xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision 259ebed4)
1 /*
2  * QEMU emulation of an Intel IOMMU (VT-d)
3  *   (DMA Remapping device)
4  *
5  * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com>
6  * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12 
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17 
18  * You should have received a copy of the GNU General Public License along
19  * with this program; if not, see <http://www.gnu.org/licenses/>.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/error-report.h"
24 #include "qemu/main-loop.h"
25 #include "qapi/error.h"
26 #include "hw/sysbus.h"
27 #include "intel_iommu_internal.h"
28 #include "hw/pci/pci.h"
29 #include "hw/pci/pci_bus.h"
30 #include "hw/qdev-properties.h"
31 #include "hw/i386/pc.h"
32 #include "hw/i386/apic-msidef.h"
33 #include "hw/i386/x86-iommu.h"
34 #include "hw/pci-host/q35.h"
35 #include "sysemu/kvm.h"
36 #include "sysemu/dma.h"
37 #include "sysemu/sysemu.h"
38 #include "hw/i386/apic_internal.h"
39 #include "kvm/kvm_i386.h"
40 #include "migration/vmstate.h"
41 #include "trace.h"
42 
43 /* context entry operations */
44 #define VTD_CE_GET_RID2PASID(ce) \
45     ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK)
46 #define VTD_CE_GET_PASID_DIR_TABLE(ce) \
47     ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK)
48 
49 /* pe operations */
50 #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT)
51 #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW))
52 
53 /*
54  * PCI bus number (or SID) is not reliable since the device is usaully
55  * initialized before guest can configure the PCI bridge
56  * (SECONDARY_BUS_NUMBER).
57  */
58 struct vtd_as_key {
59     PCIBus *bus;
60     uint8_t devfn;
61     uint32_t pasid;
62 };
63 
64 struct vtd_iotlb_key {
65     uint64_t gfn;
66     uint32_t pasid;
67     uint16_t sid;
68     uint8_t level;
69 };
70 
71 static void vtd_address_space_refresh_all(IntelIOMMUState *s);
72 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n);
73 
74 static void vtd_panic_require_caching_mode(void)
75 {
76     error_report("We need to set caching-mode=on for intel-iommu to enable "
77                  "device assignment with IOMMU protection.");
78     exit(1);
79 }
80 
81 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val,
82                             uint64_t wmask, uint64_t w1cmask)
83 {
84     stq_le_p(&s->csr[addr], val);
85     stq_le_p(&s->wmask[addr], wmask);
86     stq_le_p(&s->w1cmask[addr], w1cmask);
87 }
88 
89 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask)
90 {
91     stq_le_p(&s->womask[addr], mask);
92 }
93 
94 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val,
95                             uint32_t wmask, uint32_t w1cmask)
96 {
97     stl_le_p(&s->csr[addr], val);
98     stl_le_p(&s->wmask[addr], wmask);
99     stl_le_p(&s->w1cmask[addr], w1cmask);
100 }
101 
102 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask)
103 {
104     stl_le_p(&s->womask[addr], mask);
105 }
106 
107 /* "External" get/set operations */
108 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
109 {
110     uint64_t oldval = ldq_le_p(&s->csr[addr]);
111     uint64_t wmask = ldq_le_p(&s->wmask[addr]);
112     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
113     stq_le_p(&s->csr[addr],
114              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
115 }
116 
117 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
118 {
119     uint32_t oldval = ldl_le_p(&s->csr[addr]);
120     uint32_t wmask = ldl_le_p(&s->wmask[addr]);
121     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
122     stl_le_p(&s->csr[addr],
123              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
124 }
125 
126 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr)
127 {
128     uint64_t val = ldq_le_p(&s->csr[addr]);
129     uint64_t womask = ldq_le_p(&s->womask[addr]);
130     return val & ~womask;
131 }
132 
133 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr)
134 {
135     uint32_t val = ldl_le_p(&s->csr[addr]);
136     uint32_t womask = ldl_le_p(&s->womask[addr]);
137     return val & ~womask;
138 }
139 
140 /* "Internal" get/set operations */
141 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr)
142 {
143     return ldq_le_p(&s->csr[addr]);
144 }
145 
146 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr)
147 {
148     return ldl_le_p(&s->csr[addr]);
149 }
150 
151 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val)
152 {
153     stq_le_p(&s->csr[addr], val);
154 }
155 
156 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
157                                         uint32_t clear, uint32_t mask)
158 {
159     uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
160     stl_le_p(&s->csr[addr], new_val);
161     return new_val;
162 }
163 
164 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
165                                         uint64_t clear, uint64_t mask)
166 {
167     uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
168     stq_le_p(&s->csr[addr], new_val);
169     return new_val;
170 }
171 
172 static inline void vtd_iommu_lock(IntelIOMMUState *s)
173 {
174     qemu_mutex_lock(&s->iommu_lock);
175 }
176 
177 static inline void vtd_iommu_unlock(IntelIOMMUState *s)
178 {
179     qemu_mutex_unlock(&s->iommu_lock);
180 }
181 
182 static void vtd_update_scalable_state(IntelIOMMUState *s)
183 {
184     uint64_t val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
185 
186     if (s->scalable_mode) {
187         s->root_scalable = val & VTD_RTADDR_SMT;
188     }
189 }
190 
191 static void vtd_update_iq_dw(IntelIOMMUState *s)
192 {
193     uint64_t val = vtd_get_quad_raw(s, DMAR_IQA_REG);
194 
195     if (s->ecap & VTD_ECAP_SMTS &&
196         val & VTD_IQA_DW_MASK) {
197         s->iq_dw = true;
198     } else {
199         s->iq_dw = false;
200     }
201 }
202 
203 /* Whether the address space needs to notify new mappings */
204 static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as)
205 {
206     return as->notifier_flags & IOMMU_NOTIFIER_MAP;
207 }
208 
209 /* GHashTable functions */
210 static gboolean vtd_iotlb_equal(gconstpointer v1, gconstpointer v2)
211 {
212     const struct vtd_iotlb_key *key1 = v1;
213     const struct vtd_iotlb_key *key2 = v2;
214 
215     return key1->sid == key2->sid &&
216            key1->pasid == key2->pasid &&
217            key1->level == key2->level &&
218            key1->gfn == key2->gfn;
219 }
220 
221 static guint vtd_iotlb_hash(gconstpointer v)
222 {
223     const struct vtd_iotlb_key *key = v;
224     uint64_t hash64 = key->gfn | ((uint64_t)(key->sid) << VTD_IOTLB_SID_SHIFT) |
225         (uint64_t)(key->level - 1) << VTD_IOTLB_LVL_SHIFT |
226         (uint64_t)(key->pasid) << VTD_IOTLB_PASID_SHIFT;
227 
228     return (guint)((hash64 >> 32) ^ (hash64 & 0xffffffffU));
229 }
230 
231 static gboolean vtd_as_equal(gconstpointer v1, gconstpointer v2)
232 {
233     const struct vtd_as_key *key1 = v1;
234     const struct vtd_as_key *key2 = v2;
235 
236     return (key1->bus == key2->bus) && (key1->devfn == key2->devfn) &&
237            (key1->pasid == key2->pasid);
238 }
239 
240 /*
241  * Note that we use pointer to PCIBus as the key, so hashing/shifting
242  * based on the pointer value is intended. Note that we deal with
243  * collisions through vtd_as_equal().
244  */
245 static guint vtd_as_hash(gconstpointer v)
246 {
247     const struct vtd_as_key *key = v;
248     guint value = (guint)(uintptr_t)key->bus;
249 
250     return (guint)(value << 8 | key->devfn);
251 }
252 
253 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value,
254                                           gpointer user_data)
255 {
256     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
257     uint16_t domain_id = *(uint16_t *)user_data;
258     return entry->domain_id == domain_id;
259 }
260 
261 /* The shift of an addr for a certain level of paging structure */
262 static inline uint32_t vtd_slpt_level_shift(uint32_t level)
263 {
264     assert(level != 0);
265     return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
266 }
267 
268 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level)
269 {
270     return ~((1ULL << vtd_slpt_level_shift(level)) - 1);
271 }
272 
273 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value,
274                                         gpointer user_data)
275 {
276     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
277     VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data;
278     uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask;
279     uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K;
280     return (entry->domain_id == info->domain_id) &&
281             (((entry->gfn & info->mask) == gfn) ||
282              (entry->gfn == gfn_tlb));
283 }
284 
285 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
286  * IntelIOMMUState to 1.  Must be called with IOMMU lock held.
287  */
288 static void vtd_reset_context_cache_locked(IntelIOMMUState *s)
289 {
290     VTDAddressSpace *vtd_as;
291     GHashTableIter as_it;
292 
293     trace_vtd_context_cache_reset();
294 
295     g_hash_table_iter_init(&as_it, s->vtd_address_spaces);
296 
297     while (g_hash_table_iter_next(&as_it, NULL, (void **)&vtd_as)) {
298         vtd_as->context_cache_entry.context_cache_gen = 0;
299     }
300     s->context_cache_gen = 1;
301 }
302 
303 /* Must be called with IOMMU lock held. */
304 static void vtd_reset_iotlb_locked(IntelIOMMUState *s)
305 {
306     assert(s->iotlb);
307     g_hash_table_remove_all(s->iotlb);
308 }
309 
310 static void vtd_reset_iotlb(IntelIOMMUState *s)
311 {
312     vtd_iommu_lock(s);
313     vtd_reset_iotlb_locked(s);
314     vtd_iommu_unlock(s);
315 }
316 
317 static void vtd_reset_caches(IntelIOMMUState *s)
318 {
319     vtd_iommu_lock(s);
320     vtd_reset_iotlb_locked(s);
321     vtd_reset_context_cache_locked(s);
322     vtd_iommu_unlock(s);
323 }
324 
325 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level)
326 {
327     return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K;
328 }
329 
330 /* Must be called with IOMMU lock held */
331 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id,
332                                        uint32_t pasid, hwaddr addr)
333 {
334     struct vtd_iotlb_key key;
335     VTDIOTLBEntry *entry;
336     int level;
337 
338     for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) {
339         key.gfn = vtd_get_iotlb_gfn(addr, level);
340         key.level = level;
341         key.sid = source_id;
342         key.pasid = pasid;
343         entry = g_hash_table_lookup(s->iotlb, &key);
344         if (entry) {
345             goto out;
346         }
347     }
348 
349 out:
350     return entry;
351 }
352 
353 /* Must be with IOMMU lock held */
354 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id,
355                              uint16_t domain_id, hwaddr addr, uint64_t slpte,
356                              uint8_t access_flags, uint32_t level,
357                              uint32_t pasid)
358 {
359     VTDIOTLBEntry *entry = g_malloc(sizeof(*entry));
360     struct vtd_iotlb_key *key = g_malloc(sizeof(*key));
361     uint64_t gfn = vtd_get_iotlb_gfn(addr, level);
362 
363     trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id);
364     if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) {
365         trace_vtd_iotlb_reset("iotlb exceeds size limit");
366         vtd_reset_iotlb_locked(s);
367     }
368 
369     entry->gfn = gfn;
370     entry->domain_id = domain_id;
371     entry->slpte = slpte;
372     entry->access_flags = access_flags;
373     entry->mask = vtd_slpt_level_page_mask(level);
374     entry->pasid = pasid;
375 
376     key->gfn = gfn;
377     key->sid = source_id;
378     key->level = level;
379     key->pasid = pasid;
380 
381     g_hash_table_replace(s->iotlb, key, entry);
382 }
383 
384 /* Given the reg addr of both the message data and address, generate an
385  * interrupt via MSI.
386  */
387 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
388                                    hwaddr mesg_data_reg)
389 {
390     MSIMessage msi;
391 
392     assert(mesg_data_reg < DMAR_REG_SIZE);
393     assert(mesg_addr_reg < DMAR_REG_SIZE);
394 
395     msi.address = vtd_get_long_raw(s, mesg_addr_reg);
396     msi.data = vtd_get_long_raw(s, mesg_data_reg);
397 
398     trace_vtd_irq_generate(msi.address, msi.data);
399 
400     apic_get_class(NULL)->send_msi(&msi);
401 }
402 
403 /* Generate a fault event to software via MSI if conditions are met.
404  * Notice that the value of FSTS_REG being passed to it should be the one
405  * before any update.
406  */
407 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
408 {
409     if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
410         pre_fsts & VTD_FSTS_IQE) {
411         error_report_once("There are previous interrupt conditions "
412                           "to be serviced by software, fault event "
413                           "is not generated");
414         return;
415     }
416     vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
417     if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
418         error_report_once("Interrupt Mask set, irq is not generated");
419     } else {
420         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
421         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
422     }
423 }
424 
425 /* Check if the Fault (F) field of the Fault Recording Register referenced by
426  * @index is Set.
427  */
428 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index)
429 {
430     /* Each reg is 128-bit */
431     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
432     addr += 8; /* Access the high 64-bit half */
433 
434     assert(index < DMAR_FRCD_REG_NR);
435 
436     return vtd_get_quad_raw(s, addr) & VTD_FRCD_F;
437 }
438 
439 /* Update the PPF field of Fault Status Register.
440  * Should be called whenever change the F field of any fault recording
441  * registers.
442  */
443 static void vtd_update_fsts_ppf(IntelIOMMUState *s)
444 {
445     uint32_t i;
446     uint32_t ppf_mask = 0;
447 
448     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
449         if (vtd_is_frcd_set(s, i)) {
450             ppf_mask = VTD_FSTS_PPF;
451             break;
452         }
453     }
454     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
455     trace_vtd_fsts_ppf(!!ppf_mask);
456 }
457 
458 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index)
459 {
460     /* Each reg is 128-bit */
461     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
462     addr += 8; /* Access the high 64-bit half */
463 
464     assert(index < DMAR_FRCD_REG_NR);
465 
466     vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
467     vtd_update_fsts_ppf(s);
468 }
469 
470 /* Must not update F field now, should be done later */
471 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index,
472                             uint64_t hi, uint64_t lo)
473 {
474     hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
475 
476     assert(index < DMAR_FRCD_REG_NR);
477 
478     vtd_set_quad_raw(s, frcd_reg_addr, lo);
479     vtd_set_quad_raw(s, frcd_reg_addr + 8, hi);
480 
481     trace_vtd_frr_new(index, hi, lo);
482 }
483 
484 /* Try to collapse multiple pending faults from the same requester */
485 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id)
486 {
487     uint32_t i;
488     uint64_t frcd_reg;
489     hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
490 
491     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
492         frcd_reg = vtd_get_quad_raw(s, addr);
493         if ((frcd_reg & VTD_FRCD_F) &&
494             ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
495             return true;
496         }
497         addr += 16; /* 128-bit for each */
498     }
499     return false;
500 }
501 
502 /* Log and report an DMAR (address translation) fault to software */
503 static void vtd_report_frcd_fault(IntelIOMMUState *s, uint64_t source_id,
504                                   uint64_t hi, uint64_t lo)
505 {
506     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
507 
508     if (fsts_reg & VTD_FSTS_PFO) {
509         error_report_once("New fault is not recorded due to "
510                           "Primary Fault Overflow");
511         return;
512     }
513 
514     if (vtd_try_collapse_fault(s, source_id)) {
515         error_report_once("New fault is not recorded due to "
516                           "compression of faults");
517         return;
518     }
519 
520     if (vtd_is_frcd_set(s, s->next_frcd_reg)) {
521         error_report_once("Next Fault Recording Reg is used, "
522                           "new fault is not recorded, set PFO field");
523         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
524         return;
525     }
526 
527     vtd_record_frcd(s, s->next_frcd_reg, hi, lo);
528 
529     if (fsts_reg & VTD_FSTS_PPF) {
530         error_report_once("There are pending faults already, "
531                           "fault event is not generated");
532         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg);
533         s->next_frcd_reg++;
534         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
535             s->next_frcd_reg = 0;
536         }
537     } else {
538         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
539                                 VTD_FSTS_FRI(s->next_frcd_reg));
540         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */
541         s->next_frcd_reg++;
542         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
543             s->next_frcd_reg = 0;
544         }
545         /* This case actually cause the PPF to be Set.
546          * So generate fault event (interrupt).
547          */
548          vtd_generate_fault_event(s, fsts_reg);
549     }
550 }
551 
552 /* Log and report an DMAR (address translation) fault to software */
553 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
554                                   hwaddr addr, VTDFaultReason fault,
555                                   bool is_write, bool is_pasid,
556                                   uint32_t pasid)
557 {
558     uint64_t hi, lo;
559 
560     assert(fault < VTD_FR_MAX);
561 
562     trace_vtd_dmar_fault(source_id, fault, addr, is_write);
563 
564     lo = VTD_FRCD_FI(addr);
565     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault) |
566          VTD_FRCD_PV(pasid) | VTD_FRCD_PP(is_pasid);
567     if (!is_write) {
568         hi |= VTD_FRCD_T;
569     }
570 
571     vtd_report_frcd_fault(s, source_id, hi, lo);
572 }
573 
574 
575 static void vtd_report_ir_fault(IntelIOMMUState *s, uint64_t source_id,
576                                 VTDFaultReason fault, uint16_t index)
577 {
578     uint64_t hi, lo;
579 
580     lo = VTD_FRCD_IR_IDX(index);
581     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
582 
583     vtd_report_frcd_fault(s, source_id, hi, lo);
584 }
585 
586 /* Handle Invalidation Queue Errors of queued invalidation interface error
587  * conditions.
588  */
589 static void vtd_handle_inv_queue_error(IntelIOMMUState *s)
590 {
591     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
592 
593     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE);
594     vtd_generate_fault_event(s, fsts_reg);
595 }
596 
597 /* Set the IWC field and try to generate an invalidation completion interrupt */
598 static void vtd_generate_completion_event(IntelIOMMUState *s)
599 {
600     if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) {
601         trace_vtd_inv_desc_wait_irq("One pending, skip current");
602         return;
603     }
604     vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC);
605     vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP);
606     if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) {
607         trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
608                                     "new event not generated");
609         return;
610     } else {
611         /* Generate the interrupt event */
612         trace_vtd_inv_desc_wait_irq("Generating complete event");
613         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
614         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
615     }
616 }
617 
618 static inline bool vtd_root_entry_present(IntelIOMMUState *s,
619                                           VTDRootEntry *re,
620                                           uint8_t devfn)
621 {
622     if (s->root_scalable && devfn > UINT8_MAX / 2) {
623         return re->hi & VTD_ROOT_ENTRY_P;
624     }
625 
626     return re->lo & VTD_ROOT_ENTRY_P;
627 }
628 
629 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index,
630                               VTDRootEntry *re)
631 {
632     dma_addr_t addr;
633 
634     addr = s->root + index * sizeof(*re);
635     if (dma_memory_read(&address_space_memory, addr,
636                         re, sizeof(*re), MEMTXATTRS_UNSPECIFIED)) {
637         re->lo = 0;
638         return -VTD_FR_ROOT_TABLE_INV;
639     }
640     re->lo = le64_to_cpu(re->lo);
641     re->hi = le64_to_cpu(re->hi);
642     return 0;
643 }
644 
645 static inline bool vtd_ce_present(VTDContextEntry *context)
646 {
647     return context->lo & VTD_CONTEXT_ENTRY_P;
648 }
649 
650 static int vtd_get_context_entry_from_root(IntelIOMMUState *s,
651                                            VTDRootEntry *re,
652                                            uint8_t index,
653                                            VTDContextEntry *ce)
654 {
655     dma_addr_t addr, ce_size;
656 
657     /* we have checked that root entry is present */
658     ce_size = s->root_scalable ? VTD_CTX_ENTRY_SCALABLE_SIZE :
659               VTD_CTX_ENTRY_LEGACY_SIZE;
660 
661     if (s->root_scalable && index > UINT8_MAX / 2) {
662         index = index & (~VTD_DEVFN_CHECK_MASK);
663         addr = re->hi & VTD_ROOT_ENTRY_CTP;
664     } else {
665         addr = re->lo & VTD_ROOT_ENTRY_CTP;
666     }
667 
668     addr = addr + index * ce_size;
669     if (dma_memory_read(&address_space_memory, addr,
670                         ce, ce_size, MEMTXATTRS_UNSPECIFIED)) {
671         return -VTD_FR_CONTEXT_TABLE_INV;
672     }
673 
674     ce->lo = le64_to_cpu(ce->lo);
675     ce->hi = le64_to_cpu(ce->hi);
676     if (ce_size == VTD_CTX_ENTRY_SCALABLE_SIZE) {
677         ce->val[2] = le64_to_cpu(ce->val[2]);
678         ce->val[3] = le64_to_cpu(ce->val[3]);
679     }
680     return 0;
681 }
682 
683 static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce)
684 {
685     return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
686 }
687 
688 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw)
689 {
690     return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw);
691 }
692 
693 /* Whether the pte indicates the address of the page frame */
694 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level)
695 {
696     return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
697 }
698 
699 /* Get the content of a spte located in @base_addr[@index] */
700 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index)
701 {
702     uint64_t slpte;
703 
704     assert(index < VTD_SL_PT_ENTRY_NR);
705 
706     if (dma_memory_read(&address_space_memory,
707                         base_addr + index * sizeof(slpte),
708                         &slpte, sizeof(slpte), MEMTXATTRS_UNSPECIFIED)) {
709         slpte = (uint64_t)-1;
710         return slpte;
711     }
712     slpte = le64_to_cpu(slpte);
713     return slpte;
714 }
715 
716 /* Given an iova and the level of paging structure, return the offset
717  * of current level.
718  */
719 static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level)
720 {
721     return (iova >> vtd_slpt_level_shift(level)) &
722             ((1ULL << VTD_SL_LEVEL_BITS) - 1);
723 }
724 
725 /* Check Capability Register to see if the @level of page-table is supported */
726 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level)
727 {
728     return VTD_CAP_SAGAW_MASK & s->cap &
729            (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
730 }
731 
732 /* Return true if check passed, otherwise false */
733 static inline bool vtd_pe_type_check(X86IOMMUState *x86_iommu,
734                                      VTDPASIDEntry *pe)
735 {
736     switch (VTD_PE_GET_TYPE(pe)) {
737     case VTD_SM_PASID_ENTRY_FLT:
738     case VTD_SM_PASID_ENTRY_SLT:
739     case VTD_SM_PASID_ENTRY_NESTED:
740         break;
741     case VTD_SM_PASID_ENTRY_PT:
742         if (!x86_iommu->pt_supported) {
743             return false;
744         }
745         break;
746     default:
747         /* Unknown type */
748         return false;
749     }
750     return true;
751 }
752 
753 static inline bool vtd_pdire_present(VTDPASIDDirEntry *pdire)
754 {
755     return pdire->val & 1;
756 }
757 
758 /**
759  * Caller of this function should check present bit if wants
760  * to use pdir entry for further usage except for fpd bit check.
761  */
762 static int vtd_get_pdire_from_pdir_table(dma_addr_t pasid_dir_base,
763                                          uint32_t pasid,
764                                          VTDPASIDDirEntry *pdire)
765 {
766     uint32_t index;
767     dma_addr_t addr, entry_size;
768 
769     index = VTD_PASID_DIR_INDEX(pasid);
770     entry_size = VTD_PASID_DIR_ENTRY_SIZE;
771     addr = pasid_dir_base + index * entry_size;
772     if (dma_memory_read(&address_space_memory, addr,
773                         pdire, entry_size, MEMTXATTRS_UNSPECIFIED)) {
774         return -VTD_FR_PASID_TABLE_INV;
775     }
776 
777     pdire->val = le64_to_cpu(pdire->val);
778 
779     return 0;
780 }
781 
782 static inline bool vtd_pe_present(VTDPASIDEntry *pe)
783 {
784     return pe->val[0] & VTD_PASID_ENTRY_P;
785 }
786 
787 static int vtd_get_pe_in_pasid_leaf_table(IntelIOMMUState *s,
788                                           uint32_t pasid,
789                                           dma_addr_t addr,
790                                           VTDPASIDEntry *pe)
791 {
792     uint32_t index;
793     dma_addr_t entry_size;
794     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
795 
796     index = VTD_PASID_TABLE_INDEX(pasid);
797     entry_size = VTD_PASID_ENTRY_SIZE;
798     addr = addr + index * entry_size;
799     if (dma_memory_read(&address_space_memory, addr,
800                         pe, entry_size, MEMTXATTRS_UNSPECIFIED)) {
801         return -VTD_FR_PASID_TABLE_INV;
802     }
803     for (size_t i = 0; i < ARRAY_SIZE(pe->val); i++) {
804         pe->val[i] = le64_to_cpu(pe->val[i]);
805     }
806 
807     /* Do translation type check */
808     if (!vtd_pe_type_check(x86_iommu, pe)) {
809         return -VTD_FR_PASID_TABLE_INV;
810     }
811 
812     if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) {
813         return -VTD_FR_PASID_TABLE_INV;
814     }
815 
816     return 0;
817 }
818 
819 /**
820  * Caller of this function should check present bit if wants
821  * to use pasid entry for further usage except for fpd bit check.
822  */
823 static int vtd_get_pe_from_pdire(IntelIOMMUState *s,
824                                  uint32_t pasid,
825                                  VTDPASIDDirEntry *pdire,
826                                  VTDPASIDEntry *pe)
827 {
828     dma_addr_t addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK;
829 
830     return vtd_get_pe_in_pasid_leaf_table(s, pasid, addr, pe);
831 }
832 
833 /**
834  * This function gets a pasid entry from a specified pasid
835  * table (includes dir and leaf table) with a specified pasid.
836  * Sanity check should be done to ensure return a present
837  * pasid entry to caller.
838  */
839 static int vtd_get_pe_from_pasid_table(IntelIOMMUState *s,
840                                        dma_addr_t pasid_dir_base,
841                                        uint32_t pasid,
842                                        VTDPASIDEntry *pe)
843 {
844     int ret;
845     VTDPASIDDirEntry pdire;
846 
847     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base,
848                                         pasid, &pdire);
849     if (ret) {
850         return ret;
851     }
852 
853     if (!vtd_pdire_present(&pdire)) {
854         return -VTD_FR_PASID_TABLE_INV;
855     }
856 
857     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, pe);
858     if (ret) {
859         return ret;
860     }
861 
862     if (!vtd_pe_present(pe)) {
863         return -VTD_FR_PASID_TABLE_INV;
864     }
865 
866     return 0;
867 }
868 
869 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s,
870                                       VTDContextEntry *ce,
871                                       VTDPASIDEntry *pe,
872                                       uint32_t pasid)
873 {
874     dma_addr_t pasid_dir_base;
875     int ret = 0;
876 
877     if (pasid == PCI_NO_PASID) {
878         pasid = VTD_CE_GET_RID2PASID(ce);
879     }
880     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
881     ret = vtd_get_pe_from_pasid_table(s, pasid_dir_base, pasid, pe);
882 
883     return ret;
884 }
885 
886 static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s,
887                                 VTDContextEntry *ce,
888                                 bool *pe_fpd_set,
889                                 uint32_t pasid)
890 {
891     int ret;
892     dma_addr_t pasid_dir_base;
893     VTDPASIDDirEntry pdire;
894     VTDPASIDEntry pe;
895 
896     if (pasid == PCI_NO_PASID) {
897         pasid = VTD_CE_GET_RID2PASID(ce);
898     }
899     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
900 
901     /*
902      * No present bit check since fpd is meaningful even
903      * if the present bit is clear.
904      */
905     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base, pasid, &pdire);
906     if (ret) {
907         return ret;
908     }
909 
910     if (pdire.val & VTD_PASID_DIR_FPD) {
911         *pe_fpd_set = true;
912         return 0;
913     }
914 
915     if (!vtd_pdire_present(&pdire)) {
916         return -VTD_FR_PASID_TABLE_INV;
917     }
918 
919     /*
920      * No present bit check since fpd is meaningful even
921      * if the present bit is clear.
922      */
923     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, &pe);
924     if (ret) {
925         return ret;
926     }
927 
928     if (pe.val[0] & VTD_PASID_ENTRY_FPD) {
929         *pe_fpd_set = true;
930     }
931 
932     return 0;
933 }
934 
935 /* Get the page-table level that hardware should use for the second-level
936  * page-table walk from the Address Width field of context-entry.
937  */
938 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
939 {
940     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
941 }
942 
943 static uint32_t vtd_get_iova_level(IntelIOMMUState *s,
944                                    VTDContextEntry *ce,
945                                    uint32_t pasid)
946 {
947     VTDPASIDEntry pe;
948 
949     if (s->root_scalable) {
950         vtd_ce_get_rid2pasid_entry(s, ce, &pe, pasid);
951         return VTD_PE_GET_LEVEL(&pe);
952     }
953 
954     return vtd_ce_get_level(ce);
955 }
956 
957 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
958 {
959     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
960 }
961 
962 static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s,
963                                   VTDContextEntry *ce,
964                                   uint32_t pasid)
965 {
966     VTDPASIDEntry pe;
967 
968     if (s->root_scalable) {
969         vtd_ce_get_rid2pasid_entry(s, ce, &pe, pasid);
970         return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9;
971     }
972 
973     return vtd_ce_get_agaw(ce);
974 }
975 
976 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
977 {
978     return ce->lo & VTD_CONTEXT_ENTRY_TT;
979 }
980 
981 /* Only for Legacy Mode. Return true if check passed, otherwise false */
982 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
983                                      VTDContextEntry *ce)
984 {
985     switch (vtd_ce_get_type(ce)) {
986     case VTD_CONTEXT_TT_MULTI_LEVEL:
987         /* Always supported */
988         break;
989     case VTD_CONTEXT_TT_DEV_IOTLB:
990         if (!x86_iommu->dt_supported) {
991             error_report_once("%s: DT specified but not supported", __func__);
992             return false;
993         }
994         break;
995     case VTD_CONTEXT_TT_PASS_THROUGH:
996         if (!x86_iommu->pt_supported) {
997             error_report_once("%s: PT specified but not supported", __func__);
998             return false;
999         }
1000         break;
1001     default:
1002         /* Unknown type */
1003         error_report_once("%s: unknown ce type: %"PRIu32, __func__,
1004                           vtd_ce_get_type(ce));
1005         return false;
1006     }
1007     return true;
1008 }
1009 
1010 static inline uint64_t vtd_iova_limit(IntelIOMMUState *s,
1011                                       VTDContextEntry *ce, uint8_t aw,
1012                                       uint32_t pasid)
1013 {
1014     uint32_t ce_agaw = vtd_get_iova_agaw(s, ce, pasid);
1015     return 1ULL << MIN(ce_agaw, aw);
1016 }
1017 
1018 /* Return true if IOVA passes range check, otherwise false. */
1019 static inline bool vtd_iova_range_check(IntelIOMMUState *s,
1020                                         uint64_t iova, VTDContextEntry *ce,
1021                                         uint8_t aw, uint32_t pasid)
1022 {
1023     /*
1024      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
1025      * in CAP_REG and AW in context-entry.
1026      */
1027     return !(iova & ~(vtd_iova_limit(s, ce, aw, pasid) - 1));
1028 }
1029 
1030 static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s,
1031                                           VTDContextEntry *ce,
1032                                           uint32_t pasid)
1033 {
1034     VTDPASIDEntry pe;
1035 
1036     if (s->root_scalable) {
1037         vtd_ce_get_rid2pasid_entry(s, ce, &pe, pasid);
1038         return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR;
1039     }
1040 
1041     return vtd_ce_get_slpt_base(ce);
1042 }
1043 
1044 /*
1045  * Rsvd field masks for spte:
1046  *     vtd_spte_rsvd 4k pages
1047  *     vtd_spte_rsvd_large large pages
1048  */
1049 static uint64_t vtd_spte_rsvd[5];
1050 static uint64_t vtd_spte_rsvd_large[5];
1051 
1052 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
1053 {
1054     uint64_t rsvd_mask = vtd_spte_rsvd[level];
1055 
1056     if ((level == VTD_SL_PD_LEVEL || level == VTD_SL_PDP_LEVEL) &&
1057         (slpte & VTD_SL_PT_PAGE_SIZE_MASK)) {
1058         /* large page */
1059         rsvd_mask = vtd_spte_rsvd_large[level];
1060     }
1061 
1062     return slpte & rsvd_mask;
1063 }
1064 
1065 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
1066  * of the translation, can be used for deciding the size of large page.
1067  */
1068 static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce,
1069                              uint64_t iova, bool is_write,
1070                              uint64_t *slptep, uint32_t *slpte_level,
1071                              bool *reads, bool *writes, uint8_t aw_bits,
1072                              uint32_t pasid)
1073 {
1074     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce, pasid);
1075     uint32_t level = vtd_get_iova_level(s, ce, pasid);
1076     uint32_t offset;
1077     uint64_t slpte;
1078     uint64_t access_right_check;
1079     uint64_t xlat, size;
1080 
1081     if (!vtd_iova_range_check(s, iova, ce, aw_bits, pasid)) {
1082         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ","
1083                           "pasid=0x%" PRIx32 ")", __func__, iova, pasid);
1084         return -VTD_FR_ADDR_BEYOND_MGAW;
1085     }
1086 
1087     /* FIXME: what is the Atomics request here? */
1088     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
1089 
1090     while (true) {
1091         offset = vtd_iova_level_offset(iova, level);
1092         slpte = vtd_get_slpte(addr, offset);
1093 
1094         if (slpte == (uint64_t)-1) {
1095             error_report_once("%s: detected read error on DMAR slpte "
1096                               "(iova=0x%" PRIx64 ", pasid=0x%" PRIx32 ")",
1097                               __func__, iova, pasid);
1098             if (level == vtd_get_iova_level(s, ce, pasid)) {
1099                 /* Invalid programming of context-entry */
1100                 return -VTD_FR_CONTEXT_ENTRY_INV;
1101             } else {
1102                 return -VTD_FR_PAGING_ENTRY_INV;
1103             }
1104         }
1105         *reads = (*reads) && (slpte & VTD_SL_R);
1106         *writes = (*writes) && (slpte & VTD_SL_W);
1107         if (!(slpte & access_right_check)) {
1108             error_report_once("%s: detected slpte permission error "
1109                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
1110                               "slpte=0x%" PRIx64 ", write=%d, pasid=0x%"
1111                               PRIx32 ")", __func__, iova, level,
1112                               slpte, is_write, pasid);
1113             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
1114         }
1115         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1116             error_report_once("%s: detected splte reserve non-zero "
1117                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
1118                               "slpte=0x%" PRIx64 ", pasid=0x%" PRIX32 ")",
1119                               __func__, iova, level, slpte, pasid);
1120             return -VTD_FR_PAGING_ENTRY_RSVD;
1121         }
1122 
1123         if (vtd_is_last_slpte(slpte, level)) {
1124             *slptep = slpte;
1125             *slpte_level = level;
1126             break;
1127         }
1128         addr = vtd_get_slpte_addr(slpte, aw_bits);
1129         level--;
1130     }
1131 
1132     xlat = vtd_get_slpte_addr(*slptep, aw_bits);
1133     size = ~vtd_slpt_level_page_mask(level) + 1;
1134 
1135     /*
1136      * From VT-d spec 3.14: Untranslated requests and translation
1137      * requests that result in an address in the interrupt range will be
1138      * blocked with condition code LGN.4 or SGN.8.
1139      */
1140     if ((xlat > VTD_INTERRUPT_ADDR_LAST ||
1141          xlat + size - 1 < VTD_INTERRUPT_ADDR_FIRST)) {
1142         return 0;
1143     } else {
1144         error_report_once("%s: xlat address is in interrupt range "
1145                           "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
1146                           "slpte=0x%" PRIx64 ", write=%d, "
1147                           "xlat=0x%" PRIx64 ", size=0x%" PRIx64 ", "
1148                           "pasid=0x%" PRIx32 ")",
1149                           __func__, iova, level, slpte, is_write,
1150                           xlat, size, pasid);
1151         return s->scalable_mode ? -VTD_FR_SM_INTERRUPT_ADDR :
1152                                   -VTD_FR_INTERRUPT_ADDR;
1153     }
1154 }
1155 
1156 typedef int (*vtd_page_walk_hook)(IOMMUTLBEvent *event, void *private);
1157 
1158 /**
1159  * Constant information used during page walking
1160  *
1161  * @hook_fn: hook func to be called when detected page
1162  * @private: private data to be passed into hook func
1163  * @notify_unmap: whether we should notify invalid entries
1164  * @as: VT-d address space of the device
1165  * @aw: maximum address width
1166  * @domain: domain ID of the page walk
1167  */
1168 typedef struct {
1169     VTDAddressSpace *as;
1170     vtd_page_walk_hook hook_fn;
1171     void *private;
1172     bool notify_unmap;
1173     uint8_t aw;
1174     uint16_t domain_id;
1175 } vtd_page_walk_info;
1176 
1177 static int vtd_page_walk_one(IOMMUTLBEvent *event, vtd_page_walk_info *info)
1178 {
1179     VTDAddressSpace *as = info->as;
1180     vtd_page_walk_hook hook_fn = info->hook_fn;
1181     void *private = info->private;
1182     IOMMUTLBEntry *entry = &event->entry;
1183     DMAMap target = {
1184         .iova = entry->iova,
1185         .size = entry->addr_mask,
1186         .translated_addr = entry->translated_addr,
1187         .perm = entry->perm,
1188     };
1189     const DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
1190 
1191     if (event->type == IOMMU_NOTIFIER_UNMAP && !info->notify_unmap) {
1192         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1193         return 0;
1194     }
1195 
1196     assert(hook_fn);
1197 
1198     /* Update local IOVA mapped ranges */
1199     if (event->type == IOMMU_NOTIFIER_MAP) {
1200         if (mapped) {
1201             /* If it's exactly the same translation, skip */
1202             if (!memcmp(mapped, &target, sizeof(target))) {
1203                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
1204                                                  entry->translated_addr);
1205                 return 0;
1206             } else {
1207                 /*
1208                  * Translation changed.  Normally this should not
1209                  * happen, but it can happen when with buggy guest
1210                  * OSes.  Note that there will be a small window that
1211                  * we don't have map at all.  But that's the best
1212                  * effort we can do.  The ideal way to emulate this is
1213                  * atomically modify the PTE to follow what has
1214                  * changed, but we can't.  One example is that vfio
1215                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1216                  * interface to modify a mapping (meanwhile it seems
1217                  * meaningless to even provide one).  Anyway, let's
1218                  * mark this as a TODO in case one day we'll have
1219                  * a better solution.
1220                  */
1221                 IOMMUAccessFlags cache_perm = entry->perm;
1222                 int ret;
1223 
1224                 /* Emulate an UNMAP */
1225                 event->type = IOMMU_NOTIFIER_UNMAP;
1226                 entry->perm = IOMMU_NONE;
1227                 trace_vtd_page_walk_one(info->domain_id,
1228                                         entry->iova,
1229                                         entry->translated_addr,
1230                                         entry->addr_mask,
1231                                         entry->perm);
1232                 ret = hook_fn(event, private);
1233                 if (ret) {
1234                     return ret;
1235                 }
1236                 /* Drop any existing mapping */
1237                 iova_tree_remove(as->iova_tree, target);
1238                 /* Recover the correct type */
1239                 event->type = IOMMU_NOTIFIER_MAP;
1240                 entry->perm = cache_perm;
1241             }
1242         }
1243         iova_tree_insert(as->iova_tree, &target);
1244     } else {
1245         if (!mapped) {
1246             /* Skip since we didn't map this range at all */
1247             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1248             return 0;
1249         }
1250         iova_tree_remove(as->iova_tree, target);
1251     }
1252 
1253     trace_vtd_page_walk_one(info->domain_id, entry->iova,
1254                             entry->translated_addr, entry->addr_mask,
1255                             entry->perm);
1256     return hook_fn(event, private);
1257 }
1258 
1259 /**
1260  * vtd_page_walk_level - walk over specific level for IOVA range
1261  *
1262  * @addr: base GPA addr to start the walk
1263  * @start: IOVA range start address
1264  * @end: IOVA range end address (start <= addr < end)
1265  * @read: whether parent level has read permission
1266  * @write: whether parent level has write permission
1267  * @info: constant information for the page walk
1268  */
1269 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
1270                                uint64_t end, uint32_t level, bool read,
1271                                bool write, vtd_page_walk_info *info)
1272 {
1273     bool read_cur, write_cur, entry_valid;
1274     uint32_t offset;
1275     uint64_t slpte;
1276     uint64_t subpage_size, subpage_mask;
1277     IOMMUTLBEvent event;
1278     uint64_t iova = start;
1279     uint64_t iova_next;
1280     int ret = 0;
1281 
1282     trace_vtd_page_walk_level(addr, level, start, end);
1283 
1284     subpage_size = 1ULL << vtd_slpt_level_shift(level);
1285     subpage_mask = vtd_slpt_level_page_mask(level);
1286 
1287     while (iova < end) {
1288         iova_next = (iova & subpage_mask) + subpage_size;
1289 
1290         offset = vtd_iova_level_offset(iova, level);
1291         slpte = vtd_get_slpte(addr, offset);
1292 
1293         if (slpte == (uint64_t)-1) {
1294             trace_vtd_page_walk_skip_read(iova, iova_next);
1295             goto next;
1296         }
1297 
1298         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1299             trace_vtd_page_walk_skip_reserve(iova, iova_next);
1300             goto next;
1301         }
1302 
1303         /* Permissions are stacked with parents' */
1304         read_cur = read && (slpte & VTD_SL_R);
1305         write_cur = write && (slpte & VTD_SL_W);
1306 
1307         /*
1308          * As long as we have either read/write permission, this is a
1309          * valid entry. The rule works for both page entries and page
1310          * table entries.
1311          */
1312         entry_valid = read_cur | write_cur;
1313 
1314         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
1315             /*
1316              * This is a valid PDE (or even bigger than PDE).  We need
1317              * to walk one further level.
1318              */
1319             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
1320                                       iova, MIN(iova_next, end), level - 1,
1321                                       read_cur, write_cur, info);
1322         } else {
1323             /*
1324              * This means we are either:
1325              *
1326              * (1) the real page entry (either 4K page, or huge page)
1327              * (2) the whole range is invalid
1328              *
1329              * In either case, we send an IOTLB notification down.
1330              */
1331             event.entry.target_as = &address_space_memory;
1332             event.entry.iova = iova & subpage_mask;
1333             event.entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
1334             event.entry.addr_mask = ~subpage_mask;
1335             /* NOTE: this is only meaningful if entry_valid == true */
1336             event.entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
1337             event.type = event.entry.perm ? IOMMU_NOTIFIER_MAP :
1338                                             IOMMU_NOTIFIER_UNMAP;
1339             ret = vtd_page_walk_one(&event, info);
1340         }
1341 
1342         if (ret < 0) {
1343             return ret;
1344         }
1345 
1346 next:
1347         iova = iova_next;
1348     }
1349 
1350     return 0;
1351 }
1352 
1353 /**
1354  * vtd_page_walk - walk specific IOVA range, and call the hook
1355  *
1356  * @s: intel iommu state
1357  * @ce: context entry to walk upon
1358  * @start: IOVA address to start the walk
1359  * @end: IOVA range end address (start <= addr < end)
1360  * @info: page walking information struct
1361  */
1362 static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce,
1363                          uint64_t start, uint64_t end,
1364                          vtd_page_walk_info *info,
1365                          uint32_t pasid)
1366 {
1367     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce, pasid);
1368     uint32_t level = vtd_get_iova_level(s, ce, pasid);
1369 
1370     if (!vtd_iova_range_check(s, start, ce, info->aw, pasid)) {
1371         return -VTD_FR_ADDR_BEYOND_MGAW;
1372     }
1373 
1374     if (!vtd_iova_range_check(s, end, ce, info->aw, pasid)) {
1375         /* Fix end so that it reaches the maximum */
1376         end = vtd_iova_limit(s, ce, info->aw, pasid);
1377     }
1378 
1379     return vtd_page_walk_level(addr, start, end, level, true, true, info);
1380 }
1381 
1382 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s,
1383                                           VTDRootEntry *re)
1384 {
1385     /* Legacy Mode reserved bits check */
1386     if (!s->root_scalable &&
1387         (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1388         goto rsvd_err;
1389 
1390     /* Scalable Mode reserved bits check */
1391     if (s->root_scalable &&
1392         ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) ||
1393          (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1394         goto rsvd_err;
1395 
1396     return 0;
1397 
1398 rsvd_err:
1399     error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1400                       ", lo=0x%"PRIx64,
1401                       __func__, re->hi, re->lo);
1402     return -VTD_FR_ROOT_ENTRY_RSVD;
1403 }
1404 
1405 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s,
1406                                                     VTDContextEntry *ce)
1407 {
1408     if (!s->root_scalable &&
1409         (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI ||
1410          ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1411         error_report_once("%s: invalid context entry: hi=%"PRIx64
1412                           ", lo=%"PRIx64" (reserved nonzero)",
1413                           __func__, ce->hi, ce->lo);
1414         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1415     }
1416 
1417     if (s->root_scalable &&
1418         (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) ||
1419          ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 ||
1420          ce->val[2] ||
1421          ce->val[3])) {
1422         error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1423                           ", val[2]=%"PRIx64
1424                           ", val[1]=%"PRIx64
1425                           ", val[0]=%"PRIx64" (reserved nonzero)",
1426                           __func__, ce->val[3], ce->val[2],
1427                           ce->val[1], ce->val[0]);
1428         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1429     }
1430 
1431     return 0;
1432 }
1433 
1434 static int vtd_ce_rid2pasid_check(IntelIOMMUState *s,
1435                                   VTDContextEntry *ce)
1436 {
1437     VTDPASIDEntry pe;
1438 
1439     /*
1440      * Make sure in Scalable Mode, a present context entry
1441      * has valid rid2pasid setting, which includes valid
1442      * rid2pasid field and corresponding pasid entry setting
1443      */
1444     return vtd_ce_get_rid2pasid_entry(s, ce, &pe, PCI_NO_PASID);
1445 }
1446 
1447 /* Map a device to its corresponding domain (context-entry) */
1448 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
1449                                     uint8_t devfn, VTDContextEntry *ce)
1450 {
1451     VTDRootEntry re;
1452     int ret_fr;
1453     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
1454 
1455     ret_fr = vtd_get_root_entry(s, bus_num, &re);
1456     if (ret_fr) {
1457         return ret_fr;
1458     }
1459 
1460     if (!vtd_root_entry_present(s, &re, devfn)) {
1461         /* Not error - it's okay we don't have root entry. */
1462         trace_vtd_re_not_present(bus_num);
1463         return -VTD_FR_ROOT_ENTRY_P;
1464     }
1465 
1466     ret_fr = vtd_root_entry_rsvd_bits_check(s, &re);
1467     if (ret_fr) {
1468         return ret_fr;
1469     }
1470 
1471     ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce);
1472     if (ret_fr) {
1473         return ret_fr;
1474     }
1475 
1476     if (!vtd_ce_present(ce)) {
1477         /* Not error - it's okay we don't have context entry. */
1478         trace_vtd_ce_not_present(bus_num, devfn);
1479         return -VTD_FR_CONTEXT_ENTRY_P;
1480     }
1481 
1482     ret_fr = vtd_context_entry_rsvd_bits_check(s, ce);
1483     if (ret_fr) {
1484         return ret_fr;
1485     }
1486 
1487     /* Check if the programming of context-entry is valid */
1488     if (!s->root_scalable &&
1489         !vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1490         error_report_once("%s: invalid context entry: hi=%"PRIx64
1491                           ", lo=%"PRIx64" (level %d not supported)",
1492                           __func__, ce->hi, ce->lo,
1493                           vtd_ce_get_level(ce));
1494         return -VTD_FR_CONTEXT_ENTRY_INV;
1495     }
1496 
1497     if (!s->root_scalable) {
1498         /* Do translation type check */
1499         if (!vtd_ce_type_check(x86_iommu, ce)) {
1500             /* Errors dumped in vtd_ce_type_check() */
1501             return -VTD_FR_CONTEXT_ENTRY_INV;
1502         }
1503     } else {
1504         /*
1505          * Check if the programming of context-entry.rid2pasid
1506          * and corresponding pasid setting is valid, and thus
1507          * avoids to check pasid entry fetching result in future
1508          * helper function calling.
1509          */
1510         ret_fr = vtd_ce_rid2pasid_check(s, ce);
1511         if (ret_fr) {
1512             return ret_fr;
1513         }
1514     }
1515 
1516     return 0;
1517 }
1518 
1519 static int vtd_sync_shadow_page_hook(IOMMUTLBEvent *event,
1520                                      void *private)
1521 {
1522     memory_region_notify_iommu(private, 0, *event);
1523     return 0;
1524 }
1525 
1526 static uint16_t vtd_get_domain_id(IntelIOMMUState *s,
1527                                   VTDContextEntry *ce,
1528                                   uint32_t pasid)
1529 {
1530     VTDPASIDEntry pe;
1531 
1532     if (s->root_scalable) {
1533         vtd_ce_get_rid2pasid_entry(s, ce, &pe, pasid);
1534         return VTD_SM_PASID_ENTRY_DID(pe.val[1]);
1535     }
1536 
1537     return VTD_CONTEXT_ENTRY_DID(ce->hi);
1538 }
1539 
1540 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1541                                             VTDContextEntry *ce,
1542                                             hwaddr addr, hwaddr size)
1543 {
1544     IntelIOMMUState *s = vtd_as->iommu_state;
1545     vtd_page_walk_info info = {
1546         .hook_fn = vtd_sync_shadow_page_hook,
1547         .private = (void *)&vtd_as->iommu,
1548         .notify_unmap = true,
1549         .aw = s->aw_bits,
1550         .as = vtd_as,
1551         .domain_id = vtd_get_domain_id(s, ce, vtd_as->pasid),
1552     };
1553 
1554     return vtd_page_walk(s, ce, addr, addr + size, &info, vtd_as->pasid);
1555 }
1556 
1557 static int vtd_address_space_sync(VTDAddressSpace *vtd_as)
1558 {
1559     int ret;
1560     VTDContextEntry ce;
1561     IOMMUNotifier *n;
1562 
1563     /* If no MAP notifier registered, we simply invalidate all the cache */
1564     if (!vtd_as_has_map_notifier(vtd_as)) {
1565         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1566             memory_region_unmap_iommu_notifier_range(n);
1567         }
1568         return 0;
1569     }
1570 
1571     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1572                                    pci_bus_num(vtd_as->bus),
1573                                    vtd_as->devfn, &ce);
1574     if (ret) {
1575         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1576             /*
1577              * It's a valid scenario to have a context entry that is
1578              * not present.  For example, when a device is removed
1579              * from an existing domain then the context entry will be
1580              * zeroed by the guest before it was put into another
1581              * domain.  When this happens, instead of synchronizing
1582              * the shadow pages we should invalidate all existing
1583              * mappings and notify the backends.
1584              */
1585             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1586                 vtd_address_space_unmap(vtd_as, n);
1587             }
1588             ret = 0;
1589         }
1590         return ret;
1591     }
1592 
1593     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1594 }
1595 
1596 /*
1597  * Check if specific device is configured to bypass address
1598  * translation for DMA requests. In Scalable Mode, bypass
1599  * 1st-level translation or 2nd-level translation, it depends
1600  * on PGTT setting.
1601  */
1602 static bool vtd_dev_pt_enabled(IntelIOMMUState *s, VTDContextEntry *ce,
1603                                uint32_t pasid)
1604 {
1605     VTDPASIDEntry pe;
1606     int ret;
1607 
1608     if (s->root_scalable) {
1609         ret = vtd_ce_get_rid2pasid_entry(s, ce, &pe, pasid);
1610         if (ret) {
1611             /*
1612              * This error is guest triggerable. We should assumt PT
1613              * not enabled for safety.
1614              */
1615             return false;
1616         }
1617         return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT);
1618     }
1619 
1620     return (vtd_ce_get_type(ce) == VTD_CONTEXT_TT_PASS_THROUGH);
1621 
1622 }
1623 
1624 static bool vtd_as_pt_enabled(VTDAddressSpace *as)
1625 {
1626     IntelIOMMUState *s;
1627     VTDContextEntry ce;
1628 
1629     assert(as);
1630 
1631     s = as->iommu_state;
1632     if (vtd_dev_to_context_entry(s, pci_bus_num(as->bus), as->devfn,
1633                                  &ce)) {
1634         /*
1635          * Possibly failed to parse the context entry for some reason
1636          * (e.g., during init, or any guest configuration errors on
1637          * context entries). We should assume PT not enabled for
1638          * safety.
1639          */
1640         return false;
1641     }
1642 
1643     return vtd_dev_pt_enabled(s, &ce, as->pasid);
1644 }
1645 
1646 /* Return whether the device is using IOMMU translation. */
1647 static bool vtd_switch_address_space(VTDAddressSpace *as)
1648 {
1649     bool use_iommu, pt;
1650     /* Whether we need to take the BQL on our own */
1651     bool take_bql = !qemu_mutex_iothread_locked();
1652 
1653     assert(as);
1654 
1655     use_iommu = as->iommu_state->dmar_enabled && !vtd_as_pt_enabled(as);
1656     pt = as->iommu_state->dmar_enabled && vtd_as_pt_enabled(as);
1657 
1658     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1659                                    VTD_PCI_SLOT(as->devfn),
1660                                    VTD_PCI_FUNC(as->devfn),
1661                                    use_iommu);
1662 
1663     /*
1664      * It's possible that we reach here without BQL, e.g., when called
1665      * from vtd_pt_enable_fast_path(). However the memory APIs need
1666      * it. We'd better make sure we have had it already, or, take it.
1667      */
1668     if (take_bql) {
1669         qemu_mutex_lock_iothread();
1670     }
1671 
1672     /* Turn off first then on the other */
1673     if (use_iommu) {
1674         memory_region_set_enabled(&as->nodmar, false);
1675         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1676         /*
1677          * vt-d spec v3.4 3.14:
1678          *
1679          * """
1680          * Requests-with-PASID with input address in range 0xFEEx_xxxx
1681          * are translated normally like any other request-with-PASID
1682          * through DMA-remapping hardware.
1683          * """
1684          *
1685          * Need to disable ir for as with PASID.
1686          */
1687         if (as->pasid != PCI_NO_PASID) {
1688             memory_region_set_enabled(&as->iommu_ir, false);
1689         } else {
1690             memory_region_set_enabled(&as->iommu_ir, true);
1691         }
1692     } else {
1693         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1694         memory_region_set_enabled(&as->nodmar, true);
1695     }
1696 
1697     /*
1698      * vtd-spec v3.4 3.14:
1699      *
1700      * """
1701      * Requests-with-PASID with input address in range 0xFEEx_xxxx are
1702      * translated normally like any other request-with-PASID through
1703      * DMA-remapping hardware. However, if such a request is processed
1704      * using pass-through translation, it will be blocked as described
1705      * in the paragraph below.
1706      *
1707      * Software must not program paging-structure entries to remap any
1708      * address to the interrupt address range. Untranslated requests
1709      * and translation requests that result in an address in the
1710      * interrupt range will be blocked with condition code LGN.4 or
1711      * SGN.8.
1712      * """
1713      *
1714      * We enable per as memory region (iommu_ir_fault) for catching
1715      * the translation for interrupt range through PASID + PT.
1716      */
1717     if (pt && as->pasid != PCI_NO_PASID) {
1718         memory_region_set_enabled(&as->iommu_ir_fault, true);
1719     } else {
1720         memory_region_set_enabled(&as->iommu_ir_fault, false);
1721     }
1722 
1723     if (take_bql) {
1724         qemu_mutex_unlock_iothread();
1725     }
1726 
1727     return use_iommu;
1728 }
1729 
1730 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1731 {
1732     VTDAddressSpace *vtd_as;
1733     GHashTableIter iter;
1734 
1735     g_hash_table_iter_init(&iter, s->vtd_address_spaces);
1736     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_as)) {
1737         vtd_switch_address_space(vtd_as);
1738     }
1739 }
1740 
1741 static const bool vtd_qualified_faults[] = {
1742     [VTD_FR_RESERVED] = false,
1743     [VTD_FR_ROOT_ENTRY_P] = false,
1744     [VTD_FR_CONTEXT_ENTRY_P] = true,
1745     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1746     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1747     [VTD_FR_WRITE] = true,
1748     [VTD_FR_READ] = true,
1749     [VTD_FR_PAGING_ENTRY_INV] = true,
1750     [VTD_FR_ROOT_TABLE_INV] = false,
1751     [VTD_FR_CONTEXT_TABLE_INV] = false,
1752     [VTD_FR_INTERRUPT_ADDR] = true,
1753     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1754     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1755     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1756     [VTD_FR_PASID_TABLE_INV] = false,
1757     [VTD_FR_SM_INTERRUPT_ADDR] = true,
1758     [VTD_FR_MAX] = false,
1759 };
1760 
1761 /* To see if a fault condition is "qualified", which is reported to software
1762  * only if the FPD field in the context-entry used to process the faulting
1763  * request is 0.
1764  */
1765 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1766 {
1767     return vtd_qualified_faults[fault];
1768 }
1769 
1770 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1771 {
1772     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1773 }
1774 
1775 static gboolean vtd_find_as_by_sid(gpointer key, gpointer value,
1776                                    gpointer user_data)
1777 {
1778     struct vtd_as_key *as_key = (struct vtd_as_key *)key;
1779     uint16_t target_sid = *(uint16_t *)user_data;
1780     uint16_t sid = PCI_BUILD_BDF(pci_bus_num(as_key->bus), as_key->devfn);
1781     return sid == target_sid;
1782 }
1783 
1784 static VTDAddressSpace *vtd_get_as_by_sid(IntelIOMMUState *s, uint16_t sid)
1785 {
1786     uint8_t bus_num = PCI_BUS_NUM(sid);
1787     VTDAddressSpace *vtd_as = s->vtd_as_cache[bus_num];
1788 
1789     if (vtd_as &&
1790         (sid == PCI_BUILD_BDF(pci_bus_num(vtd_as->bus), vtd_as->devfn))) {
1791         return vtd_as;
1792     }
1793 
1794     vtd_as = g_hash_table_find(s->vtd_address_spaces, vtd_find_as_by_sid, &sid);
1795     s->vtd_as_cache[bus_num] = vtd_as;
1796 
1797     return vtd_as;
1798 }
1799 
1800 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1801 {
1802     VTDAddressSpace *vtd_as;
1803     bool success = false;
1804 
1805     vtd_as = vtd_get_as_by_sid(s, source_id);
1806     if (!vtd_as) {
1807         goto out;
1808     }
1809 
1810     if (vtd_switch_address_space(vtd_as) == false) {
1811         /* We switched off IOMMU region successfully. */
1812         success = true;
1813     }
1814 
1815 out:
1816     trace_vtd_pt_enable_fast_path(source_id, success);
1817 }
1818 
1819 static void vtd_report_fault(IntelIOMMUState *s,
1820                              int err, bool is_fpd_set,
1821                              uint16_t source_id,
1822                              hwaddr addr,
1823                              bool is_write,
1824                              bool is_pasid,
1825                              uint32_t pasid)
1826 {
1827     if (is_fpd_set && vtd_is_qualified_fault(err)) {
1828         trace_vtd_fault_disabled();
1829     } else {
1830         vtd_report_dmar_fault(s, source_id, addr, err, is_write,
1831                               is_pasid, pasid);
1832     }
1833 }
1834 
1835 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1836  * translation.
1837  *
1838  * Called from RCU critical section.
1839  *
1840  * @bus_num: The bus number
1841  * @devfn: The devfn, which is the  combined of device and function number
1842  * @is_write: The access is a write operation
1843  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1844  *
1845  * Returns true if translation is successful, otherwise false.
1846  */
1847 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1848                                    uint8_t devfn, hwaddr addr, bool is_write,
1849                                    IOMMUTLBEntry *entry)
1850 {
1851     IntelIOMMUState *s = vtd_as->iommu_state;
1852     VTDContextEntry ce;
1853     uint8_t bus_num = pci_bus_num(bus);
1854     VTDContextCacheEntry *cc_entry;
1855     uint64_t slpte, page_mask;
1856     uint32_t level, pasid = vtd_as->pasid;
1857     uint16_t source_id = PCI_BUILD_BDF(bus_num, devfn);
1858     int ret_fr;
1859     bool is_fpd_set = false;
1860     bool reads = true;
1861     bool writes = true;
1862     uint8_t access_flags;
1863     bool rid2pasid = (pasid == PCI_NO_PASID) && s->root_scalable;
1864     VTDIOTLBEntry *iotlb_entry;
1865 
1866     /*
1867      * We have standalone memory region for interrupt addresses, we
1868      * should never receive translation requests in this region.
1869      */
1870     assert(!vtd_is_interrupt_addr(addr));
1871 
1872     vtd_iommu_lock(s);
1873 
1874     cc_entry = &vtd_as->context_cache_entry;
1875 
1876     /* Try to fetch slpte form IOTLB, we don't need RID2PASID logic */
1877     if (!rid2pasid) {
1878         iotlb_entry = vtd_lookup_iotlb(s, source_id, pasid, addr);
1879         if (iotlb_entry) {
1880             trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1881                                      iotlb_entry->domain_id);
1882             slpte = iotlb_entry->slpte;
1883             access_flags = iotlb_entry->access_flags;
1884             page_mask = iotlb_entry->mask;
1885             goto out;
1886         }
1887     }
1888 
1889     /* Try to fetch context-entry from cache first */
1890     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1891         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1892                                cc_entry->context_entry.lo,
1893                                cc_entry->context_cache_gen);
1894         ce = cc_entry->context_entry;
1895         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1896         if (!is_fpd_set && s->root_scalable) {
1897             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set, pasid);
1898             if (ret_fr) {
1899                 vtd_report_fault(s, -ret_fr, is_fpd_set,
1900                                  source_id, addr, is_write,
1901                                  false, 0);
1902                 goto error;
1903             }
1904         }
1905     } else {
1906         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1907         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1908         if (!ret_fr && !is_fpd_set && s->root_scalable) {
1909             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set, pasid);
1910         }
1911         if (ret_fr) {
1912             vtd_report_fault(s, -ret_fr, is_fpd_set,
1913                              source_id, addr, is_write,
1914                              false, 0);
1915             goto error;
1916         }
1917         /* Update context-cache */
1918         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1919                                   cc_entry->context_cache_gen,
1920                                   s->context_cache_gen);
1921         cc_entry->context_entry = ce;
1922         cc_entry->context_cache_gen = s->context_cache_gen;
1923     }
1924 
1925     if (rid2pasid) {
1926         pasid = VTD_CE_GET_RID2PASID(&ce);
1927     }
1928 
1929     /*
1930      * We don't need to translate for pass-through context entries.
1931      * Also, let's ignore IOTLB caching as well for PT devices.
1932      */
1933     if (vtd_dev_pt_enabled(s, &ce, pasid)) {
1934         entry->iova = addr & VTD_PAGE_MASK_4K;
1935         entry->translated_addr = entry->iova;
1936         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1937         entry->perm = IOMMU_RW;
1938         trace_vtd_translate_pt(source_id, entry->iova);
1939 
1940         /*
1941          * When this happens, it means firstly caching-mode is not
1942          * enabled, and this is the first passthrough translation for
1943          * the device. Let's enable the fast path for passthrough.
1944          *
1945          * When passthrough is disabled again for the device, we can
1946          * capture it via the context entry invalidation, then the
1947          * IOMMU region can be swapped back.
1948          */
1949         vtd_pt_enable_fast_path(s, source_id);
1950         vtd_iommu_unlock(s);
1951         return true;
1952     }
1953 
1954     /* Try to fetch slpte form IOTLB for RID2PASID slow path */
1955     if (rid2pasid) {
1956         iotlb_entry = vtd_lookup_iotlb(s, source_id, pasid, addr);
1957         if (iotlb_entry) {
1958             trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1959                                      iotlb_entry->domain_id);
1960             slpte = iotlb_entry->slpte;
1961             access_flags = iotlb_entry->access_flags;
1962             page_mask = iotlb_entry->mask;
1963             goto out;
1964         }
1965     }
1966 
1967     ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level,
1968                                &reads, &writes, s->aw_bits, pasid);
1969     if (ret_fr) {
1970         vtd_report_fault(s, -ret_fr, is_fpd_set, source_id,
1971                          addr, is_write, pasid != PCI_NO_PASID, pasid);
1972         goto error;
1973     }
1974 
1975     page_mask = vtd_slpt_level_page_mask(level);
1976     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1977     vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce, pasid),
1978                      addr, slpte, access_flags, level, pasid);
1979 out:
1980     vtd_iommu_unlock(s);
1981     entry->iova = addr & page_mask;
1982     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1983     entry->addr_mask = ~page_mask;
1984     entry->perm = access_flags;
1985     return true;
1986 
1987 error:
1988     vtd_iommu_unlock(s);
1989     entry->iova = 0;
1990     entry->translated_addr = 0;
1991     entry->addr_mask = 0;
1992     entry->perm = IOMMU_NONE;
1993     return false;
1994 }
1995 
1996 static void vtd_root_table_setup(IntelIOMMUState *s)
1997 {
1998     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1999     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
2000 
2001     vtd_update_scalable_state(s);
2002 
2003     trace_vtd_reg_dmar_root(s->root, s->root_scalable);
2004 }
2005 
2006 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
2007                                uint32_t index, uint32_t mask)
2008 {
2009     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
2010 }
2011 
2012 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
2013 {
2014     uint64_t value = 0;
2015     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
2016     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
2017     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
2018     s->intr_eime = value & VTD_IRTA_EIME;
2019 
2020     /* Notify global invalidation */
2021     vtd_iec_notify_all(s, true, 0, 0);
2022 
2023     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
2024 }
2025 
2026 static void vtd_iommu_replay_all(IntelIOMMUState *s)
2027 {
2028     VTDAddressSpace *vtd_as;
2029 
2030     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
2031         vtd_address_space_sync(vtd_as);
2032     }
2033 }
2034 
2035 static void vtd_context_global_invalidate(IntelIOMMUState *s)
2036 {
2037     trace_vtd_inv_desc_cc_global();
2038     /* Protects context cache */
2039     vtd_iommu_lock(s);
2040     s->context_cache_gen++;
2041     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
2042         vtd_reset_context_cache_locked(s);
2043     }
2044     vtd_iommu_unlock(s);
2045     vtd_address_space_refresh_all(s);
2046     /*
2047      * From VT-d spec 6.5.2.1, a global context entry invalidation
2048      * should be followed by a IOTLB global invalidation, so we should
2049      * be safe even without this. Hoewever, let's replay the region as
2050      * well to be safer, and go back here when we need finer tunes for
2051      * VT-d emulation codes.
2052      */
2053     vtd_iommu_replay_all(s);
2054 }
2055 
2056 /* Do a context-cache device-selective invalidation.
2057  * @func_mask: FM field after shifting
2058  */
2059 static void vtd_context_device_invalidate(IntelIOMMUState *s,
2060                                           uint16_t source_id,
2061                                           uint16_t func_mask)
2062 {
2063     GHashTableIter as_it;
2064     uint16_t mask;
2065     VTDAddressSpace *vtd_as;
2066     uint8_t bus_n, devfn;
2067 
2068     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
2069 
2070     switch (func_mask & 3) {
2071     case 0:
2072         mask = 0;   /* No bits in the SID field masked */
2073         break;
2074     case 1:
2075         mask = 4;   /* Mask bit 2 in the SID field */
2076         break;
2077     case 2:
2078         mask = 6;   /* Mask bit 2:1 in the SID field */
2079         break;
2080     case 3:
2081         mask = 7;   /* Mask bit 2:0 in the SID field */
2082         break;
2083     default:
2084         g_assert_not_reached();
2085     }
2086     mask = ~mask;
2087 
2088     bus_n = VTD_SID_TO_BUS(source_id);
2089     devfn = VTD_SID_TO_DEVFN(source_id);
2090 
2091     g_hash_table_iter_init(&as_it, s->vtd_address_spaces);
2092     while (g_hash_table_iter_next(&as_it, NULL, (void **)&vtd_as)) {
2093         if ((pci_bus_num(vtd_as->bus) == bus_n) &&
2094             (vtd_as->devfn & mask) == (devfn & mask)) {
2095             trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(vtd_as->devfn),
2096                                          VTD_PCI_FUNC(vtd_as->devfn));
2097             vtd_iommu_lock(s);
2098             vtd_as->context_cache_entry.context_cache_gen = 0;
2099             vtd_iommu_unlock(s);
2100             /*
2101              * Do switch address space when needed, in case if the
2102              * device passthrough bit is switched.
2103              */
2104             vtd_switch_address_space(vtd_as);
2105             /*
2106              * So a device is moving out of (or moving into) a
2107              * domain, resync the shadow page table.
2108              * This won't bring bad even if we have no such
2109              * notifier registered - the IOMMU notification
2110              * framework will skip MAP notifications if that
2111              * happened.
2112              */
2113             vtd_address_space_sync(vtd_as);
2114         }
2115     }
2116 }
2117 
2118 /* Context-cache invalidation
2119  * Returns the Context Actual Invalidation Granularity.
2120  * @val: the content of the CCMD_REG
2121  */
2122 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
2123 {
2124     uint64_t caig;
2125     uint64_t type = val & VTD_CCMD_CIRG_MASK;
2126 
2127     switch (type) {
2128     case VTD_CCMD_DOMAIN_INVL:
2129         /* Fall through */
2130     case VTD_CCMD_GLOBAL_INVL:
2131         caig = VTD_CCMD_GLOBAL_INVL_A;
2132         vtd_context_global_invalidate(s);
2133         break;
2134 
2135     case VTD_CCMD_DEVICE_INVL:
2136         caig = VTD_CCMD_DEVICE_INVL_A;
2137         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
2138         break;
2139 
2140     default:
2141         error_report_once("%s: invalid context: 0x%" PRIx64,
2142                           __func__, val);
2143         caig = 0;
2144     }
2145     return caig;
2146 }
2147 
2148 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
2149 {
2150     trace_vtd_inv_desc_iotlb_global();
2151     vtd_reset_iotlb(s);
2152     vtd_iommu_replay_all(s);
2153 }
2154 
2155 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
2156 {
2157     VTDContextEntry ce;
2158     VTDAddressSpace *vtd_as;
2159 
2160     trace_vtd_inv_desc_iotlb_domain(domain_id);
2161 
2162     vtd_iommu_lock(s);
2163     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
2164                                 &domain_id);
2165     vtd_iommu_unlock(s);
2166 
2167     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
2168         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
2169                                       vtd_as->devfn, &ce) &&
2170             domain_id == vtd_get_domain_id(s, &ce, vtd_as->pasid)) {
2171             vtd_address_space_sync(vtd_as);
2172         }
2173     }
2174 }
2175 
2176 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
2177                                            uint16_t domain_id, hwaddr addr,
2178                                              uint8_t am, uint32_t pasid)
2179 {
2180     VTDAddressSpace *vtd_as;
2181     VTDContextEntry ce;
2182     int ret;
2183     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
2184 
2185     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
2186         if (pasid != PCI_NO_PASID && pasid != vtd_as->pasid) {
2187             continue;
2188         }
2189         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
2190                                        vtd_as->devfn, &ce);
2191         if (!ret && domain_id == vtd_get_domain_id(s, &ce, vtd_as->pasid)) {
2192             if (vtd_as_has_map_notifier(vtd_as)) {
2193                 /*
2194                  * As long as we have MAP notifications registered in
2195                  * any of our IOMMU notifiers, we need to sync the
2196                  * shadow page table.
2197                  */
2198                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
2199             } else {
2200                 /*
2201                  * For UNMAP-only notifiers, we don't need to walk the
2202                  * page tables.  We just deliver the PSI down to
2203                  * invalidate caches.
2204                  */
2205                 IOMMUTLBEvent event = {
2206                     .type = IOMMU_NOTIFIER_UNMAP,
2207                     .entry = {
2208                         .target_as = &address_space_memory,
2209                         .iova = addr,
2210                         .translated_addr = 0,
2211                         .addr_mask = size - 1,
2212                         .perm = IOMMU_NONE,
2213                     },
2214                 };
2215                 memory_region_notify_iommu(&vtd_as->iommu, 0, event);
2216             }
2217         }
2218     }
2219 }
2220 
2221 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
2222                                       hwaddr addr, uint8_t am)
2223 {
2224     VTDIOTLBPageInvInfo info;
2225 
2226     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
2227 
2228     assert(am <= VTD_MAMV);
2229     info.domain_id = domain_id;
2230     info.addr = addr;
2231     info.mask = ~((1 << am) - 1);
2232     vtd_iommu_lock(s);
2233     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
2234     vtd_iommu_unlock(s);
2235     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am, PCI_NO_PASID);
2236 }
2237 
2238 /* Flush IOTLB
2239  * Returns the IOTLB Actual Invalidation Granularity.
2240  * @val: the content of the IOTLB_REG
2241  */
2242 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
2243 {
2244     uint64_t iaig;
2245     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
2246     uint16_t domain_id;
2247     hwaddr addr;
2248     uint8_t am;
2249 
2250     switch (type) {
2251     case VTD_TLB_GLOBAL_FLUSH:
2252         iaig = VTD_TLB_GLOBAL_FLUSH_A;
2253         vtd_iotlb_global_invalidate(s);
2254         break;
2255 
2256     case VTD_TLB_DSI_FLUSH:
2257         domain_id = VTD_TLB_DID(val);
2258         iaig = VTD_TLB_DSI_FLUSH_A;
2259         vtd_iotlb_domain_invalidate(s, domain_id);
2260         break;
2261 
2262     case VTD_TLB_PSI_FLUSH:
2263         domain_id = VTD_TLB_DID(val);
2264         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
2265         am = VTD_IVA_AM(addr);
2266         addr = VTD_IVA_ADDR(addr);
2267         if (am > VTD_MAMV) {
2268             error_report_once("%s: address mask overflow: 0x%" PRIx64,
2269                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
2270             iaig = 0;
2271             break;
2272         }
2273         iaig = VTD_TLB_PSI_FLUSH_A;
2274         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2275         break;
2276 
2277     default:
2278         error_report_once("%s: invalid granularity: 0x%" PRIx64,
2279                           __func__, val);
2280         iaig = 0;
2281     }
2282     return iaig;
2283 }
2284 
2285 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
2286 
2287 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
2288 {
2289     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
2290            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
2291 }
2292 
2293 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
2294 {
2295     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
2296 
2297     trace_vtd_inv_qi_enable(en);
2298 
2299     if (en) {
2300         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
2301         /* 2^(x+8) entries */
2302         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0));
2303         s->qi_enabled = true;
2304         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
2305         /* Ok - report back to driver */
2306         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
2307 
2308         if (s->iq_tail != 0) {
2309             /*
2310              * This is a spec violation but Windows guests are known to set up
2311              * Queued Invalidation this way so we allow the write and process
2312              * Invalidation Descriptors right away.
2313              */
2314             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
2315             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2316                 vtd_fetch_inv_desc(s);
2317             }
2318         }
2319     } else {
2320         if (vtd_queued_inv_disable_check(s)) {
2321             /* disable Queued Invalidation */
2322             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
2323             s->iq_head = 0;
2324             s->qi_enabled = false;
2325             /* Ok - report back to driver */
2326             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
2327         } else {
2328             error_report_once("%s: detected improper state when disable QI "
2329                               "(head=0x%x, tail=0x%x, last_type=%d)",
2330                               __func__,
2331                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
2332         }
2333     }
2334 }
2335 
2336 /* Set Root Table Pointer */
2337 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
2338 {
2339     vtd_root_table_setup(s);
2340     /* Ok - report back to driver */
2341     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
2342     vtd_reset_caches(s);
2343     vtd_address_space_refresh_all(s);
2344 }
2345 
2346 /* Set Interrupt Remap Table Pointer */
2347 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
2348 {
2349     vtd_interrupt_remap_table_setup(s);
2350     /* Ok - report back to driver */
2351     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
2352 }
2353 
2354 /* Handle Translation Enable/Disable */
2355 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
2356 {
2357     if (s->dmar_enabled == en) {
2358         return;
2359     }
2360 
2361     trace_vtd_dmar_enable(en);
2362 
2363     if (en) {
2364         s->dmar_enabled = true;
2365         /* Ok - report back to driver */
2366         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
2367     } else {
2368         s->dmar_enabled = false;
2369 
2370         /* Clear the index of Fault Recording Register */
2371         s->next_frcd_reg = 0;
2372         /* Ok - report back to driver */
2373         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
2374     }
2375 
2376     vtd_reset_caches(s);
2377     vtd_address_space_refresh_all(s);
2378 }
2379 
2380 /* Handle Interrupt Remap Enable/Disable */
2381 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
2382 {
2383     trace_vtd_ir_enable(en);
2384 
2385     if (en) {
2386         s->intr_enabled = true;
2387         /* Ok - report back to driver */
2388         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
2389     } else {
2390         s->intr_enabled = false;
2391         /* Ok - report back to driver */
2392         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
2393     }
2394 }
2395 
2396 /* Handle write to Global Command Register */
2397 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
2398 {
2399     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
2400     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
2401     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
2402     uint32_t changed = status ^ val;
2403 
2404     trace_vtd_reg_write_gcmd(status, val);
2405     if ((changed & VTD_GCMD_TE) && s->dma_translation) {
2406         /* Translation enable/disable */
2407         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
2408     }
2409     if (val & VTD_GCMD_SRTP) {
2410         /* Set/update the root-table pointer */
2411         vtd_handle_gcmd_srtp(s);
2412     }
2413     if (changed & VTD_GCMD_QIE) {
2414         /* Queued Invalidation Enable */
2415         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
2416     }
2417     if (val & VTD_GCMD_SIRTP) {
2418         /* Set/update the interrupt remapping root-table pointer */
2419         vtd_handle_gcmd_sirtp(s);
2420     }
2421     if ((changed & VTD_GCMD_IRE) &&
2422         x86_iommu_ir_supported(x86_iommu)) {
2423         /* Interrupt remap enable/disable */
2424         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
2425     }
2426 }
2427 
2428 /* Handle write to Context Command Register */
2429 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
2430 {
2431     uint64_t ret;
2432     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
2433 
2434     /* Context-cache invalidation request */
2435     if (val & VTD_CCMD_ICC) {
2436         if (s->qi_enabled) {
2437             error_report_once("Queued Invalidation enabled, "
2438                               "should not use register-based invalidation");
2439             return;
2440         }
2441         ret = vtd_context_cache_invalidate(s, val);
2442         /* Invalidation completed. Change something to show */
2443         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
2444         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
2445                                       ret);
2446     }
2447 }
2448 
2449 /* Handle write to IOTLB Invalidation Register */
2450 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
2451 {
2452     uint64_t ret;
2453     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
2454 
2455     /* IOTLB invalidation request */
2456     if (val & VTD_TLB_IVT) {
2457         if (s->qi_enabled) {
2458             error_report_once("Queued Invalidation enabled, "
2459                               "should not use register-based invalidation");
2460             return;
2461         }
2462         ret = vtd_iotlb_flush(s, val);
2463         /* Invalidation completed. Change something to show */
2464         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
2465         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
2466                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
2467     }
2468 }
2469 
2470 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2471 static bool vtd_get_inv_desc(IntelIOMMUState *s,
2472                              VTDInvDesc *inv_desc)
2473 {
2474     dma_addr_t base_addr = s->iq;
2475     uint32_t offset = s->iq_head;
2476     uint32_t dw = s->iq_dw ? 32 : 16;
2477     dma_addr_t addr = base_addr + offset * dw;
2478 
2479     if (dma_memory_read(&address_space_memory, addr,
2480                         inv_desc, dw, MEMTXATTRS_UNSPECIFIED)) {
2481         error_report_once("Read INV DESC failed.");
2482         return false;
2483     }
2484     inv_desc->lo = le64_to_cpu(inv_desc->lo);
2485     inv_desc->hi = le64_to_cpu(inv_desc->hi);
2486     if (dw == 32) {
2487         inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]);
2488         inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]);
2489     }
2490     return true;
2491 }
2492 
2493 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2494 {
2495     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
2496         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
2497         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2498                           " (reserved nonzero)", __func__, inv_desc->hi,
2499                           inv_desc->lo);
2500         return false;
2501     }
2502     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
2503         /* Status Write */
2504         uint32_t status_data = (uint32_t)(inv_desc->lo >>
2505                                VTD_INV_DESC_WAIT_DATA_SHIFT);
2506 
2507         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
2508 
2509         /* FIXME: need to be masked with HAW? */
2510         dma_addr_t status_addr = inv_desc->hi;
2511         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
2512         status_data = cpu_to_le32(status_data);
2513         if (dma_memory_write(&address_space_memory, status_addr,
2514                              &status_data, sizeof(status_data),
2515                              MEMTXATTRS_UNSPECIFIED)) {
2516             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
2517             return false;
2518         }
2519     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
2520         /* Interrupt flag */
2521         vtd_generate_completion_event(s);
2522     } else {
2523         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2524                           " (unknown type)", __func__, inv_desc->hi,
2525                           inv_desc->lo);
2526         return false;
2527     }
2528     return true;
2529 }
2530 
2531 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
2532                                            VTDInvDesc *inv_desc)
2533 {
2534     uint16_t sid, fmask;
2535 
2536     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
2537         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2538                           " (reserved nonzero)", __func__, inv_desc->hi,
2539                           inv_desc->lo);
2540         return false;
2541     }
2542     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
2543     case VTD_INV_DESC_CC_DOMAIN:
2544         trace_vtd_inv_desc_cc_domain(
2545             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
2546         /* Fall through */
2547     case VTD_INV_DESC_CC_GLOBAL:
2548         vtd_context_global_invalidate(s);
2549         break;
2550 
2551     case VTD_INV_DESC_CC_DEVICE:
2552         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
2553         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
2554         vtd_context_device_invalidate(s, sid, fmask);
2555         break;
2556 
2557     default:
2558         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2559                           " (invalid type)", __func__, inv_desc->hi,
2560                           inv_desc->lo);
2561         return false;
2562     }
2563     return true;
2564 }
2565 
2566 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2567 {
2568     uint16_t domain_id;
2569     uint8_t am;
2570     hwaddr addr;
2571 
2572     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
2573         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
2574         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2575                           ", lo=0x%"PRIx64" (reserved bits unzero)",
2576                           __func__, inv_desc->hi, inv_desc->lo);
2577         return false;
2578     }
2579 
2580     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
2581     case VTD_INV_DESC_IOTLB_GLOBAL:
2582         vtd_iotlb_global_invalidate(s);
2583         break;
2584 
2585     case VTD_INV_DESC_IOTLB_DOMAIN:
2586         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2587         vtd_iotlb_domain_invalidate(s, domain_id);
2588         break;
2589 
2590     case VTD_INV_DESC_IOTLB_PAGE:
2591         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2592         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
2593         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
2594         if (am > VTD_MAMV) {
2595             error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2596                               ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)",
2597                               __func__, inv_desc->hi, inv_desc->lo,
2598                               am, (unsigned)VTD_MAMV);
2599             return false;
2600         }
2601         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2602         break;
2603 
2604     default:
2605         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2606                           ", lo=0x%"PRIx64" (type mismatch: 0x%llx)",
2607                           __func__, inv_desc->hi, inv_desc->lo,
2608                           inv_desc->lo & VTD_INV_DESC_IOTLB_G);
2609         return false;
2610     }
2611     return true;
2612 }
2613 
2614 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
2615                                      VTDInvDesc *inv_desc)
2616 {
2617     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
2618                            inv_desc->iec.index,
2619                            inv_desc->iec.index_mask);
2620 
2621     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
2622                        inv_desc->iec.index,
2623                        inv_desc->iec.index_mask);
2624     return true;
2625 }
2626 
2627 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
2628                                           VTDInvDesc *inv_desc)
2629 {
2630     VTDAddressSpace *vtd_dev_as;
2631     IOMMUTLBEvent event;
2632     hwaddr addr;
2633     uint64_t sz;
2634     uint16_t sid;
2635     bool size;
2636 
2637     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2638     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2639     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2640 
2641     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2642         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2643         error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2644                           ", lo=%"PRIx64" (reserved nonzero)", __func__,
2645                           inv_desc->hi, inv_desc->lo);
2646         return false;
2647     }
2648 
2649     /*
2650      * Using sid is OK since the guest should have finished the
2651      * initialization of both the bus and device.
2652      */
2653     vtd_dev_as = vtd_get_as_by_sid(s, sid);
2654     if (!vtd_dev_as) {
2655         goto done;
2656     }
2657 
2658     /* According to ATS spec table 2.4:
2659      * S = 0, bits 15:12 = xxxx     range size: 4K
2660      * S = 1, bits 15:12 = xxx0     range size: 8K
2661      * S = 1, bits 15:12 = xx01     range size: 16K
2662      * S = 1, bits 15:12 = x011     range size: 32K
2663      * S = 1, bits 15:12 = 0111     range size: 64K
2664      * ...
2665      */
2666     if (size) {
2667         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2668         addr &= ~(sz - 1);
2669     } else {
2670         sz = VTD_PAGE_SIZE;
2671     }
2672 
2673     event.type = IOMMU_NOTIFIER_DEVIOTLB_UNMAP;
2674     event.entry.target_as = &vtd_dev_as->as;
2675     event.entry.addr_mask = sz - 1;
2676     event.entry.iova = addr;
2677     event.entry.perm = IOMMU_NONE;
2678     event.entry.translated_addr = 0;
2679     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, event);
2680 
2681 done:
2682     return true;
2683 }
2684 
2685 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2686 {
2687     VTDInvDesc inv_desc;
2688     uint8_t desc_type;
2689 
2690     trace_vtd_inv_qi_head(s->iq_head);
2691     if (!vtd_get_inv_desc(s, &inv_desc)) {
2692         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2693         return false;
2694     }
2695 
2696     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2697     /* FIXME: should update at first or at last? */
2698     s->iq_last_desc_type = desc_type;
2699 
2700     switch (desc_type) {
2701     case VTD_INV_DESC_CC:
2702         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2703         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2704             return false;
2705         }
2706         break;
2707 
2708     case VTD_INV_DESC_IOTLB:
2709         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2710         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2711             return false;
2712         }
2713         break;
2714 
2715     /*
2716      * TODO: the entity of below two cases will be implemented in future series.
2717      * To make guest (which integrates scalable mode support patch set in
2718      * iommu driver) work, just return true is enough so far.
2719      */
2720     case VTD_INV_DESC_PC:
2721         break;
2722 
2723     case VTD_INV_DESC_PIOTLB:
2724         break;
2725 
2726     case VTD_INV_DESC_WAIT:
2727         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2728         if (!vtd_process_wait_desc(s, &inv_desc)) {
2729             return false;
2730         }
2731         break;
2732 
2733     case VTD_INV_DESC_IEC:
2734         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2735         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2736             return false;
2737         }
2738         break;
2739 
2740     case VTD_INV_DESC_DEVICE:
2741         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2742         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2743             return false;
2744         }
2745         break;
2746 
2747     default:
2748         error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64
2749                           " (unknown type)", __func__, inv_desc.hi,
2750                           inv_desc.lo);
2751         return false;
2752     }
2753     s->iq_head++;
2754     if (s->iq_head == s->iq_size) {
2755         s->iq_head = 0;
2756     }
2757     return true;
2758 }
2759 
2760 /* Try to fetch and process more Invalidation Descriptors */
2761 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2762 {
2763     int qi_shift;
2764 
2765     /* Refer to 10.4.23 of VT-d spec 3.0 */
2766     qi_shift = s->iq_dw ? VTD_IQH_QH_SHIFT_5 : VTD_IQH_QH_SHIFT_4;
2767 
2768     trace_vtd_inv_qi_fetch();
2769 
2770     if (s->iq_tail >= s->iq_size) {
2771         /* Detects an invalid Tail pointer */
2772         error_report_once("%s: detected invalid QI tail "
2773                           "(tail=0x%x, size=0x%x)",
2774                           __func__, s->iq_tail, s->iq_size);
2775         vtd_handle_inv_queue_error(s);
2776         return;
2777     }
2778     while (s->iq_head != s->iq_tail) {
2779         if (!vtd_process_inv_desc(s)) {
2780             /* Invalidation Queue Errors */
2781             vtd_handle_inv_queue_error(s);
2782             break;
2783         }
2784         /* Must update the IQH_REG in time */
2785         vtd_set_quad_raw(s, DMAR_IQH_REG,
2786                          (((uint64_t)(s->iq_head)) << qi_shift) &
2787                          VTD_IQH_QH_MASK);
2788     }
2789 }
2790 
2791 /* Handle write to Invalidation Queue Tail Register */
2792 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2793 {
2794     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2795 
2796     if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) {
2797         error_report_once("%s: RSV bit is set: val=0x%"PRIx64,
2798                           __func__, val);
2799         return;
2800     }
2801     s->iq_tail = VTD_IQT_QT(s->iq_dw, val);
2802     trace_vtd_inv_qi_tail(s->iq_tail);
2803 
2804     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2805         /* Process Invalidation Queue here */
2806         vtd_fetch_inv_desc(s);
2807     }
2808 }
2809 
2810 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2811 {
2812     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2813     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2814     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2815 
2816     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2817         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2818         trace_vtd_fsts_clear_ip();
2819     }
2820     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2821      * Descriptors if there are any when Queued Invalidation is enabled?
2822      */
2823 }
2824 
2825 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2826 {
2827     uint32_t fectl_reg;
2828     /* FIXME: when software clears the IM field, check the IP field. But do we
2829      * need to compare the old value and the new value to conclude that
2830      * software clears the IM field? Or just check if the IM field is zero?
2831      */
2832     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2833 
2834     trace_vtd_reg_write_fectl(fectl_reg);
2835 
2836     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2837         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2838         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2839     }
2840 }
2841 
2842 static void vtd_handle_ics_write(IntelIOMMUState *s)
2843 {
2844     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2845     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2846 
2847     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2848         trace_vtd_reg_ics_clear_ip();
2849         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2850     }
2851 }
2852 
2853 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2854 {
2855     uint32_t iectl_reg;
2856     /* FIXME: when software clears the IM field, check the IP field. But do we
2857      * need to compare the old value and the new value to conclude that
2858      * software clears the IM field? Or just check if the IM field is zero?
2859      */
2860     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2861 
2862     trace_vtd_reg_write_iectl(iectl_reg);
2863 
2864     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2865         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2866         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2867     }
2868 }
2869 
2870 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2871 {
2872     IntelIOMMUState *s = opaque;
2873     uint64_t val;
2874 
2875     trace_vtd_reg_read(addr, size);
2876 
2877     if (addr + size > DMAR_REG_SIZE) {
2878         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2879                           " size=0x%x", __func__, addr, size);
2880         return (uint64_t)-1;
2881     }
2882 
2883     switch (addr) {
2884     /* Root Table Address Register, 64-bit */
2885     case DMAR_RTADDR_REG:
2886         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
2887         if (size == 4) {
2888             val = val & ((1ULL << 32) - 1);
2889         }
2890         break;
2891 
2892     case DMAR_RTADDR_REG_HI:
2893         assert(size == 4);
2894         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG) >> 32;
2895         break;
2896 
2897     /* Invalidation Queue Address Register, 64-bit */
2898     case DMAR_IQA_REG:
2899         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2900         if (size == 4) {
2901             val = val & ((1ULL << 32) - 1);
2902         }
2903         break;
2904 
2905     case DMAR_IQA_REG_HI:
2906         assert(size == 4);
2907         val = s->iq >> 32;
2908         break;
2909 
2910     default:
2911         if (size == 4) {
2912             val = vtd_get_long(s, addr);
2913         } else {
2914             val = vtd_get_quad(s, addr);
2915         }
2916     }
2917 
2918     return val;
2919 }
2920 
2921 static void vtd_mem_write(void *opaque, hwaddr addr,
2922                           uint64_t val, unsigned size)
2923 {
2924     IntelIOMMUState *s = opaque;
2925 
2926     trace_vtd_reg_write(addr, size, val);
2927 
2928     if (addr + size > DMAR_REG_SIZE) {
2929         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2930                           " size=0x%x", __func__, addr, size);
2931         return;
2932     }
2933 
2934     switch (addr) {
2935     /* Global Command Register, 32-bit */
2936     case DMAR_GCMD_REG:
2937         vtd_set_long(s, addr, val);
2938         vtd_handle_gcmd_write(s);
2939         break;
2940 
2941     /* Context Command Register, 64-bit */
2942     case DMAR_CCMD_REG:
2943         if (size == 4) {
2944             vtd_set_long(s, addr, val);
2945         } else {
2946             vtd_set_quad(s, addr, val);
2947             vtd_handle_ccmd_write(s);
2948         }
2949         break;
2950 
2951     case DMAR_CCMD_REG_HI:
2952         assert(size == 4);
2953         vtd_set_long(s, addr, val);
2954         vtd_handle_ccmd_write(s);
2955         break;
2956 
2957     /* IOTLB Invalidation Register, 64-bit */
2958     case DMAR_IOTLB_REG:
2959         if (size == 4) {
2960             vtd_set_long(s, addr, val);
2961         } else {
2962             vtd_set_quad(s, addr, val);
2963             vtd_handle_iotlb_write(s);
2964         }
2965         break;
2966 
2967     case DMAR_IOTLB_REG_HI:
2968         assert(size == 4);
2969         vtd_set_long(s, addr, val);
2970         vtd_handle_iotlb_write(s);
2971         break;
2972 
2973     /* Invalidate Address Register, 64-bit */
2974     case DMAR_IVA_REG:
2975         if (size == 4) {
2976             vtd_set_long(s, addr, val);
2977         } else {
2978             vtd_set_quad(s, addr, val);
2979         }
2980         break;
2981 
2982     case DMAR_IVA_REG_HI:
2983         assert(size == 4);
2984         vtd_set_long(s, addr, val);
2985         break;
2986 
2987     /* Fault Status Register, 32-bit */
2988     case DMAR_FSTS_REG:
2989         assert(size == 4);
2990         vtd_set_long(s, addr, val);
2991         vtd_handle_fsts_write(s);
2992         break;
2993 
2994     /* Fault Event Control Register, 32-bit */
2995     case DMAR_FECTL_REG:
2996         assert(size == 4);
2997         vtd_set_long(s, addr, val);
2998         vtd_handle_fectl_write(s);
2999         break;
3000 
3001     /* Fault Event Data Register, 32-bit */
3002     case DMAR_FEDATA_REG:
3003         assert(size == 4);
3004         vtd_set_long(s, addr, val);
3005         break;
3006 
3007     /* Fault Event Address Register, 32-bit */
3008     case DMAR_FEADDR_REG:
3009         if (size == 4) {
3010             vtd_set_long(s, addr, val);
3011         } else {
3012             /*
3013              * While the register is 32-bit only, some guests (Xen...) write to
3014              * it with 64-bit.
3015              */
3016             vtd_set_quad(s, addr, val);
3017         }
3018         break;
3019 
3020     /* Fault Event Upper Address Register, 32-bit */
3021     case DMAR_FEUADDR_REG:
3022         assert(size == 4);
3023         vtd_set_long(s, addr, val);
3024         break;
3025 
3026     /* Protected Memory Enable Register, 32-bit */
3027     case DMAR_PMEN_REG:
3028         assert(size == 4);
3029         vtd_set_long(s, addr, val);
3030         break;
3031 
3032     /* Root Table Address Register, 64-bit */
3033     case DMAR_RTADDR_REG:
3034         if (size == 4) {
3035             vtd_set_long(s, addr, val);
3036         } else {
3037             vtd_set_quad(s, addr, val);
3038         }
3039         break;
3040 
3041     case DMAR_RTADDR_REG_HI:
3042         assert(size == 4);
3043         vtd_set_long(s, addr, val);
3044         break;
3045 
3046     /* Invalidation Queue Tail Register, 64-bit */
3047     case DMAR_IQT_REG:
3048         if (size == 4) {
3049             vtd_set_long(s, addr, val);
3050         } else {
3051             vtd_set_quad(s, addr, val);
3052         }
3053         vtd_handle_iqt_write(s);
3054         break;
3055 
3056     case DMAR_IQT_REG_HI:
3057         assert(size == 4);
3058         vtd_set_long(s, addr, val);
3059         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
3060         break;
3061 
3062     /* Invalidation Queue Address Register, 64-bit */
3063     case DMAR_IQA_REG:
3064         if (size == 4) {
3065             vtd_set_long(s, addr, val);
3066         } else {
3067             vtd_set_quad(s, addr, val);
3068         }
3069         vtd_update_iq_dw(s);
3070         break;
3071 
3072     case DMAR_IQA_REG_HI:
3073         assert(size == 4);
3074         vtd_set_long(s, addr, val);
3075         break;
3076 
3077     /* Invalidation Completion Status Register, 32-bit */
3078     case DMAR_ICS_REG:
3079         assert(size == 4);
3080         vtd_set_long(s, addr, val);
3081         vtd_handle_ics_write(s);
3082         break;
3083 
3084     /* Invalidation Event Control Register, 32-bit */
3085     case DMAR_IECTL_REG:
3086         assert(size == 4);
3087         vtd_set_long(s, addr, val);
3088         vtd_handle_iectl_write(s);
3089         break;
3090 
3091     /* Invalidation Event Data Register, 32-bit */
3092     case DMAR_IEDATA_REG:
3093         assert(size == 4);
3094         vtd_set_long(s, addr, val);
3095         break;
3096 
3097     /* Invalidation Event Address Register, 32-bit */
3098     case DMAR_IEADDR_REG:
3099         assert(size == 4);
3100         vtd_set_long(s, addr, val);
3101         break;
3102 
3103     /* Invalidation Event Upper Address Register, 32-bit */
3104     case DMAR_IEUADDR_REG:
3105         assert(size == 4);
3106         vtd_set_long(s, addr, val);
3107         break;
3108 
3109     /* Fault Recording Registers, 128-bit */
3110     case DMAR_FRCD_REG_0_0:
3111         if (size == 4) {
3112             vtd_set_long(s, addr, val);
3113         } else {
3114             vtd_set_quad(s, addr, val);
3115         }
3116         break;
3117 
3118     case DMAR_FRCD_REG_0_1:
3119         assert(size == 4);
3120         vtd_set_long(s, addr, val);
3121         break;
3122 
3123     case DMAR_FRCD_REG_0_2:
3124         if (size == 4) {
3125             vtd_set_long(s, addr, val);
3126         } else {
3127             vtd_set_quad(s, addr, val);
3128             /* May clear bit 127 (Fault), update PPF */
3129             vtd_update_fsts_ppf(s);
3130         }
3131         break;
3132 
3133     case DMAR_FRCD_REG_0_3:
3134         assert(size == 4);
3135         vtd_set_long(s, addr, val);
3136         /* May clear bit 127 (Fault), update PPF */
3137         vtd_update_fsts_ppf(s);
3138         break;
3139 
3140     case DMAR_IRTA_REG:
3141         if (size == 4) {
3142             vtd_set_long(s, addr, val);
3143         } else {
3144             vtd_set_quad(s, addr, val);
3145         }
3146         break;
3147 
3148     case DMAR_IRTA_REG_HI:
3149         assert(size == 4);
3150         vtd_set_long(s, addr, val);
3151         break;
3152 
3153     default:
3154         if (size == 4) {
3155             vtd_set_long(s, addr, val);
3156         } else {
3157             vtd_set_quad(s, addr, val);
3158         }
3159     }
3160 }
3161 
3162 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
3163                                          IOMMUAccessFlags flag, int iommu_idx)
3164 {
3165     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
3166     IntelIOMMUState *s = vtd_as->iommu_state;
3167     IOMMUTLBEntry iotlb = {
3168         /* We'll fill in the rest later. */
3169         .target_as = &address_space_memory,
3170     };
3171     bool success;
3172 
3173     if (likely(s->dmar_enabled)) {
3174         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
3175                                          addr, flag & IOMMU_WO, &iotlb);
3176     } else {
3177         /* DMAR disabled, passthrough, use 4k-page*/
3178         iotlb.iova = addr & VTD_PAGE_MASK_4K;
3179         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
3180         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
3181         iotlb.perm = IOMMU_RW;
3182         success = true;
3183     }
3184 
3185     if (likely(success)) {
3186         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
3187                                  VTD_PCI_SLOT(vtd_as->devfn),
3188                                  VTD_PCI_FUNC(vtd_as->devfn),
3189                                  iotlb.iova, iotlb.translated_addr,
3190                                  iotlb.addr_mask);
3191     } else {
3192         error_report_once("%s: detected translation failure "
3193                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
3194                           __func__, pci_bus_num(vtd_as->bus),
3195                           VTD_PCI_SLOT(vtd_as->devfn),
3196                           VTD_PCI_FUNC(vtd_as->devfn),
3197                           addr);
3198     }
3199 
3200     return iotlb;
3201 }
3202 
3203 static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
3204                                          IOMMUNotifierFlag old,
3205                                          IOMMUNotifierFlag new,
3206                                          Error **errp)
3207 {
3208     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
3209     IntelIOMMUState *s = vtd_as->iommu_state;
3210     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3211 
3212     /* TODO: add support for VFIO and vhost users */
3213     if (s->snoop_control) {
3214         error_setg_errno(errp, ENOTSUP,
3215                          "Snoop Control with vhost or VFIO is not supported");
3216         return -ENOTSUP;
3217     }
3218     if (!s->caching_mode && (new & IOMMU_NOTIFIER_MAP)) {
3219         error_setg_errno(errp, ENOTSUP,
3220                          "device %02x.%02x.%x requires caching mode",
3221                          pci_bus_num(vtd_as->bus), PCI_SLOT(vtd_as->devfn),
3222                          PCI_FUNC(vtd_as->devfn));
3223         return -ENOTSUP;
3224     }
3225     if (!x86_iommu->dt_supported && (new & IOMMU_NOTIFIER_DEVIOTLB_UNMAP)) {
3226         error_setg_errno(errp, ENOTSUP,
3227                          "device %02x.%02x.%x requires device IOTLB mode",
3228                          pci_bus_num(vtd_as->bus), PCI_SLOT(vtd_as->devfn),
3229                          PCI_FUNC(vtd_as->devfn));
3230         return -ENOTSUP;
3231     }
3232 
3233     /* Update per-address-space notifier flags */
3234     vtd_as->notifier_flags = new;
3235 
3236     if (old == IOMMU_NOTIFIER_NONE) {
3237         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
3238     } else if (new == IOMMU_NOTIFIER_NONE) {
3239         QLIST_REMOVE(vtd_as, next);
3240     }
3241     return 0;
3242 }
3243 
3244 static int vtd_post_load(void *opaque, int version_id)
3245 {
3246     IntelIOMMUState *iommu = opaque;
3247 
3248     /*
3249      * We don't need to migrate the root_scalable because we can
3250      * simply do the calculation after the loading is complete.  We
3251      * can actually do similar things with root, dmar_enabled, etc.
3252      * however since we've had them already so we'd better keep them
3253      * for compatibility of migration.
3254      */
3255     vtd_update_scalable_state(iommu);
3256 
3257     vtd_update_iq_dw(iommu);
3258 
3259     /*
3260      * Memory regions are dynamically turned on/off depending on
3261      * context entry configurations from the guest. After migration,
3262      * we need to make sure the memory regions are still correct.
3263      */
3264     vtd_switch_address_space_all(iommu);
3265 
3266     return 0;
3267 }
3268 
3269 static const VMStateDescription vtd_vmstate = {
3270     .name = "iommu-intel",
3271     .version_id = 1,
3272     .minimum_version_id = 1,
3273     .priority = MIG_PRI_IOMMU,
3274     .post_load = vtd_post_load,
3275     .fields = (VMStateField[]) {
3276         VMSTATE_UINT64(root, IntelIOMMUState),
3277         VMSTATE_UINT64(intr_root, IntelIOMMUState),
3278         VMSTATE_UINT64(iq, IntelIOMMUState),
3279         VMSTATE_UINT32(intr_size, IntelIOMMUState),
3280         VMSTATE_UINT16(iq_head, IntelIOMMUState),
3281         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
3282         VMSTATE_UINT16(iq_size, IntelIOMMUState),
3283         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
3284         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
3285         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
3286         VMSTATE_UNUSED(1),      /* bool root_extended is obsolete by VT-d */
3287         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
3288         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
3289         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
3290         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
3291         VMSTATE_END_OF_LIST()
3292     }
3293 };
3294 
3295 static const MemoryRegionOps vtd_mem_ops = {
3296     .read = vtd_mem_read,
3297     .write = vtd_mem_write,
3298     .endianness = DEVICE_LITTLE_ENDIAN,
3299     .impl = {
3300         .min_access_size = 4,
3301         .max_access_size = 8,
3302     },
3303     .valid = {
3304         .min_access_size = 4,
3305         .max_access_size = 8,
3306     },
3307 };
3308 
3309 static Property vtd_properties[] = {
3310     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
3311     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
3312                             ON_OFF_AUTO_AUTO),
3313     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
3314     DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits,
3315                       VTD_HOST_ADDRESS_WIDTH),
3316     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
3317     DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE),
3318     DEFINE_PROP_BOOL("snoop-control", IntelIOMMUState, snoop_control, false),
3319     DEFINE_PROP_BOOL("x-pasid-mode", IntelIOMMUState, pasid, false),
3320     DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true),
3321     DEFINE_PROP_BOOL("dma-translation", IntelIOMMUState, dma_translation, true),
3322     DEFINE_PROP_END_OF_LIST(),
3323 };
3324 
3325 /* Read IRTE entry with specific index */
3326 static bool vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
3327                          VTD_IR_TableEntry *entry, uint16_t sid,
3328                          bool do_fault)
3329 {
3330     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
3331         {0xffff, 0xfffb, 0xfff9, 0xfff8};
3332     dma_addr_t addr = 0x00;
3333     uint16_t mask, source_id;
3334     uint8_t bus, bus_max, bus_min;
3335 
3336     if (index >= iommu->intr_size) {
3337         error_report_once("%s: index too large: ind=0x%x",
3338                           __func__, index);
3339         if (do_fault) {
3340             vtd_report_ir_fault(iommu, sid, VTD_FR_IR_INDEX_OVER, index);
3341         }
3342         return false;
3343     }
3344 
3345     addr = iommu->intr_root + index * sizeof(*entry);
3346     if (dma_memory_read(&address_space_memory, addr,
3347                         entry, sizeof(*entry), MEMTXATTRS_UNSPECIFIED)) {
3348         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
3349                           __func__, index, addr);
3350         if (do_fault) {
3351             vtd_report_ir_fault(iommu, sid, VTD_FR_IR_ROOT_INVAL, index);
3352         }
3353         return false;
3354     }
3355 
3356     entry->data[0] = le64_to_cpu(entry->data[0]);
3357     entry->data[1] = le64_to_cpu(entry->data[1]);
3358 
3359     trace_vtd_ir_irte_get(index, entry->data[1], entry->data[0]);
3360 
3361     /*
3362      * The remaining potential fault conditions are "qualified" by the
3363      * Fault Processing Disable bit in the IRTE. Even "not present".
3364      * So just clear the do_fault flag if PFD is set, which will
3365      * prevent faults being raised.
3366      */
3367     if (entry->irte.fault_disable) {
3368         do_fault = false;
3369     }
3370 
3371     if (!entry->irte.present) {
3372         error_report_once("%s: detected non-present IRTE "
3373                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3374                           __func__, index, entry->data[1], entry->data[0]);
3375         if (do_fault) {
3376             vtd_report_ir_fault(iommu, sid, VTD_FR_IR_ENTRY_P, index);
3377         }
3378         return false;
3379     }
3380 
3381     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
3382         entry->irte.__reserved_2) {
3383         error_report_once("%s: detected non-zero reserved IRTE "
3384                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3385                           __func__, index, entry->data[1], entry->data[0]);
3386         if (do_fault) {
3387             vtd_report_ir_fault(iommu, sid, VTD_FR_IR_IRTE_RSVD, index);
3388         }
3389         return false;
3390     }
3391 
3392     if (sid != X86_IOMMU_SID_INVALID) {
3393         /* Validate IRTE SID */
3394         source_id = entry->irte.source_id;
3395         switch (entry->irte.sid_vtype) {
3396         case VTD_SVT_NONE:
3397             break;
3398 
3399         case VTD_SVT_ALL:
3400             mask = vtd_svt_mask[entry->irte.sid_q];
3401             if ((source_id & mask) != (sid & mask)) {
3402                 error_report_once("%s: invalid IRTE SID "
3403                                   "(index=%u, sid=%u, source_id=%u)",
3404                                   __func__, index, sid, source_id);
3405                 if (do_fault) {
3406                     vtd_report_ir_fault(iommu, sid, VTD_FR_IR_SID_ERR, index);
3407                 }
3408                 return false;
3409             }
3410             break;
3411 
3412         case VTD_SVT_BUS:
3413             bus_max = source_id >> 8;
3414             bus_min = source_id & 0xff;
3415             bus = sid >> 8;
3416             if (bus > bus_max || bus < bus_min) {
3417                 error_report_once("%s: invalid SVT_BUS "
3418                                   "(index=%u, bus=%u, min=%u, max=%u)",
3419                                   __func__, index, bus, bus_min, bus_max);
3420                 if (do_fault) {
3421                     vtd_report_ir_fault(iommu, sid, VTD_FR_IR_SID_ERR, index);
3422                 }
3423                 return false;
3424             }
3425             break;
3426 
3427         default:
3428             error_report_once("%s: detected invalid IRTE SVT "
3429                               "(index=%u, type=%d)", __func__,
3430                               index, entry->irte.sid_vtype);
3431             /* Take this as verification failure. */
3432             if (do_fault) {
3433                 vtd_report_ir_fault(iommu, sid, VTD_FR_IR_SID_ERR, index);
3434             }
3435             return false;
3436         }
3437     }
3438 
3439     return true;
3440 }
3441 
3442 /* Fetch IRQ information of specific IR index */
3443 static bool vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
3444                               X86IOMMUIrq *irq, uint16_t sid, bool do_fault)
3445 {
3446     VTD_IR_TableEntry irte = {};
3447 
3448     if (!vtd_irte_get(iommu, index, &irte, sid, do_fault)) {
3449         return false;
3450     }
3451 
3452     irq->trigger_mode = irte.irte.trigger_mode;
3453     irq->vector = irte.irte.vector;
3454     irq->delivery_mode = irte.irte.delivery_mode;
3455     irq->dest = irte.irte.dest_id;
3456     if (!iommu->intr_eime) {
3457 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
3458 #define  VTD_IR_APIC_DEST_SHIFT        (8)
3459         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
3460             VTD_IR_APIC_DEST_SHIFT;
3461     }
3462     irq->dest_mode = irte.irte.dest_mode;
3463     irq->redir_hint = irte.irte.redir_hint;
3464 
3465     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
3466                        irq->delivery_mode, irq->dest, irq->dest_mode);
3467 
3468     return true;
3469 }
3470 
3471 /* Interrupt remapping for MSI/MSI-X entry */
3472 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
3473                                    MSIMessage *origin,
3474                                    MSIMessage *translated,
3475                                    uint16_t sid, bool do_fault)
3476 {
3477     VTD_IR_MSIAddress addr;
3478     uint16_t index;
3479     X86IOMMUIrq irq = {};
3480 
3481     assert(origin && translated);
3482 
3483     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
3484 
3485     if (!iommu || !iommu->intr_enabled) {
3486         memcpy(translated, origin, sizeof(*origin));
3487         goto out;
3488     }
3489 
3490     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
3491         error_report_once("%s: MSI address high 32 bits non-zero detected: "
3492                           "address=0x%" PRIx64, __func__, origin->address);
3493         if (do_fault) {
3494             vtd_report_ir_fault(iommu, sid, VTD_FR_IR_REQ_RSVD, 0);
3495         }
3496         return -EINVAL;
3497     }
3498 
3499     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
3500     if (addr.addr.__head != 0xfee) {
3501         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
3502                           __func__, addr.data);
3503         if (do_fault) {
3504             vtd_report_ir_fault(iommu, sid, VTD_FR_IR_REQ_RSVD, 0);
3505         }
3506         return -EINVAL;
3507     }
3508 
3509     /* This is compatible mode. */
3510     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
3511         memcpy(translated, origin, sizeof(*origin));
3512         goto out;
3513     }
3514 
3515     index = addr.addr.index_h << 15 | addr.addr.index_l;
3516 
3517 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
3518 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
3519 
3520     if (addr.addr.sub_valid) {
3521         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3522         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
3523     }
3524 
3525     if (!vtd_remap_irq_get(iommu, index, &irq, sid, do_fault)) {
3526         return -EINVAL;
3527     }
3528 
3529     if (addr.addr.sub_valid) {
3530         trace_vtd_ir_remap_type("MSI");
3531         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
3532             error_report_once("%s: invalid IR MSI "
3533                               "(sid=%u, address=0x%" PRIx64
3534                               ", data=0x%" PRIx32 ")",
3535                               __func__, sid, origin->address, origin->data);
3536             if (do_fault) {
3537                 vtd_report_ir_fault(iommu, sid, VTD_FR_IR_REQ_RSVD, 0);
3538             }
3539             return -EINVAL;
3540         }
3541     } else {
3542         uint8_t vector = origin->data & 0xff;
3543         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
3544 
3545         trace_vtd_ir_remap_type("IOAPIC");
3546         /* IOAPIC entry vector should be aligned with IRTE vector
3547          * (see vt-d spec 5.1.5.1). */
3548         if (vector != irq.vector) {
3549             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
3550         }
3551 
3552         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3553          * (see vt-d spec 5.1.5.1). */
3554         if (trigger_mode != irq.trigger_mode) {
3555             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
3556                                       irq.trigger_mode);
3557         }
3558     }
3559 
3560     /*
3561      * We'd better keep the last two bits, assuming that guest OS
3562      * might modify it. Keep it does not hurt after all.
3563      */
3564     irq.msi_addr_last_bits = addr.addr.__not_care;
3565 
3566     /* Translate X86IOMMUIrq to MSI message */
3567     x86_iommu_irq_to_msi_message(&irq, translated);
3568 
3569 out:
3570     trace_vtd_ir_remap_msi(origin->address, origin->data,
3571                            translated->address, translated->data);
3572     return 0;
3573 }
3574 
3575 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
3576                          MSIMessage *dst, uint16_t sid)
3577 {
3578     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
3579                                    src, dst, sid, false);
3580 }
3581 
3582 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
3583                                    uint64_t *data, unsigned size,
3584                                    MemTxAttrs attrs)
3585 {
3586     return MEMTX_OK;
3587 }
3588 
3589 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
3590                                     uint64_t value, unsigned size,
3591                                     MemTxAttrs attrs)
3592 {
3593     int ret = 0;
3594     MSIMessage from = {}, to = {};
3595     uint16_t sid = X86_IOMMU_SID_INVALID;
3596 
3597     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
3598     from.data = (uint32_t) value;
3599 
3600     if (!attrs.unspecified) {
3601         /* We have explicit Source ID */
3602         sid = attrs.requester_id;
3603     }
3604 
3605     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid, true);
3606     if (ret) {
3607         /* Drop this interrupt */
3608         return MEMTX_ERROR;
3609     }
3610 
3611     apic_get_class(NULL)->send_msi(&to);
3612 
3613     return MEMTX_OK;
3614 }
3615 
3616 static const MemoryRegionOps vtd_mem_ir_ops = {
3617     .read_with_attrs = vtd_mem_ir_read,
3618     .write_with_attrs = vtd_mem_ir_write,
3619     .endianness = DEVICE_LITTLE_ENDIAN,
3620     .impl = {
3621         .min_access_size = 4,
3622         .max_access_size = 4,
3623     },
3624     .valid = {
3625         .min_access_size = 4,
3626         .max_access_size = 4,
3627     },
3628 };
3629 
3630 static void vtd_report_ir_illegal_access(VTDAddressSpace *vtd_as,
3631                                          hwaddr addr, bool is_write)
3632 {
3633     IntelIOMMUState *s = vtd_as->iommu_state;
3634     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3635     uint16_t sid = PCI_BUILD_BDF(bus_n, vtd_as->devfn);
3636     bool is_fpd_set = false;
3637     VTDContextEntry ce;
3638 
3639     assert(vtd_as->pasid != PCI_NO_PASID);
3640 
3641     /* Try out best to fetch FPD, we can't do anything more */
3642     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3643         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
3644         if (!is_fpd_set && s->root_scalable) {
3645             vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set, vtd_as->pasid);
3646         }
3647     }
3648 
3649     vtd_report_fault(s, VTD_FR_SM_INTERRUPT_ADDR,
3650                      is_fpd_set, sid, addr, is_write,
3651                      true, vtd_as->pasid);
3652 }
3653 
3654 static MemTxResult vtd_mem_ir_fault_read(void *opaque, hwaddr addr,
3655                                          uint64_t *data, unsigned size,
3656                                          MemTxAttrs attrs)
3657 {
3658     vtd_report_ir_illegal_access(opaque, addr, false);
3659 
3660     return MEMTX_ERROR;
3661 }
3662 
3663 static MemTxResult vtd_mem_ir_fault_write(void *opaque, hwaddr addr,
3664                                           uint64_t value, unsigned size,
3665                                           MemTxAttrs attrs)
3666 {
3667     vtd_report_ir_illegal_access(opaque, addr, true);
3668 
3669     return MEMTX_ERROR;
3670 }
3671 
3672 static const MemoryRegionOps vtd_mem_ir_fault_ops = {
3673     .read_with_attrs = vtd_mem_ir_fault_read,
3674     .write_with_attrs = vtd_mem_ir_fault_write,
3675     .endianness = DEVICE_LITTLE_ENDIAN,
3676     .impl = {
3677         .min_access_size = 1,
3678         .max_access_size = 8,
3679     },
3680     .valid = {
3681         .min_access_size = 1,
3682         .max_access_size = 8,
3683     },
3684 };
3685 
3686 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus,
3687                                  int devfn, unsigned int pasid)
3688 {
3689     /*
3690      * We can't simply use sid here since the bus number might not be
3691      * initialized by the guest.
3692      */
3693     struct vtd_as_key key = {
3694         .bus = bus,
3695         .devfn = devfn,
3696         .pasid = pasid,
3697     };
3698     VTDAddressSpace *vtd_dev_as;
3699     char name[128];
3700 
3701     vtd_dev_as = g_hash_table_lookup(s->vtd_address_spaces, &key);
3702     if (!vtd_dev_as) {
3703         struct vtd_as_key *new_key = g_malloc(sizeof(*new_key));
3704 
3705         new_key->bus = bus;
3706         new_key->devfn = devfn;
3707         new_key->pasid = pasid;
3708 
3709         if (pasid == PCI_NO_PASID) {
3710             snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn),
3711                      PCI_FUNC(devfn));
3712         } else {
3713             snprintf(name, sizeof(name), "vtd-%02x.%x-pasid-%x", PCI_SLOT(devfn),
3714                      PCI_FUNC(devfn), pasid);
3715         }
3716 
3717         vtd_dev_as = g_new0(VTDAddressSpace, 1);
3718 
3719         vtd_dev_as->bus = bus;
3720         vtd_dev_as->devfn = (uint8_t)devfn;
3721         vtd_dev_as->pasid = pasid;
3722         vtd_dev_as->iommu_state = s;
3723         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
3724         vtd_dev_as->iova_tree = iova_tree_new();
3725 
3726         memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX);
3727         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root");
3728 
3729         /*
3730          * Build the DMAR-disabled container with aliases to the
3731          * shared MRs.  Note that aliasing to a shared memory region
3732          * could help the memory API to detect same FlatViews so we
3733          * can have devices to share the same FlatView when DMAR is
3734          * disabled (either by not providing "intel_iommu=on" or with
3735          * "iommu=pt").  It will greatly reduce the total number of
3736          * FlatViews of the system hence VM runs faster.
3737          */
3738         memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s),
3739                                  "vtd-nodmar", &s->mr_nodmar, 0,
3740                                  memory_region_size(&s->mr_nodmar));
3741 
3742         /*
3743          * Build the per-device DMAR-enabled container.
3744          *
3745          * TODO: currently we have per-device IOMMU memory region only
3746          * because we have per-device IOMMU notifiers for devices.  If
3747          * one day we can abstract the IOMMU notifiers out of the
3748          * memory regions then we can also share the same memory
3749          * region here just like what we've done above with the nodmar
3750          * region.
3751          */
3752         strcat(name, "-dmar");
3753         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
3754                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
3755                                  name, UINT64_MAX);
3756         memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir",
3757                                  &s->mr_ir, 0, memory_region_size(&s->mr_ir));
3758         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
3759                                             VTD_INTERRUPT_ADDR_FIRST,
3760                                             &vtd_dev_as->iommu_ir, 1);
3761 
3762         /*
3763          * This region is used for catching fault to access interrupt
3764          * range via passthrough + PASID. See also
3765          * vtd_switch_address_space(). We can't use alias since we
3766          * need to know the sid which is valid for MSI who uses
3767          * bus_master_as (see msi_send_message()).
3768          */
3769         memory_region_init_io(&vtd_dev_as->iommu_ir_fault, OBJECT(s),
3770                               &vtd_mem_ir_fault_ops, vtd_dev_as, "vtd-no-ir",
3771                               VTD_INTERRUPT_ADDR_SIZE);
3772         /*
3773          * Hook to root since when PT is enabled vtd_dev_as->iommu
3774          * will be disabled.
3775          */
3776         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->root),
3777                                             VTD_INTERRUPT_ADDR_FIRST,
3778                                             &vtd_dev_as->iommu_ir_fault, 2);
3779 
3780         /*
3781          * Hook both the containers under the root container, we
3782          * switch between DMAR & noDMAR by enable/disable
3783          * corresponding sub-containers
3784          */
3785         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3786                                             MEMORY_REGION(&vtd_dev_as->iommu),
3787                                             0);
3788         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3789                                             &vtd_dev_as->nodmar, 0);
3790 
3791         vtd_switch_address_space(vtd_dev_as);
3792 
3793         g_hash_table_insert(s->vtd_address_spaces, new_key, vtd_dev_as);
3794     }
3795     return vtd_dev_as;
3796 }
3797 
3798 /* Unmap the whole range in the notifier's scope. */
3799 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
3800 {
3801     hwaddr total, remain;
3802     hwaddr start = n->start;
3803     hwaddr end = n->end;
3804     IntelIOMMUState *s = as->iommu_state;
3805     DMAMap map;
3806 
3807     /*
3808      * Note: all the codes in this function has a assumption that IOVA
3809      * bits are no more than VTD_MGAW bits (which is restricted by
3810      * VT-d spec), otherwise we need to consider overflow of 64 bits.
3811      */
3812 
3813     if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) {
3814         /*
3815          * Don't need to unmap regions that is bigger than the whole
3816          * VT-d supported address space size
3817          */
3818         end = VTD_ADDRESS_SIZE(s->aw_bits) - 1;
3819     }
3820 
3821     assert(start <= end);
3822     total = remain = end - start + 1;
3823 
3824     while (remain >= VTD_PAGE_SIZE) {
3825         IOMMUTLBEvent event;
3826         uint64_t mask = dma_aligned_pow2_mask(start, end, s->aw_bits);
3827         uint64_t size = mask + 1;
3828 
3829         assert(size);
3830 
3831         event.type = IOMMU_NOTIFIER_UNMAP;
3832         event.entry.iova = start;
3833         event.entry.addr_mask = mask;
3834         event.entry.target_as = &address_space_memory;
3835         event.entry.perm = IOMMU_NONE;
3836         /* This field is meaningless for unmap */
3837         event.entry.translated_addr = 0;
3838 
3839         memory_region_notify_iommu_one(n, &event);
3840 
3841         start += size;
3842         remain -= size;
3843     }
3844 
3845     assert(!remain);
3846 
3847     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3848                              VTD_PCI_SLOT(as->devfn),
3849                              VTD_PCI_FUNC(as->devfn),
3850                              n->start, total);
3851 
3852     map.iova = n->start;
3853     map.size = total - 1; /* Inclusive */
3854     iova_tree_remove(as->iova_tree, map);
3855 }
3856 
3857 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3858 {
3859     VTDAddressSpace *vtd_as;
3860     IOMMUNotifier *n;
3861 
3862     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3863         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3864             vtd_address_space_unmap(vtd_as, n);
3865         }
3866     }
3867 }
3868 
3869 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3870 {
3871     vtd_address_space_unmap_all(s);
3872     vtd_switch_address_space_all(s);
3873 }
3874 
3875 static int vtd_replay_hook(IOMMUTLBEvent *event, void *private)
3876 {
3877     memory_region_notify_iommu_one(private, event);
3878     return 0;
3879 }
3880 
3881 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3882 {
3883     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3884     IntelIOMMUState *s = vtd_as->iommu_state;
3885     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3886     VTDContextEntry ce;
3887     DMAMap map = { .iova = 0, .size = HWADDR_MAX };
3888 
3889     /* replay is protected by BQL, page walk will re-setup it safely */
3890     iova_tree_remove(vtd_as->iova_tree, map);
3891 
3892     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3893         trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" :
3894                                   "legacy mode",
3895                                   bus_n, PCI_SLOT(vtd_as->devfn),
3896                                   PCI_FUNC(vtd_as->devfn),
3897                                   vtd_get_domain_id(s, &ce, vtd_as->pasid),
3898                                   ce.hi, ce.lo);
3899         if (n->notifier_flags & IOMMU_NOTIFIER_MAP) {
3900             /* This is required only for MAP typed notifiers */
3901             vtd_page_walk_info info = {
3902                 .hook_fn = vtd_replay_hook,
3903                 .private = (void *)n,
3904                 .notify_unmap = false,
3905                 .aw = s->aw_bits,
3906                 .as = vtd_as,
3907                 .domain_id = vtd_get_domain_id(s, &ce, vtd_as->pasid),
3908             };
3909 
3910             vtd_page_walk(s, &ce, 0, ~0ULL, &info, vtd_as->pasid);
3911         }
3912     } else {
3913         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3914                                     PCI_FUNC(vtd_as->devfn));
3915     }
3916 
3917     return;
3918 }
3919 
3920 /* Do the initialization. It will also be called when reset, so pay
3921  * attention when adding new initialization stuff.
3922  */
3923 static void vtd_init(IntelIOMMUState *s)
3924 {
3925     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3926 
3927     memset(s->csr, 0, DMAR_REG_SIZE);
3928     memset(s->wmask, 0, DMAR_REG_SIZE);
3929     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3930     memset(s->womask, 0, DMAR_REG_SIZE);
3931 
3932     s->root = 0;
3933     s->root_scalable = false;
3934     s->dmar_enabled = false;
3935     s->intr_enabled = false;
3936     s->iq_head = 0;
3937     s->iq_tail = 0;
3938     s->iq = 0;
3939     s->iq_size = 0;
3940     s->qi_enabled = false;
3941     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3942     s->iq_dw = false;
3943     s->next_frcd_reg = 0;
3944     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3945              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3946              VTD_CAP_MGAW(s->aw_bits);
3947     if (s->dma_drain) {
3948         s->cap |= VTD_CAP_DRAIN;
3949     }
3950     if (s->dma_translation) {
3951             if (s->aw_bits >= VTD_HOST_AW_39BIT) {
3952                     s->cap |= VTD_CAP_SAGAW_39bit;
3953             }
3954             if (s->aw_bits >= VTD_HOST_AW_48BIT) {
3955                     s->cap |= VTD_CAP_SAGAW_48bit;
3956             }
3957     }
3958     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3959 
3960     /*
3961      * Rsvd field masks for spte
3962      */
3963     vtd_spte_rsvd[0] = ~0ULL;
3964     vtd_spte_rsvd[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits,
3965                                                   x86_iommu->dt_supported);
3966     vtd_spte_rsvd[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3967     vtd_spte_rsvd[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3968     vtd_spte_rsvd[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3969 
3970     vtd_spte_rsvd_large[2] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits,
3971                                                          x86_iommu->dt_supported);
3972     vtd_spte_rsvd_large[3] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits,
3973                                                          x86_iommu->dt_supported);
3974 
3975     if (s->scalable_mode || s->snoop_control) {
3976         vtd_spte_rsvd[1] &= ~VTD_SPTE_SNP;
3977         vtd_spte_rsvd_large[2] &= ~VTD_SPTE_SNP;
3978         vtd_spte_rsvd_large[3] &= ~VTD_SPTE_SNP;
3979     }
3980 
3981     if (x86_iommu_ir_supported(x86_iommu)) {
3982         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3983         if (s->intr_eim == ON_OFF_AUTO_ON) {
3984             s->ecap |= VTD_ECAP_EIM;
3985         }
3986         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3987     }
3988 
3989     if (x86_iommu->dt_supported) {
3990         s->ecap |= VTD_ECAP_DT;
3991     }
3992 
3993     if (x86_iommu->pt_supported) {
3994         s->ecap |= VTD_ECAP_PT;
3995     }
3996 
3997     if (s->caching_mode) {
3998         s->cap |= VTD_CAP_CM;
3999     }
4000 
4001     /* TODO: read cap/ecap from host to decide which cap to be exposed. */
4002     if (s->scalable_mode) {
4003         s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS;
4004     }
4005 
4006     if (s->snoop_control) {
4007         s->ecap |= VTD_ECAP_SC;
4008     }
4009 
4010     if (s->pasid) {
4011         s->ecap |= VTD_ECAP_PASID;
4012     }
4013 
4014     vtd_reset_caches(s);
4015 
4016     /* Define registers with default values and bit semantics */
4017     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
4018     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
4019     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
4020     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
4021     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
4022     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
4023     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0);
4024     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
4025     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
4026 
4027     /* Advanced Fault Logging not supported */
4028     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
4029     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
4030     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
4031     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
4032 
4033     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
4034      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
4035      */
4036     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
4037 
4038     /* Treated as RO for implementations that PLMR and PHMR fields reported
4039      * as Clear in the CAP_REG.
4040      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
4041      */
4042     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
4043 
4044     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
4045     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
4046     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0);
4047     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
4048     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
4049     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
4050     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
4051     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
4052     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
4053 
4054     /* IOTLB registers */
4055     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
4056     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
4057     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
4058 
4059     /* Fault Recording Registers, 128-bit */
4060     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
4061     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
4062 
4063     /*
4064      * Interrupt remapping registers.
4065      */
4066     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
4067 }
4068 
4069 /* Should not reset address_spaces when reset because devices will still use
4070  * the address space they got at first (won't ask the bus again).
4071  */
4072 static void vtd_reset(DeviceState *dev)
4073 {
4074     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
4075 
4076     vtd_init(s);
4077     vtd_address_space_refresh_all(s);
4078 }
4079 
4080 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
4081 {
4082     IntelIOMMUState *s = opaque;
4083     VTDAddressSpace *vtd_as;
4084 
4085     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
4086 
4087     vtd_as = vtd_find_add_as(s, bus, devfn, PCI_NO_PASID);
4088     return &vtd_as->as;
4089 }
4090 
4091 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
4092 {
4093     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
4094 
4095     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) {
4096         error_setg(errp, "eim=on cannot be selected without intremap=on");
4097         return false;
4098     }
4099 
4100     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
4101         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
4102                       && x86_iommu_ir_supported(x86_iommu) ?
4103                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
4104     }
4105     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
4106         if (!kvm_irqchip_is_split()) {
4107             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
4108             return false;
4109         }
4110         if (kvm_enabled() && !kvm_enable_x2apic()) {
4111             error_setg(errp, "eim=on requires support on the KVM side"
4112                              "(X2APIC_API, first shipped in v4.7)");
4113             return false;
4114         }
4115     }
4116 
4117     /* Currently only address widths supported are 39 and 48 bits */
4118     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
4119         (s->aw_bits != VTD_HOST_AW_48BIT)) {
4120         error_setg(errp, "Supported values for aw-bits are: %d, %d",
4121                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
4122         return false;
4123     }
4124 
4125     if (s->scalable_mode && !s->dma_drain) {
4126         error_setg(errp, "Need to set dma_drain for scalable mode");
4127         return false;
4128     }
4129 
4130     if (s->pasid && !s->scalable_mode) {
4131         error_setg(errp, "Need to set scalable mode for PASID");
4132         return false;
4133     }
4134 
4135     return true;
4136 }
4137 
4138 static int vtd_machine_done_notify_one(Object *child, void *unused)
4139 {
4140     IntelIOMMUState *iommu = INTEL_IOMMU_DEVICE(x86_iommu_get_default());
4141 
4142     /*
4143      * We hard-coded here because vfio-pci is the only special case
4144      * here.  Let's be more elegant in the future when we can, but so
4145      * far there seems to be no better way.
4146      */
4147     if (object_dynamic_cast(child, "vfio-pci") && !iommu->caching_mode) {
4148         vtd_panic_require_caching_mode();
4149     }
4150 
4151     return 0;
4152 }
4153 
4154 static void vtd_machine_done_hook(Notifier *notifier, void *unused)
4155 {
4156     object_child_foreach_recursive(object_get_root(),
4157                                    vtd_machine_done_notify_one, NULL);
4158 }
4159 
4160 static Notifier vtd_machine_done_notify = {
4161     .notify = vtd_machine_done_hook,
4162 };
4163 
4164 static void vtd_realize(DeviceState *dev, Error **errp)
4165 {
4166     MachineState *ms = MACHINE(qdev_get_machine());
4167     PCMachineState *pcms = PC_MACHINE(ms);
4168     X86MachineState *x86ms = X86_MACHINE(ms);
4169     PCIBus *bus = pcms->bus;
4170     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
4171     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
4172 
4173     if (s->pasid && x86_iommu->dt_supported) {
4174         /*
4175          * PASID-based-Device-TLB Invalidate Descriptor is not
4176          * implemented and it requires support from vhost layer which
4177          * needs to be implemented in the future.
4178          */
4179         error_setg(errp, "PASID based device IOTLB is not supported");
4180         return;
4181     }
4182 
4183     if (!vtd_decide_config(s, errp)) {
4184         return;
4185     }
4186 
4187     QLIST_INIT(&s->vtd_as_with_notifiers);
4188     qemu_mutex_init(&s->iommu_lock);
4189     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
4190                           "intel_iommu", DMAR_REG_SIZE);
4191     memory_region_add_subregion(get_system_memory(),
4192                                 Q35_HOST_BRIDGE_IOMMU_ADDR, &s->csrmem);
4193 
4194     /* Create the shared memory regions by all devices */
4195     memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar",
4196                        UINT64_MAX);
4197     memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops,
4198                           s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE);
4199     memory_region_init_alias(&s->mr_sys_alias, OBJECT(s),
4200                              "vtd-sys-alias", get_system_memory(), 0,
4201                              memory_region_size(get_system_memory()));
4202     memory_region_add_subregion_overlap(&s->mr_nodmar, 0,
4203                                         &s->mr_sys_alias, 0);
4204     memory_region_add_subregion_overlap(&s->mr_nodmar,
4205                                         VTD_INTERRUPT_ADDR_FIRST,
4206                                         &s->mr_ir, 1);
4207     /* No corresponding destroy */
4208     s->iotlb = g_hash_table_new_full(vtd_iotlb_hash, vtd_iotlb_equal,
4209                                      g_free, g_free);
4210     s->vtd_address_spaces = g_hash_table_new_full(vtd_as_hash, vtd_as_equal,
4211                                       g_free, g_free);
4212     vtd_init(s);
4213     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
4214     /* Pseudo address space under root PCI bus. */
4215     x86ms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
4216     qemu_add_machine_init_done_notifier(&vtd_machine_done_notify);
4217 }
4218 
4219 static void vtd_class_init(ObjectClass *klass, void *data)
4220 {
4221     DeviceClass *dc = DEVICE_CLASS(klass);
4222     X86IOMMUClass *x86_class = X86_IOMMU_DEVICE_CLASS(klass);
4223 
4224     dc->reset = vtd_reset;
4225     dc->vmsd = &vtd_vmstate;
4226     device_class_set_props(dc, vtd_properties);
4227     dc->hotpluggable = false;
4228     x86_class->realize = vtd_realize;
4229     x86_class->int_remap = vtd_int_remap;
4230     /* Supported by the pc-q35-* machine types */
4231     dc->user_creatable = true;
4232     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
4233     dc->desc = "Intel IOMMU (VT-d) DMA Remapping device";
4234 }
4235 
4236 static const TypeInfo vtd_info = {
4237     .name          = TYPE_INTEL_IOMMU_DEVICE,
4238     .parent        = TYPE_X86_IOMMU_DEVICE,
4239     .instance_size = sizeof(IntelIOMMUState),
4240     .class_init    = vtd_class_init,
4241 };
4242 
4243 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
4244                                                      void *data)
4245 {
4246     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
4247 
4248     imrc->translate = vtd_iommu_translate;
4249     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
4250     imrc->replay = vtd_iommu_replay;
4251 }
4252 
4253 static const TypeInfo vtd_iommu_memory_region_info = {
4254     .parent = TYPE_IOMMU_MEMORY_REGION,
4255     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
4256     .class_init = vtd_iommu_memory_region_class_init,
4257 };
4258 
4259 static void vtd_register_types(void)
4260 {
4261     type_register_static(&vtd_info);
4262     type_register_static(&vtd_iommu_memory_region_info);
4263 }
4264 
4265 type_init(vtd_register_types)
4266