFFULL;
1019 
1020         if (cpu->vhyp) {
1021             /*
1022              * In virtual hypervisor mode, there's nothing to do:
1023              *   EA == GPA == qemu guest address
1024              */
1025             return raddr;
1026         } else if ((msr_hv || !env->has_hv_mode) && !(addr >> 63)) {
1027             /* In HV mode, add HRMOR if top EA bit is clear */
1028             return raddr | env->spr[SPR_HRMOR];
1029         } else if (ppc_hash64_use_vrma(env)) {
1030             /* Emulated VRMA mode */
1031             slb = &vrma_slbe;
1032             if (build_vrma_slbe(cpu, slb) != 0) {
1033                 return -1;
1034             }
1035         } else {
1036             target_ulong limit = rmls_limit(cpu);
1037 
1038             /* Emulated old-style RMO mode, bounds check against RMLS */
1039             if (raddr >= limit) {
1040                 return -1;
1041             }
1042             return raddr | env->spr[SPR_RMOR];
1043         }
1044     } else {
1045         slb = slb_lookup(cpu, addr);
1046         if (!slb) {
1047             return -1;
1048         }
1049     }
1050 
1051     ptex = ppc_hash64_htab_lookup(cpu, slb, addr, &pte, &apshift);
1052     if (ptex == -1) {
1053         return -1;
1054     }
1055 
1056     return deposit64(pte.pte1 & HPTE64_R_RPN, 0, apshift, addr)
1057         & TARGET_PAGE_MASK;
1058 }
1059 
1060 void ppc_hash64_tlb_flush_hpte(PowerPCCPU *cpu, target_ulong ptex,
1061                                target_ulong pte0, target_ulong pte1)
1062 {
1063     /*
1064      * XXX: given the fact that there are too many segments to
1065      * invalidate, and we still don't have a tlb_flush_mask(env, n,
1066      * mask) in QEMU, we just invalidate all TLBs
1067      */
1068     cpu->env.tlb_need_flush = TLB_NEED_GLOBAL_FLUSH | TLB_NEED_LOCAL_FLUSH;
1069 }
1070 
1071 void ppc_store_lpcr(PowerPCCPU *cpu, target_ulong val)
1072 {
1073     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
1074     CPUPPCState *env = &cpu->env;
1075 
1076     env->spr[SPR_LPCR] = val & pcc->lpcr_mask;
1077 }
1078 
1079 void helper_store_lpcr(CPUPPCState *env, target_ulong val)
1080 {
1081     PowerPCCPU *cpu = env_archcpu(env);
1082 
1083     ppc_store_lpcr(cpu, val);
1084 }
1085 
1086 void ppc_hash64_init(PowerPCCPU *cpu)
1087 {
1088     CPUPPCState *env = &cpu->env;
1089     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
1090 
1091     if (!pcc->hash64_opts) {
1092         assert(!(env->mmu_model & POWERPC_MMU_64));
1093         return;
1094     }
1095 
1096     cpu->hash64_opts = g_memdup(pcc->hash64_opts, sizeof(*cpu->hash64_opts));
1097 }
1098 
1099 void ppc_hash64_finalize(PowerPCCPU *cpu)
1100 {
1101     g_free(cpu->hash64_opts);
1102 }
1103 
1104 const PPCHash64Options ppc_hash64_opts_basic = {
1105     .flags = 0,
1106     .slb_size = 64,
1107     .sps = {
1108         { .page_shift = 12, /* 4K */
1109           .slb_enc = 0,
1110           .enc = { { .page_shift = 12, .pte_enc = 0 } }
1111         },
1112         { .page_shift = 24, /* 16M */
1113           .slb_enc = 0x100,
1114           .enc = { { .page_shift = 24, .pte_enc = 0 } }
1115         },
1116     },
1117 };
1118 
1119 const PPCHash64Options ppc_hash64_opts_POWER7 = {
1120     .flags = PPC_HASH64_1TSEG | PPC_HASH64_AMR | PPC_HASH64_CI_LARGEPAGE,
1121     .slb_size = 32,
1122     .sps = {
1123         {
1124             .page_shift = 12, /* 4K */
1125             .slb_enc = 0,
1126             .enc = { { .page_shift = 12, .pte_enc = 0 },
1127                      { .page_shift = 16, .pte_enc = 0x7 },
1128                      { .page_shift = 24, .pte_enc = 0x38 }, },
1129         },
1130         {
1131             .page_shift = 16, /* 64K */
1132             .slb_enc = SLB_VSID_64K,
1133             .enc = { { .page_shift = 16, .pte_enc = 0x1 },
1134                      { .page_shift = 24, .pte_enc = 0x8 }, },
1135         },
1136         {
1137             .page_shift = 24, /* 16M */
1138             .slb_enc = SLB_VSID_16M,
1139             .enc = { { .page_shift = 24, .pte_enc = 0 }, },
1140         },
1141         {
1142             .page_shift = 34, /* 16G */
1143             .slb_enc = SLB_VSID_16G,
1144             .enc = { { .page_shift = 34, .pte_enc = 0x3 }, },
1145         },
1146     }
1147 };
1148 
1149 void ppc_hash64_filter_pagesizes(PowerPCCPU *cpu,
1150                                  bool (*cb)(void *, uint32_t, uint32_t),
1151                                  void *opaque)
1152 {
1153     PPCHash64Options *opts = cpu->hash64_opts;
1154     int i;
1155     int n = 0;
1156     bool ci_largepage = false;
1157 
1158     assert(opts);
1159 
1160     n = 0;
1161     for (i = 0; i < ARRAY_SIZE(opts->sps); i++) {
1162         PPCHash64SegmentPageSizes *sps = &opts->sps[i];
1163         int j;
1164         int m = 0;
1165 
1166         assert(n <= i);
1167 
1168         if (!sps->page_shift) {
1169             break;
1170         }
1171 
1172         for (j = 0; j < ARRAY_SIZE(sps->enc); j++) {
1173             PPCHash64PageSize *ps = &sps->enc[j];
1174 
1175             assert(m <= j);
1176             if (!ps->page_shift) {
1177                 break;
1178             }
1179 
1180             if (cb(opaque, sps->page_shift, ps->page_shift)) {
1181                 if (ps->page_shift >= 16) {
1182                     ci_largepage = true;
1183                 }
1184                 sps->enc[m++] = *ps;
1185             }
1186         }
1187 
1188         /* Clear rest of the row */
1189         for (j = m; j < ARRAY_SIZE(sps->enc); j++) {
1190             memset(&sps->enc[j], 0, sizeof(sps->enc[j]));
1191         }
1192 
1193         if (m) {
1194             n++;
1195         }
1196     }
1197 
1198     /* Clear the rest of the table */
1199     for (i = n; i < ARRAY_SIZE(opts->sps); i++) {
1200         memset(&opts->sps[i], 0, sizeof(opts->sps[i]));
1201     }
1202 
1203     if (!ci_largepage) {
1204         opts->flags &= ~PPC_HASH64_CI_LARGEPAGE;
1205     }
1206 }
1207