xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision e1ecf8c8)
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 "exec/address-spaces.h"
28 #include "intel_iommu_internal.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bus.h"
31 #include "hw/qdev-properties.h"
32 #include "hw/i386/pc.h"
33 #include "hw/i386/apic-msidef.h"
34 #include "hw/boards.h"
35 #include "hw/i386/x86-iommu.h"
36 #include "hw/pci-host/q35.h"
37 #include "sysemu/kvm.h"
38 #include "sysemu/sysemu.h"
39 #include "hw/i386/apic_internal.h"
40 #include "kvm_i386.h"
41 #include "migration/vmstate.h"
42 #include "trace.h"
43 
44 /* context entry operations */
45 #define VTD_CE_GET_RID2PASID(ce) \
46     ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK)
47 #define VTD_CE_GET_PASID_DIR_TABLE(ce) \
48     ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK)
49 
50 /* pe operations */
51 #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT)
52 #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW))
53 #define VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write) {\
54     if (ret_fr) {                                                             \
55         ret_fr = -ret_fr;                                                     \
56         if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) {                   \
57             trace_vtd_fault_disabled();                                       \
58         } else {                                                              \
59             vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);      \
60         }                                                                     \
61         goto error;                                                           \
62     }                                                                         \
63 }
64 
65 static void vtd_address_space_refresh_all(IntelIOMMUState *s);
66 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n);
67 
68 static void vtd_panic_require_caching_mode(void)
69 {
70     error_report("We need to set caching-mode=on for intel-iommu to enable "
71                  "device assignment with IOMMU protection.");
72     exit(1);
73 }
74 
75 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val,
76                             uint64_t wmask, uint64_t w1cmask)
77 {
78     stq_le_p(&s->csr[addr], val);
79     stq_le_p(&s->wmask[addr], wmask);
80     stq_le_p(&s->w1cmask[addr], w1cmask);
81 }
82 
83 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask)
84 {
85     stq_le_p(&s->womask[addr], mask);
86 }
87 
88 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val,
89                             uint32_t wmask, uint32_t w1cmask)
90 {
91     stl_le_p(&s->csr[addr], val);
92     stl_le_p(&s->wmask[addr], wmask);
93     stl_le_p(&s->w1cmask[addr], w1cmask);
94 }
95 
96 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask)
97 {
98     stl_le_p(&s->womask[addr], mask);
99 }
100 
101 /* "External" get/set operations */
102 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
103 {
104     uint64_t oldval = ldq_le_p(&s->csr[addr]);
105     uint64_t wmask = ldq_le_p(&s->wmask[addr]);
106     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
107     stq_le_p(&s->csr[addr],
108              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
109 }
110 
111 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
112 {
113     uint32_t oldval = ldl_le_p(&s->csr[addr]);
114     uint32_t wmask = ldl_le_p(&s->wmask[addr]);
115     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
116     stl_le_p(&s->csr[addr],
117              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
118 }
119 
120 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr)
121 {
122     uint64_t val = ldq_le_p(&s->csr[addr]);
123     uint64_t womask = ldq_le_p(&s->womask[addr]);
124     return val & ~womask;
125 }
126 
127 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr)
128 {
129     uint32_t val = ldl_le_p(&s->csr[addr]);
130     uint32_t womask = ldl_le_p(&s->womask[addr]);
131     return val & ~womask;
132 }
133 
134 /* "Internal" get/set operations */
135 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr)
136 {
137     return ldq_le_p(&s->csr[addr]);
138 }
139 
140 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr)
141 {
142     return ldl_le_p(&s->csr[addr]);
143 }
144 
145 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val)
146 {
147     stq_le_p(&s->csr[addr], val);
148 }
149 
150 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
151                                         uint32_t clear, uint32_t mask)
152 {
153     uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
154     stl_le_p(&s->csr[addr], new_val);
155     return new_val;
156 }
157 
158 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
159                                         uint64_t clear, uint64_t mask)
160 {
161     uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
162     stq_le_p(&s->csr[addr], new_val);
163     return new_val;
164 }
165 
166 static inline void vtd_iommu_lock(IntelIOMMUState *s)
167 {
168     qemu_mutex_lock(&s->iommu_lock);
169 }
170 
171 static inline void vtd_iommu_unlock(IntelIOMMUState *s)
172 {
173     qemu_mutex_unlock(&s->iommu_lock);
174 }
175 
176 static void vtd_update_scalable_state(IntelIOMMUState *s)
177 {
178     uint64_t val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
179 
180     if (s->scalable_mode) {
181         s->root_scalable = val & VTD_RTADDR_SMT;
182     }
183 }
184 
185 /* Whether the address space needs to notify new mappings */
186 static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as)
187 {
188     return as->notifier_flags & IOMMU_NOTIFIER_MAP;
189 }
190 
191 /* GHashTable functions */
192 static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2)
193 {
194     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
195 }
196 
197 static guint vtd_uint64_hash(gconstpointer v)
198 {
199     return (guint)*(const uint64_t *)v;
200 }
201 
202 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value,
203                                           gpointer user_data)
204 {
205     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
206     uint16_t domain_id = *(uint16_t *)user_data;
207     return entry->domain_id == domain_id;
208 }
209 
210 /* The shift of an addr for a certain level of paging structure */
211 static inline uint32_t vtd_slpt_level_shift(uint32_t level)
212 {
213     assert(level != 0);
214     return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
215 }
216 
217 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level)
218 {
219     return ~((1ULL << vtd_slpt_level_shift(level)) - 1);
220 }
221 
222 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value,
223                                         gpointer user_data)
224 {
225     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
226     VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data;
227     uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask;
228     uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K;
229     return (entry->domain_id == info->domain_id) &&
230             (((entry->gfn & info->mask) == gfn) ||
231              (entry->gfn == gfn_tlb));
232 }
233 
234 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
235  * IntelIOMMUState to 1.  Must be called with IOMMU lock held.
236  */
237 static void vtd_reset_context_cache_locked(IntelIOMMUState *s)
238 {
239     VTDAddressSpace *vtd_as;
240     VTDBus *vtd_bus;
241     GHashTableIter bus_it;
242     uint32_t devfn_it;
243 
244     trace_vtd_context_cache_reset();
245 
246     g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr);
247 
248     while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) {
249         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
250             vtd_as = vtd_bus->dev_as[devfn_it];
251             if (!vtd_as) {
252                 continue;
253             }
254             vtd_as->context_cache_entry.context_cache_gen = 0;
255         }
256     }
257     s->context_cache_gen = 1;
258 }
259 
260 /* Must be called with IOMMU lock held. */
261 static void vtd_reset_iotlb_locked(IntelIOMMUState *s)
262 {
263     assert(s->iotlb);
264     g_hash_table_remove_all(s->iotlb);
265 }
266 
267 static void vtd_reset_iotlb(IntelIOMMUState *s)
268 {
269     vtd_iommu_lock(s);
270     vtd_reset_iotlb_locked(s);
271     vtd_iommu_unlock(s);
272 }
273 
274 static void vtd_reset_caches(IntelIOMMUState *s)
275 {
276     vtd_iommu_lock(s);
277     vtd_reset_iotlb_locked(s);
278     vtd_reset_context_cache_locked(s);
279     vtd_iommu_unlock(s);
280 }
281 
282 static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id,
283                                   uint32_t level)
284 {
285     return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) |
286            ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT);
287 }
288 
289 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level)
290 {
291     return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K;
292 }
293 
294 /* Must be called with IOMMU lock held */
295 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id,
296                                        hwaddr addr)
297 {
298     VTDIOTLBEntry *entry;
299     uint64_t key;
300     int level;
301 
302     for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) {
303         key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level),
304                                 source_id, level);
305         entry = g_hash_table_lookup(s->iotlb, &key);
306         if (entry) {
307             goto out;
308         }
309     }
310 
311 out:
312     return entry;
313 }
314 
315 /* Must be with IOMMU lock held */
316 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id,
317                              uint16_t domain_id, hwaddr addr, uint64_t slpte,
318                              uint8_t access_flags, uint32_t level)
319 {
320     VTDIOTLBEntry *entry = g_malloc(sizeof(*entry));
321     uint64_t *key = g_malloc(sizeof(*key));
322     uint64_t gfn = vtd_get_iotlb_gfn(addr, level);
323 
324     trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id);
325     if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) {
326         trace_vtd_iotlb_reset("iotlb exceeds size limit");
327         vtd_reset_iotlb_locked(s);
328     }
329 
330     entry->gfn = gfn;
331     entry->domain_id = domain_id;
332     entry->slpte = slpte;
333     entry->access_flags = access_flags;
334     entry->mask = vtd_slpt_level_page_mask(level);
335     *key = vtd_get_iotlb_key(gfn, source_id, level);
336     g_hash_table_replace(s->iotlb, key, entry);
337 }
338 
339 /* Given the reg addr of both the message data and address, generate an
340  * interrupt via MSI.
341  */
342 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
343                                    hwaddr mesg_data_reg)
344 {
345     MSIMessage msi;
346 
347     assert(mesg_data_reg < DMAR_REG_SIZE);
348     assert(mesg_addr_reg < DMAR_REG_SIZE);
349 
350     msi.address = vtd_get_long_raw(s, mesg_addr_reg);
351     msi.data = vtd_get_long_raw(s, mesg_data_reg);
352 
353     trace_vtd_irq_generate(msi.address, msi.data);
354 
355     apic_get_class()->send_msi(&msi);
356 }
357 
358 /* Generate a fault event to software via MSI if conditions are met.
359  * Notice that the value of FSTS_REG being passed to it should be the one
360  * before any update.
361  */
362 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
363 {
364     if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
365         pre_fsts & VTD_FSTS_IQE) {
366         error_report_once("There are previous interrupt conditions "
367                           "to be serviced by software, fault event "
368                           "is not generated");
369         return;
370     }
371     vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
372     if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
373         error_report_once("Interrupt Mask set, irq is not generated");
374     } else {
375         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
376         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
377     }
378 }
379 
380 /* Check if the Fault (F) field of the Fault Recording Register referenced by
381  * @index is Set.
382  */
383 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index)
384 {
385     /* Each reg is 128-bit */
386     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
387     addr += 8; /* Access the high 64-bit half */
388 
389     assert(index < DMAR_FRCD_REG_NR);
390 
391     return vtd_get_quad_raw(s, addr) & VTD_FRCD_F;
392 }
393 
394 /* Update the PPF field of Fault Status Register.
395  * Should be called whenever change the F field of any fault recording
396  * registers.
397  */
398 static void vtd_update_fsts_ppf(IntelIOMMUState *s)
399 {
400     uint32_t i;
401     uint32_t ppf_mask = 0;
402 
403     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
404         if (vtd_is_frcd_set(s, i)) {
405             ppf_mask = VTD_FSTS_PPF;
406             break;
407         }
408     }
409     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
410     trace_vtd_fsts_ppf(!!ppf_mask);
411 }
412 
413 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index)
414 {
415     /* Each reg is 128-bit */
416     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
417     addr += 8; /* Access the high 64-bit half */
418 
419     assert(index < DMAR_FRCD_REG_NR);
420 
421     vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
422     vtd_update_fsts_ppf(s);
423 }
424 
425 /* Must not update F field now, should be done later */
426 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index,
427                             uint16_t source_id, hwaddr addr,
428                             VTDFaultReason fault, bool is_write)
429 {
430     uint64_t hi = 0, lo;
431     hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
432 
433     assert(index < DMAR_FRCD_REG_NR);
434 
435     lo = VTD_FRCD_FI(addr);
436     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
437     if (!is_write) {
438         hi |= VTD_FRCD_T;
439     }
440     vtd_set_quad_raw(s, frcd_reg_addr, lo);
441     vtd_set_quad_raw(s, frcd_reg_addr + 8, hi);
442 
443     trace_vtd_frr_new(index, hi, lo);
444 }
445 
446 /* Try to collapse multiple pending faults from the same requester */
447 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id)
448 {
449     uint32_t i;
450     uint64_t frcd_reg;
451     hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
452 
453     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
454         frcd_reg = vtd_get_quad_raw(s, addr);
455         if ((frcd_reg & VTD_FRCD_F) &&
456             ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
457             return true;
458         }
459         addr += 16; /* 128-bit for each */
460     }
461     return false;
462 }
463 
464 /* Log and report an DMAR (address translation) fault to software */
465 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
466                                   hwaddr addr, VTDFaultReason fault,
467                                   bool is_write)
468 {
469     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
470 
471     assert(fault < VTD_FR_MAX);
472 
473     if (fault == VTD_FR_RESERVED_ERR) {
474         /* This is not a normal fault reason case. Drop it. */
475         return;
476     }
477 
478     trace_vtd_dmar_fault(source_id, fault, addr, is_write);
479 
480     if (fsts_reg & VTD_FSTS_PFO) {
481         error_report_once("New fault is not recorded due to "
482                           "Primary Fault Overflow");
483         return;
484     }
485 
486     if (vtd_try_collapse_fault(s, source_id)) {
487         error_report_once("New fault is not recorded due to "
488                           "compression of faults");
489         return;
490     }
491 
492     if (vtd_is_frcd_set(s, s->next_frcd_reg)) {
493         error_report_once("Next Fault Recording Reg is used, "
494                           "new fault is not recorded, set PFO field");
495         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
496         return;
497     }
498 
499     vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write);
500 
501     if (fsts_reg & VTD_FSTS_PPF) {
502         error_report_once("There are pending faults already, "
503                           "fault event is not generated");
504         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg);
505         s->next_frcd_reg++;
506         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
507             s->next_frcd_reg = 0;
508         }
509     } else {
510         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
511                                 VTD_FSTS_FRI(s->next_frcd_reg));
512         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */
513         s->next_frcd_reg++;
514         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
515             s->next_frcd_reg = 0;
516         }
517         /* This case actually cause the PPF to be Set.
518          * So generate fault event (interrupt).
519          */
520          vtd_generate_fault_event(s, fsts_reg);
521     }
522 }
523 
524 /* Handle Invalidation Queue Errors of queued invalidation interface error
525  * conditions.
526  */
527 static void vtd_handle_inv_queue_error(IntelIOMMUState *s)
528 {
529     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
530 
531     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE);
532     vtd_generate_fault_event(s, fsts_reg);
533 }
534 
535 /* Set the IWC field and try to generate an invalidation completion interrupt */
536 static void vtd_generate_completion_event(IntelIOMMUState *s)
537 {
538     if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) {
539         trace_vtd_inv_desc_wait_irq("One pending, skip current");
540         return;
541     }
542     vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC);
543     vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP);
544     if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) {
545         trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
546                                     "new event not generated");
547         return;
548     } else {
549         /* Generate the interrupt event */
550         trace_vtd_inv_desc_wait_irq("Generating complete event");
551         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
552         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
553     }
554 }
555 
556 static inline bool vtd_root_entry_present(IntelIOMMUState *s,
557                                           VTDRootEntry *re,
558                                           uint8_t devfn)
559 {
560     if (s->root_scalable && devfn > UINT8_MAX / 2) {
561         return re->hi & VTD_ROOT_ENTRY_P;
562     }
563 
564     return re->lo & VTD_ROOT_ENTRY_P;
565 }
566 
567 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index,
568                               VTDRootEntry *re)
569 {
570     dma_addr_t addr;
571 
572     addr = s->root + index * sizeof(*re);
573     if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) {
574         re->lo = 0;
575         return -VTD_FR_ROOT_TABLE_INV;
576     }
577     re->lo = le64_to_cpu(re->lo);
578     re->hi = le64_to_cpu(re->hi);
579     return 0;
580 }
581 
582 static inline bool vtd_ce_present(VTDContextEntry *context)
583 {
584     return context->lo & VTD_CONTEXT_ENTRY_P;
585 }
586 
587 static int vtd_get_context_entry_from_root(IntelIOMMUState *s,
588                                            VTDRootEntry *re,
589                                            uint8_t index,
590                                            VTDContextEntry *ce)
591 {
592     dma_addr_t addr, ce_size;
593 
594     /* we have checked that root entry is present */
595     ce_size = s->root_scalable ? VTD_CTX_ENTRY_SCALABLE_SIZE :
596               VTD_CTX_ENTRY_LEGACY_SIZE;
597 
598     if (s->root_scalable && index > UINT8_MAX / 2) {
599         index = index & (~VTD_DEVFN_CHECK_MASK);
600         addr = re->hi & VTD_ROOT_ENTRY_CTP;
601     } else {
602         addr = re->lo & VTD_ROOT_ENTRY_CTP;
603     }
604 
605     addr = addr + index * ce_size;
606     if (dma_memory_read(&address_space_memory, addr, ce, ce_size)) {
607         return -VTD_FR_CONTEXT_TABLE_INV;
608     }
609 
610     ce->lo = le64_to_cpu(ce->lo);
611     ce->hi = le64_to_cpu(ce->hi);
612     if (ce_size == VTD_CTX_ENTRY_SCALABLE_SIZE) {
613         ce->val[2] = le64_to_cpu(ce->val[2]);
614         ce->val[3] = le64_to_cpu(ce->val[3]);
615     }
616     return 0;
617 }
618 
619 static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce)
620 {
621     return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
622 }
623 
624 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw)
625 {
626     return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw);
627 }
628 
629 /* Whether the pte indicates the address of the page frame */
630 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level)
631 {
632     return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
633 }
634 
635 /* Get the content of a spte located in @base_addr[@index] */
636 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index)
637 {
638     uint64_t slpte;
639 
640     assert(index < VTD_SL_PT_ENTRY_NR);
641 
642     if (dma_memory_read(&address_space_memory,
643                         base_addr + index * sizeof(slpte), &slpte,
644                         sizeof(slpte))) {
645         slpte = (uint64_t)-1;
646         return slpte;
647     }
648     slpte = le64_to_cpu(slpte);
649     return slpte;
650 }
651 
652 /* Given an iova and the level of paging structure, return the offset
653  * of current level.
654  */
655 static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level)
656 {
657     return (iova >> vtd_slpt_level_shift(level)) &
658             ((1ULL << VTD_SL_LEVEL_BITS) - 1);
659 }
660 
661 /* Check Capability Register to see if the @level of page-table is supported */
662 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level)
663 {
664     return VTD_CAP_SAGAW_MASK & s->cap &
665            (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
666 }
667 
668 /* Return true if check passed, otherwise false */
669 static inline bool vtd_pe_type_check(X86IOMMUState *x86_iommu,
670                                      VTDPASIDEntry *pe)
671 {
672     switch (VTD_PE_GET_TYPE(pe)) {
673     case VTD_SM_PASID_ENTRY_FLT:
674     case VTD_SM_PASID_ENTRY_SLT:
675     case VTD_SM_PASID_ENTRY_NESTED:
676         break;
677     case VTD_SM_PASID_ENTRY_PT:
678         if (!x86_iommu->pt_supported) {
679             return false;
680         }
681         break;
682     default:
683         /* Unknwon type */
684         return false;
685     }
686     return true;
687 }
688 
689 static int vtd_get_pasid_dire(dma_addr_t pasid_dir_base,
690                               uint32_t pasid,
691                               VTDPASIDDirEntry *pdire)
692 {
693     uint32_t index;
694     dma_addr_t addr, entry_size;
695 
696     index = VTD_PASID_DIR_INDEX(pasid);
697     entry_size = VTD_PASID_DIR_ENTRY_SIZE;
698     addr = pasid_dir_base + index * entry_size;
699     if (dma_memory_read(&address_space_memory, addr, pdire, entry_size)) {
700         return -VTD_FR_PASID_TABLE_INV;
701     }
702 
703     return 0;
704 }
705 
706 static int vtd_get_pasid_entry(IntelIOMMUState *s,
707                                uint32_t pasid,
708                                VTDPASIDDirEntry *pdire,
709                                VTDPASIDEntry *pe)
710 {
711     uint32_t index;
712     dma_addr_t addr, entry_size;
713     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
714 
715     index = VTD_PASID_TABLE_INDEX(pasid);
716     entry_size = VTD_PASID_ENTRY_SIZE;
717     addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK;
718     addr = addr + index * entry_size;
719     if (dma_memory_read(&address_space_memory, addr, pe, entry_size)) {
720         return -VTD_FR_PASID_TABLE_INV;
721     }
722 
723     /* Do translation type check */
724     if (!vtd_pe_type_check(x86_iommu, pe)) {
725         return -VTD_FR_PASID_TABLE_INV;
726     }
727 
728     if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) {
729         return -VTD_FR_PASID_TABLE_INV;
730     }
731 
732     return 0;
733 }
734 
735 static int vtd_get_pasid_entry_from_pasid(IntelIOMMUState *s,
736                                           dma_addr_t pasid_dir_base,
737                                           uint32_t pasid,
738                                           VTDPASIDEntry *pe)
739 {
740     int ret;
741     VTDPASIDDirEntry pdire;
742 
743     ret = vtd_get_pasid_dire(pasid_dir_base, pasid, &pdire);
744     if (ret) {
745         return ret;
746     }
747 
748     ret = vtd_get_pasid_entry(s, pasid, &pdire, pe);
749     if (ret) {
750         return ret;
751     }
752 
753     return ret;
754 }
755 
756 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s,
757                                       VTDContextEntry *ce,
758                                       VTDPASIDEntry *pe)
759 {
760     uint32_t pasid;
761     dma_addr_t pasid_dir_base;
762     int ret = 0;
763 
764     pasid = VTD_CE_GET_RID2PASID(ce);
765     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
766     ret = vtd_get_pasid_entry_from_pasid(s, pasid_dir_base, pasid, pe);
767 
768     return ret;
769 }
770 
771 static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s,
772                                 VTDContextEntry *ce,
773                                 bool *pe_fpd_set)
774 {
775     int ret;
776     uint32_t pasid;
777     dma_addr_t pasid_dir_base;
778     VTDPASIDDirEntry pdire;
779     VTDPASIDEntry pe;
780 
781     pasid = VTD_CE_GET_RID2PASID(ce);
782     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
783 
784     ret = vtd_get_pasid_dire(pasid_dir_base, pasid, &pdire);
785     if (ret) {
786         return ret;
787     }
788 
789     if (pdire.val & VTD_PASID_DIR_FPD) {
790         *pe_fpd_set = true;
791         return 0;
792     }
793 
794     ret = vtd_get_pasid_entry(s, pasid, &pdire, &pe);
795     if (ret) {
796         return ret;
797     }
798 
799     if (pe.val[0] & VTD_PASID_ENTRY_FPD) {
800         *pe_fpd_set = true;
801     }
802 
803     return 0;
804 }
805 
806 /* Get the page-table level that hardware should use for the second-level
807  * page-table walk from the Address Width field of context-entry.
808  */
809 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
810 {
811     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
812 }
813 
814 static uint32_t vtd_get_iova_level(IntelIOMMUState *s,
815                                    VTDContextEntry *ce)
816 {
817     VTDPASIDEntry pe;
818 
819     if (s->root_scalable) {
820         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
821         return VTD_PE_GET_LEVEL(&pe);
822     }
823 
824     return vtd_ce_get_level(ce);
825 }
826 
827 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
828 {
829     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
830 }
831 
832 static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s,
833                                   VTDContextEntry *ce)
834 {
835     VTDPASIDEntry pe;
836 
837     if (s->root_scalable) {
838         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
839         return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9;
840     }
841 
842     return vtd_ce_get_agaw(ce);
843 }
844 
845 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
846 {
847     return ce->lo & VTD_CONTEXT_ENTRY_TT;
848 }
849 
850 /* Only for Legacy Mode. Return true if check passed, otherwise false */
851 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
852                                      VTDContextEntry *ce)
853 {
854     switch (vtd_ce_get_type(ce)) {
855     case VTD_CONTEXT_TT_MULTI_LEVEL:
856         /* Always supported */
857         break;
858     case VTD_CONTEXT_TT_DEV_IOTLB:
859         if (!x86_iommu->dt_supported) {
860             error_report_once("%s: DT specified but not supported", __func__);
861             return false;
862         }
863         break;
864     case VTD_CONTEXT_TT_PASS_THROUGH:
865         if (!x86_iommu->pt_supported) {
866             error_report_once("%s: PT specified but not supported", __func__);
867             return false;
868         }
869         break;
870     default:
871         /* Unknown type */
872         error_report_once("%s: unknown ce type: %"PRIu32, __func__,
873                           vtd_ce_get_type(ce));
874         return false;
875     }
876     return true;
877 }
878 
879 static inline uint64_t vtd_iova_limit(IntelIOMMUState *s,
880                                       VTDContextEntry *ce, uint8_t aw)
881 {
882     uint32_t ce_agaw = vtd_get_iova_agaw(s, ce);
883     return 1ULL << MIN(ce_agaw, aw);
884 }
885 
886 /* Return true if IOVA passes range check, otherwise false. */
887 static inline bool vtd_iova_range_check(IntelIOMMUState *s,
888                                         uint64_t iova, VTDContextEntry *ce,
889                                         uint8_t aw)
890 {
891     /*
892      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
893      * in CAP_REG and AW in context-entry.
894      */
895     return !(iova & ~(vtd_iova_limit(s, ce, aw) - 1));
896 }
897 
898 static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s,
899                                           VTDContextEntry *ce)
900 {
901     VTDPASIDEntry pe;
902 
903     if (s->root_scalable) {
904         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
905         return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR;
906     }
907 
908     return vtd_ce_get_slpt_base(ce);
909 }
910 
911 /*
912  * Rsvd field masks for spte:
913  *     Index [1] to [4] 4k pages
914  *     Index [5] to [8] large pages
915  */
916 static uint64_t vtd_paging_entry_rsvd_field[9];
917 
918 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
919 {
920     if (slpte & VTD_SL_PT_PAGE_SIZE_MASK) {
921         /* Maybe large page */
922         return slpte & vtd_paging_entry_rsvd_field[level + 4];
923     } else {
924         return slpte & vtd_paging_entry_rsvd_field[level];
925     }
926 }
927 
928 /* Find the VTD address space associated with a given bus number */
929 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num)
930 {
931     VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num];
932     if (!vtd_bus) {
933         /*
934          * Iterate over the registered buses to find the one which
935          * currently hold this bus number, and update the bus_num
936          * lookup table:
937          */
938         GHashTableIter iter;
939 
940         g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
941         while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
942             if (pci_bus_num(vtd_bus->bus) == bus_num) {
943                 s->vtd_as_by_bus_num[bus_num] = vtd_bus;
944                 return vtd_bus;
945             }
946         }
947     }
948     return vtd_bus;
949 }
950 
951 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
952  * of the translation, can be used for deciding the size of large page.
953  */
954 static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce,
955                              uint64_t iova, bool is_write,
956                              uint64_t *slptep, uint32_t *slpte_level,
957                              bool *reads, bool *writes, uint8_t aw_bits)
958 {
959     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
960     uint32_t level = vtd_get_iova_level(s, ce);
961     uint32_t offset;
962     uint64_t slpte;
963     uint64_t access_right_check;
964 
965     if (!vtd_iova_range_check(s, iova, ce, aw_bits)) {
966         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")",
967                           __func__, iova);
968         return -VTD_FR_ADDR_BEYOND_MGAW;
969     }
970 
971     /* FIXME: what is the Atomics request here? */
972     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
973 
974     while (true) {
975         offset = vtd_iova_level_offset(iova, level);
976         slpte = vtd_get_slpte(addr, offset);
977 
978         if (slpte == (uint64_t)-1) {
979             error_report_once("%s: detected read error on DMAR slpte "
980                               "(iova=0x%" PRIx64 ")", __func__, iova);
981             if (level == vtd_get_iova_level(s, ce)) {
982                 /* Invalid programming of context-entry */
983                 return -VTD_FR_CONTEXT_ENTRY_INV;
984             } else {
985                 return -VTD_FR_PAGING_ENTRY_INV;
986             }
987         }
988         *reads = (*reads) && (slpte & VTD_SL_R);
989         *writes = (*writes) && (slpte & VTD_SL_W);
990         if (!(slpte & access_right_check)) {
991             error_report_once("%s: detected slpte permission error "
992                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
993                               "slpte=0x%" PRIx64 ", write=%d)", __func__,
994                               iova, level, slpte, is_write);
995             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
996         }
997         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
998             error_report_once("%s: detected splte reserve non-zero "
999                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
1000                               "slpte=0x%" PRIx64 ")", __func__, iova,
1001                               level, slpte);
1002             return -VTD_FR_PAGING_ENTRY_RSVD;
1003         }
1004 
1005         if (vtd_is_last_slpte(slpte, level)) {
1006             *slptep = slpte;
1007             *slpte_level = level;
1008             return 0;
1009         }
1010         addr = vtd_get_slpte_addr(slpte, aw_bits);
1011         level--;
1012     }
1013 }
1014 
1015 typedef int (*vtd_page_walk_hook)(IOMMUTLBEntry *entry, void *private);
1016 
1017 /**
1018  * Constant information used during page walking
1019  *
1020  * @hook_fn: hook func to be called when detected page
1021  * @private: private data to be passed into hook func
1022  * @notify_unmap: whether we should notify invalid entries
1023  * @as: VT-d address space of the device
1024  * @aw: maximum address width
1025  * @domain: domain ID of the page walk
1026  */
1027 typedef struct {
1028     VTDAddressSpace *as;
1029     vtd_page_walk_hook hook_fn;
1030     void *private;
1031     bool notify_unmap;
1032     uint8_t aw;
1033     uint16_t domain_id;
1034 } vtd_page_walk_info;
1035 
1036 static int vtd_page_walk_one(IOMMUTLBEntry *entry, vtd_page_walk_info *info)
1037 {
1038     VTDAddressSpace *as = info->as;
1039     vtd_page_walk_hook hook_fn = info->hook_fn;
1040     void *private = info->private;
1041     DMAMap target = {
1042         .iova = entry->iova,
1043         .size = entry->addr_mask,
1044         .translated_addr = entry->translated_addr,
1045         .perm = entry->perm,
1046     };
1047     DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
1048 
1049     if (entry->perm == IOMMU_NONE && !info->notify_unmap) {
1050         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1051         return 0;
1052     }
1053 
1054     assert(hook_fn);
1055 
1056     /* Update local IOVA mapped ranges */
1057     if (entry->perm) {
1058         if (mapped) {
1059             /* If it's exactly the same translation, skip */
1060             if (!memcmp(mapped, &target, sizeof(target))) {
1061                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
1062                                                  entry->translated_addr);
1063                 return 0;
1064             } else {
1065                 /*
1066                  * Translation changed.  Normally this should not
1067                  * happen, but it can happen when with buggy guest
1068                  * OSes.  Note that there will be a small window that
1069                  * we don't have map at all.  But that's the best
1070                  * effort we can do.  The ideal way to emulate this is
1071                  * atomically modify the PTE to follow what has
1072                  * changed, but we can't.  One example is that vfio
1073                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1074                  * interface to modify a mapping (meanwhile it seems
1075                  * meaningless to even provide one).  Anyway, let's
1076                  * mark this as a TODO in case one day we'll have
1077                  * a better solution.
1078                  */
1079                 IOMMUAccessFlags cache_perm = entry->perm;
1080                 int ret;
1081 
1082                 /* Emulate an UNMAP */
1083                 entry->perm = IOMMU_NONE;
1084                 trace_vtd_page_walk_one(info->domain_id,
1085                                         entry->iova,
1086                                         entry->translated_addr,
1087                                         entry->addr_mask,
1088                                         entry->perm);
1089                 ret = hook_fn(entry, private);
1090                 if (ret) {
1091                     return ret;
1092                 }
1093                 /* Drop any existing mapping */
1094                 iova_tree_remove(as->iova_tree, &target);
1095                 /* Recover the correct permission */
1096                 entry->perm = cache_perm;
1097             }
1098         }
1099         iova_tree_insert(as->iova_tree, &target);
1100     } else {
1101         if (!mapped) {
1102             /* Skip since we didn't map this range at all */
1103             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1104             return 0;
1105         }
1106         iova_tree_remove(as->iova_tree, &target);
1107     }
1108 
1109     trace_vtd_page_walk_one(info->domain_id, entry->iova,
1110                             entry->translated_addr, entry->addr_mask,
1111                             entry->perm);
1112     return hook_fn(entry, private);
1113 }
1114 
1115 /**
1116  * vtd_page_walk_level - walk over specific level for IOVA range
1117  *
1118  * @addr: base GPA addr to start the walk
1119  * @start: IOVA range start address
1120  * @end: IOVA range end address (start <= addr < end)
1121  * @read: whether parent level has read permission
1122  * @write: whether parent level has write permission
1123  * @info: constant information for the page walk
1124  */
1125 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
1126                                uint64_t end, uint32_t level, bool read,
1127                                bool write, vtd_page_walk_info *info)
1128 {
1129     bool read_cur, write_cur, entry_valid;
1130     uint32_t offset;
1131     uint64_t slpte;
1132     uint64_t subpage_size, subpage_mask;
1133     IOMMUTLBEntry entry;
1134     uint64_t iova = start;
1135     uint64_t iova_next;
1136     int ret = 0;
1137 
1138     trace_vtd_page_walk_level(addr, level, start, end);
1139 
1140     subpage_size = 1ULL << vtd_slpt_level_shift(level);
1141     subpage_mask = vtd_slpt_level_page_mask(level);
1142 
1143     while (iova < end) {
1144         iova_next = (iova & subpage_mask) + subpage_size;
1145 
1146         offset = vtd_iova_level_offset(iova, level);
1147         slpte = vtd_get_slpte(addr, offset);
1148 
1149         if (slpte == (uint64_t)-1) {
1150             trace_vtd_page_walk_skip_read(iova, iova_next);
1151             goto next;
1152         }
1153 
1154         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1155             trace_vtd_page_walk_skip_reserve(iova, iova_next);
1156             goto next;
1157         }
1158 
1159         /* Permissions are stacked with parents' */
1160         read_cur = read && (slpte & VTD_SL_R);
1161         write_cur = write && (slpte & VTD_SL_W);
1162 
1163         /*
1164          * As long as we have either read/write permission, this is a
1165          * valid entry. The rule works for both page entries and page
1166          * table entries.
1167          */
1168         entry_valid = read_cur | write_cur;
1169 
1170         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
1171             /*
1172              * This is a valid PDE (or even bigger than PDE).  We need
1173              * to walk one further level.
1174              */
1175             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
1176                                       iova, MIN(iova_next, end), level - 1,
1177                                       read_cur, write_cur, info);
1178         } else {
1179             /*
1180              * This means we are either:
1181              *
1182              * (1) the real page entry (either 4K page, or huge page)
1183              * (2) the whole range is invalid
1184              *
1185              * In either case, we send an IOTLB notification down.
1186              */
1187             entry.target_as = &address_space_memory;
1188             entry.iova = iova & subpage_mask;
1189             entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
1190             entry.addr_mask = ~subpage_mask;
1191             /* NOTE: this is only meaningful if entry_valid == true */
1192             entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
1193             ret = vtd_page_walk_one(&entry, info);
1194         }
1195 
1196         if (ret < 0) {
1197             return ret;
1198         }
1199 
1200 next:
1201         iova = iova_next;
1202     }
1203 
1204     return 0;
1205 }
1206 
1207 /**
1208  * vtd_page_walk - walk specific IOVA range, and call the hook
1209  *
1210  * @s: intel iommu state
1211  * @ce: context entry to walk upon
1212  * @start: IOVA address to start the walk
1213  * @end: IOVA range end address (start <= addr < end)
1214  * @info: page walking information struct
1215  */
1216 static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce,
1217                          uint64_t start, uint64_t end,
1218                          vtd_page_walk_info *info)
1219 {
1220     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1221     uint32_t level = vtd_get_iova_level(s, ce);
1222 
1223     if (!vtd_iova_range_check(s, start, ce, info->aw)) {
1224         return -VTD_FR_ADDR_BEYOND_MGAW;
1225     }
1226 
1227     if (!vtd_iova_range_check(s, end, ce, info->aw)) {
1228         /* Fix end so that it reaches the maximum */
1229         end = vtd_iova_limit(s, ce, info->aw);
1230     }
1231 
1232     return vtd_page_walk_level(addr, start, end, level, true, true, info);
1233 }
1234 
1235 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s,
1236                                           VTDRootEntry *re)
1237 {
1238     /* Legacy Mode reserved bits check */
1239     if (!s->root_scalable &&
1240         (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1241         goto rsvd_err;
1242 
1243     /* Scalable Mode reserved bits check */
1244     if (s->root_scalable &&
1245         ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) ||
1246          (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1247         goto rsvd_err;
1248 
1249     return 0;
1250 
1251 rsvd_err:
1252     error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1253                       ", lo=0x%"PRIx64,
1254                       __func__, re->hi, re->lo);
1255     return -VTD_FR_ROOT_ENTRY_RSVD;
1256 }
1257 
1258 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s,
1259                                                     VTDContextEntry *ce)
1260 {
1261     if (!s->root_scalable &&
1262         (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI ||
1263          ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1264         error_report_once("%s: invalid context entry: hi=%"PRIx64
1265                           ", lo=%"PRIx64" (reserved nonzero)",
1266                           __func__, ce->hi, ce->lo);
1267         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1268     }
1269 
1270     if (s->root_scalable &&
1271         (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) ||
1272          ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 ||
1273          ce->val[2] ||
1274          ce->val[3])) {
1275         error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1276                           ", val[2]=%"PRIx64
1277                           ", val[1]=%"PRIx64
1278                           ", val[0]=%"PRIx64" (reserved nonzero)",
1279                           __func__, ce->val[3], ce->val[2],
1280                           ce->val[1], ce->val[0]);
1281         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1282     }
1283 
1284     return 0;
1285 }
1286 
1287 static int vtd_ce_rid2pasid_check(IntelIOMMUState *s,
1288                                   VTDContextEntry *ce)
1289 {
1290     VTDPASIDEntry pe;
1291 
1292     /*
1293      * Make sure in Scalable Mode, a present context entry
1294      * has valid rid2pasid setting, which includes valid
1295      * rid2pasid field and corresponding pasid entry setting
1296      */
1297     return vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1298 }
1299 
1300 /* Map a device to its corresponding domain (context-entry) */
1301 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
1302                                     uint8_t devfn, VTDContextEntry *ce)
1303 {
1304     VTDRootEntry re;
1305     int ret_fr;
1306     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
1307 
1308     ret_fr = vtd_get_root_entry(s, bus_num, &re);
1309     if (ret_fr) {
1310         return ret_fr;
1311     }
1312 
1313     if (!vtd_root_entry_present(s, &re, devfn)) {
1314         /* Not error - it's okay we don't have root entry. */
1315         trace_vtd_re_not_present(bus_num);
1316         return -VTD_FR_ROOT_ENTRY_P;
1317     }
1318 
1319     ret_fr = vtd_root_entry_rsvd_bits_check(s, &re);
1320     if (ret_fr) {
1321         return ret_fr;
1322     }
1323 
1324     ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce);
1325     if (ret_fr) {
1326         return ret_fr;
1327     }
1328 
1329     if (!vtd_ce_present(ce)) {
1330         /* Not error - it's okay we don't have context entry. */
1331         trace_vtd_ce_not_present(bus_num, devfn);
1332         return -VTD_FR_CONTEXT_ENTRY_P;
1333     }
1334 
1335     ret_fr = vtd_context_entry_rsvd_bits_check(s, ce);
1336     if (ret_fr) {
1337         return ret_fr;
1338     }
1339 
1340     /* Check if the programming of context-entry is valid */
1341     if (!s->root_scalable &&
1342         !vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1343         error_report_once("%s: invalid context entry: hi=%"PRIx64
1344                           ", lo=%"PRIx64" (level %d not supported)",
1345                           __func__, ce->hi, ce->lo,
1346                           vtd_ce_get_level(ce));
1347         return -VTD_FR_CONTEXT_ENTRY_INV;
1348     }
1349 
1350     if (!s->root_scalable) {
1351         /* Do translation type check */
1352         if (!vtd_ce_type_check(x86_iommu, ce)) {
1353             /* Errors dumped in vtd_ce_type_check() */
1354             return -VTD_FR_CONTEXT_ENTRY_INV;
1355         }
1356     } else {
1357         /*
1358          * Check if the programming of context-entry.rid2pasid
1359          * and corresponding pasid setting is valid, and thus
1360          * avoids to check pasid entry fetching result in future
1361          * helper function calling.
1362          */
1363         ret_fr = vtd_ce_rid2pasid_check(s, ce);
1364         if (ret_fr) {
1365             return ret_fr;
1366         }
1367     }
1368 
1369     return 0;
1370 }
1371 
1372 static int vtd_sync_shadow_page_hook(IOMMUTLBEntry *entry,
1373                                      void *private)
1374 {
1375     memory_region_notify_iommu((IOMMUMemoryRegion *)private, 0, *entry);
1376     return 0;
1377 }
1378 
1379 static uint16_t vtd_get_domain_id(IntelIOMMUState *s,
1380                                   VTDContextEntry *ce)
1381 {
1382     VTDPASIDEntry pe;
1383 
1384     if (s->root_scalable) {
1385         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1386         return VTD_SM_PASID_ENTRY_DID(pe.val[1]);
1387     }
1388 
1389     return VTD_CONTEXT_ENTRY_DID(ce->hi);
1390 }
1391 
1392 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1393                                             VTDContextEntry *ce,
1394                                             hwaddr addr, hwaddr size)
1395 {
1396     IntelIOMMUState *s = vtd_as->iommu_state;
1397     vtd_page_walk_info info = {
1398         .hook_fn = vtd_sync_shadow_page_hook,
1399         .private = (void *)&vtd_as->iommu,
1400         .notify_unmap = true,
1401         .aw = s->aw_bits,
1402         .as = vtd_as,
1403         .domain_id = vtd_get_domain_id(s, ce),
1404     };
1405 
1406     return vtd_page_walk(s, ce, addr, addr + size, &info);
1407 }
1408 
1409 static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
1410 {
1411     int ret;
1412     VTDContextEntry ce;
1413     IOMMUNotifier *n;
1414 
1415     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1416                                    pci_bus_num(vtd_as->bus),
1417                                    vtd_as->devfn, &ce);
1418     if (ret) {
1419         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1420             /*
1421              * It's a valid scenario to have a context entry that is
1422              * not present.  For example, when a device is removed
1423              * from an existing domain then the context entry will be
1424              * zeroed by the guest before it was put into another
1425              * domain.  When this happens, instead of synchronizing
1426              * the shadow pages we should invalidate all existing
1427              * mappings and notify the backends.
1428              */
1429             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1430                 vtd_address_space_unmap(vtd_as, n);
1431             }
1432             ret = 0;
1433         }
1434         return ret;
1435     }
1436 
1437     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1438 }
1439 
1440 /*
1441  * Check if specific device is configed to bypass address
1442  * translation for DMA requests. In Scalable Mode, bypass
1443  * 1st-level translation or 2nd-level translation, it depends
1444  * on PGTT setting.
1445  */
1446 static bool vtd_dev_pt_enabled(VTDAddressSpace *as)
1447 {
1448     IntelIOMMUState *s;
1449     VTDContextEntry ce;
1450     VTDPASIDEntry pe;
1451     int ret;
1452 
1453     assert(as);
1454 
1455     s = as->iommu_state;
1456     ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus),
1457                                    as->devfn, &ce);
1458     if (ret) {
1459         /*
1460          * Possibly failed to parse the context entry for some reason
1461          * (e.g., during init, or any guest configuration errors on
1462          * context entries). We should assume PT not enabled for
1463          * safety.
1464          */
1465         return false;
1466     }
1467 
1468     if (s->root_scalable) {
1469         ret = vtd_ce_get_rid2pasid_entry(s, &ce, &pe);
1470         if (ret) {
1471             error_report_once("%s: vtd_ce_get_rid2pasid_entry error: %"PRId32,
1472                               __func__, ret);
1473             return false;
1474         }
1475         return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT);
1476     }
1477 
1478     return (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH);
1479 }
1480 
1481 /* Return whether the device is using IOMMU translation. */
1482 static bool vtd_switch_address_space(VTDAddressSpace *as)
1483 {
1484     bool use_iommu;
1485     /* Whether we need to take the BQL on our own */
1486     bool take_bql = !qemu_mutex_iothread_locked();
1487 
1488     assert(as);
1489 
1490     use_iommu = as->iommu_state->dmar_enabled && !vtd_dev_pt_enabled(as);
1491 
1492     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1493                                    VTD_PCI_SLOT(as->devfn),
1494                                    VTD_PCI_FUNC(as->devfn),
1495                                    use_iommu);
1496 
1497     /*
1498      * It's possible that we reach here without BQL, e.g., when called
1499      * from vtd_pt_enable_fast_path(). However the memory APIs need
1500      * it. We'd better make sure we have had it already, or, take it.
1501      */
1502     if (take_bql) {
1503         qemu_mutex_lock_iothread();
1504     }
1505 
1506     /* Turn off first then on the other */
1507     if (use_iommu) {
1508         memory_region_set_enabled(&as->nodmar, false);
1509         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1510     } else {
1511         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1512         memory_region_set_enabled(&as->nodmar, true);
1513     }
1514 
1515     if (take_bql) {
1516         qemu_mutex_unlock_iothread();
1517     }
1518 
1519     return use_iommu;
1520 }
1521 
1522 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1523 {
1524     GHashTableIter iter;
1525     VTDBus *vtd_bus;
1526     int i;
1527 
1528     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1529     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1530         for (i = 0; i < PCI_DEVFN_MAX; i++) {
1531             if (!vtd_bus->dev_as[i]) {
1532                 continue;
1533             }
1534             vtd_switch_address_space(vtd_bus->dev_as[i]);
1535         }
1536     }
1537 }
1538 
1539 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn)
1540 {
1541     return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
1542 }
1543 
1544 static const bool vtd_qualified_faults[] = {
1545     [VTD_FR_RESERVED] = false,
1546     [VTD_FR_ROOT_ENTRY_P] = false,
1547     [VTD_FR_CONTEXT_ENTRY_P] = true,
1548     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1549     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1550     [VTD_FR_WRITE] = true,
1551     [VTD_FR_READ] = true,
1552     [VTD_FR_PAGING_ENTRY_INV] = true,
1553     [VTD_FR_ROOT_TABLE_INV] = false,
1554     [VTD_FR_CONTEXT_TABLE_INV] = false,
1555     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1556     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1557     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1558     [VTD_FR_PASID_TABLE_INV] = false,
1559     [VTD_FR_RESERVED_ERR] = false,
1560     [VTD_FR_MAX] = false,
1561 };
1562 
1563 /* To see if a fault condition is "qualified", which is reported to software
1564  * only if the FPD field in the context-entry used to process the faulting
1565  * request is 0.
1566  */
1567 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1568 {
1569     return vtd_qualified_faults[fault];
1570 }
1571 
1572 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1573 {
1574     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1575 }
1576 
1577 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1578 {
1579     VTDBus *vtd_bus;
1580     VTDAddressSpace *vtd_as;
1581     bool success = false;
1582 
1583     vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id));
1584     if (!vtd_bus) {
1585         goto out;
1586     }
1587 
1588     vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)];
1589     if (!vtd_as) {
1590         goto out;
1591     }
1592 
1593     if (vtd_switch_address_space(vtd_as) == false) {
1594         /* We switched off IOMMU region successfully. */
1595         success = true;
1596     }
1597 
1598 out:
1599     trace_vtd_pt_enable_fast_path(source_id, success);
1600 }
1601 
1602 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1603  * translation.
1604  *
1605  * Called from RCU critical section.
1606  *
1607  * @bus_num: The bus number
1608  * @devfn: The devfn, which is the  combined of device and function number
1609  * @is_write: The access is a write operation
1610  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1611  *
1612  * Returns true if translation is successful, otherwise false.
1613  */
1614 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1615                                    uint8_t devfn, hwaddr addr, bool is_write,
1616                                    IOMMUTLBEntry *entry)
1617 {
1618     IntelIOMMUState *s = vtd_as->iommu_state;
1619     VTDContextEntry ce;
1620     uint8_t bus_num = pci_bus_num(bus);
1621     VTDContextCacheEntry *cc_entry;
1622     uint64_t slpte, page_mask;
1623     uint32_t level;
1624     uint16_t source_id = vtd_make_source_id(bus_num, devfn);
1625     int ret_fr;
1626     bool is_fpd_set = false;
1627     bool reads = true;
1628     bool writes = true;
1629     uint8_t access_flags;
1630     VTDIOTLBEntry *iotlb_entry;
1631 
1632     /*
1633      * We have standalone memory region for interrupt addresses, we
1634      * should never receive translation requests in this region.
1635      */
1636     assert(!vtd_is_interrupt_addr(addr));
1637 
1638     vtd_iommu_lock(s);
1639 
1640     cc_entry = &vtd_as->context_cache_entry;
1641 
1642     /* Try to fetch slpte form IOTLB */
1643     iotlb_entry = vtd_lookup_iotlb(s, source_id, addr);
1644     if (iotlb_entry) {
1645         trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1646                                  iotlb_entry->domain_id);
1647         slpte = iotlb_entry->slpte;
1648         access_flags = iotlb_entry->access_flags;
1649         page_mask = iotlb_entry->mask;
1650         goto out;
1651     }
1652 
1653     /* Try to fetch context-entry from cache first */
1654     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1655         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1656                                cc_entry->context_entry.lo,
1657                                cc_entry->context_cache_gen);
1658         ce = cc_entry->context_entry;
1659         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1660         if (!is_fpd_set && s->root_scalable) {
1661             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1662             VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1663         }
1664     } else {
1665         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1666         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1667         if (!ret_fr && !is_fpd_set && s->root_scalable) {
1668             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1669         }
1670         VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1671         /* Update context-cache */
1672         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1673                                   cc_entry->context_cache_gen,
1674                                   s->context_cache_gen);
1675         cc_entry->context_entry = ce;
1676         cc_entry->context_cache_gen = s->context_cache_gen;
1677     }
1678 
1679     /*
1680      * We don't need to translate for pass-through context entries.
1681      * Also, let's ignore IOTLB caching as well for PT devices.
1682      */
1683     if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) {
1684         entry->iova = addr & VTD_PAGE_MASK_4K;
1685         entry->translated_addr = entry->iova;
1686         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1687         entry->perm = IOMMU_RW;
1688         trace_vtd_translate_pt(source_id, entry->iova);
1689 
1690         /*
1691          * When this happens, it means firstly caching-mode is not
1692          * enabled, and this is the first passthrough translation for
1693          * the device. Let's enable the fast path for passthrough.
1694          *
1695          * When passthrough is disabled again for the device, we can
1696          * capture it via the context entry invalidation, then the
1697          * IOMMU region can be swapped back.
1698          */
1699         vtd_pt_enable_fast_path(s, source_id);
1700         vtd_iommu_unlock(s);
1701         return true;
1702     }
1703 
1704     ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level,
1705                                &reads, &writes, s->aw_bits);
1706     VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1707 
1708     page_mask = vtd_slpt_level_page_mask(level);
1709     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1710     vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce), addr, slpte,
1711                      access_flags, level);
1712 out:
1713     vtd_iommu_unlock(s);
1714     entry->iova = addr & page_mask;
1715     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1716     entry->addr_mask = ~page_mask;
1717     entry->perm = access_flags;
1718     return true;
1719 
1720 error:
1721     vtd_iommu_unlock(s);
1722     entry->iova = 0;
1723     entry->translated_addr = 0;
1724     entry->addr_mask = 0;
1725     entry->perm = IOMMU_NONE;
1726     return false;
1727 }
1728 
1729 static void vtd_root_table_setup(IntelIOMMUState *s)
1730 {
1731     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1732     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
1733 
1734     vtd_update_scalable_state(s);
1735 
1736     trace_vtd_reg_dmar_root(s->root, s->root_scalable);
1737 }
1738 
1739 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
1740                                uint32_t index, uint32_t mask)
1741 {
1742     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
1743 }
1744 
1745 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
1746 {
1747     uint64_t value = 0;
1748     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
1749     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
1750     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
1751     s->intr_eime = value & VTD_IRTA_EIME;
1752 
1753     /* Notify global invalidation */
1754     vtd_iec_notify_all(s, true, 0, 0);
1755 
1756     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
1757 }
1758 
1759 static void vtd_iommu_replay_all(IntelIOMMUState *s)
1760 {
1761     VTDAddressSpace *vtd_as;
1762 
1763     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1764         vtd_sync_shadow_page_table(vtd_as);
1765     }
1766 }
1767 
1768 static void vtd_context_global_invalidate(IntelIOMMUState *s)
1769 {
1770     trace_vtd_inv_desc_cc_global();
1771     /* Protects context cache */
1772     vtd_iommu_lock(s);
1773     s->context_cache_gen++;
1774     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
1775         vtd_reset_context_cache_locked(s);
1776     }
1777     vtd_iommu_unlock(s);
1778     vtd_address_space_refresh_all(s);
1779     /*
1780      * From VT-d spec 6.5.2.1, a global context entry invalidation
1781      * should be followed by a IOTLB global invalidation, so we should
1782      * be safe even without this. Hoewever, let's replay the region as
1783      * well to be safer, and go back here when we need finer tunes for
1784      * VT-d emulation codes.
1785      */
1786     vtd_iommu_replay_all(s);
1787 }
1788 
1789 /* Do a context-cache device-selective invalidation.
1790  * @func_mask: FM field after shifting
1791  */
1792 static void vtd_context_device_invalidate(IntelIOMMUState *s,
1793                                           uint16_t source_id,
1794                                           uint16_t func_mask)
1795 {
1796     uint16_t mask;
1797     VTDBus *vtd_bus;
1798     VTDAddressSpace *vtd_as;
1799     uint8_t bus_n, devfn;
1800     uint16_t devfn_it;
1801 
1802     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
1803 
1804     switch (func_mask & 3) {
1805     case 0:
1806         mask = 0;   /* No bits in the SID field masked */
1807         break;
1808     case 1:
1809         mask = 4;   /* Mask bit 2 in the SID field */
1810         break;
1811     case 2:
1812         mask = 6;   /* Mask bit 2:1 in the SID field */
1813         break;
1814     case 3:
1815         mask = 7;   /* Mask bit 2:0 in the SID field */
1816         break;
1817     }
1818     mask = ~mask;
1819 
1820     bus_n = VTD_SID_TO_BUS(source_id);
1821     vtd_bus = vtd_find_as_from_bus_num(s, bus_n);
1822     if (vtd_bus) {
1823         devfn = VTD_SID_TO_DEVFN(source_id);
1824         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
1825             vtd_as = vtd_bus->dev_as[devfn_it];
1826             if (vtd_as && ((devfn_it & mask) == (devfn & mask))) {
1827                 trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it),
1828                                              VTD_PCI_FUNC(devfn_it));
1829                 vtd_iommu_lock(s);
1830                 vtd_as->context_cache_entry.context_cache_gen = 0;
1831                 vtd_iommu_unlock(s);
1832                 /*
1833                  * Do switch address space when needed, in case if the
1834                  * device passthrough bit is switched.
1835                  */
1836                 vtd_switch_address_space(vtd_as);
1837                 /*
1838                  * So a device is moving out of (or moving into) a
1839                  * domain, resync the shadow page table.
1840                  * This won't bring bad even if we have no such
1841                  * notifier registered - the IOMMU notification
1842                  * framework will skip MAP notifications if that
1843                  * happened.
1844                  */
1845                 vtd_sync_shadow_page_table(vtd_as);
1846             }
1847         }
1848     }
1849 }
1850 
1851 /* Context-cache invalidation
1852  * Returns the Context Actual Invalidation Granularity.
1853  * @val: the content of the CCMD_REG
1854  */
1855 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
1856 {
1857     uint64_t caig;
1858     uint64_t type = val & VTD_CCMD_CIRG_MASK;
1859 
1860     switch (type) {
1861     case VTD_CCMD_DOMAIN_INVL:
1862         /* Fall through */
1863     case VTD_CCMD_GLOBAL_INVL:
1864         caig = VTD_CCMD_GLOBAL_INVL_A;
1865         vtd_context_global_invalidate(s);
1866         break;
1867 
1868     case VTD_CCMD_DEVICE_INVL:
1869         caig = VTD_CCMD_DEVICE_INVL_A;
1870         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
1871         break;
1872 
1873     default:
1874         error_report_once("%s: invalid context: 0x%" PRIx64,
1875                           __func__, val);
1876         caig = 0;
1877     }
1878     return caig;
1879 }
1880 
1881 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
1882 {
1883     trace_vtd_inv_desc_iotlb_global();
1884     vtd_reset_iotlb(s);
1885     vtd_iommu_replay_all(s);
1886 }
1887 
1888 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
1889 {
1890     VTDContextEntry ce;
1891     VTDAddressSpace *vtd_as;
1892 
1893     trace_vtd_inv_desc_iotlb_domain(domain_id);
1894 
1895     vtd_iommu_lock(s);
1896     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
1897                                 &domain_id);
1898     vtd_iommu_unlock(s);
1899 
1900     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1901         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1902                                       vtd_as->devfn, &ce) &&
1903             domain_id == vtd_get_domain_id(s, &ce)) {
1904             vtd_sync_shadow_page_table(vtd_as);
1905         }
1906     }
1907 }
1908 
1909 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
1910                                            uint16_t domain_id, hwaddr addr,
1911                                            uint8_t am)
1912 {
1913     VTDAddressSpace *vtd_as;
1914     VTDContextEntry ce;
1915     int ret;
1916     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
1917 
1918     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
1919         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1920                                        vtd_as->devfn, &ce);
1921         if (!ret && domain_id == vtd_get_domain_id(s, &ce)) {
1922             if (vtd_as_has_map_notifier(vtd_as)) {
1923                 /*
1924                  * As long as we have MAP notifications registered in
1925                  * any of our IOMMU notifiers, we need to sync the
1926                  * shadow page table.
1927                  */
1928                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
1929             } else {
1930                 /*
1931                  * For UNMAP-only notifiers, we don't need to walk the
1932                  * page tables.  We just deliver the PSI down to
1933                  * invalidate caches.
1934                  */
1935                 IOMMUTLBEntry entry = {
1936                     .target_as = &address_space_memory,
1937                     .iova = addr,
1938                     .translated_addr = 0,
1939                     .addr_mask = size - 1,
1940                     .perm = IOMMU_NONE,
1941                 };
1942                 memory_region_notify_iommu(&vtd_as->iommu, 0, entry);
1943             }
1944         }
1945     }
1946 }
1947 
1948 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
1949                                       hwaddr addr, uint8_t am)
1950 {
1951     VTDIOTLBPageInvInfo info;
1952 
1953     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
1954 
1955     assert(am <= VTD_MAMV);
1956     info.domain_id = domain_id;
1957     info.addr = addr;
1958     info.mask = ~((1 << am) - 1);
1959     vtd_iommu_lock(s);
1960     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
1961     vtd_iommu_unlock(s);
1962     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am);
1963 }
1964 
1965 /* Flush IOTLB
1966  * Returns the IOTLB Actual Invalidation Granularity.
1967  * @val: the content of the IOTLB_REG
1968  */
1969 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
1970 {
1971     uint64_t iaig;
1972     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
1973     uint16_t domain_id;
1974     hwaddr addr;
1975     uint8_t am;
1976 
1977     switch (type) {
1978     case VTD_TLB_GLOBAL_FLUSH:
1979         iaig = VTD_TLB_GLOBAL_FLUSH_A;
1980         vtd_iotlb_global_invalidate(s);
1981         break;
1982 
1983     case VTD_TLB_DSI_FLUSH:
1984         domain_id = VTD_TLB_DID(val);
1985         iaig = VTD_TLB_DSI_FLUSH_A;
1986         vtd_iotlb_domain_invalidate(s, domain_id);
1987         break;
1988 
1989     case VTD_TLB_PSI_FLUSH:
1990         domain_id = VTD_TLB_DID(val);
1991         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
1992         am = VTD_IVA_AM(addr);
1993         addr = VTD_IVA_ADDR(addr);
1994         if (am > VTD_MAMV) {
1995             error_report_once("%s: address mask overflow: 0x%" PRIx64,
1996                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
1997             iaig = 0;
1998             break;
1999         }
2000         iaig = VTD_TLB_PSI_FLUSH_A;
2001         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2002         break;
2003 
2004     default:
2005         error_report_once("%s: invalid granularity: 0x%" PRIx64,
2006                           __func__, val);
2007         iaig = 0;
2008     }
2009     return iaig;
2010 }
2011 
2012 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
2013 
2014 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
2015 {
2016     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
2017            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
2018 }
2019 
2020 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
2021 {
2022     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
2023 
2024     trace_vtd_inv_qi_enable(en);
2025 
2026     if (en) {
2027         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
2028         /* 2^(x+8) entries */
2029         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0));
2030         s->qi_enabled = true;
2031         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
2032         /* Ok - report back to driver */
2033         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
2034 
2035         if (s->iq_tail != 0) {
2036             /*
2037              * This is a spec violation but Windows guests are known to set up
2038              * Queued Invalidation this way so we allow the write and process
2039              * Invalidation Descriptors right away.
2040              */
2041             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
2042             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2043                 vtd_fetch_inv_desc(s);
2044             }
2045         }
2046     } else {
2047         if (vtd_queued_inv_disable_check(s)) {
2048             /* disable Queued Invalidation */
2049             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
2050             s->iq_head = 0;
2051             s->qi_enabled = false;
2052             /* Ok - report back to driver */
2053             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
2054         } else {
2055             error_report_once("%s: detected improper state when disable QI "
2056                               "(head=0x%x, tail=0x%x, last_type=%d)",
2057                               __func__,
2058                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
2059         }
2060     }
2061 }
2062 
2063 /* Set Root Table Pointer */
2064 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
2065 {
2066     vtd_root_table_setup(s);
2067     /* Ok - report back to driver */
2068     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
2069     vtd_reset_caches(s);
2070     vtd_address_space_refresh_all(s);
2071 }
2072 
2073 /* Set Interrupt Remap Table Pointer */
2074 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
2075 {
2076     vtd_interrupt_remap_table_setup(s);
2077     /* Ok - report back to driver */
2078     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
2079 }
2080 
2081 /* Handle Translation Enable/Disable */
2082 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
2083 {
2084     if (s->dmar_enabled == en) {
2085         return;
2086     }
2087 
2088     trace_vtd_dmar_enable(en);
2089 
2090     if (en) {
2091         s->dmar_enabled = true;
2092         /* Ok - report back to driver */
2093         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
2094     } else {
2095         s->dmar_enabled = false;
2096 
2097         /* Clear the index of Fault Recording Register */
2098         s->next_frcd_reg = 0;
2099         /* Ok - report back to driver */
2100         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
2101     }
2102 
2103     vtd_reset_caches(s);
2104     vtd_address_space_refresh_all(s);
2105 }
2106 
2107 /* Handle Interrupt Remap Enable/Disable */
2108 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
2109 {
2110     trace_vtd_ir_enable(en);
2111 
2112     if (en) {
2113         s->intr_enabled = true;
2114         /* Ok - report back to driver */
2115         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
2116     } else {
2117         s->intr_enabled = false;
2118         /* Ok - report back to driver */
2119         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
2120     }
2121 }
2122 
2123 /* Handle write to Global Command Register */
2124 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
2125 {
2126     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
2127     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
2128     uint32_t changed = status ^ val;
2129 
2130     trace_vtd_reg_write_gcmd(status, val);
2131     if (changed & VTD_GCMD_TE) {
2132         /* Translation enable/disable */
2133         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
2134     }
2135     if (val & VTD_GCMD_SRTP) {
2136         /* Set/update the root-table pointer */
2137         vtd_handle_gcmd_srtp(s);
2138     }
2139     if (changed & VTD_GCMD_QIE) {
2140         /* Queued Invalidation Enable */
2141         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
2142     }
2143     if (val & VTD_GCMD_SIRTP) {
2144         /* Set/update the interrupt remapping root-table pointer */
2145         vtd_handle_gcmd_sirtp(s);
2146     }
2147     if (changed & VTD_GCMD_IRE) {
2148         /* Interrupt remap enable/disable */
2149         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
2150     }
2151 }
2152 
2153 /* Handle write to Context Command Register */
2154 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
2155 {
2156     uint64_t ret;
2157     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
2158 
2159     /* Context-cache invalidation request */
2160     if (val & VTD_CCMD_ICC) {
2161         if (s->qi_enabled) {
2162             error_report_once("Queued Invalidation enabled, "
2163                               "should not use register-based invalidation");
2164             return;
2165         }
2166         ret = vtd_context_cache_invalidate(s, val);
2167         /* Invalidation completed. Change something to show */
2168         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
2169         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
2170                                       ret);
2171     }
2172 }
2173 
2174 /* Handle write to IOTLB Invalidation Register */
2175 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
2176 {
2177     uint64_t ret;
2178     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
2179 
2180     /* IOTLB invalidation request */
2181     if (val & VTD_TLB_IVT) {
2182         if (s->qi_enabled) {
2183             error_report_once("Queued Invalidation enabled, "
2184                               "should not use register-based invalidation");
2185             return;
2186         }
2187         ret = vtd_iotlb_flush(s, val);
2188         /* Invalidation completed. Change something to show */
2189         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
2190         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
2191                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
2192     }
2193 }
2194 
2195 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2196 static bool vtd_get_inv_desc(IntelIOMMUState *s,
2197                              VTDInvDesc *inv_desc)
2198 {
2199     dma_addr_t base_addr = s->iq;
2200     uint32_t offset = s->iq_head;
2201     uint32_t dw = s->iq_dw ? 32 : 16;
2202     dma_addr_t addr = base_addr + offset * dw;
2203 
2204     if (dma_memory_read(&address_space_memory, addr, inv_desc, dw)) {
2205         error_report_once("Read INV DESC failed.");
2206         return false;
2207     }
2208     inv_desc->lo = le64_to_cpu(inv_desc->lo);
2209     inv_desc->hi = le64_to_cpu(inv_desc->hi);
2210     if (dw == 32) {
2211         inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]);
2212         inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]);
2213     }
2214     return true;
2215 }
2216 
2217 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2218 {
2219     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
2220         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
2221         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2222                           " (reserved nonzero)", __func__, inv_desc->hi,
2223                           inv_desc->lo);
2224         return false;
2225     }
2226     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
2227         /* Status Write */
2228         uint32_t status_data = (uint32_t)(inv_desc->lo >>
2229                                VTD_INV_DESC_WAIT_DATA_SHIFT);
2230 
2231         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
2232 
2233         /* FIXME: need to be masked with HAW? */
2234         dma_addr_t status_addr = inv_desc->hi;
2235         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
2236         status_data = cpu_to_le32(status_data);
2237         if (dma_memory_write(&address_space_memory, status_addr, &status_data,
2238                              sizeof(status_data))) {
2239             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
2240             return false;
2241         }
2242     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
2243         /* Interrupt flag */
2244         vtd_generate_completion_event(s);
2245     } else {
2246         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2247                           " (unknown type)", __func__, inv_desc->hi,
2248                           inv_desc->lo);
2249         return false;
2250     }
2251     return true;
2252 }
2253 
2254 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
2255                                            VTDInvDesc *inv_desc)
2256 {
2257     uint16_t sid, fmask;
2258 
2259     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
2260         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2261                           " (reserved nonzero)", __func__, inv_desc->hi,
2262                           inv_desc->lo);
2263         return false;
2264     }
2265     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
2266     case VTD_INV_DESC_CC_DOMAIN:
2267         trace_vtd_inv_desc_cc_domain(
2268             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
2269         /* Fall through */
2270     case VTD_INV_DESC_CC_GLOBAL:
2271         vtd_context_global_invalidate(s);
2272         break;
2273 
2274     case VTD_INV_DESC_CC_DEVICE:
2275         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
2276         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
2277         vtd_context_device_invalidate(s, sid, fmask);
2278         break;
2279 
2280     default:
2281         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2282                           " (invalid type)", __func__, inv_desc->hi,
2283                           inv_desc->lo);
2284         return false;
2285     }
2286     return true;
2287 }
2288 
2289 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2290 {
2291     uint16_t domain_id;
2292     uint8_t am;
2293     hwaddr addr;
2294 
2295     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
2296         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
2297         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2298                           ", lo=0x%"PRIx64" (reserved bits unzero)\n",
2299                           __func__, inv_desc->hi, inv_desc->lo);
2300         return false;
2301     }
2302 
2303     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
2304     case VTD_INV_DESC_IOTLB_GLOBAL:
2305         vtd_iotlb_global_invalidate(s);
2306         break;
2307 
2308     case VTD_INV_DESC_IOTLB_DOMAIN:
2309         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2310         vtd_iotlb_domain_invalidate(s, domain_id);
2311         break;
2312 
2313     case VTD_INV_DESC_IOTLB_PAGE:
2314         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2315         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
2316         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
2317         if (am > VTD_MAMV) {
2318             error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2319                               ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)\n",
2320                               __func__, inv_desc->hi, inv_desc->lo,
2321                               am, (unsigned)VTD_MAMV);
2322             return false;
2323         }
2324         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2325         break;
2326 
2327     default:
2328         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2329                           ", lo=0x%"PRIx64" (type mismatch: 0x%llx)\n",
2330                           __func__, inv_desc->hi, inv_desc->lo,
2331                           inv_desc->lo & VTD_INV_DESC_IOTLB_G);
2332         return false;
2333     }
2334     return true;
2335 }
2336 
2337 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
2338                                      VTDInvDesc *inv_desc)
2339 {
2340     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
2341                            inv_desc->iec.index,
2342                            inv_desc->iec.index_mask);
2343 
2344     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
2345                        inv_desc->iec.index,
2346                        inv_desc->iec.index_mask);
2347     return true;
2348 }
2349 
2350 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
2351                                           VTDInvDesc *inv_desc)
2352 {
2353     VTDAddressSpace *vtd_dev_as;
2354     IOMMUTLBEntry entry;
2355     struct VTDBus *vtd_bus;
2356     hwaddr addr;
2357     uint64_t sz;
2358     uint16_t sid;
2359     uint8_t devfn;
2360     bool size;
2361     uint8_t bus_num;
2362 
2363     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2364     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2365     devfn = sid & 0xff;
2366     bus_num = sid >> 8;
2367     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2368 
2369     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2370         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2371         error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2372                           ", lo=%"PRIx64" (reserved nonzero)", __func__,
2373                           inv_desc->hi, inv_desc->lo);
2374         return false;
2375     }
2376 
2377     vtd_bus = vtd_find_as_from_bus_num(s, bus_num);
2378     if (!vtd_bus) {
2379         goto done;
2380     }
2381 
2382     vtd_dev_as = vtd_bus->dev_as[devfn];
2383     if (!vtd_dev_as) {
2384         goto done;
2385     }
2386 
2387     /* According to ATS spec table 2.4:
2388      * S = 0, bits 15:12 = xxxx     range size: 4K
2389      * S = 1, bits 15:12 = xxx0     range size: 8K
2390      * S = 1, bits 15:12 = xx01     range size: 16K
2391      * S = 1, bits 15:12 = x011     range size: 32K
2392      * S = 1, bits 15:12 = 0111     range size: 64K
2393      * ...
2394      */
2395     if (size) {
2396         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2397         addr &= ~(sz - 1);
2398     } else {
2399         sz = VTD_PAGE_SIZE;
2400     }
2401 
2402     entry.target_as = &vtd_dev_as->as;
2403     entry.addr_mask = sz - 1;
2404     entry.iova = addr;
2405     entry.perm = IOMMU_NONE;
2406     entry.translated_addr = 0;
2407     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry);
2408 
2409 done:
2410     return true;
2411 }
2412 
2413 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2414 {
2415     VTDInvDesc inv_desc;
2416     uint8_t desc_type;
2417 
2418     trace_vtd_inv_qi_head(s->iq_head);
2419     if (!vtd_get_inv_desc(s, &inv_desc)) {
2420         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2421         return false;
2422     }
2423 
2424     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2425     /* FIXME: should update at first or at last? */
2426     s->iq_last_desc_type = desc_type;
2427 
2428     switch (desc_type) {
2429     case VTD_INV_DESC_CC:
2430         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2431         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2432             return false;
2433         }
2434         break;
2435 
2436     case VTD_INV_DESC_IOTLB:
2437         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2438         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2439             return false;
2440         }
2441         break;
2442 
2443     /*
2444      * TODO: the entity of below two cases will be implemented in future series.
2445      * To make guest (which integrates scalable mode support patch set in
2446      * iommu driver) work, just return true is enough so far.
2447      */
2448     case VTD_INV_DESC_PC:
2449         break;
2450 
2451     case VTD_INV_DESC_PIOTLB:
2452         break;
2453 
2454     case VTD_INV_DESC_WAIT:
2455         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2456         if (!vtd_process_wait_desc(s, &inv_desc)) {
2457             return false;
2458         }
2459         break;
2460 
2461     case VTD_INV_DESC_IEC:
2462         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2463         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2464             return false;
2465         }
2466         break;
2467 
2468     case VTD_INV_DESC_DEVICE:
2469         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2470         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2471             return false;
2472         }
2473         break;
2474 
2475     default:
2476         error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64
2477                           " (unknown type)", __func__, inv_desc.hi,
2478                           inv_desc.lo);
2479         return false;
2480     }
2481     s->iq_head++;
2482     if (s->iq_head == s->iq_size) {
2483         s->iq_head = 0;
2484     }
2485     return true;
2486 }
2487 
2488 /* Try to fetch and process more Invalidation Descriptors */
2489 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2490 {
2491     trace_vtd_inv_qi_fetch();
2492 
2493     if (s->iq_tail >= s->iq_size) {
2494         /* Detects an invalid Tail pointer */
2495         error_report_once("%s: detected invalid QI tail "
2496                           "(tail=0x%x, size=0x%x)",
2497                           __func__, s->iq_tail, s->iq_size);
2498         vtd_handle_inv_queue_error(s);
2499         return;
2500     }
2501     while (s->iq_head != s->iq_tail) {
2502         if (!vtd_process_inv_desc(s)) {
2503             /* Invalidation Queue Errors */
2504             vtd_handle_inv_queue_error(s);
2505             break;
2506         }
2507         /* Must update the IQH_REG in time */
2508         vtd_set_quad_raw(s, DMAR_IQH_REG,
2509                          (((uint64_t)(s->iq_head)) << VTD_IQH_QH_SHIFT) &
2510                          VTD_IQH_QH_MASK);
2511     }
2512 }
2513 
2514 /* Handle write to Invalidation Queue Tail Register */
2515 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2516 {
2517     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2518 
2519     if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) {
2520         error_report_once("%s: RSV bit is set: val=0x%"PRIx64,
2521                           __func__, val);
2522         return;
2523     }
2524     s->iq_tail = VTD_IQT_QT(s->iq_dw, val);
2525     trace_vtd_inv_qi_tail(s->iq_tail);
2526 
2527     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2528         /* Process Invalidation Queue here */
2529         vtd_fetch_inv_desc(s);
2530     }
2531 }
2532 
2533 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2534 {
2535     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2536     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2537     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2538 
2539     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2540         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2541         trace_vtd_fsts_clear_ip();
2542     }
2543     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2544      * Descriptors if there are any when Queued Invalidation is enabled?
2545      */
2546 }
2547 
2548 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2549 {
2550     uint32_t fectl_reg;
2551     /* FIXME: when software clears the IM field, check the IP field. But do we
2552      * need to compare the old value and the new value to conclude that
2553      * software clears the IM field? Or just check if the IM field is zero?
2554      */
2555     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2556 
2557     trace_vtd_reg_write_fectl(fectl_reg);
2558 
2559     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2560         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2561         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2562     }
2563 }
2564 
2565 static void vtd_handle_ics_write(IntelIOMMUState *s)
2566 {
2567     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2568     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2569 
2570     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2571         trace_vtd_reg_ics_clear_ip();
2572         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2573     }
2574 }
2575 
2576 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2577 {
2578     uint32_t iectl_reg;
2579     /* FIXME: when software clears the IM field, check the IP field. But do we
2580      * need to compare the old value and the new value to conclude that
2581      * software clears the IM field? Or just check if the IM field is zero?
2582      */
2583     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2584 
2585     trace_vtd_reg_write_iectl(iectl_reg);
2586 
2587     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2588         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2589         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2590     }
2591 }
2592 
2593 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2594 {
2595     IntelIOMMUState *s = opaque;
2596     uint64_t val;
2597 
2598     trace_vtd_reg_read(addr, size);
2599 
2600     if (addr + size > DMAR_REG_SIZE) {
2601         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2602                           " size=0x%u", __func__, addr, size);
2603         return (uint64_t)-1;
2604     }
2605 
2606     switch (addr) {
2607     /* Root Table Address Register, 64-bit */
2608     case DMAR_RTADDR_REG:
2609         if (size == 4) {
2610             val = s->root & ((1ULL << 32) - 1);
2611         } else {
2612             val = s->root;
2613         }
2614         break;
2615 
2616     case DMAR_RTADDR_REG_HI:
2617         assert(size == 4);
2618         val = s->root >> 32;
2619         break;
2620 
2621     /* Invalidation Queue Address Register, 64-bit */
2622     case DMAR_IQA_REG:
2623         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2624         if (size == 4) {
2625             val = val & ((1ULL << 32) - 1);
2626         }
2627         break;
2628 
2629     case DMAR_IQA_REG_HI:
2630         assert(size == 4);
2631         val = s->iq >> 32;
2632         break;
2633 
2634     default:
2635         if (size == 4) {
2636             val = vtd_get_long(s, addr);
2637         } else {
2638             val = vtd_get_quad(s, addr);
2639         }
2640     }
2641 
2642     return val;
2643 }
2644 
2645 static void vtd_mem_write(void *opaque, hwaddr addr,
2646                           uint64_t val, unsigned size)
2647 {
2648     IntelIOMMUState *s = opaque;
2649 
2650     trace_vtd_reg_write(addr, size, val);
2651 
2652     if (addr + size > DMAR_REG_SIZE) {
2653         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2654                           " size=0x%u", __func__, addr, size);
2655         return;
2656     }
2657 
2658     switch (addr) {
2659     /* Global Command Register, 32-bit */
2660     case DMAR_GCMD_REG:
2661         vtd_set_long(s, addr, val);
2662         vtd_handle_gcmd_write(s);
2663         break;
2664 
2665     /* Context Command Register, 64-bit */
2666     case DMAR_CCMD_REG:
2667         if (size == 4) {
2668             vtd_set_long(s, addr, val);
2669         } else {
2670             vtd_set_quad(s, addr, val);
2671             vtd_handle_ccmd_write(s);
2672         }
2673         break;
2674 
2675     case DMAR_CCMD_REG_HI:
2676         assert(size == 4);
2677         vtd_set_long(s, addr, val);
2678         vtd_handle_ccmd_write(s);
2679         break;
2680 
2681     /* IOTLB Invalidation Register, 64-bit */
2682     case DMAR_IOTLB_REG:
2683         if (size == 4) {
2684             vtd_set_long(s, addr, val);
2685         } else {
2686             vtd_set_quad(s, addr, val);
2687             vtd_handle_iotlb_write(s);
2688         }
2689         break;
2690 
2691     case DMAR_IOTLB_REG_HI:
2692         assert(size == 4);
2693         vtd_set_long(s, addr, val);
2694         vtd_handle_iotlb_write(s);
2695         break;
2696 
2697     /* Invalidate Address Register, 64-bit */
2698     case DMAR_IVA_REG:
2699         if (size == 4) {
2700             vtd_set_long(s, addr, val);
2701         } else {
2702             vtd_set_quad(s, addr, val);
2703         }
2704         break;
2705 
2706     case DMAR_IVA_REG_HI:
2707         assert(size == 4);
2708         vtd_set_long(s, addr, val);
2709         break;
2710 
2711     /* Fault Status Register, 32-bit */
2712     case DMAR_FSTS_REG:
2713         assert(size == 4);
2714         vtd_set_long(s, addr, val);
2715         vtd_handle_fsts_write(s);
2716         break;
2717 
2718     /* Fault Event Control Register, 32-bit */
2719     case DMAR_FECTL_REG:
2720         assert(size == 4);
2721         vtd_set_long(s, addr, val);
2722         vtd_handle_fectl_write(s);
2723         break;
2724 
2725     /* Fault Event Data Register, 32-bit */
2726     case DMAR_FEDATA_REG:
2727         assert(size == 4);
2728         vtd_set_long(s, addr, val);
2729         break;
2730 
2731     /* Fault Event Address Register, 32-bit */
2732     case DMAR_FEADDR_REG:
2733         if (size == 4) {
2734             vtd_set_long(s, addr, val);
2735         } else {
2736             /*
2737              * While the register is 32-bit only, some guests (Xen...) write to
2738              * it with 64-bit.
2739              */
2740             vtd_set_quad(s, addr, val);
2741         }
2742         break;
2743 
2744     /* Fault Event Upper Address Register, 32-bit */
2745     case DMAR_FEUADDR_REG:
2746         assert(size == 4);
2747         vtd_set_long(s, addr, val);
2748         break;
2749 
2750     /* Protected Memory Enable Register, 32-bit */
2751     case DMAR_PMEN_REG:
2752         assert(size == 4);
2753         vtd_set_long(s, addr, val);
2754         break;
2755 
2756     /* Root Table Address Register, 64-bit */
2757     case DMAR_RTADDR_REG:
2758         if (size == 4) {
2759             vtd_set_long(s, addr, val);
2760         } else {
2761             vtd_set_quad(s, addr, val);
2762         }
2763         break;
2764 
2765     case DMAR_RTADDR_REG_HI:
2766         assert(size == 4);
2767         vtd_set_long(s, addr, val);
2768         break;
2769 
2770     /* Invalidation Queue Tail Register, 64-bit */
2771     case DMAR_IQT_REG:
2772         if (size == 4) {
2773             vtd_set_long(s, addr, val);
2774         } else {
2775             vtd_set_quad(s, addr, val);
2776         }
2777         vtd_handle_iqt_write(s);
2778         break;
2779 
2780     case DMAR_IQT_REG_HI:
2781         assert(size == 4);
2782         vtd_set_long(s, addr, val);
2783         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
2784         break;
2785 
2786     /* Invalidation Queue Address Register, 64-bit */
2787     case DMAR_IQA_REG:
2788         if (size == 4) {
2789             vtd_set_long(s, addr, val);
2790         } else {
2791             vtd_set_quad(s, addr, val);
2792         }
2793         if (s->ecap & VTD_ECAP_SMTS &&
2794             val & VTD_IQA_DW_MASK) {
2795             s->iq_dw = true;
2796         } else {
2797             s->iq_dw = false;
2798         }
2799         break;
2800 
2801     case DMAR_IQA_REG_HI:
2802         assert(size == 4);
2803         vtd_set_long(s, addr, val);
2804         break;
2805 
2806     /* Invalidation Completion Status Register, 32-bit */
2807     case DMAR_ICS_REG:
2808         assert(size == 4);
2809         vtd_set_long(s, addr, val);
2810         vtd_handle_ics_write(s);
2811         break;
2812 
2813     /* Invalidation Event Control Register, 32-bit */
2814     case DMAR_IECTL_REG:
2815         assert(size == 4);
2816         vtd_set_long(s, addr, val);
2817         vtd_handle_iectl_write(s);
2818         break;
2819 
2820     /* Invalidation Event Data Register, 32-bit */
2821     case DMAR_IEDATA_REG:
2822         assert(size == 4);
2823         vtd_set_long(s, addr, val);
2824         break;
2825 
2826     /* Invalidation Event Address Register, 32-bit */
2827     case DMAR_IEADDR_REG:
2828         assert(size == 4);
2829         vtd_set_long(s, addr, val);
2830         break;
2831 
2832     /* Invalidation Event Upper Address Register, 32-bit */
2833     case DMAR_IEUADDR_REG:
2834         assert(size == 4);
2835         vtd_set_long(s, addr, val);
2836         break;
2837 
2838     /* Fault Recording Registers, 128-bit */
2839     case DMAR_FRCD_REG_0_0:
2840         if (size == 4) {
2841             vtd_set_long(s, addr, val);
2842         } else {
2843             vtd_set_quad(s, addr, val);
2844         }
2845         break;
2846 
2847     case DMAR_FRCD_REG_0_1:
2848         assert(size == 4);
2849         vtd_set_long(s, addr, val);
2850         break;
2851 
2852     case DMAR_FRCD_REG_0_2:
2853         if (size == 4) {
2854             vtd_set_long(s, addr, val);
2855         } else {
2856             vtd_set_quad(s, addr, val);
2857             /* May clear bit 127 (Fault), update PPF */
2858             vtd_update_fsts_ppf(s);
2859         }
2860         break;
2861 
2862     case DMAR_FRCD_REG_0_3:
2863         assert(size == 4);
2864         vtd_set_long(s, addr, val);
2865         /* May clear bit 127 (Fault), update PPF */
2866         vtd_update_fsts_ppf(s);
2867         break;
2868 
2869     case DMAR_IRTA_REG:
2870         if (size == 4) {
2871             vtd_set_long(s, addr, val);
2872         } else {
2873             vtd_set_quad(s, addr, val);
2874         }
2875         break;
2876 
2877     case DMAR_IRTA_REG_HI:
2878         assert(size == 4);
2879         vtd_set_long(s, addr, val);
2880         break;
2881 
2882     default:
2883         if (size == 4) {
2884             vtd_set_long(s, addr, val);
2885         } else {
2886             vtd_set_quad(s, addr, val);
2887         }
2888     }
2889 }
2890 
2891 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
2892                                          IOMMUAccessFlags flag, int iommu_idx)
2893 {
2894     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2895     IntelIOMMUState *s = vtd_as->iommu_state;
2896     IOMMUTLBEntry iotlb = {
2897         /* We'll fill in the rest later. */
2898         .target_as = &address_space_memory,
2899     };
2900     bool success;
2901 
2902     if (likely(s->dmar_enabled)) {
2903         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
2904                                          addr, flag & IOMMU_WO, &iotlb);
2905     } else {
2906         /* DMAR disabled, passthrough, use 4k-page*/
2907         iotlb.iova = addr & VTD_PAGE_MASK_4K;
2908         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
2909         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
2910         iotlb.perm = IOMMU_RW;
2911         success = true;
2912     }
2913 
2914     if (likely(success)) {
2915         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
2916                                  VTD_PCI_SLOT(vtd_as->devfn),
2917                                  VTD_PCI_FUNC(vtd_as->devfn),
2918                                  iotlb.iova, iotlb.translated_addr,
2919                                  iotlb.addr_mask);
2920     } else {
2921         error_report_once("%s: detected translation failure "
2922                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
2923                           __func__, pci_bus_num(vtd_as->bus),
2924                           VTD_PCI_SLOT(vtd_as->devfn),
2925                           VTD_PCI_FUNC(vtd_as->devfn),
2926                           addr);
2927     }
2928 
2929     return iotlb;
2930 }
2931 
2932 static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
2933                                          IOMMUNotifierFlag old,
2934                                          IOMMUNotifierFlag new,
2935                                          Error **errp)
2936 {
2937     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2938     IntelIOMMUState *s = vtd_as->iommu_state;
2939 
2940     /* Update per-address-space notifier flags */
2941     vtd_as->notifier_flags = new;
2942 
2943     if (old == IOMMU_NOTIFIER_NONE) {
2944         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
2945     } else if (new == IOMMU_NOTIFIER_NONE) {
2946         QLIST_REMOVE(vtd_as, next);
2947     }
2948     return 0;
2949 }
2950 
2951 static int vtd_post_load(void *opaque, int version_id)
2952 {
2953     IntelIOMMUState *iommu = opaque;
2954 
2955     /*
2956      * Memory regions are dynamically turned on/off depending on
2957      * context entry configurations from the guest. After migration,
2958      * we need to make sure the memory regions are still correct.
2959      */
2960     vtd_switch_address_space_all(iommu);
2961 
2962     /*
2963      * We don't need to migrate the root_scalable because we can
2964      * simply do the calculation after the loading is complete.  We
2965      * can actually do similar things with root, dmar_enabled, etc.
2966      * however since we've had them already so we'd better keep them
2967      * for compatibility of migration.
2968      */
2969     vtd_update_scalable_state(iommu);
2970 
2971     return 0;
2972 }
2973 
2974 static const VMStateDescription vtd_vmstate = {
2975     .name = "iommu-intel",
2976     .version_id = 1,
2977     .minimum_version_id = 1,
2978     .priority = MIG_PRI_IOMMU,
2979     .post_load = vtd_post_load,
2980     .fields = (VMStateField[]) {
2981         VMSTATE_UINT64(root, IntelIOMMUState),
2982         VMSTATE_UINT64(intr_root, IntelIOMMUState),
2983         VMSTATE_UINT64(iq, IntelIOMMUState),
2984         VMSTATE_UINT32(intr_size, IntelIOMMUState),
2985         VMSTATE_UINT16(iq_head, IntelIOMMUState),
2986         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
2987         VMSTATE_UINT16(iq_size, IntelIOMMUState),
2988         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
2989         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
2990         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
2991         VMSTATE_UNUSED(1),      /* bool root_extended is obsolete by VT-d */
2992         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
2993         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
2994         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
2995         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
2996         VMSTATE_END_OF_LIST()
2997     }
2998 };
2999 
3000 static const MemoryRegionOps vtd_mem_ops = {
3001     .read = vtd_mem_read,
3002     .write = vtd_mem_write,
3003     .endianness = DEVICE_LITTLE_ENDIAN,
3004     .impl = {
3005         .min_access_size = 4,
3006         .max_access_size = 8,
3007     },
3008     .valid = {
3009         .min_access_size = 4,
3010         .max_access_size = 8,
3011     },
3012 };
3013 
3014 static Property vtd_properties[] = {
3015     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
3016     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
3017                             ON_OFF_AUTO_AUTO),
3018     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
3019     DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits,
3020                       VTD_HOST_ADDRESS_WIDTH),
3021     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
3022     DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE),
3023     DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true),
3024     DEFINE_PROP_END_OF_LIST(),
3025 };
3026 
3027 /* Read IRTE entry with specific index */
3028 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
3029                         VTD_IR_TableEntry *entry, uint16_t sid)
3030 {
3031     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
3032         {0xffff, 0xfffb, 0xfff9, 0xfff8};
3033     dma_addr_t addr = 0x00;
3034     uint16_t mask, source_id;
3035     uint8_t bus, bus_max, bus_min;
3036 
3037     addr = iommu->intr_root + index * sizeof(*entry);
3038     if (dma_memory_read(&address_space_memory, addr, entry,
3039                         sizeof(*entry))) {
3040         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
3041                           __func__, index, addr);
3042         return -VTD_FR_IR_ROOT_INVAL;
3043     }
3044 
3045     trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]),
3046                           le64_to_cpu(entry->data[0]));
3047 
3048     if (!entry->irte.present) {
3049         error_report_once("%s: detected non-present IRTE "
3050                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3051                           __func__, index, le64_to_cpu(entry->data[1]),
3052                           le64_to_cpu(entry->data[0]));
3053         return -VTD_FR_IR_ENTRY_P;
3054     }
3055 
3056     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
3057         entry->irte.__reserved_2) {
3058         error_report_once("%s: detected non-zero reserved IRTE "
3059                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3060                           __func__, index, le64_to_cpu(entry->data[1]),
3061                           le64_to_cpu(entry->data[0]));
3062         return -VTD_FR_IR_IRTE_RSVD;
3063     }
3064 
3065     if (sid != X86_IOMMU_SID_INVALID) {
3066         /* Validate IRTE SID */
3067         source_id = le32_to_cpu(entry->irte.source_id);
3068         switch (entry->irte.sid_vtype) {
3069         case VTD_SVT_NONE:
3070             break;
3071 
3072         case VTD_SVT_ALL:
3073             mask = vtd_svt_mask[entry->irte.sid_q];
3074             if ((source_id & mask) != (sid & mask)) {
3075                 error_report_once("%s: invalid IRTE SID "
3076                                   "(index=%u, sid=%u, source_id=%u)",
3077                                   __func__, index, sid, source_id);
3078                 return -VTD_FR_IR_SID_ERR;
3079             }
3080             break;
3081 
3082         case VTD_SVT_BUS:
3083             bus_max = source_id >> 8;
3084             bus_min = source_id & 0xff;
3085             bus = sid >> 8;
3086             if (bus > bus_max || bus < bus_min) {
3087                 error_report_once("%s: invalid SVT_BUS "
3088                                   "(index=%u, bus=%u, min=%u, max=%u)",
3089                                   __func__, index, bus, bus_min, bus_max);
3090                 return -VTD_FR_IR_SID_ERR;
3091             }
3092             break;
3093 
3094         default:
3095             error_report_once("%s: detected invalid IRTE SVT "
3096                               "(index=%u, type=%d)", __func__,
3097                               index, entry->irte.sid_vtype);
3098             /* Take this as verification failure. */
3099             return -VTD_FR_IR_SID_ERR;
3100             break;
3101         }
3102     }
3103 
3104     return 0;
3105 }
3106 
3107 /* Fetch IRQ information of specific IR index */
3108 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
3109                              X86IOMMUIrq *irq, uint16_t sid)
3110 {
3111     VTD_IR_TableEntry irte = {};
3112     int ret = 0;
3113 
3114     ret = vtd_irte_get(iommu, index, &irte, sid);
3115     if (ret) {
3116         return ret;
3117     }
3118 
3119     irq->trigger_mode = irte.irte.trigger_mode;
3120     irq->vector = irte.irte.vector;
3121     irq->delivery_mode = irte.irte.delivery_mode;
3122     irq->dest = le32_to_cpu(irte.irte.dest_id);
3123     if (!iommu->intr_eime) {
3124 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
3125 #define  VTD_IR_APIC_DEST_SHIFT        (8)
3126         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
3127             VTD_IR_APIC_DEST_SHIFT;
3128     }
3129     irq->dest_mode = irte.irte.dest_mode;
3130     irq->redir_hint = irte.irte.redir_hint;
3131 
3132     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
3133                        irq->delivery_mode, irq->dest, irq->dest_mode);
3134 
3135     return 0;
3136 }
3137 
3138 /* Interrupt remapping for MSI/MSI-X entry */
3139 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
3140                                    MSIMessage *origin,
3141                                    MSIMessage *translated,
3142                                    uint16_t sid)
3143 {
3144     int ret = 0;
3145     VTD_IR_MSIAddress addr;
3146     uint16_t index;
3147     X86IOMMUIrq irq = {};
3148 
3149     assert(origin && translated);
3150 
3151     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
3152 
3153     if (!iommu || !iommu->intr_enabled) {
3154         memcpy(translated, origin, sizeof(*origin));
3155         goto out;
3156     }
3157 
3158     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
3159         error_report_once("%s: MSI address high 32 bits non-zero detected: "
3160                           "address=0x%" PRIx64, __func__, origin->address);
3161         return -VTD_FR_IR_REQ_RSVD;
3162     }
3163 
3164     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
3165     if (addr.addr.__head != 0xfee) {
3166         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
3167                           __func__, addr.data);
3168         return -VTD_FR_IR_REQ_RSVD;
3169     }
3170 
3171     /* This is compatible mode. */
3172     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
3173         memcpy(translated, origin, sizeof(*origin));
3174         goto out;
3175     }
3176 
3177     index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l);
3178 
3179 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
3180 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
3181 
3182     if (addr.addr.sub_valid) {
3183         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3184         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
3185     }
3186 
3187     ret = vtd_remap_irq_get(iommu, index, &irq, sid);
3188     if (ret) {
3189         return ret;
3190     }
3191 
3192     if (addr.addr.sub_valid) {
3193         trace_vtd_ir_remap_type("MSI");
3194         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
3195             error_report_once("%s: invalid IR MSI "
3196                               "(sid=%u, address=0x%" PRIx64
3197                               ", data=0x%" PRIx32 ")",
3198                               __func__, sid, origin->address, origin->data);
3199             return -VTD_FR_IR_REQ_RSVD;
3200         }
3201     } else {
3202         uint8_t vector = origin->data & 0xff;
3203         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
3204 
3205         trace_vtd_ir_remap_type("IOAPIC");
3206         /* IOAPIC entry vector should be aligned with IRTE vector
3207          * (see vt-d spec 5.1.5.1). */
3208         if (vector != irq.vector) {
3209             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
3210         }
3211 
3212         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3213          * (see vt-d spec 5.1.5.1). */
3214         if (trigger_mode != irq.trigger_mode) {
3215             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
3216                                       irq.trigger_mode);
3217         }
3218     }
3219 
3220     /*
3221      * We'd better keep the last two bits, assuming that guest OS
3222      * might modify it. Keep it does not hurt after all.
3223      */
3224     irq.msi_addr_last_bits = addr.addr.__not_care;
3225 
3226     /* Translate X86IOMMUIrq to MSI message */
3227     x86_iommu_irq_to_msi_message(&irq, translated);
3228 
3229 out:
3230     trace_vtd_ir_remap_msi(origin->address, origin->data,
3231                            translated->address, translated->data);
3232     return 0;
3233 }
3234 
3235 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
3236                          MSIMessage *dst, uint16_t sid)
3237 {
3238     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
3239                                    src, dst, sid);
3240 }
3241 
3242 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
3243                                    uint64_t *data, unsigned size,
3244                                    MemTxAttrs attrs)
3245 {
3246     return MEMTX_OK;
3247 }
3248 
3249 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
3250                                     uint64_t value, unsigned size,
3251                                     MemTxAttrs attrs)
3252 {
3253     int ret = 0;
3254     MSIMessage from = {}, to = {};
3255     uint16_t sid = X86_IOMMU_SID_INVALID;
3256 
3257     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
3258     from.data = (uint32_t) value;
3259 
3260     if (!attrs.unspecified) {
3261         /* We have explicit Source ID */
3262         sid = attrs.requester_id;
3263     }
3264 
3265     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid);
3266     if (ret) {
3267         /* TODO: report error */
3268         /* Drop this interrupt */
3269         return MEMTX_ERROR;
3270     }
3271 
3272     apic_get_class()->send_msi(&to);
3273 
3274     return MEMTX_OK;
3275 }
3276 
3277 static const MemoryRegionOps vtd_mem_ir_ops = {
3278     .read_with_attrs = vtd_mem_ir_read,
3279     .write_with_attrs = vtd_mem_ir_write,
3280     .endianness = DEVICE_LITTLE_ENDIAN,
3281     .impl = {
3282         .min_access_size = 4,
3283         .max_access_size = 4,
3284     },
3285     .valid = {
3286         .min_access_size = 4,
3287         .max_access_size = 4,
3288     },
3289 };
3290 
3291 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
3292 {
3293     uintptr_t key = (uintptr_t)bus;
3294     VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
3295     VTDAddressSpace *vtd_dev_as;
3296     char name[128];
3297 
3298     if (!vtd_bus) {
3299         uintptr_t *new_key = g_malloc(sizeof(*new_key));
3300         *new_key = (uintptr_t)bus;
3301         /* No corresponding free() */
3302         vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
3303                             PCI_DEVFN_MAX);
3304         vtd_bus->bus = bus;
3305         g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
3306     }
3307 
3308     vtd_dev_as = vtd_bus->dev_as[devfn];
3309 
3310     if (!vtd_dev_as) {
3311         snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn),
3312                  PCI_FUNC(devfn));
3313         vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace));
3314 
3315         vtd_dev_as->bus = bus;
3316         vtd_dev_as->devfn = (uint8_t)devfn;
3317         vtd_dev_as->iommu_state = s;
3318         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
3319         vtd_dev_as->iova_tree = iova_tree_new();
3320 
3321         memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX);
3322         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root");
3323 
3324         /*
3325          * Build the DMAR-disabled container with aliases to the
3326          * shared MRs.  Note that aliasing to a shared memory region
3327          * could help the memory API to detect same FlatViews so we
3328          * can have devices to share the same FlatView when DMAR is
3329          * disabled (either by not providing "intel_iommu=on" or with
3330          * "iommu=pt").  It will greatly reduce the total number of
3331          * FlatViews of the system hence VM runs faster.
3332          */
3333         memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s),
3334                                  "vtd-nodmar", &s->mr_nodmar, 0,
3335                                  memory_region_size(&s->mr_nodmar));
3336 
3337         /*
3338          * Build the per-device DMAR-enabled container.
3339          *
3340          * TODO: currently we have per-device IOMMU memory region only
3341          * because we have per-device IOMMU notifiers for devices.  If
3342          * one day we can abstract the IOMMU notifiers out of the
3343          * memory regions then we can also share the same memory
3344          * region here just like what we've done above with the nodmar
3345          * region.
3346          */
3347         strcat(name, "-dmar");
3348         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
3349                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
3350                                  name, UINT64_MAX);
3351         memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir",
3352                                  &s->mr_ir, 0, memory_region_size(&s->mr_ir));
3353         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
3354                                             VTD_INTERRUPT_ADDR_FIRST,
3355                                             &vtd_dev_as->iommu_ir, 1);
3356 
3357         /*
3358          * Hook both the containers under the root container, we
3359          * switch between DMAR & noDMAR by enable/disable
3360          * corresponding sub-containers
3361          */
3362         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3363                                             MEMORY_REGION(&vtd_dev_as->iommu),
3364                                             0);
3365         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3366                                             &vtd_dev_as->nodmar, 0);
3367 
3368         vtd_switch_address_space(vtd_dev_as);
3369     }
3370     return vtd_dev_as;
3371 }
3372 
3373 static uint64_t get_naturally_aligned_size(uint64_t start,
3374                                            uint64_t size, int gaw)
3375 {
3376     uint64_t max_mask = 1ULL << gaw;
3377     uint64_t alignment = start ? start & -start : max_mask;
3378 
3379     alignment = MIN(alignment, max_mask);
3380     size = MIN(size, max_mask);
3381 
3382     if (alignment <= size) {
3383         /* Increase the alignment of start */
3384         return alignment;
3385     } else {
3386         /* Find the largest page mask from size */
3387         return 1ULL << (63 - clz64(size));
3388     }
3389 }
3390 
3391 /* Unmap the whole range in the notifier's scope. */
3392 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
3393 {
3394     hwaddr size, remain;
3395     hwaddr start = n->start;
3396     hwaddr end = n->end;
3397     IntelIOMMUState *s = as->iommu_state;
3398     DMAMap map;
3399 
3400     /*
3401      * Note: all the codes in this function has a assumption that IOVA
3402      * bits are no more than VTD_MGAW bits (which is restricted by
3403      * VT-d spec), otherwise we need to consider overflow of 64 bits.
3404      */
3405 
3406     if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) {
3407         /*
3408          * Don't need to unmap regions that is bigger than the whole
3409          * VT-d supported address space size
3410          */
3411         end = VTD_ADDRESS_SIZE(s->aw_bits) - 1;
3412     }
3413 
3414     assert(start <= end);
3415     size = remain = end - start + 1;
3416 
3417     while (remain >= VTD_PAGE_SIZE) {
3418         IOMMUTLBEntry entry;
3419         uint64_t mask = get_naturally_aligned_size(start, remain, s->aw_bits);
3420 
3421         assert(mask);
3422 
3423         entry.iova = start;
3424         entry.addr_mask = mask - 1;
3425         entry.target_as = &address_space_memory;
3426         entry.perm = IOMMU_NONE;
3427         /* This field is meaningless for unmap */
3428         entry.translated_addr = 0;
3429 
3430         memory_region_notify_one(n, &entry);
3431 
3432         start += mask;
3433         remain -= mask;
3434     }
3435 
3436     assert(!remain);
3437 
3438     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3439                              VTD_PCI_SLOT(as->devfn),
3440                              VTD_PCI_FUNC(as->devfn),
3441                              n->start, size);
3442 
3443     map.iova = n->start;
3444     map.size = size;
3445     iova_tree_remove(as->iova_tree, &map);
3446 }
3447 
3448 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3449 {
3450     VTDAddressSpace *vtd_as;
3451     IOMMUNotifier *n;
3452 
3453     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3454         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3455             vtd_address_space_unmap(vtd_as, n);
3456         }
3457     }
3458 }
3459 
3460 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3461 {
3462     vtd_address_space_unmap_all(s);
3463     vtd_switch_address_space_all(s);
3464 }
3465 
3466 static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private)
3467 {
3468     memory_region_notify_one((IOMMUNotifier *)private, entry);
3469     return 0;
3470 }
3471 
3472 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3473 {
3474     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3475     IntelIOMMUState *s = vtd_as->iommu_state;
3476     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3477     VTDContextEntry ce;
3478 
3479     /*
3480      * The replay can be triggered by either a invalidation or a newly
3481      * created entry. No matter what, we release existing mappings
3482      * (it means flushing caches for UNMAP-only registers).
3483      */
3484     vtd_address_space_unmap(vtd_as, n);
3485 
3486     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3487         trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" :
3488                                   "legacy mode",
3489                                   bus_n, PCI_SLOT(vtd_as->devfn),
3490                                   PCI_FUNC(vtd_as->devfn),
3491                                   vtd_get_domain_id(s, &ce),
3492                                   ce.hi, ce.lo);
3493         if (vtd_as_has_map_notifier(vtd_as)) {
3494             /* This is required only for MAP typed notifiers */
3495             vtd_page_walk_info info = {
3496                 .hook_fn = vtd_replay_hook,
3497                 .private = (void *)n,
3498                 .notify_unmap = false,
3499                 .aw = s->aw_bits,
3500                 .as = vtd_as,
3501                 .domain_id = vtd_get_domain_id(s, &ce),
3502             };
3503 
3504             vtd_page_walk(s, &ce, 0, ~0ULL, &info);
3505         }
3506     } else {
3507         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3508                                     PCI_FUNC(vtd_as->devfn));
3509     }
3510 
3511     return;
3512 }
3513 
3514 /* Do the initialization. It will also be called when reset, so pay
3515  * attention when adding new initialization stuff.
3516  */
3517 static void vtd_init(IntelIOMMUState *s)
3518 {
3519     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3520 
3521     memset(s->csr, 0, DMAR_REG_SIZE);
3522     memset(s->wmask, 0, DMAR_REG_SIZE);
3523     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3524     memset(s->womask, 0, DMAR_REG_SIZE);
3525 
3526     s->root = 0;
3527     s->root_scalable = false;
3528     s->dmar_enabled = false;
3529     s->intr_enabled = false;
3530     s->iq_head = 0;
3531     s->iq_tail = 0;
3532     s->iq = 0;
3533     s->iq_size = 0;
3534     s->qi_enabled = false;
3535     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3536     s->iq_dw = false;
3537     s->next_frcd_reg = 0;
3538     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3539              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3540              VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits);
3541     if (s->dma_drain) {
3542         s->cap |= VTD_CAP_DRAIN;
3543     }
3544     if (s->aw_bits == VTD_HOST_AW_48BIT) {
3545         s->cap |= VTD_CAP_SAGAW_48bit;
3546     }
3547     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3548 
3549     /*
3550      * Rsvd field masks for spte
3551      */
3552     vtd_paging_entry_rsvd_field[0] = ~0ULL;
3553     vtd_paging_entry_rsvd_field[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits);
3554     vtd_paging_entry_rsvd_field[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3555     vtd_paging_entry_rsvd_field[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3556     vtd_paging_entry_rsvd_field[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3557     vtd_paging_entry_rsvd_field[5] = VTD_SPTE_LPAGE_L1_RSVD_MASK(s->aw_bits);
3558     vtd_paging_entry_rsvd_field[6] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits);
3559     vtd_paging_entry_rsvd_field[7] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits);
3560     vtd_paging_entry_rsvd_field[8] = VTD_SPTE_LPAGE_L4_RSVD_MASK(s->aw_bits);
3561 
3562     if (x86_iommu_ir_supported(x86_iommu)) {
3563         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3564         if (s->intr_eim == ON_OFF_AUTO_ON) {
3565             s->ecap |= VTD_ECAP_EIM;
3566         }
3567         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3568     }
3569 
3570     if (x86_iommu->dt_supported) {
3571         s->ecap |= VTD_ECAP_DT;
3572     }
3573 
3574     if (x86_iommu->pt_supported) {
3575         s->ecap |= VTD_ECAP_PT;
3576     }
3577 
3578     if (s->caching_mode) {
3579         s->cap |= VTD_CAP_CM;
3580     }
3581 
3582     /* TODO: read cap/ecap from host to decide which cap to be exposed. */
3583     if (s->scalable_mode) {
3584         s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS;
3585     }
3586 
3587     vtd_reset_caches(s);
3588 
3589     /* Define registers with default values and bit semantics */
3590     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
3591     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
3592     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
3593     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
3594     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
3595     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
3596     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0);
3597     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
3598     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
3599 
3600     /* Advanced Fault Logging not supported */
3601     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
3602     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3603     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
3604     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
3605 
3606     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3607      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3608      */
3609     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
3610 
3611     /* Treated as RO for implementations that PLMR and PHMR fields reported
3612      * as Clear in the CAP_REG.
3613      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3614      */
3615     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
3616 
3617     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
3618     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
3619     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0);
3620     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
3621     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3622     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
3623     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
3624     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3625     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
3626 
3627     /* IOTLB registers */
3628     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
3629     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
3630     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
3631 
3632     /* Fault Recording Registers, 128-bit */
3633     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
3634     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
3635 
3636     /*
3637      * Interrupt remapping registers.
3638      */
3639     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
3640 }
3641 
3642 /* Should not reset address_spaces when reset because devices will still use
3643  * the address space they got at first (won't ask the bus again).
3644  */
3645 static void vtd_reset(DeviceState *dev)
3646 {
3647     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3648 
3649     vtd_init(s);
3650     vtd_address_space_refresh_all(s);
3651 }
3652 
3653 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
3654 {
3655     IntelIOMMUState *s = opaque;
3656     VTDAddressSpace *vtd_as;
3657 
3658     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
3659 
3660     vtd_as = vtd_find_add_as(s, bus, devfn);
3661     return &vtd_as->as;
3662 }
3663 
3664 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
3665 {
3666     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3667 
3668     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) {
3669         error_setg(errp, "eim=on cannot be selected without intremap=on");
3670         return false;
3671     }
3672 
3673     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
3674         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
3675                       && x86_iommu_ir_supported(x86_iommu) ?
3676                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
3677     }
3678     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
3679         if (!kvm_irqchip_in_kernel()) {
3680             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
3681             return false;
3682         }
3683         if (!kvm_enable_x2apic()) {
3684             error_setg(errp, "eim=on requires support on the KVM side"
3685                              "(X2APIC_API, first shipped in v4.7)");
3686             return false;
3687         }
3688     }
3689 
3690     /* Currently only address widths supported are 39 and 48 bits */
3691     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
3692         (s->aw_bits != VTD_HOST_AW_48BIT)) {
3693         error_setg(errp, "Supported values for x-aw-bits are: %d, %d",
3694                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
3695         return false;
3696     }
3697 
3698     if (s->scalable_mode && !s->dma_drain) {
3699         error_setg(errp, "Need to set dma_drain for scalable mode");
3700         return false;
3701     }
3702 
3703     return true;
3704 }
3705 
3706 static int vtd_machine_done_notify_one(Object *child, void *unused)
3707 {
3708     IntelIOMMUState *iommu = INTEL_IOMMU_DEVICE(x86_iommu_get_default());
3709 
3710     /*
3711      * We hard-coded here because vfio-pci is the only special case
3712      * here.  Let's be more elegant in the future when we can, but so
3713      * far there seems to be no better way.
3714      */
3715     if (object_dynamic_cast(child, "vfio-pci") && !iommu->caching_mode) {
3716         vtd_panic_require_caching_mode();
3717     }
3718 
3719     return 0;
3720 }
3721 
3722 static void vtd_machine_done_hook(Notifier *notifier, void *unused)
3723 {
3724     object_child_foreach_recursive(object_get_root(),
3725                                    vtd_machine_done_notify_one, NULL);
3726 }
3727 
3728 static Notifier vtd_machine_done_notify = {
3729     .notify = vtd_machine_done_hook,
3730 };
3731 
3732 static void vtd_realize(DeviceState *dev, Error **errp)
3733 {
3734     MachineState *ms = MACHINE(qdev_get_machine());
3735     PCMachineState *pcms = PC_MACHINE(ms);
3736     PCIBus *bus = pcms->bus;
3737     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3738     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
3739 
3740     x86_iommu->type = TYPE_INTEL;
3741 
3742     if (!vtd_decide_config(s, errp)) {
3743         return;
3744     }
3745 
3746     QLIST_INIT(&s->vtd_as_with_notifiers);
3747     qemu_mutex_init(&s->iommu_lock);
3748     memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num));
3749     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
3750                           "intel_iommu", DMAR_REG_SIZE);
3751 
3752     /* Create the shared memory regions by all devices */
3753     memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar",
3754                        UINT64_MAX);
3755     memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops,
3756                           s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE);
3757     memory_region_init_alias(&s->mr_sys_alias, OBJECT(s),
3758                              "vtd-sys-alias", get_system_memory(), 0,
3759                              memory_region_size(get_system_memory()));
3760     memory_region_add_subregion_overlap(&s->mr_nodmar, 0,
3761                                         &s->mr_sys_alias, 0);
3762     memory_region_add_subregion_overlap(&s->mr_nodmar,
3763                                         VTD_INTERRUPT_ADDR_FIRST,
3764                                         &s->mr_ir, 1);
3765 
3766     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
3767     /* No corresponding destroy */
3768     s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3769                                      g_free, g_free);
3770     s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3771                                               g_free, g_free);
3772     vtd_init(s);
3773     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
3774     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
3775     /* Pseudo address space under root PCI bus. */
3776     pcms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
3777     qemu_add_machine_init_done_notifier(&vtd_machine_done_notify);
3778 }
3779 
3780 static void vtd_class_init(ObjectClass *klass, void *data)
3781 {
3782     DeviceClass *dc = DEVICE_CLASS(klass);
3783     X86IOMMUClass *x86_class = X86_IOMMU_CLASS(klass);
3784 
3785     dc->reset = vtd_reset;
3786     dc->vmsd = &vtd_vmstate;
3787     dc->props = vtd_properties;
3788     dc->hotpluggable = false;
3789     x86_class->realize = vtd_realize;
3790     x86_class->int_remap = vtd_int_remap;
3791     /* Supported by the pc-q35-* machine types */
3792     dc->user_creatable = true;
3793     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
3794     dc->desc = "Intel IOMMU (VT-d) DMA Remapping device";
3795 }
3796 
3797 static const TypeInfo vtd_info = {
3798     .name          = TYPE_INTEL_IOMMU_DEVICE,
3799     .parent        = TYPE_X86_IOMMU_DEVICE,
3800     .instance_size = sizeof(IntelIOMMUState),
3801     .class_init    = vtd_class_init,
3802 };
3803 
3804 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
3805                                                      void *data)
3806 {
3807     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
3808 
3809     imrc->translate = vtd_iommu_translate;
3810     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
3811     imrc->replay = vtd_iommu_replay;
3812 }
3813 
3814 static const TypeInfo vtd_iommu_memory_region_info = {
3815     .parent = TYPE_IOMMU_MEMORY_REGION,
3816     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
3817     .class_init = vtd_iommu_memory_region_class_init,
3818 };
3819 
3820 static void vtd_register_types(void)
3821 {
3822     type_register_static(&vtd_info);
3823     type_register_static(&vtd_iommu_memory_region_info);
3824 }
3825 
3826 type_init(vtd_register_types)
3827