xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision 284d697c)
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 inline bool vtd_pdire_present(VTDPASIDDirEntry *pdire)
690 {
691     return pdire->val & 1;
692 }
693 
694 /**
695  * Caller of this function should check present bit if wants
696  * to use pdir entry for futher usage except for fpd bit check.
697  */
698 static int vtd_get_pdire_from_pdir_table(dma_addr_t pasid_dir_base,
699                                          uint32_t pasid,
700                                          VTDPASIDDirEntry *pdire)
701 {
702     uint32_t index;
703     dma_addr_t addr, entry_size;
704 
705     index = VTD_PASID_DIR_INDEX(pasid);
706     entry_size = VTD_PASID_DIR_ENTRY_SIZE;
707     addr = pasid_dir_base + index * entry_size;
708     if (dma_memory_read(&address_space_memory, addr, pdire, entry_size)) {
709         return -VTD_FR_PASID_TABLE_INV;
710     }
711 
712     return 0;
713 }
714 
715 static inline bool vtd_pe_present(VTDPASIDEntry *pe)
716 {
717     return pe->val[0] & VTD_PASID_ENTRY_P;
718 }
719 
720 static int vtd_get_pe_in_pasid_leaf_table(IntelIOMMUState *s,
721                                           uint32_t pasid,
722                                           dma_addr_t addr,
723                                           VTDPASIDEntry *pe)
724 {
725     uint32_t index;
726     dma_addr_t entry_size;
727     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
728 
729     index = VTD_PASID_TABLE_INDEX(pasid);
730     entry_size = VTD_PASID_ENTRY_SIZE;
731     addr = addr + index * entry_size;
732     if (dma_memory_read(&address_space_memory, addr, pe, entry_size)) {
733         return -VTD_FR_PASID_TABLE_INV;
734     }
735 
736     /* Do translation type check */
737     if (!vtd_pe_type_check(x86_iommu, pe)) {
738         return -VTD_FR_PASID_TABLE_INV;
739     }
740 
741     if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) {
742         return -VTD_FR_PASID_TABLE_INV;
743     }
744 
745     return 0;
746 }
747 
748 /**
749  * Caller of this function should check present bit if wants
750  * to use pasid entry for futher usage except for fpd bit check.
751  */
752 static int vtd_get_pe_from_pdire(IntelIOMMUState *s,
753                                  uint32_t pasid,
754                                  VTDPASIDDirEntry *pdire,
755                                  VTDPASIDEntry *pe)
756 {
757     dma_addr_t addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK;
758 
759     return vtd_get_pe_in_pasid_leaf_table(s, pasid, addr, pe);
760 }
761 
762 /**
763  * This function gets a pasid entry from a specified pasid
764  * table (includes dir and leaf table) with a specified pasid.
765  * Sanity check should be done to ensure return a present
766  * pasid entry to caller.
767  */
768 static int vtd_get_pe_from_pasid_table(IntelIOMMUState *s,
769                                        dma_addr_t pasid_dir_base,
770                                        uint32_t pasid,
771                                        VTDPASIDEntry *pe)
772 {
773     int ret;
774     VTDPASIDDirEntry pdire;
775 
776     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base,
777                                         pasid, &pdire);
778     if (ret) {
779         return ret;
780     }
781 
782     if (!vtd_pdire_present(&pdire)) {
783         return -VTD_FR_PASID_TABLE_INV;
784     }
785 
786     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, pe);
787     if (ret) {
788         return ret;
789     }
790 
791     if (!vtd_pe_present(pe)) {
792         return -VTD_FR_PASID_TABLE_INV;
793     }
794 
795     return 0;
796 }
797 
798 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s,
799                                       VTDContextEntry *ce,
800                                       VTDPASIDEntry *pe)
801 {
802     uint32_t pasid;
803     dma_addr_t pasid_dir_base;
804     int ret = 0;
805 
806     pasid = VTD_CE_GET_RID2PASID(ce);
807     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
808     ret = vtd_get_pe_from_pasid_table(s, pasid_dir_base, pasid, pe);
809 
810     return ret;
811 }
812 
813 static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s,
814                                 VTDContextEntry *ce,
815                                 bool *pe_fpd_set)
816 {
817     int ret;
818     uint32_t pasid;
819     dma_addr_t pasid_dir_base;
820     VTDPASIDDirEntry pdire;
821     VTDPASIDEntry pe;
822 
823     pasid = VTD_CE_GET_RID2PASID(ce);
824     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
825 
826     /*
827      * No present bit check since fpd is meaningful even
828      * if the present bit is clear.
829      */
830     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base, pasid, &pdire);
831     if (ret) {
832         return ret;
833     }
834 
835     if (pdire.val & VTD_PASID_DIR_FPD) {
836         *pe_fpd_set = true;
837         return 0;
838     }
839 
840     if (!vtd_pdire_present(&pdire)) {
841         return -VTD_FR_PASID_TABLE_INV;
842     }
843 
844     /*
845      * No present bit check since fpd is meaningful even
846      * if the present bit is clear.
847      */
848     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, &pe);
849     if (ret) {
850         return ret;
851     }
852 
853     if (pe.val[0] & VTD_PASID_ENTRY_FPD) {
854         *pe_fpd_set = true;
855     }
856 
857     return 0;
858 }
859 
860 /* Get the page-table level that hardware should use for the second-level
861  * page-table walk from the Address Width field of context-entry.
862  */
863 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
864 {
865     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
866 }
867 
868 static uint32_t vtd_get_iova_level(IntelIOMMUState *s,
869                                    VTDContextEntry *ce)
870 {
871     VTDPASIDEntry pe;
872 
873     if (s->root_scalable) {
874         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
875         return VTD_PE_GET_LEVEL(&pe);
876     }
877 
878     return vtd_ce_get_level(ce);
879 }
880 
881 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
882 {
883     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
884 }
885 
886 static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s,
887                                   VTDContextEntry *ce)
888 {
889     VTDPASIDEntry pe;
890 
891     if (s->root_scalable) {
892         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
893         return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9;
894     }
895 
896     return vtd_ce_get_agaw(ce);
897 }
898 
899 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
900 {
901     return ce->lo & VTD_CONTEXT_ENTRY_TT;
902 }
903 
904 /* Only for Legacy Mode. Return true if check passed, otherwise false */
905 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
906                                      VTDContextEntry *ce)
907 {
908     switch (vtd_ce_get_type(ce)) {
909     case VTD_CONTEXT_TT_MULTI_LEVEL:
910         /* Always supported */
911         break;
912     case VTD_CONTEXT_TT_DEV_IOTLB:
913         if (!x86_iommu->dt_supported) {
914             error_report_once("%s: DT specified but not supported", __func__);
915             return false;
916         }
917         break;
918     case VTD_CONTEXT_TT_PASS_THROUGH:
919         if (!x86_iommu->pt_supported) {
920             error_report_once("%s: PT specified but not supported", __func__);
921             return false;
922         }
923         break;
924     default:
925         /* Unknown type */
926         error_report_once("%s: unknown ce type: %"PRIu32, __func__,
927                           vtd_ce_get_type(ce));
928         return false;
929     }
930     return true;
931 }
932 
933 static inline uint64_t vtd_iova_limit(IntelIOMMUState *s,
934                                       VTDContextEntry *ce, uint8_t aw)
935 {
936     uint32_t ce_agaw = vtd_get_iova_agaw(s, ce);
937     return 1ULL << MIN(ce_agaw, aw);
938 }
939 
940 /* Return true if IOVA passes range check, otherwise false. */
941 static inline bool vtd_iova_range_check(IntelIOMMUState *s,
942                                         uint64_t iova, VTDContextEntry *ce,
943                                         uint8_t aw)
944 {
945     /*
946      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
947      * in CAP_REG and AW in context-entry.
948      */
949     return !(iova & ~(vtd_iova_limit(s, ce, aw) - 1));
950 }
951 
952 static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s,
953                                           VTDContextEntry *ce)
954 {
955     VTDPASIDEntry pe;
956 
957     if (s->root_scalable) {
958         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
959         return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR;
960     }
961 
962     return vtd_ce_get_slpt_base(ce);
963 }
964 
965 /*
966  * Rsvd field masks for spte:
967  *     vtd_spte_rsvd 4k pages
968  *     vtd_spte_rsvd_large large pages
969  */
970 static uint64_t vtd_spte_rsvd[5];
971 static uint64_t vtd_spte_rsvd_large[5];
972 
973 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
974 {
975     uint64_t rsvd_mask = vtd_spte_rsvd[level];
976 
977     if ((level == VTD_SL_PD_LEVEL || level == VTD_SL_PDP_LEVEL) &&
978         (slpte & VTD_SL_PT_PAGE_SIZE_MASK)) {
979         /* large page */
980         rsvd_mask = vtd_spte_rsvd_large[level];
981     }
982 
983     return slpte & rsvd_mask;
984 }
985 
986 /* Find the VTD address space associated with a given bus number */
987 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num)
988 {
989     VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num];
990     GHashTableIter iter;
991 
992     if (vtd_bus) {
993         return vtd_bus;
994     }
995 
996     /*
997      * Iterate over the registered buses to find the one which
998      * currently holds this bus number and update the bus_num
999      * lookup table.
1000      */
1001     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1002     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1003         if (pci_bus_num(vtd_bus->bus) == bus_num) {
1004             s->vtd_as_by_bus_num[bus_num] = vtd_bus;
1005             return vtd_bus;
1006         }
1007     }
1008 
1009     return NULL;
1010 }
1011 
1012 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
1013  * of the translation, can be used for deciding the size of large page.
1014  */
1015 static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce,
1016                              uint64_t iova, bool is_write,
1017                              uint64_t *slptep, uint32_t *slpte_level,
1018                              bool *reads, bool *writes, uint8_t aw_bits)
1019 {
1020     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1021     uint32_t level = vtd_get_iova_level(s, ce);
1022     uint32_t offset;
1023     uint64_t slpte;
1024     uint64_t access_right_check;
1025 
1026     if (!vtd_iova_range_check(s, iova, ce, aw_bits)) {
1027         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")",
1028                           __func__, iova);
1029         return -VTD_FR_ADDR_BEYOND_MGAW;
1030     }
1031 
1032     /* FIXME: what is the Atomics request here? */
1033     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
1034 
1035     while (true) {
1036         offset = vtd_iova_level_offset(iova, level);
1037         slpte = vtd_get_slpte(addr, offset);
1038 
1039         if (slpte == (uint64_t)-1) {
1040             error_report_once("%s: detected read error on DMAR slpte "
1041                               "(iova=0x%" PRIx64 ")", __func__, iova);
1042             if (level == vtd_get_iova_level(s, ce)) {
1043                 /* Invalid programming of context-entry */
1044                 return -VTD_FR_CONTEXT_ENTRY_INV;
1045             } else {
1046                 return -VTD_FR_PAGING_ENTRY_INV;
1047             }
1048         }
1049         *reads = (*reads) && (slpte & VTD_SL_R);
1050         *writes = (*writes) && (slpte & VTD_SL_W);
1051         if (!(slpte & access_right_check)) {
1052             error_report_once("%s: detected slpte permission error "
1053                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
1054                               "slpte=0x%" PRIx64 ", write=%d)", __func__,
1055                               iova, level, slpte, is_write);
1056             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
1057         }
1058         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1059             error_report_once("%s: detected splte reserve non-zero "
1060                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
1061                               "slpte=0x%" PRIx64 ")", __func__, iova,
1062                               level, slpte);
1063             return -VTD_FR_PAGING_ENTRY_RSVD;
1064         }
1065 
1066         if (vtd_is_last_slpte(slpte, level)) {
1067             *slptep = slpte;
1068             *slpte_level = level;
1069             return 0;
1070         }
1071         addr = vtd_get_slpte_addr(slpte, aw_bits);
1072         level--;
1073     }
1074 }
1075 
1076 typedef int (*vtd_page_walk_hook)(IOMMUTLBEntry *entry, void *private);
1077 
1078 /**
1079  * Constant information used during page walking
1080  *
1081  * @hook_fn: hook func to be called when detected page
1082  * @private: private data to be passed into hook func
1083  * @notify_unmap: whether we should notify invalid entries
1084  * @as: VT-d address space of the device
1085  * @aw: maximum address width
1086  * @domain: domain ID of the page walk
1087  */
1088 typedef struct {
1089     VTDAddressSpace *as;
1090     vtd_page_walk_hook hook_fn;
1091     void *private;
1092     bool notify_unmap;
1093     uint8_t aw;
1094     uint16_t domain_id;
1095 } vtd_page_walk_info;
1096 
1097 static int vtd_page_walk_one(IOMMUTLBEntry *entry, vtd_page_walk_info *info)
1098 {
1099     VTDAddressSpace *as = info->as;
1100     vtd_page_walk_hook hook_fn = info->hook_fn;
1101     void *private = info->private;
1102     DMAMap target = {
1103         .iova = entry->iova,
1104         .size = entry->addr_mask,
1105         .translated_addr = entry->translated_addr,
1106         .perm = entry->perm,
1107     };
1108     DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
1109 
1110     if (entry->perm == IOMMU_NONE && !info->notify_unmap) {
1111         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1112         return 0;
1113     }
1114 
1115     assert(hook_fn);
1116 
1117     /* Update local IOVA mapped ranges */
1118     if (entry->perm) {
1119         if (mapped) {
1120             /* If it's exactly the same translation, skip */
1121             if (!memcmp(mapped, &target, sizeof(target))) {
1122                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
1123                                                  entry->translated_addr);
1124                 return 0;
1125             } else {
1126                 /*
1127                  * Translation changed.  Normally this should not
1128                  * happen, but it can happen when with buggy guest
1129                  * OSes.  Note that there will be a small window that
1130                  * we don't have map at all.  But that's the best
1131                  * effort we can do.  The ideal way to emulate this is
1132                  * atomically modify the PTE to follow what has
1133                  * changed, but we can't.  One example is that vfio
1134                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1135                  * interface to modify a mapping (meanwhile it seems
1136                  * meaningless to even provide one).  Anyway, let's
1137                  * mark this as a TODO in case one day we'll have
1138                  * a better solution.
1139                  */
1140                 IOMMUAccessFlags cache_perm = entry->perm;
1141                 int ret;
1142 
1143                 /* Emulate an UNMAP */
1144                 entry->perm = IOMMU_NONE;
1145                 trace_vtd_page_walk_one(info->domain_id,
1146                                         entry->iova,
1147                                         entry->translated_addr,
1148                                         entry->addr_mask,
1149                                         entry->perm);
1150                 ret = hook_fn(entry, private);
1151                 if (ret) {
1152                     return ret;
1153                 }
1154                 /* Drop any existing mapping */
1155                 iova_tree_remove(as->iova_tree, &target);
1156                 /* Recover the correct permission */
1157                 entry->perm = cache_perm;
1158             }
1159         }
1160         iova_tree_insert(as->iova_tree, &target);
1161     } else {
1162         if (!mapped) {
1163             /* Skip since we didn't map this range at all */
1164             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1165             return 0;
1166         }
1167         iova_tree_remove(as->iova_tree, &target);
1168     }
1169 
1170     trace_vtd_page_walk_one(info->domain_id, entry->iova,
1171                             entry->translated_addr, entry->addr_mask,
1172                             entry->perm);
1173     return hook_fn(entry, private);
1174 }
1175 
1176 /**
1177  * vtd_page_walk_level - walk over specific level for IOVA range
1178  *
1179  * @addr: base GPA addr to start the walk
1180  * @start: IOVA range start address
1181  * @end: IOVA range end address (start <= addr < end)
1182  * @read: whether parent level has read permission
1183  * @write: whether parent level has write permission
1184  * @info: constant information for the page walk
1185  */
1186 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
1187                                uint64_t end, uint32_t level, bool read,
1188                                bool write, vtd_page_walk_info *info)
1189 {
1190     bool read_cur, write_cur, entry_valid;
1191     uint32_t offset;
1192     uint64_t slpte;
1193     uint64_t subpage_size, subpage_mask;
1194     IOMMUTLBEntry entry;
1195     uint64_t iova = start;
1196     uint64_t iova_next;
1197     int ret = 0;
1198 
1199     trace_vtd_page_walk_level(addr, level, start, end);
1200 
1201     subpage_size = 1ULL << vtd_slpt_level_shift(level);
1202     subpage_mask = vtd_slpt_level_page_mask(level);
1203 
1204     while (iova < end) {
1205         iova_next = (iova & subpage_mask) + subpage_size;
1206 
1207         offset = vtd_iova_level_offset(iova, level);
1208         slpte = vtd_get_slpte(addr, offset);
1209 
1210         if (slpte == (uint64_t)-1) {
1211             trace_vtd_page_walk_skip_read(iova, iova_next);
1212             goto next;
1213         }
1214 
1215         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1216             trace_vtd_page_walk_skip_reserve(iova, iova_next);
1217             goto next;
1218         }
1219 
1220         /* Permissions are stacked with parents' */
1221         read_cur = read && (slpte & VTD_SL_R);
1222         write_cur = write && (slpte & VTD_SL_W);
1223 
1224         /*
1225          * As long as we have either read/write permission, this is a
1226          * valid entry. The rule works for both page entries and page
1227          * table entries.
1228          */
1229         entry_valid = read_cur | write_cur;
1230 
1231         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
1232             /*
1233              * This is a valid PDE (or even bigger than PDE).  We need
1234              * to walk one further level.
1235              */
1236             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
1237                                       iova, MIN(iova_next, end), level - 1,
1238                                       read_cur, write_cur, info);
1239         } else {
1240             /*
1241              * This means we are either:
1242              *
1243              * (1) the real page entry (either 4K page, or huge page)
1244              * (2) the whole range is invalid
1245              *
1246              * In either case, we send an IOTLB notification down.
1247              */
1248             entry.target_as = &address_space_memory;
1249             entry.iova = iova & subpage_mask;
1250             entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
1251             entry.addr_mask = ~subpage_mask;
1252             /* NOTE: this is only meaningful if entry_valid == true */
1253             entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
1254             ret = vtd_page_walk_one(&entry, info);
1255         }
1256 
1257         if (ret < 0) {
1258             return ret;
1259         }
1260 
1261 next:
1262         iova = iova_next;
1263     }
1264 
1265     return 0;
1266 }
1267 
1268 /**
1269  * vtd_page_walk - walk specific IOVA range, and call the hook
1270  *
1271  * @s: intel iommu state
1272  * @ce: context entry to walk upon
1273  * @start: IOVA address to start the walk
1274  * @end: IOVA range end address (start <= addr < end)
1275  * @info: page walking information struct
1276  */
1277 static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce,
1278                          uint64_t start, uint64_t end,
1279                          vtd_page_walk_info *info)
1280 {
1281     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1282     uint32_t level = vtd_get_iova_level(s, ce);
1283 
1284     if (!vtd_iova_range_check(s, start, ce, info->aw)) {
1285         return -VTD_FR_ADDR_BEYOND_MGAW;
1286     }
1287 
1288     if (!vtd_iova_range_check(s, end, ce, info->aw)) {
1289         /* Fix end so that it reaches the maximum */
1290         end = vtd_iova_limit(s, ce, info->aw);
1291     }
1292 
1293     return vtd_page_walk_level(addr, start, end, level, true, true, info);
1294 }
1295 
1296 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s,
1297                                           VTDRootEntry *re)
1298 {
1299     /* Legacy Mode reserved bits check */
1300     if (!s->root_scalable &&
1301         (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1302         goto rsvd_err;
1303 
1304     /* Scalable Mode reserved bits check */
1305     if (s->root_scalable &&
1306         ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) ||
1307          (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1308         goto rsvd_err;
1309 
1310     return 0;
1311 
1312 rsvd_err:
1313     error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1314                       ", lo=0x%"PRIx64,
1315                       __func__, re->hi, re->lo);
1316     return -VTD_FR_ROOT_ENTRY_RSVD;
1317 }
1318 
1319 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s,
1320                                                     VTDContextEntry *ce)
1321 {
1322     if (!s->root_scalable &&
1323         (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI ||
1324          ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1325         error_report_once("%s: invalid context entry: hi=%"PRIx64
1326                           ", lo=%"PRIx64" (reserved nonzero)",
1327                           __func__, ce->hi, ce->lo);
1328         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1329     }
1330 
1331     if (s->root_scalable &&
1332         (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) ||
1333          ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 ||
1334          ce->val[2] ||
1335          ce->val[3])) {
1336         error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1337                           ", val[2]=%"PRIx64
1338                           ", val[1]=%"PRIx64
1339                           ", val[0]=%"PRIx64" (reserved nonzero)",
1340                           __func__, ce->val[3], ce->val[2],
1341                           ce->val[1], ce->val[0]);
1342         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1343     }
1344 
1345     return 0;
1346 }
1347 
1348 static int vtd_ce_rid2pasid_check(IntelIOMMUState *s,
1349                                   VTDContextEntry *ce)
1350 {
1351     VTDPASIDEntry pe;
1352 
1353     /*
1354      * Make sure in Scalable Mode, a present context entry
1355      * has valid rid2pasid setting, which includes valid
1356      * rid2pasid field and corresponding pasid entry setting
1357      */
1358     return vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1359 }
1360 
1361 /* Map a device to its corresponding domain (context-entry) */
1362 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
1363                                     uint8_t devfn, VTDContextEntry *ce)
1364 {
1365     VTDRootEntry re;
1366     int ret_fr;
1367     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
1368 
1369     ret_fr = vtd_get_root_entry(s, bus_num, &re);
1370     if (ret_fr) {
1371         return ret_fr;
1372     }
1373 
1374     if (!vtd_root_entry_present(s, &re, devfn)) {
1375         /* Not error - it's okay we don't have root entry. */
1376         trace_vtd_re_not_present(bus_num);
1377         return -VTD_FR_ROOT_ENTRY_P;
1378     }
1379 
1380     ret_fr = vtd_root_entry_rsvd_bits_check(s, &re);
1381     if (ret_fr) {
1382         return ret_fr;
1383     }
1384 
1385     ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce);
1386     if (ret_fr) {
1387         return ret_fr;
1388     }
1389 
1390     if (!vtd_ce_present(ce)) {
1391         /* Not error - it's okay we don't have context entry. */
1392         trace_vtd_ce_not_present(bus_num, devfn);
1393         return -VTD_FR_CONTEXT_ENTRY_P;
1394     }
1395 
1396     ret_fr = vtd_context_entry_rsvd_bits_check(s, ce);
1397     if (ret_fr) {
1398         return ret_fr;
1399     }
1400 
1401     /* Check if the programming of context-entry is valid */
1402     if (!s->root_scalable &&
1403         !vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1404         error_report_once("%s: invalid context entry: hi=%"PRIx64
1405                           ", lo=%"PRIx64" (level %d not supported)",
1406                           __func__, ce->hi, ce->lo,
1407                           vtd_ce_get_level(ce));
1408         return -VTD_FR_CONTEXT_ENTRY_INV;
1409     }
1410 
1411     if (!s->root_scalable) {
1412         /* Do translation type check */
1413         if (!vtd_ce_type_check(x86_iommu, ce)) {
1414             /* Errors dumped in vtd_ce_type_check() */
1415             return -VTD_FR_CONTEXT_ENTRY_INV;
1416         }
1417     } else {
1418         /*
1419          * Check if the programming of context-entry.rid2pasid
1420          * and corresponding pasid setting is valid, and thus
1421          * avoids to check pasid entry fetching result in future
1422          * helper function calling.
1423          */
1424         ret_fr = vtd_ce_rid2pasid_check(s, ce);
1425         if (ret_fr) {
1426             return ret_fr;
1427         }
1428     }
1429 
1430     return 0;
1431 }
1432 
1433 static int vtd_sync_shadow_page_hook(IOMMUTLBEntry *entry,
1434                                      void *private)
1435 {
1436     memory_region_notify_iommu((IOMMUMemoryRegion *)private, 0, *entry);
1437     return 0;
1438 }
1439 
1440 static uint16_t vtd_get_domain_id(IntelIOMMUState *s,
1441                                   VTDContextEntry *ce)
1442 {
1443     VTDPASIDEntry pe;
1444 
1445     if (s->root_scalable) {
1446         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1447         return VTD_SM_PASID_ENTRY_DID(pe.val[1]);
1448     }
1449 
1450     return VTD_CONTEXT_ENTRY_DID(ce->hi);
1451 }
1452 
1453 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1454                                             VTDContextEntry *ce,
1455                                             hwaddr addr, hwaddr size)
1456 {
1457     IntelIOMMUState *s = vtd_as->iommu_state;
1458     vtd_page_walk_info info = {
1459         .hook_fn = vtd_sync_shadow_page_hook,
1460         .private = (void *)&vtd_as->iommu,
1461         .notify_unmap = true,
1462         .aw = s->aw_bits,
1463         .as = vtd_as,
1464         .domain_id = vtd_get_domain_id(s, ce),
1465     };
1466 
1467     return vtd_page_walk(s, ce, addr, addr + size, &info);
1468 }
1469 
1470 static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
1471 {
1472     int ret;
1473     VTDContextEntry ce;
1474     IOMMUNotifier *n;
1475 
1476     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1477                                    pci_bus_num(vtd_as->bus),
1478                                    vtd_as->devfn, &ce);
1479     if (ret) {
1480         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1481             /*
1482              * It's a valid scenario to have a context entry that is
1483              * not present.  For example, when a device is removed
1484              * from an existing domain then the context entry will be
1485              * zeroed by the guest before it was put into another
1486              * domain.  When this happens, instead of synchronizing
1487              * the shadow pages we should invalidate all existing
1488              * mappings and notify the backends.
1489              */
1490             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1491                 vtd_address_space_unmap(vtd_as, n);
1492             }
1493             ret = 0;
1494         }
1495         return ret;
1496     }
1497 
1498     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1499 }
1500 
1501 /*
1502  * Check if specific device is configed to bypass address
1503  * translation for DMA requests. In Scalable Mode, bypass
1504  * 1st-level translation or 2nd-level translation, it depends
1505  * on PGTT setting.
1506  */
1507 static bool vtd_dev_pt_enabled(VTDAddressSpace *as)
1508 {
1509     IntelIOMMUState *s;
1510     VTDContextEntry ce;
1511     VTDPASIDEntry pe;
1512     int ret;
1513 
1514     assert(as);
1515 
1516     s = as->iommu_state;
1517     ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus),
1518                                    as->devfn, &ce);
1519     if (ret) {
1520         /*
1521          * Possibly failed to parse the context entry for some reason
1522          * (e.g., during init, or any guest configuration errors on
1523          * context entries). We should assume PT not enabled for
1524          * safety.
1525          */
1526         return false;
1527     }
1528 
1529     if (s->root_scalable) {
1530         ret = vtd_ce_get_rid2pasid_entry(s, &ce, &pe);
1531         if (ret) {
1532             error_report_once("%s: vtd_ce_get_rid2pasid_entry error: %"PRId32,
1533                               __func__, ret);
1534             return false;
1535         }
1536         return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT);
1537     }
1538 
1539     return (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH);
1540 }
1541 
1542 /* Return whether the device is using IOMMU translation. */
1543 static bool vtd_switch_address_space(VTDAddressSpace *as)
1544 {
1545     bool use_iommu;
1546     /* Whether we need to take the BQL on our own */
1547     bool take_bql = !qemu_mutex_iothread_locked();
1548 
1549     assert(as);
1550 
1551     use_iommu = as->iommu_state->dmar_enabled && !vtd_dev_pt_enabled(as);
1552 
1553     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1554                                    VTD_PCI_SLOT(as->devfn),
1555                                    VTD_PCI_FUNC(as->devfn),
1556                                    use_iommu);
1557 
1558     /*
1559      * It's possible that we reach here without BQL, e.g., when called
1560      * from vtd_pt_enable_fast_path(). However the memory APIs need
1561      * it. We'd better make sure we have had it already, or, take it.
1562      */
1563     if (take_bql) {
1564         qemu_mutex_lock_iothread();
1565     }
1566 
1567     /* Turn off first then on the other */
1568     if (use_iommu) {
1569         memory_region_set_enabled(&as->nodmar, false);
1570         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1571     } else {
1572         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1573         memory_region_set_enabled(&as->nodmar, true);
1574     }
1575 
1576     if (take_bql) {
1577         qemu_mutex_unlock_iothread();
1578     }
1579 
1580     return use_iommu;
1581 }
1582 
1583 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1584 {
1585     GHashTableIter iter;
1586     VTDBus *vtd_bus;
1587     int i;
1588 
1589     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1590     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1591         for (i = 0; i < PCI_DEVFN_MAX; i++) {
1592             if (!vtd_bus->dev_as[i]) {
1593                 continue;
1594             }
1595             vtd_switch_address_space(vtd_bus->dev_as[i]);
1596         }
1597     }
1598 }
1599 
1600 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn)
1601 {
1602     return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
1603 }
1604 
1605 static const bool vtd_qualified_faults[] = {
1606     [VTD_FR_RESERVED] = false,
1607     [VTD_FR_ROOT_ENTRY_P] = false,
1608     [VTD_FR_CONTEXT_ENTRY_P] = true,
1609     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1610     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1611     [VTD_FR_WRITE] = true,
1612     [VTD_FR_READ] = true,
1613     [VTD_FR_PAGING_ENTRY_INV] = true,
1614     [VTD_FR_ROOT_TABLE_INV] = false,
1615     [VTD_FR_CONTEXT_TABLE_INV] = false,
1616     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1617     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1618     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1619     [VTD_FR_PASID_TABLE_INV] = false,
1620     [VTD_FR_RESERVED_ERR] = false,
1621     [VTD_FR_MAX] = false,
1622 };
1623 
1624 /* To see if a fault condition is "qualified", which is reported to software
1625  * only if the FPD field in the context-entry used to process the faulting
1626  * request is 0.
1627  */
1628 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1629 {
1630     return vtd_qualified_faults[fault];
1631 }
1632 
1633 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1634 {
1635     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1636 }
1637 
1638 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1639 {
1640     VTDBus *vtd_bus;
1641     VTDAddressSpace *vtd_as;
1642     bool success = false;
1643 
1644     vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id));
1645     if (!vtd_bus) {
1646         goto out;
1647     }
1648 
1649     vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)];
1650     if (!vtd_as) {
1651         goto out;
1652     }
1653 
1654     if (vtd_switch_address_space(vtd_as) == false) {
1655         /* We switched off IOMMU region successfully. */
1656         success = true;
1657     }
1658 
1659 out:
1660     trace_vtd_pt_enable_fast_path(source_id, success);
1661 }
1662 
1663 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1664  * translation.
1665  *
1666  * Called from RCU critical section.
1667  *
1668  * @bus_num: The bus number
1669  * @devfn: The devfn, which is the  combined of device and function number
1670  * @is_write: The access is a write operation
1671  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1672  *
1673  * Returns true if translation is successful, otherwise false.
1674  */
1675 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1676                                    uint8_t devfn, hwaddr addr, bool is_write,
1677                                    IOMMUTLBEntry *entry)
1678 {
1679     IntelIOMMUState *s = vtd_as->iommu_state;
1680     VTDContextEntry ce;
1681     uint8_t bus_num = pci_bus_num(bus);
1682     VTDContextCacheEntry *cc_entry;
1683     uint64_t slpte, page_mask;
1684     uint32_t level;
1685     uint16_t source_id = vtd_make_source_id(bus_num, devfn);
1686     int ret_fr;
1687     bool is_fpd_set = false;
1688     bool reads = true;
1689     bool writes = true;
1690     uint8_t access_flags;
1691     VTDIOTLBEntry *iotlb_entry;
1692 
1693     /*
1694      * We have standalone memory region for interrupt addresses, we
1695      * should never receive translation requests in this region.
1696      */
1697     assert(!vtd_is_interrupt_addr(addr));
1698 
1699     vtd_iommu_lock(s);
1700 
1701     cc_entry = &vtd_as->context_cache_entry;
1702 
1703     /* Try to fetch slpte form IOTLB */
1704     iotlb_entry = vtd_lookup_iotlb(s, source_id, addr);
1705     if (iotlb_entry) {
1706         trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1707                                  iotlb_entry->domain_id);
1708         slpte = iotlb_entry->slpte;
1709         access_flags = iotlb_entry->access_flags;
1710         page_mask = iotlb_entry->mask;
1711         goto out;
1712     }
1713 
1714     /* Try to fetch context-entry from cache first */
1715     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1716         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1717                                cc_entry->context_entry.lo,
1718                                cc_entry->context_cache_gen);
1719         ce = cc_entry->context_entry;
1720         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1721         if (!is_fpd_set && s->root_scalable) {
1722             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1723             VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1724         }
1725     } else {
1726         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1727         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1728         if (!ret_fr && !is_fpd_set && s->root_scalable) {
1729             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1730         }
1731         VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1732         /* Update context-cache */
1733         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1734                                   cc_entry->context_cache_gen,
1735                                   s->context_cache_gen);
1736         cc_entry->context_entry = ce;
1737         cc_entry->context_cache_gen = s->context_cache_gen;
1738     }
1739 
1740     /*
1741      * We don't need to translate for pass-through context entries.
1742      * Also, let's ignore IOTLB caching as well for PT devices.
1743      */
1744     if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) {
1745         entry->iova = addr & VTD_PAGE_MASK_4K;
1746         entry->translated_addr = entry->iova;
1747         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1748         entry->perm = IOMMU_RW;
1749         trace_vtd_translate_pt(source_id, entry->iova);
1750 
1751         /*
1752          * When this happens, it means firstly caching-mode is not
1753          * enabled, and this is the first passthrough translation for
1754          * the device. Let's enable the fast path for passthrough.
1755          *
1756          * When passthrough is disabled again for the device, we can
1757          * capture it via the context entry invalidation, then the
1758          * IOMMU region can be swapped back.
1759          */
1760         vtd_pt_enable_fast_path(s, source_id);
1761         vtd_iommu_unlock(s);
1762         return true;
1763     }
1764 
1765     ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level,
1766                                &reads, &writes, s->aw_bits);
1767     VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1768 
1769     page_mask = vtd_slpt_level_page_mask(level);
1770     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1771     vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce), addr, slpte,
1772                      access_flags, level);
1773 out:
1774     vtd_iommu_unlock(s);
1775     entry->iova = addr & page_mask;
1776     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1777     entry->addr_mask = ~page_mask;
1778     entry->perm = access_flags;
1779     return true;
1780 
1781 error:
1782     vtd_iommu_unlock(s);
1783     entry->iova = 0;
1784     entry->translated_addr = 0;
1785     entry->addr_mask = 0;
1786     entry->perm = IOMMU_NONE;
1787     return false;
1788 }
1789 
1790 static void vtd_root_table_setup(IntelIOMMUState *s)
1791 {
1792     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1793     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
1794 
1795     vtd_update_scalable_state(s);
1796 
1797     trace_vtd_reg_dmar_root(s->root, s->root_scalable);
1798 }
1799 
1800 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
1801                                uint32_t index, uint32_t mask)
1802 {
1803     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
1804 }
1805 
1806 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
1807 {
1808     uint64_t value = 0;
1809     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
1810     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
1811     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
1812     s->intr_eime = value & VTD_IRTA_EIME;
1813 
1814     /* Notify global invalidation */
1815     vtd_iec_notify_all(s, true, 0, 0);
1816 
1817     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
1818 }
1819 
1820 static void vtd_iommu_replay_all(IntelIOMMUState *s)
1821 {
1822     VTDAddressSpace *vtd_as;
1823 
1824     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1825         vtd_sync_shadow_page_table(vtd_as);
1826     }
1827 }
1828 
1829 static void vtd_context_global_invalidate(IntelIOMMUState *s)
1830 {
1831     trace_vtd_inv_desc_cc_global();
1832     /* Protects context cache */
1833     vtd_iommu_lock(s);
1834     s->context_cache_gen++;
1835     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
1836         vtd_reset_context_cache_locked(s);
1837     }
1838     vtd_iommu_unlock(s);
1839     vtd_address_space_refresh_all(s);
1840     /*
1841      * From VT-d spec 6.5.2.1, a global context entry invalidation
1842      * should be followed by a IOTLB global invalidation, so we should
1843      * be safe even without this. Hoewever, let's replay the region as
1844      * well to be safer, and go back here when we need finer tunes for
1845      * VT-d emulation codes.
1846      */
1847     vtd_iommu_replay_all(s);
1848 }
1849 
1850 /* Do a context-cache device-selective invalidation.
1851  * @func_mask: FM field after shifting
1852  */
1853 static void vtd_context_device_invalidate(IntelIOMMUState *s,
1854                                           uint16_t source_id,
1855                                           uint16_t func_mask)
1856 {
1857     uint16_t mask;
1858     VTDBus *vtd_bus;
1859     VTDAddressSpace *vtd_as;
1860     uint8_t bus_n, devfn;
1861     uint16_t devfn_it;
1862 
1863     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
1864 
1865     switch (func_mask & 3) {
1866     case 0:
1867         mask = 0;   /* No bits in the SID field masked */
1868         break;
1869     case 1:
1870         mask = 4;   /* Mask bit 2 in the SID field */
1871         break;
1872     case 2:
1873         mask = 6;   /* Mask bit 2:1 in the SID field */
1874         break;
1875     case 3:
1876         mask = 7;   /* Mask bit 2:0 in the SID field */
1877         break;
1878     }
1879     mask = ~mask;
1880 
1881     bus_n = VTD_SID_TO_BUS(source_id);
1882     vtd_bus = vtd_find_as_from_bus_num(s, bus_n);
1883     if (vtd_bus) {
1884         devfn = VTD_SID_TO_DEVFN(source_id);
1885         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
1886             vtd_as = vtd_bus->dev_as[devfn_it];
1887             if (vtd_as && ((devfn_it & mask) == (devfn & mask))) {
1888                 trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it),
1889                                              VTD_PCI_FUNC(devfn_it));
1890                 vtd_iommu_lock(s);
1891                 vtd_as->context_cache_entry.context_cache_gen = 0;
1892                 vtd_iommu_unlock(s);
1893                 /*
1894                  * Do switch address space when needed, in case if the
1895                  * device passthrough bit is switched.
1896                  */
1897                 vtd_switch_address_space(vtd_as);
1898                 /*
1899                  * So a device is moving out of (or moving into) a
1900                  * domain, resync the shadow page table.
1901                  * This won't bring bad even if we have no such
1902                  * notifier registered - the IOMMU notification
1903                  * framework will skip MAP notifications if that
1904                  * happened.
1905                  */
1906                 vtd_sync_shadow_page_table(vtd_as);
1907             }
1908         }
1909     }
1910 }
1911 
1912 /* Context-cache invalidation
1913  * Returns the Context Actual Invalidation Granularity.
1914  * @val: the content of the CCMD_REG
1915  */
1916 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
1917 {
1918     uint64_t caig;
1919     uint64_t type = val & VTD_CCMD_CIRG_MASK;
1920 
1921     switch (type) {
1922     case VTD_CCMD_DOMAIN_INVL:
1923         /* Fall through */
1924     case VTD_CCMD_GLOBAL_INVL:
1925         caig = VTD_CCMD_GLOBAL_INVL_A;
1926         vtd_context_global_invalidate(s);
1927         break;
1928 
1929     case VTD_CCMD_DEVICE_INVL:
1930         caig = VTD_CCMD_DEVICE_INVL_A;
1931         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
1932         break;
1933 
1934     default:
1935         error_report_once("%s: invalid context: 0x%" PRIx64,
1936                           __func__, val);
1937         caig = 0;
1938     }
1939     return caig;
1940 }
1941 
1942 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
1943 {
1944     trace_vtd_inv_desc_iotlb_global();
1945     vtd_reset_iotlb(s);
1946     vtd_iommu_replay_all(s);
1947 }
1948 
1949 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
1950 {
1951     VTDContextEntry ce;
1952     VTDAddressSpace *vtd_as;
1953 
1954     trace_vtd_inv_desc_iotlb_domain(domain_id);
1955 
1956     vtd_iommu_lock(s);
1957     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
1958                                 &domain_id);
1959     vtd_iommu_unlock(s);
1960 
1961     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1962         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1963                                       vtd_as->devfn, &ce) &&
1964             domain_id == vtd_get_domain_id(s, &ce)) {
1965             vtd_sync_shadow_page_table(vtd_as);
1966         }
1967     }
1968 }
1969 
1970 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
1971                                            uint16_t domain_id, hwaddr addr,
1972                                            uint8_t am)
1973 {
1974     VTDAddressSpace *vtd_as;
1975     VTDContextEntry ce;
1976     int ret;
1977     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
1978 
1979     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
1980         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1981                                        vtd_as->devfn, &ce);
1982         if (!ret && domain_id == vtd_get_domain_id(s, &ce)) {
1983             if (vtd_as_has_map_notifier(vtd_as)) {
1984                 /*
1985                  * As long as we have MAP notifications registered in
1986                  * any of our IOMMU notifiers, we need to sync the
1987                  * shadow page table.
1988                  */
1989                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
1990             } else {
1991                 /*
1992                  * For UNMAP-only notifiers, we don't need to walk the
1993                  * page tables.  We just deliver the PSI down to
1994                  * invalidate caches.
1995                  */
1996                 IOMMUTLBEntry entry = {
1997                     .target_as = &address_space_memory,
1998                     .iova = addr,
1999                     .translated_addr = 0,
2000                     .addr_mask = size - 1,
2001                     .perm = IOMMU_NONE,
2002                 };
2003                 memory_region_notify_iommu(&vtd_as->iommu, 0, entry);
2004             }
2005         }
2006     }
2007 }
2008 
2009 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
2010                                       hwaddr addr, uint8_t am)
2011 {
2012     VTDIOTLBPageInvInfo info;
2013 
2014     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
2015 
2016     assert(am <= VTD_MAMV);
2017     info.domain_id = domain_id;
2018     info.addr = addr;
2019     info.mask = ~((1 << am) - 1);
2020     vtd_iommu_lock(s);
2021     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
2022     vtd_iommu_unlock(s);
2023     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am);
2024 }
2025 
2026 /* Flush IOTLB
2027  * Returns the IOTLB Actual Invalidation Granularity.
2028  * @val: the content of the IOTLB_REG
2029  */
2030 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
2031 {
2032     uint64_t iaig;
2033     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
2034     uint16_t domain_id;
2035     hwaddr addr;
2036     uint8_t am;
2037 
2038     switch (type) {
2039     case VTD_TLB_GLOBAL_FLUSH:
2040         iaig = VTD_TLB_GLOBAL_FLUSH_A;
2041         vtd_iotlb_global_invalidate(s);
2042         break;
2043 
2044     case VTD_TLB_DSI_FLUSH:
2045         domain_id = VTD_TLB_DID(val);
2046         iaig = VTD_TLB_DSI_FLUSH_A;
2047         vtd_iotlb_domain_invalidate(s, domain_id);
2048         break;
2049 
2050     case VTD_TLB_PSI_FLUSH:
2051         domain_id = VTD_TLB_DID(val);
2052         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
2053         am = VTD_IVA_AM(addr);
2054         addr = VTD_IVA_ADDR(addr);
2055         if (am > VTD_MAMV) {
2056             error_report_once("%s: address mask overflow: 0x%" PRIx64,
2057                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
2058             iaig = 0;
2059             break;
2060         }
2061         iaig = VTD_TLB_PSI_FLUSH_A;
2062         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2063         break;
2064 
2065     default:
2066         error_report_once("%s: invalid granularity: 0x%" PRIx64,
2067                           __func__, val);
2068         iaig = 0;
2069     }
2070     return iaig;
2071 }
2072 
2073 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
2074 
2075 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
2076 {
2077     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
2078            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
2079 }
2080 
2081 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
2082 {
2083     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
2084 
2085     trace_vtd_inv_qi_enable(en);
2086 
2087     if (en) {
2088         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
2089         /* 2^(x+8) entries */
2090         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0));
2091         s->qi_enabled = true;
2092         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
2093         /* Ok - report back to driver */
2094         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
2095 
2096         if (s->iq_tail != 0) {
2097             /*
2098              * This is a spec violation but Windows guests are known to set up
2099              * Queued Invalidation this way so we allow the write and process
2100              * Invalidation Descriptors right away.
2101              */
2102             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
2103             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2104                 vtd_fetch_inv_desc(s);
2105             }
2106         }
2107     } else {
2108         if (vtd_queued_inv_disable_check(s)) {
2109             /* disable Queued Invalidation */
2110             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
2111             s->iq_head = 0;
2112             s->qi_enabled = false;
2113             /* Ok - report back to driver */
2114             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
2115         } else {
2116             error_report_once("%s: detected improper state when disable QI "
2117                               "(head=0x%x, tail=0x%x, last_type=%d)",
2118                               __func__,
2119                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
2120         }
2121     }
2122 }
2123 
2124 /* Set Root Table Pointer */
2125 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
2126 {
2127     vtd_root_table_setup(s);
2128     /* Ok - report back to driver */
2129     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
2130     vtd_reset_caches(s);
2131     vtd_address_space_refresh_all(s);
2132 }
2133 
2134 /* Set Interrupt Remap Table Pointer */
2135 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
2136 {
2137     vtd_interrupt_remap_table_setup(s);
2138     /* Ok - report back to driver */
2139     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
2140 }
2141 
2142 /* Handle Translation Enable/Disable */
2143 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
2144 {
2145     if (s->dmar_enabled == en) {
2146         return;
2147     }
2148 
2149     trace_vtd_dmar_enable(en);
2150 
2151     if (en) {
2152         s->dmar_enabled = true;
2153         /* Ok - report back to driver */
2154         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
2155     } else {
2156         s->dmar_enabled = false;
2157 
2158         /* Clear the index of Fault Recording Register */
2159         s->next_frcd_reg = 0;
2160         /* Ok - report back to driver */
2161         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
2162     }
2163 
2164     vtd_reset_caches(s);
2165     vtd_address_space_refresh_all(s);
2166 }
2167 
2168 /* Handle Interrupt Remap Enable/Disable */
2169 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
2170 {
2171     trace_vtd_ir_enable(en);
2172 
2173     if (en) {
2174         s->intr_enabled = true;
2175         /* Ok - report back to driver */
2176         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
2177     } else {
2178         s->intr_enabled = false;
2179         /* Ok - report back to driver */
2180         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
2181     }
2182 }
2183 
2184 /* Handle write to Global Command Register */
2185 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
2186 {
2187     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
2188     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
2189     uint32_t changed = status ^ val;
2190 
2191     trace_vtd_reg_write_gcmd(status, val);
2192     if (changed & VTD_GCMD_TE) {
2193         /* Translation enable/disable */
2194         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
2195     }
2196     if (val & VTD_GCMD_SRTP) {
2197         /* Set/update the root-table pointer */
2198         vtd_handle_gcmd_srtp(s);
2199     }
2200     if (changed & VTD_GCMD_QIE) {
2201         /* Queued Invalidation Enable */
2202         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
2203     }
2204     if (val & VTD_GCMD_SIRTP) {
2205         /* Set/update the interrupt remapping root-table pointer */
2206         vtd_handle_gcmd_sirtp(s);
2207     }
2208     if (changed & VTD_GCMD_IRE) {
2209         /* Interrupt remap enable/disable */
2210         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
2211     }
2212 }
2213 
2214 /* Handle write to Context Command Register */
2215 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
2216 {
2217     uint64_t ret;
2218     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
2219 
2220     /* Context-cache invalidation request */
2221     if (val & VTD_CCMD_ICC) {
2222         if (s->qi_enabled) {
2223             error_report_once("Queued Invalidation enabled, "
2224                               "should not use register-based invalidation");
2225             return;
2226         }
2227         ret = vtd_context_cache_invalidate(s, val);
2228         /* Invalidation completed. Change something to show */
2229         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
2230         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
2231                                       ret);
2232     }
2233 }
2234 
2235 /* Handle write to IOTLB Invalidation Register */
2236 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
2237 {
2238     uint64_t ret;
2239     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
2240 
2241     /* IOTLB invalidation request */
2242     if (val & VTD_TLB_IVT) {
2243         if (s->qi_enabled) {
2244             error_report_once("Queued Invalidation enabled, "
2245                               "should not use register-based invalidation");
2246             return;
2247         }
2248         ret = vtd_iotlb_flush(s, val);
2249         /* Invalidation completed. Change something to show */
2250         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
2251         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
2252                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
2253     }
2254 }
2255 
2256 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2257 static bool vtd_get_inv_desc(IntelIOMMUState *s,
2258                              VTDInvDesc *inv_desc)
2259 {
2260     dma_addr_t base_addr = s->iq;
2261     uint32_t offset = s->iq_head;
2262     uint32_t dw = s->iq_dw ? 32 : 16;
2263     dma_addr_t addr = base_addr + offset * dw;
2264 
2265     if (dma_memory_read(&address_space_memory, addr, inv_desc, dw)) {
2266         error_report_once("Read INV DESC failed.");
2267         return false;
2268     }
2269     inv_desc->lo = le64_to_cpu(inv_desc->lo);
2270     inv_desc->hi = le64_to_cpu(inv_desc->hi);
2271     if (dw == 32) {
2272         inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]);
2273         inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]);
2274     }
2275     return true;
2276 }
2277 
2278 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2279 {
2280     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
2281         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
2282         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2283                           " (reserved nonzero)", __func__, inv_desc->hi,
2284                           inv_desc->lo);
2285         return false;
2286     }
2287     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
2288         /* Status Write */
2289         uint32_t status_data = (uint32_t)(inv_desc->lo >>
2290                                VTD_INV_DESC_WAIT_DATA_SHIFT);
2291 
2292         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
2293 
2294         /* FIXME: need to be masked with HAW? */
2295         dma_addr_t status_addr = inv_desc->hi;
2296         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
2297         status_data = cpu_to_le32(status_data);
2298         if (dma_memory_write(&address_space_memory, status_addr, &status_data,
2299                              sizeof(status_data))) {
2300             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
2301             return false;
2302         }
2303     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
2304         /* Interrupt flag */
2305         vtd_generate_completion_event(s);
2306     } else {
2307         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2308                           " (unknown type)", __func__, inv_desc->hi,
2309                           inv_desc->lo);
2310         return false;
2311     }
2312     return true;
2313 }
2314 
2315 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
2316                                            VTDInvDesc *inv_desc)
2317 {
2318     uint16_t sid, fmask;
2319 
2320     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
2321         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2322                           " (reserved nonzero)", __func__, inv_desc->hi,
2323                           inv_desc->lo);
2324         return false;
2325     }
2326     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
2327     case VTD_INV_DESC_CC_DOMAIN:
2328         trace_vtd_inv_desc_cc_domain(
2329             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
2330         /* Fall through */
2331     case VTD_INV_DESC_CC_GLOBAL:
2332         vtd_context_global_invalidate(s);
2333         break;
2334 
2335     case VTD_INV_DESC_CC_DEVICE:
2336         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
2337         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
2338         vtd_context_device_invalidate(s, sid, fmask);
2339         break;
2340 
2341     default:
2342         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2343                           " (invalid type)", __func__, inv_desc->hi,
2344                           inv_desc->lo);
2345         return false;
2346     }
2347     return true;
2348 }
2349 
2350 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2351 {
2352     uint16_t domain_id;
2353     uint8_t am;
2354     hwaddr addr;
2355 
2356     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
2357         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
2358         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2359                           ", lo=0x%"PRIx64" (reserved bits unzero)",
2360                           __func__, inv_desc->hi, inv_desc->lo);
2361         return false;
2362     }
2363 
2364     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
2365     case VTD_INV_DESC_IOTLB_GLOBAL:
2366         vtd_iotlb_global_invalidate(s);
2367         break;
2368 
2369     case VTD_INV_DESC_IOTLB_DOMAIN:
2370         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2371         vtd_iotlb_domain_invalidate(s, domain_id);
2372         break;
2373 
2374     case VTD_INV_DESC_IOTLB_PAGE:
2375         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2376         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
2377         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
2378         if (am > VTD_MAMV) {
2379             error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2380                               ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)",
2381                               __func__, inv_desc->hi, inv_desc->lo,
2382                               am, (unsigned)VTD_MAMV);
2383             return false;
2384         }
2385         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2386         break;
2387 
2388     default:
2389         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2390                           ", lo=0x%"PRIx64" (type mismatch: 0x%llx)",
2391                           __func__, inv_desc->hi, inv_desc->lo,
2392                           inv_desc->lo & VTD_INV_DESC_IOTLB_G);
2393         return false;
2394     }
2395     return true;
2396 }
2397 
2398 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
2399                                      VTDInvDesc *inv_desc)
2400 {
2401     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
2402                            inv_desc->iec.index,
2403                            inv_desc->iec.index_mask);
2404 
2405     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
2406                        inv_desc->iec.index,
2407                        inv_desc->iec.index_mask);
2408     return true;
2409 }
2410 
2411 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
2412                                           VTDInvDesc *inv_desc)
2413 {
2414     VTDAddressSpace *vtd_dev_as;
2415     IOMMUTLBEntry entry;
2416     struct VTDBus *vtd_bus;
2417     hwaddr addr;
2418     uint64_t sz;
2419     uint16_t sid;
2420     uint8_t devfn;
2421     bool size;
2422     uint8_t bus_num;
2423 
2424     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2425     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2426     devfn = sid & 0xff;
2427     bus_num = sid >> 8;
2428     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2429 
2430     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2431         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2432         error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2433                           ", lo=%"PRIx64" (reserved nonzero)", __func__,
2434                           inv_desc->hi, inv_desc->lo);
2435         return false;
2436     }
2437 
2438     vtd_bus = vtd_find_as_from_bus_num(s, bus_num);
2439     if (!vtd_bus) {
2440         goto done;
2441     }
2442 
2443     vtd_dev_as = vtd_bus->dev_as[devfn];
2444     if (!vtd_dev_as) {
2445         goto done;
2446     }
2447 
2448     /* According to ATS spec table 2.4:
2449      * S = 0, bits 15:12 = xxxx     range size: 4K
2450      * S = 1, bits 15:12 = xxx0     range size: 8K
2451      * S = 1, bits 15:12 = xx01     range size: 16K
2452      * S = 1, bits 15:12 = x011     range size: 32K
2453      * S = 1, bits 15:12 = 0111     range size: 64K
2454      * ...
2455      */
2456     if (size) {
2457         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2458         addr &= ~(sz - 1);
2459     } else {
2460         sz = VTD_PAGE_SIZE;
2461     }
2462 
2463     entry.target_as = &vtd_dev_as->as;
2464     entry.addr_mask = sz - 1;
2465     entry.iova = addr;
2466     entry.perm = IOMMU_NONE;
2467     entry.translated_addr = 0;
2468     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry);
2469 
2470 done:
2471     return true;
2472 }
2473 
2474 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2475 {
2476     VTDInvDesc inv_desc;
2477     uint8_t desc_type;
2478 
2479     trace_vtd_inv_qi_head(s->iq_head);
2480     if (!vtd_get_inv_desc(s, &inv_desc)) {
2481         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2482         return false;
2483     }
2484 
2485     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2486     /* FIXME: should update at first or at last? */
2487     s->iq_last_desc_type = desc_type;
2488 
2489     switch (desc_type) {
2490     case VTD_INV_DESC_CC:
2491         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2492         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2493             return false;
2494         }
2495         break;
2496 
2497     case VTD_INV_DESC_IOTLB:
2498         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2499         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2500             return false;
2501         }
2502         break;
2503 
2504     /*
2505      * TODO: the entity of below two cases will be implemented in future series.
2506      * To make guest (which integrates scalable mode support patch set in
2507      * iommu driver) work, just return true is enough so far.
2508      */
2509     case VTD_INV_DESC_PC:
2510         break;
2511 
2512     case VTD_INV_DESC_PIOTLB:
2513         break;
2514 
2515     case VTD_INV_DESC_WAIT:
2516         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2517         if (!vtd_process_wait_desc(s, &inv_desc)) {
2518             return false;
2519         }
2520         break;
2521 
2522     case VTD_INV_DESC_IEC:
2523         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2524         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2525             return false;
2526         }
2527         break;
2528 
2529     case VTD_INV_DESC_DEVICE:
2530         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2531         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2532             return false;
2533         }
2534         break;
2535 
2536     default:
2537         error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64
2538                           " (unknown type)", __func__, inv_desc.hi,
2539                           inv_desc.lo);
2540         return false;
2541     }
2542     s->iq_head++;
2543     if (s->iq_head == s->iq_size) {
2544         s->iq_head = 0;
2545     }
2546     return true;
2547 }
2548 
2549 /* Try to fetch and process more Invalidation Descriptors */
2550 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2551 {
2552     int qi_shift;
2553 
2554     /* Refer to 10.4.23 of VT-d spec 3.0 */
2555     qi_shift = s->iq_dw ? VTD_IQH_QH_SHIFT_5 : VTD_IQH_QH_SHIFT_4;
2556 
2557     trace_vtd_inv_qi_fetch();
2558 
2559     if (s->iq_tail >= s->iq_size) {
2560         /* Detects an invalid Tail pointer */
2561         error_report_once("%s: detected invalid QI tail "
2562                           "(tail=0x%x, size=0x%x)",
2563                           __func__, s->iq_tail, s->iq_size);
2564         vtd_handle_inv_queue_error(s);
2565         return;
2566     }
2567     while (s->iq_head != s->iq_tail) {
2568         if (!vtd_process_inv_desc(s)) {
2569             /* Invalidation Queue Errors */
2570             vtd_handle_inv_queue_error(s);
2571             break;
2572         }
2573         /* Must update the IQH_REG in time */
2574         vtd_set_quad_raw(s, DMAR_IQH_REG,
2575                          (((uint64_t)(s->iq_head)) << qi_shift) &
2576                          VTD_IQH_QH_MASK);
2577     }
2578 }
2579 
2580 /* Handle write to Invalidation Queue Tail Register */
2581 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2582 {
2583     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2584 
2585     if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) {
2586         error_report_once("%s: RSV bit is set: val=0x%"PRIx64,
2587                           __func__, val);
2588         return;
2589     }
2590     s->iq_tail = VTD_IQT_QT(s->iq_dw, val);
2591     trace_vtd_inv_qi_tail(s->iq_tail);
2592 
2593     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2594         /* Process Invalidation Queue here */
2595         vtd_fetch_inv_desc(s);
2596     }
2597 }
2598 
2599 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2600 {
2601     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2602     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2603     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2604 
2605     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2606         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2607         trace_vtd_fsts_clear_ip();
2608     }
2609     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2610      * Descriptors if there are any when Queued Invalidation is enabled?
2611      */
2612 }
2613 
2614 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2615 {
2616     uint32_t fectl_reg;
2617     /* FIXME: when software clears the IM field, check the IP field. But do we
2618      * need to compare the old value and the new value to conclude that
2619      * software clears the IM field? Or just check if the IM field is zero?
2620      */
2621     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2622 
2623     trace_vtd_reg_write_fectl(fectl_reg);
2624 
2625     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2626         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2627         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2628     }
2629 }
2630 
2631 static void vtd_handle_ics_write(IntelIOMMUState *s)
2632 {
2633     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2634     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2635 
2636     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2637         trace_vtd_reg_ics_clear_ip();
2638         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2639     }
2640 }
2641 
2642 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2643 {
2644     uint32_t iectl_reg;
2645     /* FIXME: when software clears the IM field, check the IP field. But do we
2646      * need to compare the old value and the new value to conclude that
2647      * software clears the IM field? Or just check if the IM field is zero?
2648      */
2649     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2650 
2651     trace_vtd_reg_write_iectl(iectl_reg);
2652 
2653     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2654         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2655         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2656     }
2657 }
2658 
2659 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2660 {
2661     IntelIOMMUState *s = opaque;
2662     uint64_t val;
2663 
2664     trace_vtd_reg_read(addr, size);
2665 
2666     if (addr + size > DMAR_REG_SIZE) {
2667         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2668                           " size=0x%x", __func__, addr, size);
2669         return (uint64_t)-1;
2670     }
2671 
2672     switch (addr) {
2673     /* Root Table Address Register, 64-bit */
2674     case DMAR_RTADDR_REG:
2675         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
2676         if (size == 4) {
2677             val = val & ((1ULL << 32) - 1);
2678         }
2679         break;
2680 
2681     case DMAR_RTADDR_REG_HI:
2682         assert(size == 4);
2683         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG) >> 32;
2684         break;
2685 
2686     /* Invalidation Queue Address Register, 64-bit */
2687     case DMAR_IQA_REG:
2688         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2689         if (size == 4) {
2690             val = val & ((1ULL << 32) - 1);
2691         }
2692         break;
2693 
2694     case DMAR_IQA_REG_HI:
2695         assert(size == 4);
2696         val = s->iq >> 32;
2697         break;
2698 
2699     default:
2700         if (size == 4) {
2701             val = vtd_get_long(s, addr);
2702         } else {
2703             val = vtd_get_quad(s, addr);
2704         }
2705     }
2706 
2707     return val;
2708 }
2709 
2710 static void vtd_mem_write(void *opaque, hwaddr addr,
2711                           uint64_t val, unsigned size)
2712 {
2713     IntelIOMMUState *s = opaque;
2714 
2715     trace_vtd_reg_write(addr, size, val);
2716 
2717     if (addr + size > DMAR_REG_SIZE) {
2718         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2719                           " size=0x%x", __func__, addr, size);
2720         return;
2721     }
2722 
2723     switch (addr) {
2724     /* Global Command Register, 32-bit */
2725     case DMAR_GCMD_REG:
2726         vtd_set_long(s, addr, val);
2727         vtd_handle_gcmd_write(s);
2728         break;
2729 
2730     /* Context Command Register, 64-bit */
2731     case DMAR_CCMD_REG:
2732         if (size == 4) {
2733             vtd_set_long(s, addr, val);
2734         } else {
2735             vtd_set_quad(s, addr, val);
2736             vtd_handle_ccmd_write(s);
2737         }
2738         break;
2739 
2740     case DMAR_CCMD_REG_HI:
2741         assert(size == 4);
2742         vtd_set_long(s, addr, val);
2743         vtd_handle_ccmd_write(s);
2744         break;
2745 
2746     /* IOTLB Invalidation Register, 64-bit */
2747     case DMAR_IOTLB_REG:
2748         if (size == 4) {
2749             vtd_set_long(s, addr, val);
2750         } else {
2751             vtd_set_quad(s, addr, val);
2752             vtd_handle_iotlb_write(s);
2753         }
2754         break;
2755 
2756     case DMAR_IOTLB_REG_HI:
2757         assert(size == 4);
2758         vtd_set_long(s, addr, val);
2759         vtd_handle_iotlb_write(s);
2760         break;
2761 
2762     /* Invalidate Address Register, 64-bit */
2763     case DMAR_IVA_REG:
2764         if (size == 4) {
2765             vtd_set_long(s, addr, val);
2766         } else {
2767             vtd_set_quad(s, addr, val);
2768         }
2769         break;
2770 
2771     case DMAR_IVA_REG_HI:
2772         assert(size == 4);
2773         vtd_set_long(s, addr, val);
2774         break;
2775 
2776     /* Fault Status Register, 32-bit */
2777     case DMAR_FSTS_REG:
2778         assert(size == 4);
2779         vtd_set_long(s, addr, val);
2780         vtd_handle_fsts_write(s);
2781         break;
2782 
2783     /* Fault Event Control Register, 32-bit */
2784     case DMAR_FECTL_REG:
2785         assert(size == 4);
2786         vtd_set_long(s, addr, val);
2787         vtd_handle_fectl_write(s);
2788         break;
2789 
2790     /* Fault Event Data Register, 32-bit */
2791     case DMAR_FEDATA_REG:
2792         assert(size == 4);
2793         vtd_set_long(s, addr, val);
2794         break;
2795 
2796     /* Fault Event Address Register, 32-bit */
2797     case DMAR_FEADDR_REG:
2798         if (size == 4) {
2799             vtd_set_long(s, addr, val);
2800         } else {
2801             /*
2802              * While the register is 32-bit only, some guests (Xen...) write to
2803              * it with 64-bit.
2804              */
2805             vtd_set_quad(s, addr, val);
2806         }
2807         break;
2808 
2809     /* Fault Event Upper Address Register, 32-bit */
2810     case DMAR_FEUADDR_REG:
2811         assert(size == 4);
2812         vtd_set_long(s, addr, val);
2813         break;
2814 
2815     /* Protected Memory Enable Register, 32-bit */
2816     case DMAR_PMEN_REG:
2817         assert(size == 4);
2818         vtd_set_long(s, addr, val);
2819         break;
2820 
2821     /* Root Table Address Register, 64-bit */
2822     case DMAR_RTADDR_REG:
2823         if (size == 4) {
2824             vtd_set_long(s, addr, val);
2825         } else {
2826             vtd_set_quad(s, addr, val);
2827         }
2828         break;
2829 
2830     case DMAR_RTADDR_REG_HI:
2831         assert(size == 4);
2832         vtd_set_long(s, addr, val);
2833         break;
2834 
2835     /* Invalidation Queue Tail Register, 64-bit */
2836     case DMAR_IQT_REG:
2837         if (size == 4) {
2838             vtd_set_long(s, addr, val);
2839         } else {
2840             vtd_set_quad(s, addr, val);
2841         }
2842         vtd_handle_iqt_write(s);
2843         break;
2844 
2845     case DMAR_IQT_REG_HI:
2846         assert(size == 4);
2847         vtd_set_long(s, addr, val);
2848         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
2849         break;
2850 
2851     /* Invalidation Queue Address Register, 64-bit */
2852     case DMAR_IQA_REG:
2853         if (size == 4) {
2854             vtd_set_long(s, addr, val);
2855         } else {
2856             vtd_set_quad(s, addr, val);
2857         }
2858         if (s->ecap & VTD_ECAP_SMTS &&
2859             val & VTD_IQA_DW_MASK) {
2860             s->iq_dw = true;
2861         } else {
2862             s->iq_dw = false;
2863         }
2864         break;
2865 
2866     case DMAR_IQA_REG_HI:
2867         assert(size == 4);
2868         vtd_set_long(s, addr, val);
2869         break;
2870 
2871     /* Invalidation Completion Status Register, 32-bit */
2872     case DMAR_ICS_REG:
2873         assert(size == 4);
2874         vtd_set_long(s, addr, val);
2875         vtd_handle_ics_write(s);
2876         break;
2877 
2878     /* Invalidation Event Control Register, 32-bit */
2879     case DMAR_IECTL_REG:
2880         assert(size == 4);
2881         vtd_set_long(s, addr, val);
2882         vtd_handle_iectl_write(s);
2883         break;
2884 
2885     /* Invalidation Event Data Register, 32-bit */
2886     case DMAR_IEDATA_REG:
2887         assert(size == 4);
2888         vtd_set_long(s, addr, val);
2889         break;
2890 
2891     /* Invalidation Event Address Register, 32-bit */
2892     case DMAR_IEADDR_REG:
2893         assert(size == 4);
2894         vtd_set_long(s, addr, val);
2895         break;
2896 
2897     /* Invalidation Event Upper Address Register, 32-bit */
2898     case DMAR_IEUADDR_REG:
2899         assert(size == 4);
2900         vtd_set_long(s, addr, val);
2901         break;
2902 
2903     /* Fault Recording Registers, 128-bit */
2904     case DMAR_FRCD_REG_0_0:
2905         if (size == 4) {
2906             vtd_set_long(s, addr, val);
2907         } else {
2908             vtd_set_quad(s, addr, val);
2909         }
2910         break;
2911 
2912     case DMAR_FRCD_REG_0_1:
2913         assert(size == 4);
2914         vtd_set_long(s, addr, val);
2915         break;
2916 
2917     case DMAR_FRCD_REG_0_2:
2918         if (size == 4) {
2919             vtd_set_long(s, addr, val);
2920         } else {
2921             vtd_set_quad(s, addr, val);
2922             /* May clear bit 127 (Fault), update PPF */
2923             vtd_update_fsts_ppf(s);
2924         }
2925         break;
2926 
2927     case DMAR_FRCD_REG_0_3:
2928         assert(size == 4);
2929         vtd_set_long(s, addr, val);
2930         /* May clear bit 127 (Fault), update PPF */
2931         vtd_update_fsts_ppf(s);
2932         break;
2933 
2934     case DMAR_IRTA_REG:
2935         if (size == 4) {
2936             vtd_set_long(s, addr, val);
2937         } else {
2938             vtd_set_quad(s, addr, val);
2939         }
2940         break;
2941 
2942     case DMAR_IRTA_REG_HI:
2943         assert(size == 4);
2944         vtd_set_long(s, addr, val);
2945         break;
2946 
2947     default:
2948         if (size == 4) {
2949             vtd_set_long(s, addr, val);
2950         } else {
2951             vtd_set_quad(s, addr, val);
2952         }
2953     }
2954 }
2955 
2956 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
2957                                          IOMMUAccessFlags flag, int iommu_idx)
2958 {
2959     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2960     IntelIOMMUState *s = vtd_as->iommu_state;
2961     IOMMUTLBEntry iotlb = {
2962         /* We'll fill in the rest later. */
2963         .target_as = &address_space_memory,
2964     };
2965     bool success;
2966 
2967     if (likely(s->dmar_enabled)) {
2968         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
2969                                          addr, flag & IOMMU_WO, &iotlb);
2970     } else {
2971         /* DMAR disabled, passthrough, use 4k-page*/
2972         iotlb.iova = addr & VTD_PAGE_MASK_4K;
2973         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
2974         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
2975         iotlb.perm = IOMMU_RW;
2976         success = true;
2977     }
2978 
2979     if (likely(success)) {
2980         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
2981                                  VTD_PCI_SLOT(vtd_as->devfn),
2982                                  VTD_PCI_FUNC(vtd_as->devfn),
2983                                  iotlb.iova, iotlb.translated_addr,
2984                                  iotlb.addr_mask);
2985     } else {
2986         error_report_once("%s: detected translation failure "
2987                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
2988                           __func__, pci_bus_num(vtd_as->bus),
2989                           VTD_PCI_SLOT(vtd_as->devfn),
2990                           VTD_PCI_FUNC(vtd_as->devfn),
2991                           addr);
2992     }
2993 
2994     return iotlb;
2995 }
2996 
2997 static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
2998                                          IOMMUNotifierFlag old,
2999                                          IOMMUNotifierFlag new,
3000                                          Error **errp)
3001 {
3002     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
3003     IntelIOMMUState *s = vtd_as->iommu_state;
3004 
3005     /* Update per-address-space notifier flags */
3006     vtd_as->notifier_flags = new;
3007 
3008     if (old == IOMMU_NOTIFIER_NONE) {
3009         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
3010     } else if (new == IOMMU_NOTIFIER_NONE) {
3011         QLIST_REMOVE(vtd_as, next);
3012     }
3013     return 0;
3014 }
3015 
3016 static int vtd_post_load(void *opaque, int version_id)
3017 {
3018     IntelIOMMUState *iommu = opaque;
3019 
3020     /*
3021      * Memory regions are dynamically turned on/off depending on
3022      * context entry configurations from the guest. After migration,
3023      * we need to make sure the memory regions are still correct.
3024      */
3025     vtd_switch_address_space_all(iommu);
3026 
3027     /*
3028      * We don't need to migrate the root_scalable because we can
3029      * simply do the calculation after the loading is complete.  We
3030      * can actually do similar things with root, dmar_enabled, etc.
3031      * however since we've had them already so we'd better keep them
3032      * for compatibility of migration.
3033      */
3034     vtd_update_scalable_state(iommu);
3035 
3036     return 0;
3037 }
3038 
3039 static const VMStateDescription vtd_vmstate = {
3040     .name = "iommu-intel",
3041     .version_id = 1,
3042     .minimum_version_id = 1,
3043     .priority = MIG_PRI_IOMMU,
3044     .post_load = vtd_post_load,
3045     .fields = (VMStateField[]) {
3046         VMSTATE_UINT64(root, IntelIOMMUState),
3047         VMSTATE_UINT64(intr_root, IntelIOMMUState),
3048         VMSTATE_UINT64(iq, IntelIOMMUState),
3049         VMSTATE_UINT32(intr_size, IntelIOMMUState),
3050         VMSTATE_UINT16(iq_head, IntelIOMMUState),
3051         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
3052         VMSTATE_UINT16(iq_size, IntelIOMMUState),
3053         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
3054         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
3055         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
3056         VMSTATE_UNUSED(1),      /* bool root_extended is obsolete by VT-d */
3057         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
3058         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
3059         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
3060         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
3061         VMSTATE_END_OF_LIST()
3062     }
3063 };
3064 
3065 static const MemoryRegionOps vtd_mem_ops = {
3066     .read = vtd_mem_read,
3067     .write = vtd_mem_write,
3068     .endianness = DEVICE_LITTLE_ENDIAN,
3069     .impl = {
3070         .min_access_size = 4,
3071         .max_access_size = 8,
3072     },
3073     .valid = {
3074         .min_access_size = 4,
3075         .max_access_size = 8,
3076     },
3077 };
3078 
3079 static Property vtd_properties[] = {
3080     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
3081     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
3082                             ON_OFF_AUTO_AUTO),
3083     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
3084     DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits,
3085                       VTD_HOST_ADDRESS_WIDTH),
3086     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
3087     DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE),
3088     DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true),
3089     DEFINE_PROP_END_OF_LIST(),
3090 };
3091 
3092 /* Read IRTE entry with specific index */
3093 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
3094                         VTD_IR_TableEntry *entry, uint16_t sid)
3095 {
3096     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
3097         {0xffff, 0xfffb, 0xfff9, 0xfff8};
3098     dma_addr_t addr = 0x00;
3099     uint16_t mask, source_id;
3100     uint8_t bus, bus_max, bus_min;
3101 
3102     if (index >= iommu->intr_size) {
3103         error_report_once("%s: index too large: ind=0x%x",
3104                           __func__, index);
3105         return -VTD_FR_IR_INDEX_OVER;
3106     }
3107 
3108     addr = iommu->intr_root + index * sizeof(*entry);
3109     if (dma_memory_read(&address_space_memory, addr, entry,
3110                         sizeof(*entry))) {
3111         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
3112                           __func__, index, addr);
3113         return -VTD_FR_IR_ROOT_INVAL;
3114     }
3115 
3116     trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]),
3117                           le64_to_cpu(entry->data[0]));
3118 
3119     if (!entry->irte.present) {
3120         error_report_once("%s: detected non-present IRTE "
3121                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3122                           __func__, index, le64_to_cpu(entry->data[1]),
3123                           le64_to_cpu(entry->data[0]));
3124         return -VTD_FR_IR_ENTRY_P;
3125     }
3126 
3127     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
3128         entry->irte.__reserved_2) {
3129         error_report_once("%s: detected non-zero reserved IRTE "
3130                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3131                           __func__, index, le64_to_cpu(entry->data[1]),
3132                           le64_to_cpu(entry->data[0]));
3133         return -VTD_FR_IR_IRTE_RSVD;
3134     }
3135 
3136     if (sid != X86_IOMMU_SID_INVALID) {
3137         /* Validate IRTE SID */
3138         source_id = le32_to_cpu(entry->irte.source_id);
3139         switch (entry->irte.sid_vtype) {
3140         case VTD_SVT_NONE:
3141             break;
3142 
3143         case VTD_SVT_ALL:
3144             mask = vtd_svt_mask[entry->irte.sid_q];
3145             if ((source_id & mask) != (sid & mask)) {
3146                 error_report_once("%s: invalid IRTE SID "
3147                                   "(index=%u, sid=%u, source_id=%u)",
3148                                   __func__, index, sid, source_id);
3149                 return -VTD_FR_IR_SID_ERR;
3150             }
3151             break;
3152 
3153         case VTD_SVT_BUS:
3154             bus_max = source_id >> 8;
3155             bus_min = source_id & 0xff;
3156             bus = sid >> 8;
3157             if (bus > bus_max || bus < bus_min) {
3158                 error_report_once("%s: invalid SVT_BUS "
3159                                   "(index=%u, bus=%u, min=%u, max=%u)",
3160                                   __func__, index, bus, bus_min, bus_max);
3161                 return -VTD_FR_IR_SID_ERR;
3162             }
3163             break;
3164 
3165         default:
3166             error_report_once("%s: detected invalid IRTE SVT "
3167                               "(index=%u, type=%d)", __func__,
3168                               index, entry->irte.sid_vtype);
3169             /* Take this as verification failure. */
3170             return -VTD_FR_IR_SID_ERR;
3171         }
3172     }
3173 
3174     return 0;
3175 }
3176 
3177 /* Fetch IRQ information of specific IR index */
3178 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
3179                              X86IOMMUIrq *irq, uint16_t sid)
3180 {
3181     VTD_IR_TableEntry irte = {};
3182     int ret = 0;
3183 
3184     ret = vtd_irte_get(iommu, index, &irte, sid);
3185     if (ret) {
3186         return ret;
3187     }
3188 
3189     irq->trigger_mode = irte.irte.trigger_mode;
3190     irq->vector = irte.irte.vector;
3191     irq->delivery_mode = irte.irte.delivery_mode;
3192     irq->dest = le32_to_cpu(irte.irte.dest_id);
3193     if (!iommu->intr_eime) {
3194 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
3195 #define  VTD_IR_APIC_DEST_SHIFT        (8)
3196         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
3197             VTD_IR_APIC_DEST_SHIFT;
3198     }
3199     irq->dest_mode = irte.irte.dest_mode;
3200     irq->redir_hint = irte.irte.redir_hint;
3201 
3202     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
3203                        irq->delivery_mode, irq->dest, irq->dest_mode);
3204 
3205     return 0;
3206 }
3207 
3208 /* Interrupt remapping for MSI/MSI-X entry */
3209 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
3210                                    MSIMessage *origin,
3211                                    MSIMessage *translated,
3212                                    uint16_t sid)
3213 {
3214     int ret = 0;
3215     VTD_IR_MSIAddress addr;
3216     uint16_t index;
3217     X86IOMMUIrq irq = {};
3218 
3219     assert(origin && translated);
3220 
3221     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
3222 
3223     if (!iommu || !iommu->intr_enabled) {
3224         memcpy(translated, origin, sizeof(*origin));
3225         goto out;
3226     }
3227 
3228     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
3229         error_report_once("%s: MSI address high 32 bits non-zero detected: "
3230                           "address=0x%" PRIx64, __func__, origin->address);
3231         return -VTD_FR_IR_REQ_RSVD;
3232     }
3233 
3234     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
3235     if (addr.addr.__head != 0xfee) {
3236         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
3237                           __func__, addr.data);
3238         return -VTD_FR_IR_REQ_RSVD;
3239     }
3240 
3241     /* This is compatible mode. */
3242     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
3243         memcpy(translated, origin, sizeof(*origin));
3244         goto out;
3245     }
3246 
3247     index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l);
3248 
3249 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
3250 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
3251 
3252     if (addr.addr.sub_valid) {
3253         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3254         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
3255     }
3256 
3257     ret = vtd_remap_irq_get(iommu, index, &irq, sid);
3258     if (ret) {
3259         return ret;
3260     }
3261 
3262     if (addr.addr.sub_valid) {
3263         trace_vtd_ir_remap_type("MSI");
3264         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
3265             error_report_once("%s: invalid IR MSI "
3266                               "(sid=%u, address=0x%" PRIx64
3267                               ", data=0x%" PRIx32 ")",
3268                               __func__, sid, origin->address, origin->data);
3269             return -VTD_FR_IR_REQ_RSVD;
3270         }
3271     } else {
3272         uint8_t vector = origin->data & 0xff;
3273         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
3274 
3275         trace_vtd_ir_remap_type("IOAPIC");
3276         /* IOAPIC entry vector should be aligned with IRTE vector
3277          * (see vt-d spec 5.1.5.1). */
3278         if (vector != irq.vector) {
3279             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
3280         }
3281 
3282         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3283          * (see vt-d spec 5.1.5.1). */
3284         if (trigger_mode != irq.trigger_mode) {
3285             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
3286                                       irq.trigger_mode);
3287         }
3288     }
3289 
3290     /*
3291      * We'd better keep the last two bits, assuming that guest OS
3292      * might modify it. Keep it does not hurt after all.
3293      */
3294     irq.msi_addr_last_bits = addr.addr.__not_care;
3295 
3296     /* Translate X86IOMMUIrq to MSI message */
3297     x86_iommu_irq_to_msi_message(&irq, translated);
3298 
3299 out:
3300     trace_vtd_ir_remap_msi(origin->address, origin->data,
3301                            translated->address, translated->data);
3302     return 0;
3303 }
3304 
3305 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
3306                          MSIMessage *dst, uint16_t sid)
3307 {
3308     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
3309                                    src, dst, sid);
3310 }
3311 
3312 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
3313                                    uint64_t *data, unsigned size,
3314                                    MemTxAttrs attrs)
3315 {
3316     return MEMTX_OK;
3317 }
3318 
3319 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
3320                                     uint64_t value, unsigned size,
3321                                     MemTxAttrs attrs)
3322 {
3323     int ret = 0;
3324     MSIMessage from = {}, to = {};
3325     uint16_t sid = X86_IOMMU_SID_INVALID;
3326 
3327     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
3328     from.data = (uint32_t) value;
3329 
3330     if (!attrs.unspecified) {
3331         /* We have explicit Source ID */
3332         sid = attrs.requester_id;
3333     }
3334 
3335     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid);
3336     if (ret) {
3337         /* TODO: report error */
3338         /* Drop this interrupt */
3339         return MEMTX_ERROR;
3340     }
3341 
3342     apic_get_class()->send_msi(&to);
3343 
3344     return MEMTX_OK;
3345 }
3346 
3347 static const MemoryRegionOps vtd_mem_ir_ops = {
3348     .read_with_attrs = vtd_mem_ir_read,
3349     .write_with_attrs = vtd_mem_ir_write,
3350     .endianness = DEVICE_LITTLE_ENDIAN,
3351     .impl = {
3352         .min_access_size = 4,
3353         .max_access_size = 4,
3354     },
3355     .valid = {
3356         .min_access_size = 4,
3357         .max_access_size = 4,
3358     },
3359 };
3360 
3361 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
3362 {
3363     uintptr_t key = (uintptr_t)bus;
3364     VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
3365     VTDAddressSpace *vtd_dev_as;
3366     char name[128];
3367 
3368     if (!vtd_bus) {
3369         uintptr_t *new_key = g_malloc(sizeof(*new_key));
3370         *new_key = (uintptr_t)bus;
3371         /* No corresponding free() */
3372         vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
3373                             PCI_DEVFN_MAX);
3374         vtd_bus->bus = bus;
3375         g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
3376     }
3377 
3378     vtd_dev_as = vtd_bus->dev_as[devfn];
3379 
3380     if (!vtd_dev_as) {
3381         snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn),
3382                  PCI_FUNC(devfn));
3383         vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace));
3384 
3385         vtd_dev_as->bus = bus;
3386         vtd_dev_as->devfn = (uint8_t)devfn;
3387         vtd_dev_as->iommu_state = s;
3388         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
3389         vtd_dev_as->iova_tree = iova_tree_new();
3390 
3391         memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX);
3392         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root");
3393 
3394         /*
3395          * Build the DMAR-disabled container with aliases to the
3396          * shared MRs.  Note that aliasing to a shared memory region
3397          * could help the memory API to detect same FlatViews so we
3398          * can have devices to share the same FlatView when DMAR is
3399          * disabled (either by not providing "intel_iommu=on" or with
3400          * "iommu=pt").  It will greatly reduce the total number of
3401          * FlatViews of the system hence VM runs faster.
3402          */
3403         memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s),
3404                                  "vtd-nodmar", &s->mr_nodmar, 0,
3405                                  memory_region_size(&s->mr_nodmar));
3406 
3407         /*
3408          * Build the per-device DMAR-enabled container.
3409          *
3410          * TODO: currently we have per-device IOMMU memory region only
3411          * because we have per-device IOMMU notifiers for devices.  If
3412          * one day we can abstract the IOMMU notifiers out of the
3413          * memory regions then we can also share the same memory
3414          * region here just like what we've done above with the nodmar
3415          * region.
3416          */
3417         strcat(name, "-dmar");
3418         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
3419                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
3420                                  name, UINT64_MAX);
3421         memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir",
3422                                  &s->mr_ir, 0, memory_region_size(&s->mr_ir));
3423         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
3424                                             VTD_INTERRUPT_ADDR_FIRST,
3425                                             &vtd_dev_as->iommu_ir, 1);
3426 
3427         /*
3428          * Hook both the containers under the root container, we
3429          * switch between DMAR & noDMAR by enable/disable
3430          * corresponding sub-containers
3431          */
3432         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3433                                             MEMORY_REGION(&vtd_dev_as->iommu),
3434                                             0);
3435         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3436                                             &vtd_dev_as->nodmar, 0);
3437 
3438         vtd_switch_address_space(vtd_dev_as);
3439     }
3440     return vtd_dev_as;
3441 }
3442 
3443 static uint64_t get_naturally_aligned_size(uint64_t start,
3444                                            uint64_t size, int gaw)
3445 {
3446     uint64_t max_mask = 1ULL << gaw;
3447     uint64_t alignment = start ? start & -start : max_mask;
3448 
3449     alignment = MIN(alignment, max_mask);
3450     size = MIN(size, max_mask);
3451 
3452     if (alignment <= size) {
3453         /* Increase the alignment of start */
3454         return alignment;
3455     } else {
3456         /* Find the largest page mask from size */
3457         return 1ULL << (63 - clz64(size));
3458     }
3459 }
3460 
3461 /* Unmap the whole range in the notifier's scope. */
3462 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
3463 {
3464     hwaddr size, remain;
3465     hwaddr start = n->start;
3466     hwaddr end = n->end;
3467     IntelIOMMUState *s = as->iommu_state;
3468     DMAMap map;
3469 
3470     /*
3471      * Note: all the codes in this function has a assumption that IOVA
3472      * bits are no more than VTD_MGAW bits (which is restricted by
3473      * VT-d spec), otherwise we need to consider overflow of 64 bits.
3474      */
3475 
3476     if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) {
3477         /*
3478          * Don't need to unmap regions that is bigger than the whole
3479          * VT-d supported address space size
3480          */
3481         end = VTD_ADDRESS_SIZE(s->aw_bits) - 1;
3482     }
3483 
3484     assert(start <= end);
3485     size = remain = end - start + 1;
3486 
3487     while (remain >= VTD_PAGE_SIZE) {
3488         IOMMUTLBEntry entry;
3489         uint64_t mask = get_naturally_aligned_size(start, remain, s->aw_bits);
3490 
3491         assert(mask);
3492 
3493         entry.iova = start;
3494         entry.addr_mask = mask - 1;
3495         entry.target_as = &address_space_memory;
3496         entry.perm = IOMMU_NONE;
3497         /* This field is meaningless for unmap */
3498         entry.translated_addr = 0;
3499 
3500         memory_region_notify_one(n, &entry);
3501 
3502         start += mask;
3503         remain -= mask;
3504     }
3505 
3506     assert(!remain);
3507 
3508     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3509                              VTD_PCI_SLOT(as->devfn),
3510                              VTD_PCI_FUNC(as->devfn),
3511                              n->start, size);
3512 
3513     map.iova = n->start;
3514     map.size = size;
3515     iova_tree_remove(as->iova_tree, &map);
3516 }
3517 
3518 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3519 {
3520     VTDAddressSpace *vtd_as;
3521     IOMMUNotifier *n;
3522 
3523     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3524         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3525             vtd_address_space_unmap(vtd_as, n);
3526         }
3527     }
3528 }
3529 
3530 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3531 {
3532     vtd_address_space_unmap_all(s);
3533     vtd_switch_address_space_all(s);
3534 }
3535 
3536 static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private)
3537 {
3538     memory_region_notify_one((IOMMUNotifier *)private, entry);
3539     return 0;
3540 }
3541 
3542 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3543 {
3544     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3545     IntelIOMMUState *s = vtd_as->iommu_state;
3546     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3547     VTDContextEntry ce;
3548 
3549     /*
3550      * The replay can be triggered by either a invalidation or a newly
3551      * created entry. No matter what, we release existing mappings
3552      * (it means flushing caches for UNMAP-only registers).
3553      */
3554     vtd_address_space_unmap(vtd_as, n);
3555 
3556     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3557         trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" :
3558                                   "legacy mode",
3559                                   bus_n, PCI_SLOT(vtd_as->devfn),
3560                                   PCI_FUNC(vtd_as->devfn),
3561                                   vtd_get_domain_id(s, &ce),
3562                                   ce.hi, ce.lo);
3563         if (vtd_as_has_map_notifier(vtd_as)) {
3564             /* This is required only for MAP typed notifiers */
3565             vtd_page_walk_info info = {
3566                 .hook_fn = vtd_replay_hook,
3567                 .private = (void *)n,
3568                 .notify_unmap = false,
3569                 .aw = s->aw_bits,
3570                 .as = vtd_as,
3571                 .domain_id = vtd_get_domain_id(s, &ce),
3572             };
3573 
3574             vtd_page_walk(s, &ce, 0, ~0ULL, &info);
3575         }
3576     } else {
3577         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3578                                     PCI_FUNC(vtd_as->devfn));
3579     }
3580 
3581     return;
3582 }
3583 
3584 /* Do the initialization. It will also be called when reset, so pay
3585  * attention when adding new initialization stuff.
3586  */
3587 static void vtd_init(IntelIOMMUState *s)
3588 {
3589     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3590 
3591     memset(s->csr, 0, DMAR_REG_SIZE);
3592     memset(s->wmask, 0, DMAR_REG_SIZE);
3593     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3594     memset(s->womask, 0, DMAR_REG_SIZE);
3595 
3596     s->root = 0;
3597     s->root_scalable = false;
3598     s->dmar_enabled = false;
3599     s->intr_enabled = false;
3600     s->iq_head = 0;
3601     s->iq_tail = 0;
3602     s->iq = 0;
3603     s->iq_size = 0;
3604     s->qi_enabled = false;
3605     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3606     s->iq_dw = false;
3607     s->next_frcd_reg = 0;
3608     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3609              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3610              VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits);
3611     if (s->dma_drain) {
3612         s->cap |= VTD_CAP_DRAIN;
3613     }
3614     if (s->aw_bits == VTD_HOST_AW_48BIT) {
3615         s->cap |= VTD_CAP_SAGAW_48bit;
3616     }
3617     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3618 
3619     /*
3620      * Rsvd field masks for spte
3621      */
3622     vtd_spte_rsvd[0] = ~0ULL;
3623     vtd_spte_rsvd[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits,
3624                                                   x86_iommu->dt_supported);
3625     vtd_spte_rsvd[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3626     vtd_spte_rsvd[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3627     vtd_spte_rsvd[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3628 
3629     vtd_spte_rsvd_large[2] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits,
3630                                                          x86_iommu->dt_supported);
3631     vtd_spte_rsvd_large[3] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits,
3632                                                          x86_iommu->dt_supported);
3633 
3634     if (x86_iommu_ir_supported(x86_iommu)) {
3635         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3636         if (s->intr_eim == ON_OFF_AUTO_ON) {
3637             s->ecap |= VTD_ECAP_EIM;
3638         }
3639         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3640     }
3641 
3642     if (x86_iommu->dt_supported) {
3643         s->ecap |= VTD_ECAP_DT;
3644     }
3645 
3646     if (x86_iommu->pt_supported) {
3647         s->ecap |= VTD_ECAP_PT;
3648     }
3649 
3650     if (s->caching_mode) {
3651         s->cap |= VTD_CAP_CM;
3652     }
3653 
3654     /* TODO: read cap/ecap from host to decide which cap to be exposed. */
3655     if (s->scalable_mode) {
3656         s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS;
3657     }
3658 
3659     vtd_reset_caches(s);
3660 
3661     /* Define registers with default values and bit semantics */
3662     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
3663     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
3664     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
3665     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
3666     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
3667     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
3668     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0);
3669     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
3670     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
3671 
3672     /* Advanced Fault Logging not supported */
3673     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
3674     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3675     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
3676     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
3677 
3678     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3679      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3680      */
3681     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
3682 
3683     /* Treated as RO for implementations that PLMR and PHMR fields reported
3684      * as Clear in the CAP_REG.
3685      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3686      */
3687     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
3688 
3689     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
3690     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
3691     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0);
3692     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
3693     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3694     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
3695     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
3696     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3697     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
3698 
3699     /* IOTLB registers */
3700     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
3701     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
3702     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
3703 
3704     /* Fault Recording Registers, 128-bit */
3705     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
3706     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
3707 
3708     /*
3709      * Interrupt remapping registers.
3710      */
3711     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
3712 }
3713 
3714 /* Should not reset address_spaces when reset because devices will still use
3715  * the address space they got at first (won't ask the bus again).
3716  */
3717 static void vtd_reset(DeviceState *dev)
3718 {
3719     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3720 
3721     vtd_init(s);
3722     vtd_address_space_refresh_all(s);
3723 }
3724 
3725 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
3726 {
3727     IntelIOMMUState *s = opaque;
3728     VTDAddressSpace *vtd_as;
3729 
3730     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
3731 
3732     vtd_as = vtd_find_add_as(s, bus, devfn);
3733     return &vtd_as->as;
3734 }
3735 
3736 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
3737 {
3738     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3739 
3740     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) {
3741         error_setg(errp, "eim=on cannot be selected without intremap=on");
3742         return false;
3743     }
3744 
3745     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
3746         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
3747                       && x86_iommu_ir_supported(x86_iommu) ?
3748                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
3749     }
3750     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
3751         if (!kvm_irqchip_in_kernel()) {
3752             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
3753             return false;
3754         }
3755         if (!kvm_enable_x2apic()) {
3756             error_setg(errp, "eim=on requires support on the KVM side"
3757                              "(X2APIC_API, first shipped in v4.7)");
3758             return false;
3759         }
3760     }
3761 
3762     /* Currently only address widths supported are 39 and 48 bits */
3763     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
3764         (s->aw_bits != VTD_HOST_AW_48BIT)) {
3765         error_setg(errp, "Supported values for aw-bits are: %d, %d",
3766                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
3767         return false;
3768     }
3769 
3770     if (s->scalable_mode && !s->dma_drain) {
3771         error_setg(errp, "Need to set dma_drain for scalable mode");
3772         return false;
3773     }
3774 
3775     return true;
3776 }
3777 
3778 static int vtd_machine_done_notify_one(Object *child, void *unused)
3779 {
3780     IntelIOMMUState *iommu = INTEL_IOMMU_DEVICE(x86_iommu_get_default());
3781 
3782     /*
3783      * We hard-coded here because vfio-pci is the only special case
3784      * here.  Let's be more elegant in the future when we can, but so
3785      * far there seems to be no better way.
3786      */
3787     if (object_dynamic_cast(child, "vfio-pci") && !iommu->caching_mode) {
3788         vtd_panic_require_caching_mode();
3789     }
3790 
3791     return 0;
3792 }
3793 
3794 static void vtd_machine_done_hook(Notifier *notifier, void *unused)
3795 {
3796     object_child_foreach_recursive(object_get_root(),
3797                                    vtd_machine_done_notify_one, NULL);
3798 }
3799 
3800 static Notifier vtd_machine_done_notify = {
3801     .notify = vtd_machine_done_hook,
3802 };
3803 
3804 static void vtd_realize(DeviceState *dev, Error **errp)
3805 {
3806     MachineState *ms = MACHINE(qdev_get_machine());
3807     PCMachineState *pcms = PC_MACHINE(ms);
3808     X86MachineState *x86ms = X86_MACHINE(ms);
3809     PCIBus *bus = pcms->bus;
3810     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3811     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
3812 
3813     x86_iommu->type = TYPE_INTEL;
3814 
3815     if (!vtd_decide_config(s, errp)) {
3816         return;
3817     }
3818 
3819     QLIST_INIT(&s->vtd_as_with_notifiers);
3820     qemu_mutex_init(&s->iommu_lock);
3821     memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num));
3822     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
3823                           "intel_iommu", DMAR_REG_SIZE);
3824 
3825     /* Create the shared memory regions by all devices */
3826     memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar",
3827                        UINT64_MAX);
3828     memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops,
3829                           s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE);
3830     memory_region_init_alias(&s->mr_sys_alias, OBJECT(s),
3831                              "vtd-sys-alias", get_system_memory(), 0,
3832                              memory_region_size(get_system_memory()));
3833     memory_region_add_subregion_overlap(&s->mr_nodmar, 0,
3834                                         &s->mr_sys_alias, 0);
3835     memory_region_add_subregion_overlap(&s->mr_nodmar,
3836                                         VTD_INTERRUPT_ADDR_FIRST,
3837                                         &s->mr_ir, 1);
3838 
3839     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
3840     /* No corresponding destroy */
3841     s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3842                                      g_free, g_free);
3843     s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3844                                               g_free, g_free);
3845     vtd_init(s);
3846     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
3847     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
3848     /* Pseudo address space under root PCI bus. */
3849     x86ms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
3850     qemu_add_machine_init_done_notifier(&vtd_machine_done_notify);
3851 }
3852 
3853 static void vtd_class_init(ObjectClass *klass, void *data)
3854 {
3855     DeviceClass *dc = DEVICE_CLASS(klass);
3856     X86IOMMUClass *x86_class = X86_IOMMU_DEVICE_CLASS(klass);
3857 
3858     dc->reset = vtd_reset;
3859     dc->vmsd = &vtd_vmstate;
3860     device_class_set_props(dc, vtd_properties);
3861     dc->hotpluggable = false;
3862     x86_class->realize = vtd_realize;
3863     x86_class->int_remap = vtd_int_remap;
3864     /* Supported by the pc-q35-* machine types */
3865     dc->user_creatable = true;
3866     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
3867     dc->desc = "Intel IOMMU (VT-d) DMA Remapping device";
3868 }
3869 
3870 static const TypeInfo vtd_info = {
3871     .name          = TYPE_INTEL_IOMMU_DEVICE,
3872     .parent        = TYPE_X86_IOMMU_DEVICE,
3873     .instance_size = sizeof(IntelIOMMUState),
3874     .class_init    = vtd_class_init,
3875 };
3876 
3877 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
3878                                                      void *data)
3879 {
3880     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
3881 
3882     imrc->translate = vtd_iommu_translate;
3883     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
3884     imrc->replay = vtd_iommu_replay;
3885 }
3886 
3887 static const TypeInfo vtd_iommu_memory_region_info = {
3888     .parent = TYPE_IOMMU_MEMORY_REGION,
3889     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
3890     .class_init = vtd_iommu_memory_region_class_init,
3891 };
3892 
3893 static void vtd_register_types(void)
3894 {
3895     type_register_static(&vtd_info);
3896     type_register_static(&vtd_iommu_memory_region_info);
3897 }
3898 
3899 type_init(vtd_register_types)
3900