xref: /openbmc/qemu/hw/arm/smmuv3.c (revision 2e35439f)
1 /*
2  * Copyright (C) 2014-2016 Broadcom Corporation
3  * Copyright (c) 2017 Red Hat, Inc.
4  * Written by Prem Mallappa, Eric Auger
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License along
16  * with this program; if not, see <http://www.gnu.org/licenses/>.
17  */
18 
19 #include "qemu/osdep.h"
20 #include "qemu/bitops.h"
21 #include "hw/irq.h"
22 #include "hw/sysbus.h"
23 #include "migration/vmstate.h"
24 #include "hw/qdev-properties.h"
25 #include "hw/qdev-core.h"
26 #include "hw/pci/pci.h"
27 #include "cpu.h"
28 #include "trace.h"
29 #include "qemu/log.h"
30 #include "qemu/error-report.h"
31 #include "qapi/error.h"
32 
33 #include "hw/arm/smmuv3.h"
34 #include "smmuv3-internal.h"
35 #include "smmu-internal.h"
36 
37 #define PTW_RECORD_FAULT(ptw_info, cfg) (((ptw_info).stage == SMMU_STAGE_1 && \
38                                         (cfg)->record_faults) || \
39                                         ((ptw_info).stage == SMMU_STAGE_2 && \
40                                         (cfg)->s2cfg.record_faults))
41 
42 /**
43  * smmuv3_trigger_irq - pulse @irq if enabled and update
44  * GERROR register in case of GERROR interrupt
45  *
46  * @irq: irq type
47  * @gerror_mask: mask of gerrors to toggle (relevant if @irq is GERROR)
48  */
49 static void smmuv3_trigger_irq(SMMUv3State *s, SMMUIrq irq,
50                                uint32_t gerror_mask)
51 {
52 
53     bool pulse = false;
54 
55     switch (irq) {
56     case SMMU_IRQ_EVTQ:
57         pulse = smmuv3_eventq_irq_enabled(s);
58         break;
59     case SMMU_IRQ_PRIQ:
60         qemu_log_mask(LOG_UNIMP, "PRI not yet supported\n");
61         break;
62     case SMMU_IRQ_CMD_SYNC:
63         pulse = true;
64         break;
65     case SMMU_IRQ_GERROR:
66     {
67         uint32_t pending = s->gerror ^ s->gerrorn;
68         uint32_t new_gerrors = ~pending & gerror_mask;
69 
70         if (!new_gerrors) {
71             /* only toggle non pending errors */
72             return;
73         }
74         s->gerror ^= new_gerrors;
75         trace_smmuv3_write_gerror(new_gerrors, s->gerror);
76 
77         pulse = smmuv3_gerror_irq_enabled(s);
78         break;
79     }
80     }
81     if (pulse) {
82             trace_smmuv3_trigger_irq(irq);
83             qemu_irq_pulse(s->irq[irq]);
84     }
85 }
86 
87 static void smmuv3_write_gerrorn(SMMUv3State *s, uint32_t new_gerrorn)
88 {
89     uint32_t pending = s->gerror ^ s->gerrorn;
90     uint32_t toggled = s->gerrorn ^ new_gerrorn;
91 
92     if (toggled & ~pending) {
93         qemu_log_mask(LOG_GUEST_ERROR,
94                       "guest toggles non pending errors = 0x%x\n",
95                       toggled & ~pending);
96     }
97 
98     /*
99      * We do not raise any error in case guest toggles bits corresponding
100      * to not active IRQs (CONSTRAINED UNPREDICTABLE)
101      */
102     s->gerrorn = new_gerrorn;
103 
104     trace_smmuv3_write_gerrorn(toggled & pending, s->gerrorn);
105 }
106 
107 static inline MemTxResult queue_read(SMMUQueue *q, Cmd *cmd)
108 {
109     dma_addr_t addr = Q_CONS_ENTRY(q);
110     MemTxResult ret;
111     int i;
112 
113     ret = dma_memory_read(&address_space_memory, addr, cmd, sizeof(Cmd),
114                           MEMTXATTRS_UNSPECIFIED);
115     if (ret != MEMTX_OK) {
116         return ret;
117     }
118     for (i = 0; i < ARRAY_SIZE(cmd->word); i++) {
119         le32_to_cpus(&cmd->word[i]);
120     }
121     return ret;
122 }
123 
124 static MemTxResult queue_write(SMMUQueue *q, Evt *evt_in)
125 {
126     dma_addr_t addr = Q_PROD_ENTRY(q);
127     MemTxResult ret;
128     Evt evt = *evt_in;
129     int i;
130 
131     for (i = 0; i < ARRAY_SIZE(evt.word); i++) {
132         cpu_to_le32s(&evt.word[i]);
133     }
134     ret = dma_memory_write(&address_space_memory, addr, &evt, sizeof(Evt),
135                            MEMTXATTRS_UNSPECIFIED);
136     if (ret != MEMTX_OK) {
137         return ret;
138     }
139 
140     queue_prod_incr(q);
141     return MEMTX_OK;
142 }
143 
144 static MemTxResult smmuv3_write_eventq(SMMUv3State *s, Evt *evt)
145 {
146     SMMUQueue *q = &s->eventq;
147     MemTxResult r;
148 
149     if (!smmuv3_eventq_enabled(s)) {
150         return MEMTX_ERROR;
151     }
152 
153     if (smmuv3_q_full(q)) {
154         return MEMTX_ERROR;
155     }
156 
157     r = queue_write(q, evt);
158     if (r != MEMTX_OK) {
159         return r;
160     }
161 
162     if (!smmuv3_q_empty(q)) {
163         smmuv3_trigger_irq(s, SMMU_IRQ_EVTQ, 0);
164     }
165     return MEMTX_OK;
166 }
167 
168 void smmuv3_record_event(SMMUv3State *s, SMMUEventInfo *info)
169 {
170     Evt evt = {};
171     MemTxResult r;
172 
173     if (!smmuv3_eventq_enabled(s)) {
174         return;
175     }
176 
177     EVT_SET_TYPE(&evt, info->type);
178     EVT_SET_SID(&evt, info->sid);
179 
180     switch (info->type) {
181     case SMMU_EVT_NONE:
182         return;
183     case SMMU_EVT_F_UUT:
184         EVT_SET_SSID(&evt, info->u.f_uut.ssid);
185         EVT_SET_SSV(&evt,  info->u.f_uut.ssv);
186         EVT_SET_ADDR(&evt, info->u.f_uut.addr);
187         EVT_SET_RNW(&evt,  info->u.f_uut.rnw);
188         EVT_SET_PNU(&evt,  info->u.f_uut.pnu);
189         EVT_SET_IND(&evt,  info->u.f_uut.ind);
190         break;
191     case SMMU_EVT_C_BAD_STREAMID:
192         EVT_SET_SSID(&evt, info->u.c_bad_streamid.ssid);
193         EVT_SET_SSV(&evt,  info->u.c_bad_streamid.ssv);
194         break;
195     case SMMU_EVT_F_STE_FETCH:
196         EVT_SET_SSID(&evt, info->u.f_ste_fetch.ssid);
197         EVT_SET_SSV(&evt,  info->u.f_ste_fetch.ssv);
198         EVT_SET_ADDR2(&evt, info->u.f_ste_fetch.addr);
199         break;
200     case SMMU_EVT_C_BAD_STE:
201         EVT_SET_SSID(&evt, info->u.c_bad_ste.ssid);
202         EVT_SET_SSV(&evt,  info->u.c_bad_ste.ssv);
203         break;
204     case SMMU_EVT_F_STREAM_DISABLED:
205         break;
206     case SMMU_EVT_F_TRANS_FORBIDDEN:
207         EVT_SET_ADDR(&evt, info->u.f_transl_forbidden.addr);
208         EVT_SET_RNW(&evt, info->u.f_transl_forbidden.rnw);
209         break;
210     case SMMU_EVT_C_BAD_SUBSTREAMID:
211         EVT_SET_SSID(&evt, info->u.c_bad_substream.ssid);
212         break;
213     case SMMU_EVT_F_CD_FETCH:
214         EVT_SET_SSID(&evt, info->u.f_cd_fetch.ssid);
215         EVT_SET_SSV(&evt,  info->u.f_cd_fetch.ssv);
216         EVT_SET_ADDR(&evt, info->u.f_cd_fetch.addr);
217         break;
218     case SMMU_EVT_C_BAD_CD:
219         EVT_SET_SSID(&evt, info->u.c_bad_cd.ssid);
220         EVT_SET_SSV(&evt,  info->u.c_bad_cd.ssv);
221         break;
222     case SMMU_EVT_F_WALK_EABT:
223     case SMMU_EVT_F_TRANSLATION:
224     case SMMU_EVT_F_ADDR_SIZE:
225     case SMMU_EVT_F_ACCESS:
226     case SMMU_EVT_F_PERMISSION:
227         EVT_SET_STALL(&evt, info->u.f_walk_eabt.stall);
228         EVT_SET_STAG(&evt, info->u.f_walk_eabt.stag);
229         EVT_SET_SSID(&evt, info->u.f_walk_eabt.ssid);
230         EVT_SET_SSV(&evt, info->u.f_walk_eabt.ssv);
231         EVT_SET_S2(&evt, info->u.f_walk_eabt.s2);
232         EVT_SET_ADDR(&evt, info->u.f_walk_eabt.addr);
233         EVT_SET_RNW(&evt, info->u.f_walk_eabt.rnw);
234         EVT_SET_PNU(&evt, info->u.f_walk_eabt.pnu);
235         EVT_SET_IND(&evt, info->u.f_walk_eabt.ind);
236         EVT_SET_CLASS(&evt, info->u.f_walk_eabt.class);
237         EVT_SET_ADDR2(&evt, info->u.f_walk_eabt.addr2);
238         break;
239     case SMMU_EVT_F_CFG_CONFLICT:
240         EVT_SET_SSID(&evt, info->u.f_cfg_conflict.ssid);
241         EVT_SET_SSV(&evt,  info->u.f_cfg_conflict.ssv);
242         break;
243     /* rest is not implemented */
244     case SMMU_EVT_F_BAD_ATS_TREQ:
245     case SMMU_EVT_F_TLB_CONFLICT:
246     case SMMU_EVT_E_PAGE_REQ:
247     default:
248         g_assert_not_reached();
249     }
250 
251     trace_smmuv3_record_event(smmu_event_string(info->type), info->sid);
252     r = smmuv3_write_eventq(s, &evt);
253     if (r != MEMTX_OK) {
254         smmuv3_trigger_irq(s, SMMU_IRQ_GERROR, R_GERROR_EVENTQ_ABT_ERR_MASK);
255     }
256     info->recorded = true;
257 }
258 
259 static void smmuv3_init_regs(SMMUv3State *s)
260 {
261     /* Based on sys property, the stages supported in smmu will be advertised.*/
262     if (s->stage && !strcmp("2", s->stage)) {
263         s->idr[0] = FIELD_DP32(s->idr[0], IDR0, S2P, 1);
264     } else if (s->stage && !strcmp("nested", s->stage)) {
265         s->idr[0] = FIELD_DP32(s->idr[0], IDR0, S1P, 1);
266         s->idr[0] = FIELD_DP32(s->idr[0], IDR0, S2P, 1);
267     } else {
268         s->idr[0] = FIELD_DP32(s->idr[0], IDR0, S1P, 1);
269     }
270 
271     s->idr[0] = FIELD_DP32(s->idr[0], IDR0, TTF, 2); /* AArch64 PTW only */
272     s->idr[0] = FIELD_DP32(s->idr[0], IDR0, COHACC, 1); /* IO coherent */
273     s->idr[0] = FIELD_DP32(s->idr[0], IDR0, ASID16, 1); /* 16-bit ASID */
274     s->idr[0] = FIELD_DP32(s->idr[0], IDR0, VMID16, 1); /* 16-bit VMID */
275     s->idr[0] = FIELD_DP32(s->idr[0], IDR0, TTENDIAN, 2); /* little endian */
276     s->idr[0] = FIELD_DP32(s->idr[0], IDR0, STALL_MODEL, 1); /* No stall */
277     /* terminated transaction will always be aborted/error returned */
278     s->idr[0] = FIELD_DP32(s->idr[0], IDR0, TERM_MODEL, 1);
279     /* 2-level stream table supported */
280     s->idr[0] = FIELD_DP32(s->idr[0], IDR0, STLEVEL, 1);
281 
282     s->idr[1] = FIELD_DP32(s->idr[1], IDR1, SIDSIZE, SMMU_IDR1_SIDSIZE);
283     s->idr[1] = FIELD_DP32(s->idr[1], IDR1, EVENTQS, SMMU_EVENTQS);
284     s->idr[1] = FIELD_DP32(s->idr[1], IDR1, CMDQS,   SMMU_CMDQS);
285 
286     s->idr[3] = FIELD_DP32(s->idr[3], IDR3, HAD, 1);
287     if (FIELD_EX32(s->idr[0], IDR0, S2P)) {
288         /* XNX is a stage-2-specific feature */
289         s->idr[3] = FIELD_DP32(s->idr[3], IDR3, XNX, 1);
290     }
291     s->idr[3] = FIELD_DP32(s->idr[3], IDR3, RIL, 1);
292     s->idr[3] = FIELD_DP32(s->idr[3], IDR3, BBML, 2);
293 
294     s->idr[5] = FIELD_DP32(s->idr[5], IDR5, OAS, SMMU_IDR5_OAS); /* 44 bits */
295     /* 4K, 16K and 64K granule support */
296     s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN4K, 1);
297     s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN16K, 1);
298     s->idr[5] = FIELD_DP32(s->idr[5], IDR5, GRAN64K, 1);
299 
300     s->cmdq.base = deposit64(s->cmdq.base, 0, 5, SMMU_CMDQS);
301     s->cmdq.prod = 0;
302     s->cmdq.cons = 0;
303     s->cmdq.entry_size = sizeof(struct Cmd);
304     s->eventq.base = deposit64(s->eventq.base, 0, 5, SMMU_EVENTQS);
305     s->eventq.prod = 0;
306     s->eventq.cons = 0;
307     s->eventq.entry_size = sizeof(struct Evt);
308 
309     s->features = 0;
310     s->sid_split = 0;
311     s->aidr = 0x1;
312     s->cr[0] = 0;
313     s->cr0ack = 0;
314     s->irq_ctrl = 0;
315     s->gerror = 0;
316     s->gerrorn = 0;
317     s->statusr = 0;
318     s->gbpa = SMMU_GBPA_RESET_VAL;
319 }
320 
321 static int smmu_get_ste(SMMUv3State *s, dma_addr_t addr, STE *buf,
322                         SMMUEventInfo *event)
323 {
324     int ret, i;
325 
326     trace_smmuv3_get_ste(addr);
327     /* TODO: guarantee 64-bit single-copy atomicity */
328     ret = dma_memory_read(&address_space_memory, addr, buf, sizeof(*buf),
329                           MEMTXATTRS_UNSPECIFIED);
330     if (ret != MEMTX_OK) {
331         qemu_log_mask(LOG_GUEST_ERROR,
332                       "Cannot fetch pte at address=0x%"PRIx64"\n", addr);
333         event->type = SMMU_EVT_F_STE_FETCH;
334         event->u.f_ste_fetch.addr = addr;
335         return -EINVAL;
336     }
337     for (i = 0; i < ARRAY_SIZE(buf->word); i++) {
338         le32_to_cpus(&buf->word[i]);
339     }
340     return 0;
341 
342 }
343 
344 static SMMUTranslationStatus smmuv3_do_translate(SMMUv3State *s, hwaddr addr,
345                                                  SMMUTransCfg *cfg,
346                                                  SMMUEventInfo *event,
347                                                  IOMMUAccessFlags flag,
348                                                  SMMUTLBEntry **out_entry,
349                                                  SMMUTranslationClass class);
350 /* @ssid > 0 not supported yet */
351 static int smmu_get_cd(SMMUv3State *s, STE *ste, SMMUTransCfg *cfg,
352                        uint32_t ssid, CD *buf, SMMUEventInfo *event)
353 {
354     dma_addr_t addr = STE_CTXPTR(ste);
355     int ret, i;
356     SMMUTranslationStatus status;
357     SMMUTLBEntry *entry;
358 
359     trace_smmuv3_get_cd(addr);
360 
361     if (cfg->stage == SMMU_NESTED) {
362         status = smmuv3_do_translate(s, addr, cfg, event,
363                                      IOMMU_RO, &entry, SMMU_CLASS_CD);
364 
365         /* Same PTW faults are reported but with CLASS = CD. */
366         if (status != SMMU_TRANS_SUCCESS) {
367             return -EINVAL;
368         }
369 
370         addr = CACHED_ENTRY_TO_ADDR(entry, addr);
371     }
372 
373     /* TODO: guarantee 64-bit single-copy atomicity */
374     ret = dma_memory_read(&address_space_memory, addr, buf, sizeof(*buf),
375                           MEMTXATTRS_UNSPECIFIED);
376     if (ret != MEMTX_OK) {
377         qemu_log_mask(LOG_GUEST_ERROR,
378                       "Cannot fetch pte at address=0x%"PRIx64"\n", addr);
379         event->type = SMMU_EVT_F_CD_FETCH;
380         event->u.f_ste_fetch.addr = addr;
381         return -EINVAL;
382     }
383     for (i = 0; i < ARRAY_SIZE(buf->word); i++) {
384         le32_to_cpus(&buf->word[i]);
385     }
386     return 0;
387 }
388 
389 /*
390  * Max valid value is 39 when SMMU_IDR3.STT == 0.
391  * In architectures after SMMUv3.0:
392  * - If STE.S2TG selects a 4KB or 16KB granule, the minimum valid value for this
393  *   field is MAX(16, 64-IAS)
394  * - If STE.S2TG selects a 64KB granule, the minimum valid value for this field
395  *   is (64-IAS).
396  * As we only support AA64, IAS = OAS.
397  */
398 static bool s2t0sz_valid(SMMUTransCfg *cfg)
399 {
400     if (cfg->s2cfg.tsz > 39) {
401         return false;
402     }
403 
404     if (cfg->s2cfg.granule_sz == 16) {
405         return (cfg->s2cfg.tsz >= 64 - cfg->s2cfg.eff_ps);
406     }
407 
408     return (cfg->s2cfg.tsz >= MAX(64 - cfg->s2cfg.eff_ps, 16));
409 }
410 
411 /*
412  * Return true if s2 page table config is valid.
413  * This checks with the configured start level, ias_bits and granularity we can
414  * have a valid page table as described in ARM ARM D8.2 Translation process.
415  * The idea here is to see for the highest possible number of IPA bits, how
416  * many concatenated tables we would need, if it is more than 16, then this is
417  * not possible.
418  */
419 static bool s2_pgtable_config_valid(uint8_t sl0, uint8_t t0sz, uint8_t gran)
420 {
421     int level = get_start_level(sl0, gran);
422     uint64_t ipa_bits = 64 - t0sz;
423     uint64_t max_ipa = (1ULL << ipa_bits) - 1;
424     int nr_concat = pgd_concat_idx(level, gran, max_ipa) + 1;
425 
426     return nr_concat <= VMSA_MAX_S2_CONCAT;
427 }
428 
429 static int decode_ste_s2_cfg(SMMUv3State *s, SMMUTransCfg *cfg,
430                              STE *ste)
431 {
432     uint8_t oas = FIELD_EX32(s->idr[5], IDR5, OAS);
433 
434     if (STE_S2AA64(ste) == 0x0) {
435         qemu_log_mask(LOG_UNIMP,
436                       "SMMUv3 AArch32 tables not supported\n");
437         g_assert_not_reached();
438     }
439 
440     switch (STE_S2TG(ste)) {
441     case 0x0: /* 4KB */
442         cfg->s2cfg.granule_sz = 12;
443         break;
444     case 0x1: /* 64KB */
445         cfg->s2cfg.granule_sz = 16;
446         break;
447     case 0x2: /* 16KB */
448         cfg->s2cfg.granule_sz = 14;
449         break;
450     default:
451         qemu_log_mask(LOG_GUEST_ERROR,
452                       "SMMUv3 bad STE S2TG: %x\n", STE_S2TG(ste));
453         goto bad_ste;
454     }
455 
456     cfg->s2cfg.vttb = STE_S2TTB(ste);
457 
458     cfg->s2cfg.sl0 = STE_S2SL0(ste);
459     /* FEAT_TTST not supported. */
460     if (cfg->s2cfg.sl0 == 0x3) {
461         qemu_log_mask(LOG_UNIMP, "SMMUv3 S2SL0 = 0x3 has no meaning!\n");
462         goto bad_ste;
463     }
464 
465     /* For AA64, The effective S2PS size is capped to the OAS. */
466     cfg->s2cfg.eff_ps = oas2bits(MIN(STE_S2PS(ste), oas));
467     /*
468      * For SMMUv3.1 and later, when OAS == IAS == 52, the stage 2 input
469      * range is further limited to 48 bits unless STE.S2TG indicates a
470      * 64KB granule.
471      */
472     if (cfg->s2cfg.granule_sz != 16) {
473         cfg->s2cfg.eff_ps = MIN(cfg->s2cfg.eff_ps, 48);
474     }
475     /*
476      * It is ILLEGAL for the address in S2TTB to be outside the range
477      * described by the effective S2PS value.
478      */
479     if (cfg->s2cfg.vttb & ~(MAKE_64BIT_MASK(0, cfg->s2cfg.eff_ps))) {
480         qemu_log_mask(LOG_GUEST_ERROR,
481                       "SMMUv3 S2TTB too large 0x%" PRIx64
482                       ", effective PS %d bits\n",
483                       cfg->s2cfg.vttb,  cfg->s2cfg.eff_ps);
484         goto bad_ste;
485     }
486 
487     cfg->s2cfg.tsz = STE_S2T0SZ(ste);
488 
489     if (!s2t0sz_valid(cfg)) {
490         qemu_log_mask(LOG_GUEST_ERROR, "SMMUv3 bad STE S2T0SZ = %d\n",
491                       cfg->s2cfg.tsz);
492         goto bad_ste;
493     }
494 
495     if (!s2_pgtable_config_valid(cfg->s2cfg.sl0, cfg->s2cfg.tsz,
496                                     cfg->s2cfg.granule_sz)) {
497         qemu_log_mask(LOG_GUEST_ERROR,
498                       "SMMUv3 STE stage 2 config not valid!\n");
499         goto bad_ste;
500     }
501 
502     /* Only LE supported(IDR0.TTENDIAN). */
503     if (STE_S2ENDI(ste)) {
504         qemu_log_mask(LOG_GUEST_ERROR,
505                       "SMMUv3 STE_S2ENDI only supports LE!\n");
506         goto bad_ste;
507     }
508 
509     cfg->s2cfg.affd = STE_S2AFFD(ste);
510 
511     cfg->s2cfg.record_faults = STE_S2R(ste);
512     /* As stall is not supported. */
513     if (STE_S2S(ste)) {
514         qemu_log_mask(LOG_UNIMP, "SMMUv3 Stall not implemented!\n");
515         goto bad_ste;
516     }
517 
518     return 0;
519 
520 bad_ste:
521     return -EINVAL;
522 }
523 
524 static void decode_ste_config(SMMUTransCfg *cfg, uint32_t config)
525 {
526 
527     if (STE_CFG_ABORT(config)) {
528         cfg->aborted = true;
529         return;
530     }
531     if (STE_CFG_BYPASS(config)) {
532         cfg->bypassed = true;
533         return;
534     }
535 
536     if (STE_CFG_S1_ENABLED(config)) {
537         cfg->stage = SMMU_STAGE_1;
538     }
539 
540     if (STE_CFG_S2_ENABLED(config)) {
541         cfg->stage |= SMMU_STAGE_2;
542     }
543 }
544 
545 /* Returns < 0 in case of invalid STE, 0 otherwise */
546 static int decode_ste(SMMUv3State *s, SMMUTransCfg *cfg,
547                       STE *ste, SMMUEventInfo *event)
548 {
549     uint32_t config;
550     uint8_t oas = FIELD_EX32(s->idr[5], IDR5, OAS);
551     int ret;
552 
553     if (!STE_VALID(ste)) {
554         if (!event->inval_ste_allowed) {
555             qemu_log_mask(LOG_GUEST_ERROR, "invalid STE\n");
556         }
557         goto bad_ste;
558     }
559 
560     config = STE_CONFIG(ste);
561 
562     decode_ste_config(cfg, config);
563 
564     if (cfg->aborted || cfg->bypassed) {
565         return 0;
566     }
567 
568     /*
569      * If a stage is enabled in SW while not advertised, throw bad ste
570      * according to user manual(IHI0070E) "5.2 Stream Table Entry".
571      */
572     if (!STAGE1_SUPPORTED(s) && STE_CFG_S1_ENABLED(config)) {
573         qemu_log_mask(LOG_GUEST_ERROR, "SMMUv3 S1 used but not supported.\n");
574         goto bad_ste;
575     }
576     if (!STAGE2_SUPPORTED(s) && STE_CFG_S2_ENABLED(config)) {
577         qemu_log_mask(LOG_GUEST_ERROR, "SMMUv3 S2 used but not supported.\n");
578         goto bad_ste;
579     }
580 
581     if (STAGE2_SUPPORTED(s)) {
582         /* VMID is considered even if s2 is disabled. */
583         cfg->s2cfg.vmid = STE_S2VMID(ste);
584     } else {
585         /* Default to -1 */
586         cfg->s2cfg.vmid = -1;
587     }
588 
589     if (STE_CFG_S2_ENABLED(config)) {
590         /*
591          * Stage-1 OAS defaults to OAS even if not enabled as it would be used
592          * in input address check for stage-2.
593          */
594         cfg->oas = oas2bits(oas);
595         ret = decode_ste_s2_cfg(s, cfg, ste);
596         if (ret) {
597             goto bad_ste;
598         }
599     }
600 
601     if (STE_S1CDMAX(ste) != 0) {
602         qemu_log_mask(LOG_UNIMP,
603                       "SMMUv3 does not support multiple context descriptors yet\n");
604         goto bad_ste;
605     }
606 
607     if (STE_S1STALLD(ste)) {
608         qemu_log_mask(LOG_UNIMP,
609                       "SMMUv3 S1 stalling fault model not allowed yet\n");
610         goto bad_ste;
611     }
612     return 0;
613 
614 bad_ste:
615     event->type = SMMU_EVT_C_BAD_STE;
616     return -EINVAL;
617 }
618 
619 /**
620  * smmu_find_ste - Return the stream table entry associated
621  * to the sid
622  *
623  * @s: smmuv3 handle
624  * @sid: stream ID
625  * @ste: returned stream table entry
626  * @event: handle to an event info
627  *
628  * Supports linear and 2-level stream table
629  * Return 0 on success, -EINVAL otherwise
630  */
631 static int smmu_find_ste(SMMUv3State *s, uint32_t sid, STE *ste,
632                          SMMUEventInfo *event)
633 {
634     dma_addr_t addr, strtab_base;
635     uint32_t log2size;
636     int strtab_size_shift;
637     int ret;
638 
639     trace_smmuv3_find_ste(sid, s->features, s->sid_split);
640     log2size = FIELD_EX32(s->strtab_base_cfg, STRTAB_BASE_CFG, LOG2SIZE);
641     /*
642      * Check SID range against both guest-configured and implementation limits
643      */
644     if (sid >= (1 << MIN(log2size, SMMU_IDR1_SIDSIZE))) {
645         event->type = SMMU_EVT_C_BAD_STREAMID;
646         return -EINVAL;
647     }
648     if (s->features & SMMU_FEATURE_2LVL_STE) {
649         int l1_ste_offset, l2_ste_offset, max_l2_ste, span, i;
650         dma_addr_t l1ptr, l2ptr;
651         STEDesc l1std;
652 
653         /*
654          * Align strtab base address to table size. For this purpose, assume it
655          * is not bounded by SMMU_IDR1_SIDSIZE.
656          */
657         strtab_size_shift = MAX(5, (int)log2size - s->sid_split - 1 + 3);
658         strtab_base = s->strtab_base & SMMU_BASE_ADDR_MASK &
659                       ~MAKE_64BIT_MASK(0, strtab_size_shift);
660         l1_ste_offset = sid >> s->sid_split;
661         l2_ste_offset = sid & ((1 << s->sid_split) - 1);
662         l1ptr = (dma_addr_t)(strtab_base + l1_ste_offset * sizeof(l1std));
663         /* TODO: guarantee 64-bit single-copy atomicity */
664         ret = dma_memory_read(&address_space_memory, l1ptr, &l1std,
665                               sizeof(l1std), MEMTXATTRS_UNSPECIFIED);
666         if (ret != MEMTX_OK) {
667             qemu_log_mask(LOG_GUEST_ERROR,
668                           "Could not read L1PTR at 0X%"PRIx64"\n", l1ptr);
669             event->type = SMMU_EVT_F_STE_FETCH;
670             event->u.f_ste_fetch.addr = l1ptr;
671             return -EINVAL;
672         }
673         for (i = 0; i < ARRAY_SIZE(l1std.word); i++) {
674             le32_to_cpus(&l1std.word[i]);
675         }
676 
677         span = L1STD_SPAN(&l1std);
678 
679         if (!span) {
680             /* l2ptr is not valid */
681             if (!event->inval_ste_allowed) {
682                 qemu_log_mask(LOG_GUEST_ERROR,
683                               "invalid sid=%d (L1STD span=0)\n", sid);
684             }
685             event->type = SMMU_EVT_C_BAD_STREAMID;
686             return -EINVAL;
687         }
688         max_l2_ste = (1 << span) - 1;
689         l2ptr = l1std_l2ptr(&l1std);
690         trace_smmuv3_find_ste_2lvl(s->strtab_base, l1ptr, l1_ste_offset,
691                                    l2ptr, l2_ste_offset, max_l2_ste);
692         if (l2_ste_offset > max_l2_ste) {
693             qemu_log_mask(LOG_GUEST_ERROR,
694                           "l2_ste_offset=%d > max_l2_ste=%d\n",
695                           l2_ste_offset, max_l2_ste);
696             event->type = SMMU_EVT_C_BAD_STE;
697             return -EINVAL;
698         }
699         addr = l2ptr + l2_ste_offset * sizeof(*ste);
700     } else {
701         strtab_size_shift = log2size + 5;
702         strtab_base = s->strtab_base & SMMU_BASE_ADDR_MASK &
703                       ~MAKE_64BIT_MASK(0, strtab_size_shift);
704         addr = strtab_base + sid * sizeof(*ste);
705     }
706 
707     if (smmu_get_ste(s, addr, ste, event)) {
708         return -EINVAL;
709     }
710 
711     return 0;
712 }
713 
714 static int decode_cd(SMMUv3State *s, SMMUTransCfg *cfg,
715                      CD *cd, SMMUEventInfo *event)
716 {
717     int ret = -EINVAL;
718     int i;
719     SMMUTranslationStatus status;
720     SMMUTLBEntry *entry;
721     uint8_t oas = FIELD_EX32(s->idr[5], IDR5, OAS);
722 
723     if (!CD_VALID(cd) || !CD_AARCH64(cd)) {
724         goto bad_cd;
725     }
726     if (!CD_A(cd)) {
727         goto bad_cd; /* SMMU_IDR0.TERM_MODEL == 1 */
728     }
729     if (CD_S(cd)) {
730         goto bad_cd; /* !STE_SECURE && SMMU_IDR0.STALL_MODEL == 1 */
731     }
732     if (CD_HA(cd) || CD_HD(cd)) {
733         goto bad_cd; /* HTTU = 0 */
734     }
735 
736     /* we support only those at the moment */
737     cfg->aa64 = true;
738 
739     cfg->oas = oas2bits(CD_IPS(cd));
740     cfg->oas = MIN(oas2bits(oas), cfg->oas);
741     cfg->tbi = CD_TBI(cd);
742     cfg->asid = CD_ASID(cd);
743     cfg->affd = CD_AFFD(cd);
744 
745     trace_smmuv3_decode_cd(cfg->oas);
746 
747     /* decode data dependent on TT */
748     for (i = 0; i <= 1; i++) {
749         int tg, tsz;
750         SMMUTransTableInfo *tt = &cfg->tt[i];
751 
752         cfg->tt[i].disabled = CD_EPD(cd, i);
753         if (cfg->tt[i].disabled) {
754             continue;
755         }
756 
757         tsz = CD_TSZ(cd, i);
758         if (tsz < 16 || tsz > 39) {
759             goto bad_cd;
760         }
761 
762         tg = CD_TG(cd, i);
763         tt->granule_sz = tg2granule(tg, i);
764         if ((tt->granule_sz != 12 && tt->granule_sz != 14 &&
765              tt->granule_sz != 16) || CD_ENDI(cd)) {
766             goto bad_cd;
767         }
768 
769         /*
770          * An address greater than 48 bits in size can only be output from a
771          * TTD when, in SMMUv3.1 and later, the effective IPS is 52 and a 64KB
772          * granule is in use for that translation table
773          */
774         if (tt->granule_sz != 16) {
775             cfg->oas = MIN(cfg->oas, 48);
776         }
777         tt->tsz = tsz;
778         tt->ttb = CD_TTB(cd, i);
779 
780         if (tt->ttb & ~(MAKE_64BIT_MASK(0, cfg->oas))) {
781             goto bad_cd;
782         }
783 
784         /* Translate the TTBx, from IPA to PA if nesting is enabled. */
785         if (cfg->stage == SMMU_NESTED) {
786             status = smmuv3_do_translate(s, tt->ttb, cfg, event, IOMMU_RO,
787                                          &entry, SMMU_CLASS_TT);
788             /*
789              * Same PTW faults are reported but with CLASS = TT.
790              * If TTBx is larger than the effective stage 1 output addres
791              * size, it reports C_BAD_CD, which is handled by the above case.
792              */
793             if (status != SMMU_TRANS_SUCCESS) {
794                 return -EINVAL;
795             }
796             tt->ttb = CACHED_ENTRY_TO_ADDR(entry, tt->ttb);
797         }
798 
799         tt->had = CD_HAD(cd, i);
800         trace_smmuv3_decode_cd_tt(i, tt->tsz, tt->ttb, tt->granule_sz, tt->had);
801     }
802 
803     cfg->record_faults = CD_R(cd);
804 
805     return 0;
806 
807 bad_cd:
808     event->type = SMMU_EVT_C_BAD_CD;
809     return ret;
810 }
811 
812 /**
813  * smmuv3_decode_config - Prepare the translation configuration
814  * for the @mr iommu region
815  * @mr: iommu memory region the translation config must be prepared for
816  * @cfg: output translation configuration which is populated through
817  *       the different configuration decoding steps
818  * @event: must be zero'ed by the caller
819  *
820  * return < 0 in case of config decoding error (@event is filled
821  * accordingly). Return 0 otherwise.
822  */
823 static int smmuv3_decode_config(IOMMUMemoryRegion *mr, SMMUTransCfg *cfg,
824                                 SMMUEventInfo *event)
825 {
826     SMMUDevice *sdev = container_of(mr, SMMUDevice, iommu);
827     uint32_t sid = smmu_get_sid(sdev);
828     SMMUv3State *s = sdev->smmu;
829     int ret;
830     STE ste;
831     CD cd;
832 
833     /* ASID defaults to -1 (if s1 is not supported). */
834     cfg->asid = -1;
835 
836     ret = smmu_find_ste(s, sid, &ste, event);
837     if (ret) {
838         return ret;
839     }
840 
841     ret = decode_ste(s, cfg, &ste, event);
842     if (ret) {
843         return ret;
844     }
845 
846     if (cfg->aborted || cfg->bypassed || (cfg->stage == SMMU_STAGE_2)) {
847         return 0;
848     }
849 
850     ret = smmu_get_cd(s, &ste, cfg, 0 /* ssid */, &cd, event);
851     if (ret) {
852         return ret;
853     }
854 
855     return decode_cd(s, cfg, &cd, event);
856 }
857 
858 /**
859  * smmuv3_get_config - Look up for a cached copy of configuration data for
860  * @sdev and on cache miss performs a configuration structure decoding from
861  * guest RAM.
862  *
863  * @sdev: SMMUDevice handle
864  * @event: output event info
865  *
866  * The configuration cache contains data resulting from both STE and CD
867  * decoding under the form of an SMMUTransCfg struct. The hash table is indexed
868  * by the SMMUDevice handle.
869  */
870 static SMMUTransCfg *smmuv3_get_config(SMMUDevice *sdev, SMMUEventInfo *event)
871 {
872     SMMUv3State *s = sdev->smmu;
873     SMMUState *bc = &s->smmu_state;
874     SMMUTransCfg *cfg;
875 
876     cfg = g_hash_table_lookup(bc->configs, sdev);
877     if (cfg) {
878         sdev->cfg_cache_hits++;
879         trace_smmuv3_config_cache_hit(smmu_get_sid(sdev),
880                             sdev->cfg_cache_hits, sdev->cfg_cache_misses,
881                             100 * sdev->cfg_cache_hits /
882                             (sdev->cfg_cache_hits + sdev->cfg_cache_misses));
883     } else {
884         sdev->cfg_cache_misses++;
885         trace_smmuv3_config_cache_miss(smmu_get_sid(sdev),
886                             sdev->cfg_cache_hits, sdev->cfg_cache_misses,
887                             100 * sdev->cfg_cache_hits /
888                             (sdev->cfg_cache_hits + sdev->cfg_cache_misses));
889         cfg = g_new0(SMMUTransCfg, 1);
890 
891         if (!smmuv3_decode_config(&sdev->iommu, cfg, event)) {
892             g_hash_table_insert(bc->configs, sdev, cfg);
893         } else {
894             g_free(cfg);
895             cfg = NULL;
896         }
897     }
898     return cfg;
899 }
900 
901 static void smmuv3_flush_config(SMMUDevice *sdev)
902 {
903     SMMUv3State *s = sdev->smmu;
904     SMMUState *bc = &s->smmu_state;
905 
906     trace_smmuv3_config_cache_inv(smmu_get_sid(sdev));
907     g_hash_table_remove(bc->configs, sdev);
908 }
909 
910 /* Do translation with TLB lookup. */
911 static SMMUTranslationStatus smmuv3_do_translate(SMMUv3State *s, hwaddr addr,
912                                                  SMMUTransCfg *cfg,
913                                                  SMMUEventInfo *event,
914                                                  IOMMUAccessFlags flag,
915                                                  SMMUTLBEntry **out_entry,
916                                                  SMMUTranslationClass class)
917 {
918     SMMUPTWEventInfo ptw_info = {};
919     SMMUState *bs = ARM_SMMU(s);
920     SMMUTLBEntry *cached_entry = NULL;
921     int asid, stage;
922     bool desc_s2_translation = class != SMMU_CLASS_IN;
923 
924     /*
925      * The function uses the argument class to identify which stage is used:
926      * - CLASS = IN: Means an input translation, determine the stage from STE.
927      * - CLASS = CD: Means the addr is an IPA of the CD, and it would be
928      *   translated using the stage-2.
929      * - CLASS = TT: Means the addr is an IPA of the stage-1 translation table
930      *   and it would be translated using the stage-2.
931      * For the last 2 cases instead of having intrusive changes in the common
932      * logic, we modify the cfg to be a stage-2 translation only in case of
933      * nested, and then restore it after.
934      */
935     if (desc_s2_translation) {
936         asid = cfg->asid;
937         stage = cfg->stage;
938         cfg->asid = -1;
939         cfg->stage = SMMU_STAGE_2;
940     }
941 
942     cached_entry = smmu_translate(bs, cfg, addr, flag, &ptw_info);
943 
944     if (desc_s2_translation) {
945         cfg->asid = asid;
946         cfg->stage = stage;
947     }
948 
949     if (!cached_entry) {
950         /* All faults from PTW has S2 field. */
951         event->u.f_walk_eabt.s2 = (ptw_info.stage == SMMU_STAGE_2);
952         /*
953          * Fault class is set as follows based on "class" input to
954          * the function and to "ptw_info" from "smmu_translate()"
955          * For stage-1:
956          *   - EABT => CLASS_TT (hardcoded)
957          *   - other events => CLASS_IN (input to function)
958          * For stage-2 => CLASS_IN (input to function)
959          * For nested, for all events:
960          *  - CD fetch => CLASS_CD (input to function)
961          *  - walking stage 1 translation table  => CLASS_TT (from
962          *    is_ipa_descriptor or input in case of TTBx)
963          *  - s2 translation => CLASS_IN (input to function)
964          */
965         class = ptw_info.is_ipa_descriptor ? SMMU_CLASS_TT : class;
966         switch (ptw_info.type) {
967         case SMMU_PTW_ERR_WALK_EABT:
968             event->type = SMMU_EVT_F_WALK_EABT;
969             event->u.f_walk_eabt.rnw = flag & 0x1;
970             event->u.f_walk_eabt.class = (ptw_info.stage == SMMU_STAGE_2) ?
971                                           class : SMMU_CLASS_TT;
972             event->u.f_walk_eabt.addr2 = ptw_info.addr;
973             break;
974         case SMMU_PTW_ERR_TRANSLATION:
975             if (PTW_RECORD_FAULT(ptw_info, cfg)) {
976                 event->type = SMMU_EVT_F_TRANSLATION;
977                 event->u.f_translation.addr2 = ptw_info.addr;
978                 event->u.f_translation.class = class;
979                 event->u.f_translation.rnw = flag & 0x1;
980             }
981             break;
982         case SMMU_PTW_ERR_ADDR_SIZE:
983             if (PTW_RECORD_FAULT(ptw_info, cfg)) {
984                 event->type = SMMU_EVT_F_ADDR_SIZE;
985                 event->u.f_addr_size.addr2 = ptw_info.addr;
986                 event->u.f_addr_size.class = class;
987                 event->u.f_addr_size.rnw = flag & 0x1;
988             }
989             break;
990         case SMMU_PTW_ERR_ACCESS:
991             if (PTW_RECORD_FAULT(ptw_info, cfg)) {
992                 event->type = SMMU_EVT_F_ACCESS;
993                 event->u.f_access.addr2 = ptw_info.addr;
994                 event->u.f_access.class = class;
995                 event->u.f_access.rnw = flag & 0x1;
996             }
997             break;
998         case SMMU_PTW_ERR_PERMISSION:
999             if (PTW_RECORD_FAULT(ptw_info, cfg)) {
1000                 event->type = SMMU_EVT_F_PERMISSION;
1001                 event->u.f_permission.addr2 = ptw_info.addr;
1002                 event->u.f_permission.class = class;
1003                 event->u.f_permission.rnw = flag & 0x1;
1004             }
1005             break;
1006         default:
1007             g_assert_not_reached();
1008         }
1009         return SMMU_TRANS_ERROR;
1010     }
1011     *out_entry = cached_entry;
1012     return SMMU_TRANS_SUCCESS;
1013 }
1014 
1015 /*
1016  * Sets the InputAddr for an SMMU_TRANS_ERROR, as it can't be
1017  * set from all contexts, as smmuv3_get_config() can return
1018  * translation faults in case of nested translation (for CD
1019  * and TTBx). But in that case the iova is not known.
1020  */
1021 static void smmuv3_fixup_event(SMMUEventInfo *event, hwaddr iova)
1022 {
1023     switch (event->type) {
1024     case SMMU_EVT_F_WALK_EABT:
1025     case SMMU_EVT_F_TRANSLATION:
1026     case SMMU_EVT_F_ADDR_SIZE:
1027     case SMMU_EVT_F_ACCESS:
1028     case SMMU_EVT_F_PERMISSION:
1029         event->u.f_walk_eabt.addr = iova;
1030         break;
1031     default:
1032         break;
1033     }
1034 }
1035 
1036 /* Entry point to SMMU, does everything. */
1037 static IOMMUTLBEntry smmuv3_translate(IOMMUMemoryRegion *mr, hwaddr addr,
1038                                       IOMMUAccessFlags flag, int iommu_idx)
1039 {
1040     SMMUDevice *sdev = container_of(mr, SMMUDevice, iommu);
1041     SMMUv3State *s = sdev->smmu;
1042     uint32_t sid = smmu_get_sid(sdev);
1043     SMMUEventInfo event = {.type = SMMU_EVT_NONE,
1044                            .sid = sid,
1045                            .inval_ste_allowed = false};
1046     SMMUTranslationStatus status;
1047     SMMUTransCfg *cfg = NULL;
1048     IOMMUTLBEntry entry = {
1049         .target_as = &address_space_memory,
1050         .iova = addr,
1051         .translated_addr = addr,
1052         .addr_mask = ~(hwaddr)0,
1053         .perm = IOMMU_NONE,
1054     };
1055     SMMUTLBEntry *cached_entry = NULL;
1056 
1057     qemu_mutex_lock(&s->mutex);
1058 
1059     if (!smmu_enabled(s)) {
1060         if (FIELD_EX32(s->gbpa, GBPA, ABORT)) {
1061             status = SMMU_TRANS_ABORT;
1062         } else {
1063             status = SMMU_TRANS_DISABLE;
1064         }
1065         goto epilogue;
1066     }
1067 
1068     cfg = smmuv3_get_config(sdev, &event);
1069     if (!cfg) {
1070         status = SMMU_TRANS_ERROR;
1071         goto epilogue;
1072     }
1073 
1074     if (cfg->aborted) {
1075         status = SMMU_TRANS_ABORT;
1076         goto epilogue;
1077     }
1078 
1079     if (cfg->bypassed) {
1080         status = SMMU_TRANS_BYPASS;
1081         goto epilogue;
1082     }
1083 
1084     status = smmuv3_do_translate(s, addr, cfg, &event, flag,
1085                                  &cached_entry, SMMU_CLASS_IN);
1086 
1087 epilogue:
1088     qemu_mutex_unlock(&s->mutex);
1089     switch (status) {
1090     case SMMU_TRANS_SUCCESS:
1091         entry.perm = cached_entry->entry.perm;
1092         entry.translated_addr = CACHED_ENTRY_TO_ADDR(cached_entry, addr);
1093         entry.addr_mask = cached_entry->entry.addr_mask;
1094         trace_smmuv3_translate_success(mr->parent_obj.name, sid, addr,
1095                                        entry.translated_addr, entry.perm,
1096                                        cfg->stage);
1097         break;
1098     case SMMU_TRANS_DISABLE:
1099         entry.perm = flag;
1100         entry.addr_mask = ~TARGET_PAGE_MASK;
1101         trace_smmuv3_translate_disable(mr->parent_obj.name, sid, addr,
1102                                       entry.perm);
1103         break;
1104     case SMMU_TRANS_BYPASS:
1105         entry.perm = flag;
1106         entry.addr_mask = ~TARGET_PAGE_MASK;
1107         trace_smmuv3_translate_bypass(mr->parent_obj.name, sid, addr,
1108                                       entry.perm);
1109         break;
1110     case SMMU_TRANS_ABORT:
1111         /* no event is recorded on abort */
1112         trace_smmuv3_translate_abort(mr->parent_obj.name, sid, addr,
1113                                      entry.perm);
1114         break;
1115     case SMMU_TRANS_ERROR:
1116         smmuv3_fixup_event(&event, addr);
1117         qemu_log_mask(LOG_GUEST_ERROR,
1118                       "%s translation failed for iova=0x%"PRIx64" (%s)\n",
1119                       mr->parent_obj.name, addr, smmu_event_string(event.type));
1120         smmuv3_record_event(s, &event);
1121         break;
1122     }
1123 
1124     return entry;
1125 }
1126 
1127 /**
1128  * smmuv3_notify_iova - call the notifier @n for a given
1129  * @asid and @iova tuple.
1130  *
1131  * @mr: IOMMU mr region handle
1132  * @n: notifier to be called
1133  * @asid: address space ID or negative value if we don't care
1134  * @vmid: virtual machine ID or negative value if we don't care
1135  * @iova: iova
1136  * @tg: translation granule (if communicated through range invalidation)
1137  * @num_pages: number of @granule sized pages (if tg != 0), otherwise 1
1138  * @stage: Which stage(1 or 2) is used
1139  */
1140 static void smmuv3_notify_iova(IOMMUMemoryRegion *mr,
1141                                IOMMUNotifier *n,
1142                                int asid, int vmid,
1143                                dma_addr_t iova, uint8_t tg,
1144                                uint64_t num_pages, int stage)
1145 {
1146     SMMUDevice *sdev = container_of(mr, SMMUDevice, iommu);
1147     SMMUEventInfo eventinfo = {.inval_ste_allowed = true};
1148     SMMUTransCfg *cfg = smmuv3_get_config(sdev, &eventinfo);
1149     IOMMUTLBEvent event;
1150     uint8_t granule;
1151 
1152     if (!cfg) {
1153         return;
1154     }
1155 
1156     /*
1157      * stage is passed from TLB invalidation commands which can be either
1158      * stage-1 or stage-2.
1159      * However, IOMMUTLBEvent only understands IOVA, for stage-1 or stage-2
1160      * SMMU instances we consider the input address as the IOVA, but when
1161      * nesting is used, we can't mix stage-1 and stage-2 addresses, so for
1162      * nesting only stage-1 is considered the IOVA and would be notified.
1163      */
1164     if ((stage == SMMU_STAGE_2) && (cfg->stage == SMMU_NESTED))
1165         return;
1166 
1167     if (!tg) {
1168         SMMUTransTableInfo *tt;
1169 
1170         if (asid >= 0 && cfg->asid != asid) {
1171             return;
1172         }
1173 
1174         if (vmid >= 0 && cfg->s2cfg.vmid != vmid) {
1175             return;
1176         }
1177 
1178         if (stage == SMMU_STAGE_1) {
1179             tt = select_tt(cfg, iova);
1180             if (!tt) {
1181                 return;
1182             }
1183             granule = tt->granule_sz;
1184         } else {
1185             granule = cfg->s2cfg.granule_sz;
1186         }
1187 
1188     } else {
1189         granule = tg * 2 + 10;
1190     }
1191 
1192     event.type = IOMMU_NOTIFIER_UNMAP;
1193     event.entry.target_as = &address_space_memory;
1194     event.entry.iova = iova;
1195     event.entry.addr_mask = num_pages * (1 << granule) - 1;
1196     event.entry.perm = IOMMU_NONE;
1197 
1198     memory_region_notify_iommu_one(n, &event);
1199 }
1200 
1201 /* invalidate an asid/vmid/iova range tuple in all mr's */
1202 static void smmuv3_inv_notifiers_iova(SMMUState *s, int asid, int vmid,
1203                                       dma_addr_t iova, uint8_t tg,
1204                                       uint64_t num_pages, int stage)
1205 {
1206     SMMUDevice *sdev;
1207 
1208     QLIST_FOREACH(sdev, &s->devices_with_notifiers, next) {
1209         IOMMUMemoryRegion *mr = &sdev->iommu;
1210         IOMMUNotifier *n;
1211 
1212         trace_smmuv3_inv_notifiers_iova(mr->parent_obj.name, asid, vmid,
1213                                         iova, tg, num_pages, stage);
1214 
1215         IOMMU_NOTIFIER_FOREACH(n, mr) {
1216             smmuv3_notify_iova(mr, n, asid, vmid, iova, tg, num_pages, stage);
1217         }
1218     }
1219 }
1220 
1221 static void smmuv3_range_inval(SMMUState *s, Cmd *cmd, SMMUStage stage)
1222 {
1223     dma_addr_t end, addr = CMD_ADDR(cmd);
1224     uint8_t type = CMD_TYPE(cmd);
1225     int vmid = -1;
1226     uint8_t scale = CMD_SCALE(cmd);
1227     uint8_t num = CMD_NUM(cmd);
1228     uint8_t ttl = CMD_TTL(cmd);
1229     bool leaf = CMD_LEAF(cmd);
1230     uint8_t tg = CMD_TG(cmd);
1231     uint64_t num_pages;
1232     uint8_t granule;
1233     int asid = -1;
1234     SMMUv3State *smmuv3 = ARM_SMMUV3(s);
1235 
1236     /* Only consider VMID if stage-2 is supported. */
1237     if (STAGE2_SUPPORTED(smmuv3)) {
1238         vmid = CMD_VMID(cmd);
1239     }
1240 
1241     if (type == SMMU_CMD_TLBI_NH_VA) {
1242         asid = CMD_ASID(cmd);
1243     }
1244 
1245     if (!tg) {
1246         trace_smmuv3_range_inval(vmid, asid, addr, tg, 1, ttl, leaf, stage);
1247         smmuv3_inv_notifiers_iova(s, asid, vmid, addr, tg, 1, stage);
1248         if (stage == SMMU_STAGE_1) {
1249             smmu_iotlb_inv_iova(s, asid, vmid, addr, tg, 1, ttl);
1250         } else {
1251             smmu_iotlb_inv_ipa(s, vmid, addr, tg, 1, ttl);
1252         }
1253         return;
1254     }
1255 
1256     /* RIL in use */
1257 
1258     num_pages = (num + 1) * BIT_ULL(scale);
1259     granule = tg * 2 + 10;
1260 
1261     /* Split invalidations into ^2 range invalidations */
1262     end = addr + (num_pages << granule) - 1;
1263 
1264     while (addr != end + 1) {
1265         uint64_t mask = dma_aligned_pow2_mask(addr, end, 64);
1266 
1267         num_pages = (mask + 1) >> granule;
1268         trace_smmuv3_range_inval(vmid, asid, addr, tg, num_pages,
1269                                  ttl, leaf, stage);
1270         smmuv3_inv_notifiers_iova(s, asid, vmid, addr, tg, num_pages, stage);
1271         if (stage == SMMU_STAGE_1) {
1272             smmu_iotlb_inv_iova(s, asid, vmid, addr, tg, num_pages, ttl);
1273         } else {
1274             smmu_iotlb_inv_ipa(s, vmid, addr, tg, num_pages, ttl);
1275         }
1276         addr += mask + 1;
1277     }
1278 }
1279 
1280 static gboolean
1281 smmuv3_invalidate_ste(gpointer key, gpointer value, gpointer user_data)
1282 {
1283     SMMUDevice *sdev = (SMMUDevice *)key;
1284     uint32_t sid = smmu_get_sid(sdev);
1285     SMMUSIDRange *sid_range = (SMMUSIDRange *)user_data;
1286 
1287     if (sid < sid_range->start || sid > sid_range->end) {
1288         return false;
1289     }
1290     trace_smmuv3_config_cache_inv(sid);
1291     return true;
1292 }
1293 
1294 static int smmuv3_cmdq_consume(SMMUv3State *s)
1295 {
1296     SMMUState *bs = ARM_SMMU(s);
1297     SMMUCmdError cmd_error = SMMU_CERROR_NONE;
1298     SMMUQueue *q = &s->cmdq;
1299     SMMUCommandType type = 0;
1300 
1301     if (!smmuv3_cmdq_enabled(s)) {
1302         return 0;
1303     }
1304     /*
1305      * some commands depend on register values, typically CR0. In case those
1306      * register values change while handling the command, spec says it
1307      * is UNPREDICTABLE whether the command is interpreted under the new
1308      * or old value.
1309      */
1310 
1311     while (!smmuv3_q_empty(q)) {
1312         uint32_t pending = s->gerror ^ s->gerrorn;
1313         Cmd cmd;
1314 
1315         trace_smmuv3_cmdq_consume(Q_PROD(q), Q_CONS(q),
1316                                   Q_PROD_WRAP(q), Q_CONS_WRAP(q));
1317 
1318         if (FIELD_EX32(pending, GERROR, CMDQ_ERR)) {
1319             break;
1320         }
1321 
1322         if (queue_read(q, &cmd) != MEMTX_OK) {
1323             cmd_error = SMMU_CERROR_ABT;
1324             break;
1325         }
1326 
1327         type = CMD_TYPE(&cmd);
1328 
1329         trace_smmuv3_cmdq_opcode(smmu_cmd_string(type));
1330 
1331         qemu_mutex_lock(&s->mutex);
1332         switch (type) {
1333         case SMMU_CMD_SYNC:
1334             if (CMD_SYNC_CS(&cmd) & CMD_SYNC_SIG_IRQ) {
1335                 smmuv3_trigger_irq(s, SMMU_IRQ_CMD_SYNC, 0);
1336             }
1337             break;
1338         case SMMU_CMD_PREFETCH_CONFIG:
1339         case SMMU_CMD_PREFETCH_ADDR:
1340             break;
1341         case SMMU_CMD_CFGI_STE:
1342         {
1343             uint32_t sid = CMD_SID(&cmd);
1344             SMMUDevice *sdev = smmu_find_sdev(bs, sid);
1345 
1346             if (CMD_SSEC(&cmd)) {
1347                 cmd_error = SMMU_CERROR_ILL;
1348                 break;
1349             }
1350 
1351             if (!sdev) {
1352                 break;
1353             }
1354 
1355             trace_smmuv3_cmdq_cfgi_ste(sid);
1356             smmuv3_flush_config(sdev);
1357 
1358             break;
1359         }
1360         case SMMU_CMD_CFGI_STE_RANGE: /* same as SMMU_CMD_CFGI_ALL */
1361         {
1362             uint32_t sid = CMD_SID(&cmd), mask;
1363             uint8_t range = CMD_STE_RANGE(&cmd);
1364             SMMUSIDRange sid_range;
1365 
1366             if (CMD_SSEC(&cmd)) {
1367                 cmd_error = SMMU_CERROR_ILL;
1368                 break;
1369             }
1370 
1371             mask = (1ULL << (range + 1)) - 1;
1372             sid_range.start = sid & ~mask;
1373             sid_range.end = sid_range.start + mask;
1374 
1375             trace_smmuv3_cmdq_cfgi_ste_range(sid_range.start, sid_range.end);
1376             g_hash_table_foreach_remove(bs->configs, smmuv3_invalidate_ste,
1377                                         &sid_range);
1378             break;
1379         }
1380         case SMMU_CMD_CFGI_CD:
1381         case SMMU_CMD_CFGI_CD_ALL:
1382         {
1383             uint32_t sid = CMD_SID(&cmd);
1384             SMMUDevice *sdev = smmu_find_sdev(bs, sid);
1385 
1386             if (CMD_SSEC(&cmd)) {
1387                 cmd_error = SMMU_CERROR_ILL;
1388                 break;
1389             }
1390 
1391             if (!sdev) {
1392                 break;
1393             }
1394 
1395             trace_smmuv3_cmdq_cfgi_cd(sid);
1396             smmuv3_flush_config(sdev);
1397             break;
1398         }
1399         case SMMU_CMD_TLBI_NH_ASID:
1400         {
1401             int asid = CMD_ASID(&cmd);
1402             int vmid = -1;
1403 
1404             if (!STAGE1_SUPPORTED(s)) {
1405                 cmd_error = SMMU_CERROR_ILL;
1406                 break;
1407             }
1408 
1409             /*
1410              * VMID is only matched when stage 2 is supported, otherwise set it
1411              * to -1 as the value used for stage-1 only VMIDs.
1412              */
1413             if (STAGE2_SUPPORTED(s)) {
1414                 vmid = CMD_VMID(&cmd);
1415             }
1416 
1417             trace_smmuv3_cmdq_tlbi_nh_asid(asid);
1418             smmu_inv_notifiers_all(&s->smmu_state);
1419             smmu_iotlb_inv_asid_vmid(bs, asid, vmid);
1420             break;
1421         }
1422         case SMMU_CMD_TLBI_NH_ALL:
1423         {
1424             int vmid = -1;
1425 
1426             if (!STAGE1_SUPPORTED(s)) {
1427                 cmd_error = SMMU_CERROR_ILL;
1428                 break;
1429             }
1430 
1431             /*
1432              * If stage-2 is supported, invalidate for this VMID only, otherwise
1433              * invalidate the whole thing.
1434              */
1435             if (STAGE2_SUPPORTED(s)) {
1436                 vmid = CMD_VMID(&cmd);
1437                 trace_smmuv3_cmdq_tlbi_nh(vmid);
1438                 smmu_iotlb_inv_vmid_s1(bs, vmid);
1439                 break;
1440             }
1441             QEMU_FALLTHROUGH;
1442         }
1443         case SMMU_CMD_TLBI_NSNH_ALL:
1444             trace_smmuv3_cmdq_tlbi_nsnh();
1445             smmu_inv_notifiers_all(&s->smmu_state);
1446             smmu_iotlb_inv_all(bs);
1447             break;
1448         case SMMU_CMD_TLBI_NH_VAA:
1449         case SMMU_CMD_TLBI_NH_VA:
1450             if (!STAGE1_SUPPORTED(s)) {
1451                 cmd_error = SMMU_CERROR_ILL;
1452                 break;
1453             }
1454             smmuv3_range_inval(bs, &cmd, SMMU_STAGE_1);
1455             break;
1456         case SMMU_CMD_TLBI_S12_VMALL:
1457         {
1458             int vmid = CMD_VMID(&cmd);
1459 
1460             if (!STAGE2_SUPPORTED(s)) {
1461                 cmd_error = SMMU_CERROR_ILL;
1462                 break;
1463             }
1464 
1465             trace_smmuv3_cmdq_tlbi_s12_vmid(vmid);
1466             smmu_inv_notifiers_all(&s->smmu_state);
1467             smmu_iotlb_inv_vmid(bs, vmid);
1468             break;
1469         }
1470         case SMMU_CMD_TLBI_S2_IPA:
1471             if (!STAGE2_SUPPORTED(s)) {
1472                 cmd_error = SMMU_CERROR_ILL;
1473                 break;
1474             }
1475             /*
1476              * As currently only either s1 or s2 are supported
1477              * we can reuse same function for s2.
1478              */
1479             smmuv3_range_inval(bs, &cmd, SMMU_STAGE_2);
1480             break;
1481         case SMMU_CMD_TLBI_EL3_ALL:
1482         case SMMU_CMD_TLBI_EL3_VA:
1483         case SMMU_CMD_TLBI_EL2_ALL:
1484         case SMMU_CMD_TLBI_EL2_ASID:
1485         case SMMU_CMD_TLBI_EL2_VA:
1486         case SMMU_CMD_TLBI_EL2_VAA:
1487         case SMMU_CMD_ATC_INV:
1488         case SMMU_CMD_PRI_RESP:
1489         case SMMU_CMD_RESUME:
1490         case SMMU_CMD_STALL_TERM:
1491             trace_smmuv3_unhandled_cmd(type);
1492             break;
1493         default:
1494             cmd_error = SMMU_CERROR_ILL;
1495             break;
1496         }
1497         qemu_mutex_unlock(&s->mutex);
1498         if (cmd_error) {
1499             if (cmd_error == SMMU_CERROR_ILL) {
1500                 qemu_log_mask(LOG_GUEST_ERROR,
1501                               "Illegal command type: %d\n", CMD_TYPE(&cmd));
1502             }
1503             break;
1504         }
1505         /*
1506          * We only increment the cons index after the completion of
1507          * the command. We do that because the SYNC returns immediately
1508          * and does not check the completion of previous commands
1509          */
1510         queue_cons_incr(q);
1511     }
1512 
1513     if (cmd_error) {
1514         trace_smmuv3_cmdq_consume_error(smmu_cmd_string(type), cmd_error);
1515         smmu_write_cmdq_err(s, cmd_error);
1516         smmuv3_trigger_irq(s, SMMU_IRQ_GERROR, R_GERROR_CMDQ_ERR_MASK);
1517     }
1518 
1519     trace_smmuv3_cmdq_consume_out(Q_PROD(q), Q_CONS(q),
1520                                   Q_PROD_WRAP(q), Q_CONS_WRAP(q));
1521 
1522     return 0;
1523 }
1524 
1525 static MemTxResult smmu_writell(SMMUv3State *s, hwaddr offset,
1526                                uint64_t data, MemTxAttrs attrs)
1527 {
1528     switch (offset) {
1529     case A_GERROR_IRQ_CFG0:
1530         s->gerror_irq_cfg0 = data;
1531         return MEMTX_OK;
1532     case A_STRTAB_BASE:
1533         s->strtab_base = data;
1534         return MEMTX_OK;
1535     case A_CMDQ_BASE:
1536         s->cmdq.base = data;
1537         s->cmdq.log2size = extract64(s->cmdq.base, 0, 5);
1538         if (s->cmdq.log2size > SMMU_CMDQS) {
1539             s->cmdq.log2size = SMMU_CMDQS;
1540         }
1541         return MEMTX_OK;
1542     case A_EVENTQ_BASE:
1543         s->eventq.base = data;
1544         s->eventq.log2size = extract64(s->eventq.base, 0, 5);
1545         if (s->eventq.log2size > SMMU_EVENTQS) {
1546             s->eventq.log2size = SMMU_EVENTQS;
1547         }
1548         return MEMTX_OK;
1549     case A_EVENTQ_IRQ_CFG0:
1550         s->eventq_irq_cfg0 = data;
1551         return MEMTX_OK;
1552     default:
1553         qemu_log_mask(LOG_UNIMP,
1554                       "%s Unexpected 64-bit access to 0x%"PRIx64" (WI)\n",
1555                       __func__, offset);
1556         return MEMTX_OK;
1557     }
1558 }
1559 
1560 static MemTxResult smmu_writel(SMMUv3State *s, hwaddr offset,
1561                                uint64_t data, MemTxAttrs attrs)
1562 {
1563     switch (offset) {
1564     case A_CR0:
1565         s->cr[0] = data;
1566         s->cr0ack = data & ~SMMU_CR0_RESERVED;
1567         /* in case the command queue has been enabled */
1568         smmuv3_cmdq_consume(s);
1569         return MEMTX_OK;
1570     case A_CR1:
1571         s->cr[1] = data;
1572         return MEMTX_OK;
1573     case A_CR2:
1574         s->cr[2] = data;
1575         return MEMTX_OK;
1576     case A_IRQ_CTRL:
1577         s->irq_ctrl = data;
1578         return MEMTX_OK;
1579     case A_GERRORN:
1580         smmuv3_write_gerrorn(s, data);
1581         /*
1582          * By acknowledging the CMDQ_ERR, SW may notify cmds can
1583          * be processed again
1584          */
1585         smmuv3_cmdq_consume(s);
1586         return MEMTX_OK;
1587     case A_GERROR_IRQ_CFG0: /* 64b */
1588         s->gerror_irq_cfg0 = deposit64(s->gerror_irq_cfg0, 0, 32, data);
1589         return MEMTX_OK;
1590     case A_GERROR_IRQ_CFG0 + 4:
1591         s->gerror_irq_cfg0 = deposit64(s->gerror_irq_cfg0, 32, 32, data);
1592         return MEMTX_OK;
1593     case A_GERROR_IRQ_CFG1:
1594         s->gerror_irq_cfg1 = data;
1595         return MEMTX_OK;
1596     case A_GERROR_IRQ_CFG2:
1597         s->gerror_irq_cfg2 = data;
1598         return MEMTX_OK;
1599     case A_GBPA:
1600         /*
1601          * If UPDATE is not set, the write is ignored. This is the only
1602          * permitted behavior in SMMUv3.2 and later.
1603          */
1604         if (data & R_GBPA_UPDATE_MASK) {
1605             /* Ignore update bit as write is synchronous. */
1606             s->gbpa = data & ~R_GBPA_UPDATE_MASK;
1607         }
1608         return MEMTX_OK;
1609     case A_STRTAB_BASE: /* 64b */
1610         s->strtab_base = deposit64(s->strtab_base, 0, 32, data);
1611         return MEMTX_OK;
1612     case A_STRTAB_BASE + 4:
1613         s->strtab_base = deposit64(s->strtab_base, 32, 32, data);
1614         return MEMTX_OK;
1615     case A_STRTAB_BASE_CFG:
1616         s->strtab_base_cfg = data;
1617         if (FIELD_EX32(data, STRTAB_BASE_CFG, FMT) == 1) {
1618             s->sid_split = FIELD_EX32(data, STRTAB_BASE_CFG, SPLIT);
1619             s->features |= SMMU_FEATURE_2LVL_STE;
1620         }
1621         return MEMTX_OK;
1622     case A_CMDQ_BASE: /* 64b */
1623         s->cmdq.base = deposit64(s->cmdq.base, 0, 32, data);
1624         s->cmdq.log2size = extract64(s->cmdq.base, 0, 5);
1625         if (s->cmdq.log2size > SMMU_CMDQS) {
1626             s->cmdq.log2size = SMMU_CMDQS;
1627         }
1628         return MEMTX_OK;
1629     case A_CMDQ_BASE + 4: /* 64b */
1630         s->cmdq.base = deposit64(s->cmdq.base, 32, 32, data);
1631         return MEMTX_OK;
1632     case A_CMDQ_PROD:
1633         s->cmdq.prod = data;
1634         smmuv3_cmdq_consume(s);
1635         return MEMTX_OK;
1636     case A_CMDQ_CONS:
1637         s->cmdq.cons = data;
1638         return MEMTX_OK;
1639     case A_EVENTQ_BASE: /* 64b */
1640         s->eventq.base = deposit64(s->eventq.base, 0, 32, data);
1641         s->eventq.log2size = extract64(s->eventq.base, 0, 5);
1642         if (s->eventq.log2size > SMMU_EVENTQS) {
1643             s->eventq.log2size = SMMU_EVENTQS;
1644         }
1645         return MEMTX_OK;
1646     case A_EVENTQ_BASE + 4:
1647         s->eventq.base = deposit64(s->eventq.base, 32, 32, data);
1648         return MEMTX_OK;
1649     case A_EVENTQ_PROD:
1650         s->eventq.prod = data;
1651         return MEMTX_OK;
1652     case A_EVENTQ_CONS:
1653         s->eventq.cons = data;
1654         return MEMTX_OK;
1655     case A_EVENTQ_IRQ_CFG0: /* 64b */
1656         s->eventq_irq_cfg0 = deposit64(s->eventq_irq_cfg0, 0, 32, data);
1657         return MEMTX_OK;
1658     case A_EVENTQ_IRQ_CFG0 + 4:
1659         s->eventq_irq_cfg0 = deposit64(s->eventq_irq_cfg0, 32, 32, data);
1660         return MEMTX_OK;
1661     case A_EVENTQ_IRQ_CFG1:
1662         s->eventq_irq_cfg1 = data;
1663         return MEMTX_OK;
1664     case A_EVENTQ_IRQ_CFG2:
1665         s->eventq_irq_cfg2 = data;
1666         return MEMTX_OK;
1667     default:
1668         qemu_log_mask(LOG_UNIMP,
1669                       "%s Unexpected 32-bit access to 0x%"PRIx64" (WI)\n",
1670                       __func__, offset);
1671         return MEMTX_OK;
1672     }
1673 }
1674 
1675 static MemTxResult smmu_write_mmio(void *opaque, hwaddr offset, uint64_t data,
1676                                    unsigned size, MemTxAttrs attrs)
1677 {
1678     SMMUState *sys = opaque;
1679     SMMUv3State *s = ARM_SMMUV3(sys);
1680     MemTxResult r;
1681 
1682     /* CONSTRAINED UNPREDICTABLE choice to have page0/1 be exact aliases */
1683     offset &= ~0x10000;
1684 
1685     switch (size) {
1686     case 8:
1687         r = smmu_writell(s, offset, data, attrs);
1688         break;
1689     case 4:
1690         r = smmu_writel(s, offset, data, attrs);
1691         break;
1692     default:
1693         r = MEMTX_ERROR;
1694         break;
1695     }
1696 
1697     trace_smmuv3_write_mmio(offset, data, size, r);
1698     return r;
1699 }
1700 
1701 static MemTxResult smmu_readll(SMMUv3State *s, hwaddr offset,
1702                                uint64_t *data, MemTxAttrs attrs)
1703 {
1704     switch (offset) {
1705     case A_GERROR_IRQ_CFG0:
1706         *data = s->gerror_irq_cfg0;
1707         return MEMTX_OK;
1708     case A_STRTAB_BASE:
1709         *data = s->strtab_base;
1710         return MEMTX_OK;
1711     case A_CMDQ_BASE:
1712         *data = s->cmdq.base;
1713         return MEMTX_OK;
1714     case A_EVENTQ_BASE:
1715         *data = s->eventq.base;
1716         return MEMTX_OK;
1717     default:
1718         *data = 0;
1719         qemu_log_mask(LOG_UNIMP,
1720                       "%s Unexpected 64-bit access to 0x%"PRIx64" (RAZ)\n",
1721                       __func__, offset);
1722         return MEMTX_OK;
1723     }
1724 }
1725 
1726 static MemTxResult smmu_readl(SMMUv3State *s, hwaddr offset,
1727                               uint64_t *data, MemTxAttrs attrs)
1728 {
1729     switch (offset) {
1730     case A_IDREGS ... A_IDREGS + 0x2f:
1731         *data = smmuv3_idreg(offset - A_IDREGS);
1732         return MEMTX_OK;
1733     case A_IDR0 ... A_IDR5:
1734         *data = s->idr[(offset - A_IDR0) / 4];
1735         return MEMTX_OK;
1736     case A_IIDR:
1737         *data = s->iidr;
1738         return MEMTX_OK;
1739     case A_AIDR:
1740         *data = s->aidr;
1741         return MEMTX_OK;
1742     case A_CR0:
1743         *data = s->cr[0];
1744         return MEMTX_OK;
1745     case A_CR0ACK:
1746         *data = s->cr0ack;
1747         return MEMTX_OK;
1748     case A_CR1:
1749         *data = s->cr[1];
1750         return MEMTX_OK;
1751     case A_CR2:
1752         *data = s->cr[2];
1753         return MEMTX_OK;
1754     case A_STATUSR:
1755         *data = s->statusr;
1756         return MEMTX_OK;
1757     case A_GBPA:
1758         *data = s->gbpa;
1759         return MEMTX_OK;
1760     case A_IRQ_CTRL:
1761     case A_IRQ_CTRL_ACK:
1762         *data = s->irq_ctrl;
1763         return MEMTX_OK;
1764     case A_GERROR:
1765         *data = s->gerror;
1766         return MEMTX_OK;
1767     case A_GERRORN:
1768         *data = s->gerrorn;
1769         return MEMTX_OK;
1770     case A_GERROR_IRQ_CFG0: /* 64b */
1771         *data = extract64(s->gerror_irq_cfg0, 0, 32);
1772         return MEMTX_OK;
1773     case A_GERROR_IRQ_CFG0 + 4:
1774         *data = extract64(s->gerror_irq_cfg0, 32, 32);
1775         return MEMTX_OK;
1776     case A_GERROR_IRQ_CFG1:
1777         *data = s->gerror_irq_cfg1;
1778         return MEMTX_OK;
1779     case A_GERROR_IRQ_CFG2:
1780         *data = s->gerror_irq_cfg2;
1781         return MEMTX_OK;
1782     case A_STRTAB_BASE: /* 64b */
1783         *data = extract64(s->strtab_base, 0, 32);
1784         return MEMTX_OK;
1785     case A_STRTAB_BASE + 4: /* 64b */
1786         *data = extract64(s->strtab_base, 32, 32);
1787         return MEMTX_OK;
1788     case A_STRTAB_BASE_CFG:
1789         *data = s->strtab_base_cfg;
1790         return MEMTX_OK;
1791     case A_CMDQ_BASE: /* 64b */
1792         *data = extract64(s->cmdq.base, 0, 32);
1793         return MEMTX_OK;
1794     case A_CMDQ_BASE + 4:
1795         *data = extract64(s->cmdq.base, 32, 32);
1796         return MEMTX_OK;
1797     case A_CMDQ_PROD:
1798         *data = s->cmdq.prod;
1799         return MEMTX_OK;
1800     case A_CMDQ_CONS:
1801         *data = s->cmdq.cons;
1802         return MEMTX_OK;
1803     case A_EVENTQ_BASE: /* 64b */
1804         *data = extract64(s->eventq.base, 0, 32);
1805         return MEMTX_OK;
1806     case A_EVENTQ_BASE + 4: /* 64b */
1807         *data = extract64(s->eventq.base, 32, 32);
1808         return MEMTX_OK;
1809     case A_EVENTQ_PROD:
1810         *data = s->eventq.prod;
1811         return MEMTX_OK;
1812     case A_EVENTQ_CONS:
1813         *data = s->eventq.cons;
1814         return MEMTX_OK;
1815     default:
1816         *data = 0;
1817         qemu_log_mask(LOG_UNIMP,
1818                       "%s unhandled 32-bit access at 0x%"PRIx64" (RAZ)\n",
1819                       __func__, offset);
1820         return MEMTX_OK;
1821     }
1822 }
1823 
1824 static MemTxResult smmu_read_mmio(void *opaque, hwaddr offset, uint64_t *data,
1825                                   unsigned size, MemTxAttrs attrs)
1826 {
1827     SMMUState *sys = opaque;
1828     SMMUv3State *s = ARM_SMMUV3(sys);
1829     MemTxResult r;
1830 
1831     /* CONSTRAINED UNPREDICTABLE choice to have page0/1 be exact aliases */
1832     offset &= ~0x10000;
1833 
1834     switch (size) {
1835     case 8:
1836         r = smmu_readll(s, offset, data, attrs);
1837         break;
1838     case 4:
1839         r = smmu_readl(s, offset, data, attrs);
1840         break;
1841     default:
1842         r = MEMTX_ERROR;
1843         break;
1844     }
1845 
1846     trace_smmuv3_read_mmio(offset, *data, size, r);
1847     return r;
1848 }
1849 
1850 static const MemoryRegionOps smmu_mem_ops = {
1851     .read_with_attrs = smmu_read_mmio,
1852     .write_with_attrs = smmu_write_mmio,
1853     .endianness = DEVICE_LITTLE_ENDIAN,
1854     .valid = {
1855         .min_access_size = 4,
1856         .max_access_size = 8,
1857     },
1858     .impl = {
1859         .min_access_size = 4,
1860         .max_access_size = 8,
1861     },
1862 };
1863 
1864 static void smmu_init_irq(SMMUv3State *s, SysBusDevice *dev)
1865 {
1866     int i;
1867 
1868     for (i = 0; i < ARRAY_SIZE(s->irq); i++) {
1869         sysbus_init_irq(dev, &s->irq[i]);
1870     }
1871 }
1872 
1873 static void smmu_reset_hold(Object *obj, ResetType type)
1874 {
1875     SMMUv3State *s = ARM_SMMUV3(obj);
1876     SMMUv3Class *c = ARM_SMMUV3_GET_CLASS(s);
1877 
1878     if (c->parent_phases.hold) {
1879         c->parent_phases.hold(obj, type);
1880     }
1881 
1882     smmuv3_init_regs(s);
1883 }
1884 
1885 static void smmu_realize(DeviceState *d, Error **errp)
1886 {
1887     SMMUState *sys = ARM_SMMU(d);
1888     SMMUv3State *s = ARM_SMMUV3(sys);
1889     SMMUv3Class *c = ARM_SMMUV3_GET_CLASS(s);
1890     SysBusDevice *dev = SYS_BUS_DEVICE(d);
1891     Error *local_err = NULL;
1892 
1893     c->parent_realize(d, &local_err);
1894     if (local_err) {
1895         error_propagate(errp, local_err);
1896         return;
1897     }
1898 
1899     qemu_mutex_init(&s->mutex);
1900 
1901     memory_region_init_io(&sys->iomem, OBJECT(s),
1902                           &smmu_mem_ops, sys, TYPE_ARM_SMMUV3, 0x20000);
1903 
1904     sys->mrtypename = TYPE_SMMUV3_IOMMU_MEMORY_REGION;
1905 
1906     sysbus_init_mmio(dev, &sys->iomem);
1907 
1908     smmu_init_irq(s, dev);
1909 }
1910 
1911 static const VMStateDescription vmstate_smmuv3_queue = {
1912     .name = "smmuv3_queue",
1913     .version_id = 1,
1914     .minimum_version_id = 1,
1915     .fields = (const VMStateField[]) {
1916         VMSTATE_UINT64(base, SMMUQueue),
1917         VMSTATE_UINT32(prod, SMMUQueue),
1918         VMSTATE_UINT32(cons, SMMUQueue),
1919         VMSTATE_UINT8(log2size, SMMUQueue),
1920         VMSTATE_END_OF_LIST(),
1921     },
1922 };
1923 
1924 static bool smmuv3_gbpa_needed(void *opaque)
1925 {
1926     SMMUv3State *s = opaque;
1927 
1928     /* Only migrate GBPA if it has different reset value. */
1929     return s->gbpa != SMMU_GBPA_RESET_VAL;
1930 }
1931 
1932 static const VMStateDescription vmstate_gbpa = {
1933     .name = "smmuv3/gbpa",
1934     .version_id = 1,
1935     .minimum_version_id = 1,
1936     .needed = smmuv3_gbpa_needed,
1937     .fields = (const VMStateField[]) {
1938         VMSTATE_UINT32(gbpa, SMMUv3State),
1939         VMSTATE_END_OF_LIST()
1940     }
1941 };
1942 
1943 static const VMStateDescription vmstate_smmuv3 = {
1944     .name = "smmuv3",
1945     .version_id = 1,
1946     .minimum_version_id = 1,
1947     .priority = MIG_PRI_IOMMU,
1948     .fields = (const VMStateField[]) {
1949         VMSTATE_UINT32(features, SMMUv3State),
1950         VMSTATE_UINT8(sid_size, SMMUv3State),
1951         VMSTATE_UINT8(sid_split, SMMUv3State),
1952 
1953         VMSTATE_UINT32_ARRAY(cr, SMMUv3State, 3),
1954         VMSTATE_UINT32(cr0ack, SMMUv3State),
1955         VMSTATE_UINT32(statusr, SMMUv3State),
1956         VMSTATE_UINT32(irq_ctrl, SMMUv3State),
1957         VMSTATE_UINT32(gerror, SMMUv3State),
1958         VMSTATE_UINT32(gerrorn, SMMUv3State),
1959         VMSTATE_UINT64(gerror_irq_cfg0, SMMUv3State),
1960         VMSTATE_UINT32(gerror_irq_cfg1, SMMUv3State),
1961         VMSTATE_UINT32(gerror_irq_cfg2, SMMUv3State),
1962         VMSTATE_UINT64(strtab_base, SMMUv3State),
1963         VMSTATE_UINT32(strtab_base_cfg, SMMUv3State),
1964         VMSTATE_UINT64(eventq_irq_cfg0, SMMUv3State),
1965         VMSTATE_UINT32(eventq_irq_cfg1, SMMUv3State),
1966         VMSTATE_UINT32(eventq_irq_cfg2, SMMUv3State),
1967 
1968         VMSTATE_STRUCT(cmdq, SMMUv3State, 0, vmstate_smmuv3_queue, SMMUQueue),
1969         VMSTATE_STRUCT(eventq, SMMUv3State, 0, vmstate_smmuv3_queue, SMMUQueue),
1970 
1971         VMSTATE_END_OF_LIST(),
1972     },
1973     .subsections = (const VMStateDescription * const []) {
1974         &vmstate_gbpa,
1975         NULL
1976     }
1977 };
1978 
1979 static Property smmuv3_properties[] = {
1980     /*
1981      * Stages of translation advertised.
1982      * "1": Stage 1
1983      * "2": Stage 2
1984      * Defaults to stage 1
1985      */
1986     DEFINE_PROP_STRING("stage", SMMUv3State, stage),
1987     DEFINE_PROP_END_OF_LIST()
1988 };
1989 
1990 static void smmuv3_instance_init(Object *obj)
1991 {
1992     /* Nothing much to do here as of now */
1993 }
1994 
1995 static void smmuv3_class_init(ObjectClass *klass, void *data)
1996 {
1997     DeviceClass *dc = DEVICE_CLASS(klass);
1998     ResettableClass *rc = RESETTABLE_CLASS(klass);
1999     SMMUv3Class *c = ARM_SMMUV3_CLASS(klass);
2000 
2001     dc->vmsd = &vmstate_smmuv3;
2002     resettable_class_set_parent_phases(rc, NULL, smmu_reset_hold, NULL,
2003                                        &c->parent_phases);
2004     device_class_set_parent_realize(dc, smmu_realize,
2005                                     &c->parent_realize);
2006     device_class_set_props(dc, smmuv3_properties);
2007 }
2008 
2009 static int smmuv3_notify_flag_changed(IOMMUMemoryRegion *iommu,
2010                                       IOMMUNotifierFlag old,
2011                                       IOMMUNotifierFlag new,
2012                                       Error **errp)
2013 {
2014     SMMUDevice *sdev = container_of(iommu, SMMUDevice, iommu);
2015     SMMUv3State *s3 = sdev->smmu;
2016     SMMUState *s = &(s3->smmu_state);
2017 
2018     if (new & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
2019         error_setg(errp, "SMMUv3 does not support dev-iotlb yet");
2020         return -EINVAL;
2021     }
2022 
2023     if (new & IOMMU_NOTIFIER_MAP) {
2024         error_setg(errp,
2025                    "device %02x.%02x.%x requires iommu MAP notifier which is "
2026                    "not currently supported", pci_bus_num(sdev->bus),
2027                    PCI_SLOT(sdev->devfn), PCI_FUNC(sdev->devfn));
2028         return -EINVAL;
2029     }
2030 
2031     if (old == IOMMU_NOTIFIER_NONE) {
2032         trace_smmuv3_notify_flag_add(iommu->parent_obj.name);
2033         QLIST_INSERT_HEAD(&s->devices_with_notifiers, sdev, next);
2034     } else if (new == IOMMU_NOTIFIER_NONE) {
2035         trace_smmuv3_notify_flag_del(iommu->parent_obj.name);
2036         QLIST_REMOVE(sdev, next);
2037     }
2038     return 0;
2039 }
2040 
2041 static void smmuv3_iommu_memory_region_class_init(ObjectClass *klass,
2042                                                   void *data)
2043 {
2044     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
2045 
2046     imrc->translate = smmuv3_translate;
2047     imrc->notify_flag_changed = smmuv3_notify_flag_changed;
2048 }
2049 
2050 static const TypeInfo smmuv3_type_info = {
2051     .name          = TYPE_ARM_SMMUV3,
2052     .parent        = TYPE_ARM_SMMU,
2053     .instance_size = sizeof(SMMUv3State),
2054     .instance_init = smmuv3_instance_init,
2055     .class_size    = sizeof(SMMUv3Class),
2056     .class_init    = smmuv3_class_init,
2057 };
2058 
2059 static const TypeInfo smmuv3_iommu_memory_region_info = {
2060     .parent = TYPE_IOMMU_MEMORY_REGION,
2061     .name = TYPE_SMMUV3_IOMMU_MEMORY_REGION,
2062     .class_init = smmuv3_iommu_memory_region_class_init,
2063 };
2064 
2065 static void smmuv3_register_types(void)
2066 {
2067     type_register(&smmuv3_type_info);
2068     type_register(&smmuv3_iommu_memory_region_info);
2069 }
2070 
2071 type_init(smmuv3_register_types)
2072 
2073