xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision a27bd6c7)
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 "hw/i386/apic_internal.h"
39 #include "kvm_i386.h"
40 #include "migration/vmstate.h"
41 #include "trace.h"
42 
43 /* context entry operations */
44 #define VTD_CE_GET_RID2PASID(ce) \
45     ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK)
46 #define VTD_CE_GET_PASID_DIR_TABLE(ce) \
47     ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK)
48 
49 /* pe operations */
50 #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT)
51 #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW))
52 #define VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write) {\
53     if (ret_fr) {                                                             \
54         ret_fr = -ret_fr;                                                     \
55         if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) {                   \
56             trace_vtd_fault_disabled();                                       \
57         } else {                                                              \
58             vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);      \
59         }                                                                     \
60         goto error;                                                           \
61     }                                                                         \
62 }
63 
64 static void vtd_address_space_refresh_all(IntelIOMMUState *s);
65 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n);
66 
67 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val,
68                             uint64_t wmask, uint64_t w1cmask)
69 {
70     stq_le_p(&s->csr[addr], val);
71     stq_le_p(&s->wmask[addr], wmask);
72     stq_le_p(&s->w1cmask[addr], w1cmask);
73 }
74 
75 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask)
76 {
77     stq_le_p(&s->womask[addr], mask);
78 }
79 
80 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val,
81                             uint32_t wmask, uint32_t w1cmask)
82 {
83     stl_le_p(&s->csr[addr], val);
84     stl_le_p(&s->wmask[addr], wmask);
85     stl_le_p(&s->w1cmask[addr], w1cmask);
86 }
87 
88 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask)
89 {
90     stl_le_p(&s->womask[addr], mask);
91 }
92 
93 /* "External" get/set operations */
94 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
95 {
96     uint64_t oldval = ldq_le_p(&s->csr[addr]);
97     uint64_t wmask = ldq_le_p(&s->wmask[addr]);
98     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
99     stq_le_p(&s->csr[addr],
100              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
101 }
102 
103 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
104 {
105     uint32_t oldval = ldl_le_p(&s->csr[addr]);
106     uint32_t wmask = ldl_le_p(&s->wmask[addr]);
107     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
108     stl_le_p(&s->csr[addr],
109              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
110 }
111 
112 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr)
113 {
114     uint64_t val = ldq_le_p(&s->csr[addr]);
115     uint64_t womask = ldq_le_p(&s->womask[addr]);
116     return val & ~womask;
117 }
118 
119 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr)
120 {
121     uint32_t val = ldl_le_p(&s->csr[addr]);
122     uint32_t womask = ldl_le_p(&s->womask[addr]);
123     return val & ~womask;
124 }
125 
126 /* "Internal" get/set operations */
127 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr)
128 {
129     return ldq_le_p(&s->csr[addr]);
130 }
131 
132 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr)
133 {
134     return ldl_le_p(&s->csr[addr]);
135 }
136 
137 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val)
138 {
139     stq_le_p(&s->csr[addr], val);
140 }
141 
142 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
143                                         uint32_t clear, uint32_t mask)
144 {
145     uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
146     stl_le_p(&s->csr[addr], new_val);
147     return new_val;
148 }
149 
150 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
151                                         uint64_t clear, uint64_t mask)
152 {
153     uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
154     stq_le_p(&s->csr[addr], new_val);
155     return new_val;
156 }
157 
158 static inline void vtd_iommu_lock(IntelIOMMUState *s)
159 {
160     qemu_mutex_lock(&s->iommu_lock);
161 }
162 
163 static inline void vtd_iommu_unlock(IntelIOMMUState *s)
164 {
165     qemu_mutex_unlock(&s->iommu_lock);
166 }
167 
168 static void vtd_update_scalable_state(IntelIOMMUState *s)
169 {
170     uint64_t val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
171 
172     if (s->scalable_mode) {
173         s->root_scalable = val & VTD_RTADDR_SMT;
174     }
175 }
176 
177 /* Whether the address space needs to notify new mappings */
178 static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as)
179 {
180     return as->notifier_flags & IOMMU_NOTIFIER_MAP;
181 }
182 
183 /* GHashTable functions */
184 static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2)
185 {
186     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
187 }
188 
189 static guint vtd_uint64_hash(gconstpointer v)
190 {
191     return (guint)*(const uint64_t *)v;
192 }
193 
194 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value,
195                                           gpointer user_data)
196 {
197     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
198     uint16_t domain_id = *(uint16_t *)user_data;
199     return entry->domain_id == domain_id;
200 }
201 
202 /* The shift of an addr for a certain level of paging structure */
203 static inline uint32_t vtd_slpt_level_shift(uint32_t level)
204 {
205     assert(level != 0);
206     return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
207 }
208 
209 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level)
210 {
211     return ~((1ULL << vtd_slpt_level_shift(level)) - 1);
212 }
213 
214 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value,
215                                         gpointer user_data)
216 {
217     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
218     VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data;
219     uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask;
220     uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K;
221     return (entry->domain_id == info->domain_id) &&
222             (((entry->gfn & info->mask) == gfn) ||
223              (entry->gfn == gfn_tlb));
224 }
225 
226 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
227  * IntelIOMMUState to 1.  Must be called with IOMMU lock held.
228  */
229 static void vtd_reset_context_cache_locked(IntelIOMMUState *s)
230 {
231     VTDAddressSpace *vtd_as;
232     VTDBus *vtd_bus;
233     GHashTableIter bus_it;
234     uint32_t devfn_it;
235 
236     trace_vtd_context_cache_reset();
237 
238     g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr);
239 
240     while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) {
241         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
242             vtd_as = vtd_bus->dev_as[devfn_it];
243             if (!vtd_as) {
244                 continue;
245             }
246             vtd_as->context_cache_entry.context_cache_gen = 0;
247         }
248     }
249     s->context_cache_gen = 1;
250 }
251 
252 /* Must be called with IOMMU lock held. */
253 static void vtd_reset_iotlb_locked(IntelIOMMUState *s)
254 {
255     assert(s->iotlb);
256     g_hash_table_remove_all(s->iotlb);
257 }
258 
259 static void vtd_reset_iotlb(IntelIOMMUState *s)
260 {
261     vtd_iommu_lock(s);
262     vtd_reset_iotlb_locked(s);
263     vtd_iommu_unlock(s);
264 }
265 
266 static void vtd_reset_caches(IntelIOMMUState *s)
267 {
268     vtd_iommu_lock(s);
269     vtd_reset_iotlb_locked(s);
270     vtd_reset_context_cache_locked(s);
271     vtd_iommu_unlock(s);
272 }
273 
274 static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id,
275                                   uint32_t level)
276 {
277     return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) |
278            ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT);
279 }
280 
281 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level)
282 {
283     return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K;
284 }
285 
286 /* Must be called with IOMMU lock held */
287 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id,
288                                        hwaddr addr)
289 {
290     VTDIOTLBEntry *entry;
291     uint64_t key;
292     int level;
293 
294     for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) {
295         key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level),
296                                 source_id, level);
297         entry = g_hash_table_lookup(s->iotlb, &key);
298         if (entry) {
299             goto out;
300         }
301     }
302 
303 out:
304     return entry;
305 }
306 
307 /* Must be with IOMMU lock held */
308 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id,
309                              uint16_t domain_id, hwaddr addr, uint64_t slpte,
310                              uint8_t access_flags, uint32_t level)
311 {
312     VTDIOTLBEntry *entry = g_malloc(sizeof(*entry));
313     uint64_t *key = g_malloc(sizeof(*key));
314     uint64_t gfn = vtd_get_iotlb_gfn(addr, level);
315 
316     trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id);
317     if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) {
318         trace_vtd_iotlb_reset("iotlb exceeds size limit");
319         vtd_reset_iotlb_locked(s);
320     }
321 
322     entry->gfn = gfn;
323     entry->domain_id = domain_id;
324     entry->slpte = slpte;
325     entry->access_flags = access_flags;
326     entry->mask = vtd_slpt_level_page_mask(level);
327     *key = vtd_get_iotlb_key(gfn, source_id, level);
328     g_hash_table_replace(s->iotlb, key, entry);
329 }
330 
331 /* Given the reg addr of both the message data and address, generate an
332  * interrupt via MSI.
333  */
334 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
335                                    hwaddr mesg_data_reg)
336 {
337     MSIMessage msi;
338 
339     assert(mesg_data_reg < DMAR_REG_SIZE);
340     assert(mesg_addr_reg < DMAR_REG_SIZE);
341 
342     msi.address = vtd_get_long_raw(s, mesg_addr_reg);
343     msi.data = vtd_get_long_raw(s, mesg_data_reg);
344 
345     trace_vtd_irq_generate(msi.address, msi.data);
346 
347     apic_get_class()->send_msi(&msi);
348 }
349 
350 /* Generate a fault event to software via MSI if conditions are met.
351  * Notice that the value of FSTS_REG being passed to it should be the one
352  * before any update.
353  */
354 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
355 {
356     if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
357         pre_fsts & VTD_FSTS_IQE) {
358         error_report_once("There are previous interrupt conditions "
359                           "to be serviced by software, fault event "
360                           "is not generated");
361         return;
362     }
363     vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
364     if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
365         error_report_once("Interrupt Mask set, irq is not generated");
366     } else {
367         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
368         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
369     }
370 }
371 
372 /* Check if the Fault (F) field of the Fault Recording Register referenced by
373  * @index is Set.
374  */
375 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index)
376 {
377     /* Each reg is 128-bit */
378     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
379     addr += 8; /* Access the high 64-bit half */
380 
381     assert(index < DMAR_FRCD_REG_NR);
382 
383     return vtd_get_quad_raw(s, addr) & VTD_FRCD_F;
384 }
385 
386 /* Update the PPF field of Fault Status Register.
387  * Should be called whenever change the F field of any fault recording
388  * registers.
389  */
390 static void vtd_update_fsts_ppf(IntelIOMMUState *s)
391 {
392     uint32_t i;
393     uint32_t ppf_mask = 0;
394 
395     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
396         if (vtd_is_frcd_set(s, i)) {
397             ppf_mask = VTD_FSTS_PPF;
398             break;
399         }
400     }
401     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
402     trace_vtd_fsts_ppf(!!ppf_mask);
403 }
404 
405 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index)
406 {
407     /* Each reg is 128-bit */
408     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
409     addr += 8; /* Access the high 64-bit half */
410 
411     assert(index < DMAR_FRCD_REG_NR);
412 
413     vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
414     vtd_update_fsts_ppf(s);
415 }
416 
417 /* Must not update F field now, should be done later */
418 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index,
419                             uint16_t source_id, hwaddr addr,
420                             VTDFaultReason fault, bool is_write)
421 {
422     uint64_t hi = 0, lo;
423     hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
424 
425     assert(index < DMAR_FRCD_REG_NR);
426 
427     lo = VTD_FRCD_FI(addr);
428     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
429     if (!is_write) {
430         hi |= VTD_FRCD_T;
431     }
432     vtd_set_quad_raw(s, frcd_reg_addr, lo);
433     vtd_set_quad_raw(s, frcd_reg_addr + 8, hi);
434 
435     trace_vtd_frr_new(index, hi, lo);
436 }
437 
438 /* Try to collapse multiple pending faults from the same requester */
439 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id)
440 {
441     uint32_t i;
442     uint64_t frcd_reg;
443     hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
444 
445     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
446         frcd_reg = vtd_get_quad_raw(s, addr);
447         if ((frcd_reg & VTD_FRCD_F) &&
448             ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
449             return true;
450         }
451         addr += 16; /* 128-bit for each */
452     }
453     return false;
454 }
455 
456 /* Log and report an DMAR (address translation) fault to software */
457 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
458                                   hwaddr addr, VTDFaultReason fault,
459                                   bool is_write)
460 {
461     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
462 
463     assert(fault < VTD_FR_MAX);
464 
465     if (fault == VTD_FR_RESERVED_ERR) {
466         /* This is not a normal fault reason case. Drop it. */
467         return;
468     }
469 
470     trace_vtd_dmar_fault(source_id, fault, addr, is_write);
471 
472     if (fsts_reg & VTD_FSTS_PFO) {
473         error_report_once("New fault is not recorded due to "
474                           "Primary Fault Overflow");
475         return;
476     }
477 
478     if (vtd_try_collapse_fault(s, source_id)) {
479         error_report_once("New fault is not recorded due to "
480                           "compression of faults");
481         return;
482     }
483 
484     if (vtd_is_frcd_set(s, s->next_frcd_reg)) {
485         error_report_once("Next Fault Recording Reg is used, "
486                           "new fault is not recorded, set PFO field");
487         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
488         return;
489     }
490 
491     vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write);
492 
493     if (fsts_reg & VTD_FSTS_PPF) {
494         error_report_once("There are pending faults already, "
495                           "fault event is not generated");
496         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg);
497         s->next_frcd_reg++;
498         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
499             s->next_frcd_reg = 0;
500         }
501     } else {
502         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
503                                 VTD_FSTS_FRI(s->next_frcd_reg));
504         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */
505         s->next_frcd_reg++;
506         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
507             s->next_frcd_reg = 0;
508         }
509         /* This case actually cause the PPF to be Set.
510          * So generate fault event (interrupt).
511          */
512          vtd_generate_fault_event(s, fsts_reg);
513     }
514 }
515 
516 /* Handle Invalidation Queue Errors of queued invalidation interface error
517  * conditions.
518  */
519 static void vtd_handle_inv_queue_error(IntelIOMMUState *s)
520 {
521     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
522 
523     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE);
524     vtd_generate_fault_event(s, fsts_reg);
525 }
526 
527 /* Set the IWC field and try to generate an invalidation completion interrupt */
528 static void vtd_generate_completion_event(IntelIOMMUState *s)
529 {
530     if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) {
531         trace_vtd_inv_desc_wait_irq("One pending, skip current");
532         return;
533     }
534     vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC);
535     vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP);
536     if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) {
537         trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
538                                     "new event not generated");
539         return;
540     } else {
541         /* Generate the interrupt event */
542         trace_vtd_inv_desc_wait_irq("Generating complete event");
543         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
544         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
545     }
546 }
547 
548 static inline bool vtd_root_entry_present(IntelIOMMUState *s,
549                                           VTDRootEntry *re,
550                                           uint8_t devfn)
551 {
552     if (s->root_scalable && devfn > UINT8_MAX / 2) {
553         return re->hi & VTD_ROOT_ENTRY_P;
554     }
555 
556     return re->lo & VTD_ROOT_ENTRY_P;
557 }
558 
559 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index,
560                               VTDRootEntry *re)
561 {
562     dma_addr_t addr;
563 
564     addr = s->root + index * sizeof(*re);
565     if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) {
566         re->lo = 0;
567         return -VTD_FR_ROOT_TABLE_INV;
568     }
569     re->lo = le64_to_cpu(re->lo);
570     re->hi = le64_to_cpu(re->hi);
571     return 0;
572 }
573 
574 static inline bool vtd_ce_present(VTDContextEntry *context)
575 {
576     return context->lo & VTD_CONTEXT_ENTRY_P;
577 }
578 
579 static int vtd_get_context_entry_from_root(IntelIOMMUState *s,
580                                            VTDRootEntry *re,
581                                            uint8_t index,
582                                            VTDContextEntry *ce)
583 {
584     dma_addr_t addr, ce_size;
585 
586     /* we have checked that root entry is present */
587     ce_size = s->root_scalable ? VTD_CTX_ENTRY_SCALABLE_SIZE :
588               VTD_CTX_ENTRY_LEGACY_SIZE;
589 
590     if (s->root_scalable && index > UINT8_MAX / 2) {
591         index = index & (~VTD_DEVFN_CHECK_MASK);
592         addr = re->hi & VTD_ROOT_ENTRY_CTP;
593     } else {
594         addr = re->lo & VTD_ROOT_ENTRY_CTP;
595     }
596 
597     addr = addr + index * ce_size;
598     if (dma_memory_read(&address_space_memory, addr, ce, ce_size)) {
599         return -VTD_FR_CONTEXT_TABLE_INV;
600     }
601 
602     ce->lo = le64_to_cpu(ce->lo);
603     ce->hi = le64_to_cpu(ce->hi);
604     if (ce_size == VTD_CTX_ENTRY_SCALABLE_SIZE) {
605         ce->val[2] = le64_to_cpu(ce->val[2]);
606         ce->val[3] = le64_to_cpu(ce->val[3]);
607     }
608     return 0;
609 }
610 
611 static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce)
612 {
613     return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
614 }
615 
616 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw)
617 {
618     return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw);
619 }
620 
621 /* Whether the pte indicates the address of the page frame */
622 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level)
623 {
624     return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
625 }
626 
627 /* Get the content of a spte located in @base_addr[@index] */
628 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index)
629 {
630     uint64_t slpte;
631 
632     assert(index < VTD_SL_PT_ENTRY_NR);
633 
634     if (dma_memory_read(&address_space_memory,
635                         base_addr + index * sizeof(slpte), &slpte,
636                         sizeof(slpte))) {
637         slpte = (uint64_t)-1;
638         return slpte;
639     }
640     slpte = le64_to_cpu(slpte);
641     return slpte;
642 }
643 
644 /* Given an iova and the level of paging structure, return the offset
645  * of current level.
646  */
647 static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level)
648 {
649     return (iova >> vtd_slpt_level_shift(level)) &
650             ((1ULL << VTD_SL_LEVEL_BITS) - 1);
651 }
652 
653 /* Check Capability Register to see if the @level of page-table is supported */
654 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level)
655 {
656     return VTD_CAP_SAGAW_MASK & s->cap &
657            (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
658 }
659 
660 /* Return true if check passed, otherwise false */
661 static inline bool vtd_pe_type_check(X86IOMMUState *x86_iommu,
662                                      VTDPASIDEntry *pe)
663 {
664     switch (VTD_PE_GET_TYPE(pe)) {
665     case VTD_SM_PASID_ENTRY_FLT:
666     case VTD_SM_PASID_ENTRY_SLT:
667     case VTD_SM_PASID_ENTRY_NESTED:
668         break;
669     case VTD_SM_PASID_ENTRY_PT:
670         if (!x86_iommu->pt_supported) {
671             return false;
672         }
673         break;
674     default:
675         /* Unknwon type */
676         return false;
677     }
678     return true;
679 }
680 
681 static int vtd_get_pasid_dire(dma_addr_t pasid_dir_base,
682                               uint32_t pasid,
683                               VTDPASIDDirEntry *pdire)
684 {
685     uint32_t index;
686     dma_addr_t addr, entry_size;
687 
688     index = VTD_PASID_DIR_INDEX(pasid);
689     entry_size = VTD_PASID_DIR_ENTRY_SIZE;
690     addr = pasid_dir_base + index * entry_size;
691     if (dma_memory_read(&address_space_memory, addr, pdire, entry_size)) {
692         return -VTD_FR_PASID_TABLE_INV;
693     }
694 
695     return 0;
696 }
697 
698 static int vtd_get_pasid_entry(IntelIOMMUState *s,
699                                uint32_t pasid,
700                                VTDPASIDDirEntry *pdire,
701                                VTDPASIDEntry *pe)
702 {
703     uint32_t index;
704     dma_addr_t addr, entry_size;
705     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
706 
707     index = VTD_PASID_TABLE_INDEX(pasid);
708     entry_size = VTD_PASID_ENTRY_SIZE;
709     addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK;
710     addr = addr + index * entry_size;
711     if (dma_memory_read(&address_space_memory, addr, pe, entry_size)) {
712         return -VTD_FR_PASID_TABLE_INV;
713     }
714 
715     /* Do translation type check */
716     if (!vtd_pe_type_check(x86_iommu, pe)) {
717         return -VTD_FR_PASID_TABLE_INV;
718     }
719 
720     if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) {
721         return -VTD_FR_PASID_TABLE_INV;
722     }
723 
724     return 0;
725 }
726 
727 static int vtd_get_pasid_entry_from_pasid(IntelIOMMUState *s,
728                                           dma_addr_t pasid_dir_base,
729                                           uint32_t pasid,
730                                           VTDPASIDEntry *pe)
731 {
732     int ret;
733     VTDPASIDDirEntry pdire;
734 
735     ret = vtd_get_pasid_dire(pasid_dir_base, pasid, &pdire);
736     if (ret) {
737         return ret;
738     }
739 
740     ret = vtd_get_pasid_entry(s, pasid, &pdire, pe);
741     if (ret) {
742         return ret;
743     }
744 
745     return ret;
746 }
747 
748 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s,
749                                       VTDContextEntry *ce,
750                                       VTDPASIDEntry *pe)
751 {
752     uint32_t pasid;
753     dma_addr_t pasid_dir_base;
754     int ret = 0;
755 
756     pasid = VTD_CE_GET_RID2PASID(ce);
757     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
758     ret = vtd_get_pasid_entry_from_pasid(s, pasid_dir_base, pasid, pe);
759 
760     return ret;
761 }
762 
763 static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s,
764                                 VTDContextEntry *ce,
765                                 bool *pe_fpd_set)
766 {
767     int ret;
768     uint32_t pasid;
769     dma_addr_t pasid_dir_base;
770     VTDPASIDDirEntry pdire;
771     VTDPASIDEntry pe;
772 
773     pasid = VTD_CE_GET_RID2PASID(ce);
774     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
775 
776     ret = vtd_get_pasid_dire(pasid_dir_base, pasid, &pdire);
777     if (ret) {
778         return ret;
779     }
780 
781     if (pdire.val & VTD_PASID_DIR_FPD) {
782         *pe_fpd_set = true;
783         return 0;
784     }
785 
786     ret = vtd_get_pasid_entry(s, pasid, &pdire, &pe);
787     if (ret) {
788         return ret;
789     }
790 
791     if (pe.val[0] & VTD_PASID_ENTRY_FPD) {
792         *pe_fpd_set = true;
793     }
794 
795     return 0;
796 }
797 
798 /* Get the page-table level that hardware should use for the second-level
799  * page-table walk from the Address Width field of context-entry.
800  */
801 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
802 {
803     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
804 }
805 
806 static uint32_t vtd_get_iova_level(IntelIOMMUState *s,
807                                    VTDContextEntry *ce)
808 {
809     VTDPASIDEntry pe;
810 
811     if (s->root_scalable) {
812         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
813         return VTD_PE_GET_LEVEL(&pe);
814     }
815 
816     return vtd_ce_get_level(ce);
817 }
818 
819 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
820 {
821     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
822 }
823 
824 static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s,
825                                   VTDContextEntry *ce)
826 {
827     VTDPASIDEntry pe;
828 
829     if (s->root_scalable) {
830         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
831         return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9;
832     }
833 
834     return vtd_ce_get_agaw(ce);
835 }
836 
837 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
838 {
839     return ce->lo & VTD_CONTEXT_ENTRY_TT;
840 }
841 
842 /* Only for Legacy Mode. Return true if check passed, otherwise false */
843 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
844                                      VTDContextEntry *ce)
845 {
846     switch (vtd_ce_get_type(ce)) {
847     case VTD_CONTEXT_TT_MULTI_LEVEL:
848         /* Always supported */
849         break;
850     case VTD_CONTEXT_TT_DEV_IOTLB:
851         if (!x86_iommu->dt_supported) {
852             error_report_once("%s: DT specified but not supported", __func__);
853             return false;
854         }
855         break;
856     case VTD_CONTEXT_TT_PASS_THROUGH:
857         if (!x86_iommu->pt_supported) {
858             error_report_once("%s: PT specified but not supported", __func__);
859             return false;
860         }
861         break;
862     default:
863         /* Unknown type */
864         error_report_once("%s: unknown ce type: %"PRIu32, __func__,
865                           vtd_ce_get_type(ce));
866         return false;
867     }
868     return true;
869 }
870 
871 static inline uint64_t vtd_iova_limit(IntelIOMMUState *s,
872                                       VTDContextEntry *ce, uint8_t aw)
873 {
874     uint32_t ce_agaw = vtd_get_iova_agaw(s, ce);
875     return 1ULL << MIN(ce_agaw, aw);
876 }
877 
878 /* Return true if IOVA passes range check, otherwise false. */
879 static inline bool vtd_iova_range_check(IntelIOMMUState *s,
880                                         uint64_t iova, VTDContextEntry *ce,
881                                         uint8_t aw)
882 {
883     /*
884      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
885      * in CAP_REG and AW in context-entry.
886      */
887     return !(iova & ~(vtd_iova_limit(s, ce, aw) - 1));
888 }
889 
890 static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s,
891                                           VTDContextEntry *ce)
892 {
893     VTDPASIDEntry pe;
894 
895     if (s->root_scalable) {
896         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
897         return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR;
898     }
899 
900     return vtd_ce_get_slpt_base(ce);
901 }
902 
903 /*
904  * Rsvd field masks for spte:
905  *     Index [1] to [4] 4k pages
906  *     Index [5] to [8] large pages
907  */
908 static uint64_t vtd_paging_entry_rsvd_field[9];
909 
910 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
911 {
912     if (slpte & VTD_SL_PT_PAGE_SIZE_MASK) {
913         /* Maybe large page */
914         return slpte & vtd_paging_entry_rsvd_field[level + 4];
915     } else {
916         return slpte & vtd_paging_entry_rsvd_field[level];
917     }
918 }
919 
920 /* Find the VTD address space associated with a given bus number */
921 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num)
922 {
923     VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num];
924     if (!vtd_bus) {
925         /*
926          * Iterate over the registered buses to find the one which
927          * currently hold this bus number, and update the bus_num
928          * lookup table:
929          */
930         GHashTableIter iter;
931 
932         g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
933         while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
934             if (pci_bus_num(vtd_bus->bus) == bus_num) {
935                 s->vtd_as_by_bus_num[bus_num] = vtd_bus;
936                 return vtd_bus;
937             }
938         }
939     }
940     return vtd_bus;
941 }
942 
943 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
944  * of the translation, can be used for deciding the size of large page.
945  */
946 static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce,
947                              uint64_t iova, bool is_write,
948                              uint64_t *slptep, uint32_t *slpte_level,
949                              bool *reads, bool *writes, uint8_t aw_bits)
950 {
951     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
952     uint32_t level = vtd_get_iova_level(s, ce);
953     uint32_t offset;
954     uint64_t slpte;
955     uint64_t access_right_check;
956 
957     if (!vtd_iova_range_check(s, iova, ce, aw_bits)) {
958         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")",
959                           __func__, iova);
960         return -VTD_FR_ADDR_BEYOND_MGAW;
961     }
962 
963     /* FIXME: what is the Atomics request here? */
964     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
965 
966     while (true) {
967         offset = vtd_iova_level_offset(iova, level);
968         slpte = vtd_get_slpte(addr, offset);
969 
970         if (slpte == (uint64_t)-1) {
971             error_report_once("%s: detected read error on DMAR slpte "
972                               "(iova=0x%" PRIx64 ")", __func__, iova);
973             if (level == vtd_get_iova_level(s, ce)) {
974                 /* Invalid programming of context-entry */
975                 return -VTD_FR_CONTEXT_ENTRY_INV;
976             } else {
977                 return -VTD_FR_PAGING_ENTRY_INV;
978             }
979         }
980         *reads = (*reads) && (slpte & VTD_SL_R);
981         *writes = (*writes) && (slpte & VTD_SL_W);
982         if (!(slpte & access_right_check)) {
983             error_report_once("%s: detected slpte permission error "
984                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
985                               "slpte=0x%" PRIx64 ", write=%d)", __func__,
986                               iova, level, slpte, is_write);
987             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
988         }
989         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
990             error_report_once("%s: detected splte reserve non-zero "
991                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
992                               "slpte=0x%" PRIx64 ")", __func__, iova,
993                               level, slpte);
994             return -VTD_FR_PAGING_ENTRY_RSVD;
995         }
996 
997         if (vtd_is_last_slpte(slpte, level)) {
998             *slptep = slpte;
999             *slpte_level = level;
1000             return 0;
1001         }
1002         addr = vtd_get_slpte_addr(slpte, aw_bits);
1003         level--;
1004     }
1005 }
1006 
1007 typedef int (*vtd_page_walk_hook)(IOMMUTLBEntry *entry, void *private);
1008 
1009 /**
1010  * Constant information used during page walking
1011  *
1012  * @hook_fn: hook func to be called when detected page
1013  * @private: private data to be passed into hook func
1014  * @notify_unmap: whether we should notify invalid entries
1015  * @as: VT-d address space of the device
1016  * @aw: maximum address width
1017  * @domain: domain ID of the page walk
1018  */
1019 typedef struct {
1020     VTDAddressSpace *as;
1021     vtd_page_walk_hook hook_fn;
1022     void *private;
1023     bool notify_unmap;
1024     uint8_t aw;
1025     uint16_t domain_id;
1026 } vtd_page_walk_info;
1027 
1028 static int vtd_page_walk_one(IOMMUTLBEntry *entry, vtd_page_walk_info *info)
1029 {
1030     VTDAddressSpace *as = info->as;
1031     vtd_page_walk_hook hook_fn = info->hook_fn;
1032     void *private = info->private;
1033     DMAMap target = {
1034         .iova = entry->iova,
1035         .size = entry->addr_mask,
1036         .translated_addr = entry->translated_addr,
1037         .perm = entry->perm,
1038     };
1039     DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
1040 
1041     if (entry->perm == IOMMU_NONE && !info->notify_unmap) {
1042         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1043         return 0;
1044     }
1045 
1046     assert(hook_fn);
1047 
1048     /* Update local IOVA mapped ranges */
1049     if (entry->perm) {
1050         if (mapped) {
1051             /* If it's exactly the same translation, skip */
1052             if (!memcmp(mapped, &target, sizeof(target))) {
1053                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
1054                                                  entry->translated_addr);
1055                 return 0;
1056             } else {
1057                 /*
1058                  * Translation changed.  Normally this should not
1059                  * happen, but it can happen when with buggy guest
1060                  * OSes.  Note that there will be a small window that
1061                  * we don't have map at all.  But that's the best
1062                  * effort we can do.  The ideal way to emulate this is
1063                  * atomically modify the PTE to follow what has
1064                  * changed, but we can't.  One example is that vfio
1065                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1066                  * interface to modify a mapping (meanwhile it seems
1067                  * meaningless to even provide one).  Anyway, let's
1068                  * mark this as a TODO in case one day we'll have
1069                  * a better solution.
1070                  */
1071                 IOMMUAccessFlags cache_perm = entry->perm;
1072                 int ret;
1073 
1074                 /* Emulate an UNMAP */
1075                 entry->perm = IOMMU_NONE;
1076                 trace_vtd_page_walk_one(info->domain_id,
1077                                         entry->iova,
1078                                         entry->translated_addr,
1079                                         entry->addr_mask,
1080                                         entry->perm);
1081                 ret = hook_fn(entry, private);
1082                 if (ret) {
1083                     return ret;
1084                 }
1085                 /* Drop any existing mapping */
1086                 iova_tree_remove(as->iova_tree, &target);
1087                 /* Recover the correct permission */
1088                 entry->perm = cache_perm;
1089             }
1090         }
1091         iova_tree_insert(as->iova_tree, &target);
1092     } else {
1093         if (!mapped) {
1094             /* Skip since we didn't map this range at all */
1095             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1096             return 0;
1097         }
1098         iova_tree_remove(as->iova_tree, &target);
1099     }
1100 
1101     trace_vtd_page_walk_one(info->domain_id, entry->iova,
1102                             entry->translated_addr, entry->addr_mask,
1103                             entry->perm);
1104     return hook_fn(entry, private);
1105 }
1106 
1107 /**
1108  * vtd_page_walk_level - walk over specific level for IOVA range
1109  *
1110  * @addr: base GPA addr to start the walk
1111  * @start: IOVA range start address
1112  * @end: IOVA range end address (start <= addr < end)
1113  * @read: whether parent level has read permission
1114  * @write: whether parent level has write permission
1115  * @info: constant information for the page walk
1116  */
1117 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
1118                                uint64_t end, uint32_t level, bool read,
1119                                bool write, vtd_page_walk_info *info)
1120 {
1121     bool read_cur, write_cur, entry_valid;
1122     uint32_t offset;
1123     uint64_t slpte;
1124     uint64_t subpage_size, subpage_mask;
1125     IOMMUTLBEntry entry;
1126     uint64_t iova = start;
1127     uint64_t iova_next;
1128     int ret = 0;
1129 
1130     trace_vtd_page_walk_level(addr, level, start, end);
1131 
1132     subpage_size = 1ULL << vtd_slpt_level_shift(level);
1133     subpage_mask = vtd_slpt_level_page_mask(level);
1134 
1135     while (iova < end) {
1136         iova_next = (iova & subpage_mask) + subpage_size;
1137 
1138         offset = vtd_iova_level_offset(iova, level);
1139         slpte = vtd_get_slpte(addr, offset);
1140 
1141         if (slpte == (uint64_t)-1) {
1142             trace_vtd_page_walk_skip_read(iova, iova_next);
1143             goto next;
1144         }
1145 
1146         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1147             trace_vtd_page_walk_skip_reserve(iova, iova_next);
1148             goto next;
1149         }
1150 
1151         /* Permissions are stacked with parents' */
1152         read_cur = read && (slpte & VTD_SL_R);
1153         write_cur = write && (slpte & VTD_SL_W);
1154 
1155         /*
1156          * As long as we have either read/write permission, this is a
1157          * valid entry. The rule works for both page entries and page
1158          * table entries.
1159          */
1160         entry_valid = read_cur | write_cur;
1161 
1162         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
1163             /*
1164              * This is a valid PDE (or even bigger than PDE).  We need
1165              * to walk one further level.
1166              */
1167             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
1168                                       iova, MIN(iova_next, end), level - 1,
1169                                       read_cur, write_cur, info);
1170         } else {
1171             /*
1172              * This means we are either:
1173              *
1174              * (1) the real page entry (either 4K page, or huge page)
1175              * (2) the whole range is invalid
1176              *
1177              * In either case, we send an IOTLB notification down.
1178              */
1179             entry.target_as = &address_space_memory;
1180             entry.iova = iova & subpage_mask;
1181             entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
1182             entry.addr_mask = ~subpage_mask;
1183             /* NOTE: this is only meaningful if entry_valid == true */
1184             entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
1185             ret = vtd_page_walk_one(&entry, info);
1186         }
1187 
1188         if (ret < 0) {
1189             return ret;
1190         }
1191 
1192 next:
1193         iova = iova_next;
1194     }
1195 
1196     return 0;
1197 }
1198 
1199 /**
1200  * vtd_page_walk - walk specific IOVA range, and call the hook
1201  *
1202  * @s: intel iommu state
1203  * @ce: context entry to walk upon
1204  * @start: IOVA address to start the walk
1205  * @end: IOVA range end address (start <= addr < end)
1206  * @info: page walking information struct
1207  */
1208 static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce,
1209                          uint64_t start, uint64_t end,
1210                          vtd_page_walk_info *info)
1211 {
1212     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1213     uint32_t level = vtd_get_iova_level(s, ce);
1214 
1215     if (!vtd_iova_range_check(s, start, ce, info->aw)) {
1216         return -VTD_FR_ADDR_BEYOND_MGAW;
1217     }
1218 
1219     if (!vtd_iova_range_check(s, end, ce, info->aw)) {
1220         /* Fix end so that it reaches the maximum */
1221         end = vtd_iova_limit(s, ce, info->aw);
1222     }
1223 
1224     return vtd_page_walk_level(addr, start, end, level, true, true, info);
1225 }
1226 
1227 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s,
1228                                           VTDRootEntry *re)
1229 {
1230     /* Legacy Mode reserved bits check */
1231     if (!s->root_scalable &&
1232         (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1233         goto rsvd_err;
1234 
1235     /* Scalable Mode reserved bits check */
1236     if (s->root_scalable &&
1237         ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) ||
1238          (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1239         goto rsvd_err;
1240 
1241     return 0;
1242 
1243 rsvd_err:
1244     error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1245                       ", lo=0x%"PRIx64,
1246                       __func__, re->hi, re->lo);
1247     return -VTD_FR_ROOT_ENTRY_RSVD;
1248 }
1249 
1250 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s,
1251                                                     VTDContextEntry *ce)
1252 {
1253     if (!s->root_scalable &&
1254         (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI ||
1255          ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1256         error_report_once("%s: invalid context entry: hi=%"PRIx64
1257                           ", lo=%"PRIx64" (reserved nonzero)",
1258                           __func__, ce->hi, ce->lo);
1259         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1260     }
1261 
1262     if (s->root_scalable &&
1263         (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) ||
1264          ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 ||
1265          ce->val[2] ||
1266          ce->val[3])) {
1267         error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1268                           ", val[2]=%"PRIx64
1269                           ", val[1]=%"PRIx64
1270                           ", val[0]=%"PRIx64" (reserved nonzero)",
1271                           __func__, ce->val[3], ce->val[2],
1272                           ce->val[1], ce->val[0]);
1273         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1274     }
1275 
1276     return 0;
1277 }
1278 
1279 static int vtd_ce_rid2pasid_check(IntelIOMMUState *s,
1280                                   VTDContextEntry *ce)
1281 {
1282     VTDPASIDEntry pe;
1283 
1284     /*
1285      * Make sure in Scalable Mode, a present context entry
1286      * has valid rid2pasid setting, which includes valid
1287      * rid2pasid field and corresponding pasid entry setting
1288      */
1289     return vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1290 }
1291 
1292 /* Map a device to its corresponding domain (context-entry) */
1293 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
1294                                     uint8_t devfn, VTDContextEntry *ce)
1295 {
1296     VTDRootEntry re;
1297     int ret_fr;
1298     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
1299 
1300     ret_fr = vtd_get_root_entry(s, bus_num, &re);
1301     if (ret_fr) {
1302         return ret_fr;
1303     }
1304 
1305     if (!vtd_root_entry_present(s, &re, devfn)) {
1306         /* Not error - it's okay we don't have root entry. */
1307         trace_vtd_re_not_present(bus_num);
1308         return -VTD_FR_ROOT_ENTRY_P;
1309     }
1310 
1311     ret_fr = vtd_root_entry_rsvd_bits_check(s, &re);
1312     if (ret_fr) {
1313         return ret_fr;
1314     }
1315 
1316     ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce);
1317     if (ret_fr) {
1318         return ret_fr;
1319     }
1320 
1321     if (!vtd_ce_present(ce)) {
1322         /* Not error - it's okay we don't have context entry. */
1323         trace_vtd_ce_not_present(bus_num, devfn);
1324         return -VTD_FR_CONTEXT_ENTRY_P;
1325     }
1326 
1327     ret_fr = vtd_context_entry_rsvd_bits_check(s, ce);
1328     if (ret_fr) {
1329         return ret_fr;
1330     }
1331 
1332     /* Check if the programming of context-entry is valid */
1333     if (!s->root_scalable &&
1334         !vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1335         error_report_once("%s: invalid context entry: hi=%"PRIx64
1336                           ", lo=%"PRIx64" (level %d not supported)",
1337                           __func__, ce->hi, ce->lo,
1338                           vtd_ce_get_level(ce));
1339         return -VTD_FR_CONTEXT_ENTRY_INV;
1340     }
1341 
1342     if (!s->root_scalable) {
1343         /* Do translation type check */
1344         if (!vtd_ce_type_check(x86_iommu, ce)) {
1345             /* Errors dumped in vtd_ce_type_check() */
1346             return -VTD_FR_CONTEXT_ENTRY_INV;
1347         }
1348     } else {
1349         /*
1350          * Check if the programming of context-entry.rid2pasid
1351          * and corresponding pasid setting is valid, and thus
1352          * avoids to check pasid entry fetching result in future
1353          * helper function calling.
1354          */
1355         ret_fr = vtd_ce_rid2pasid_check(s, ce);
1356         if (ret_fr) {
1357             return ret_fr;
1358         }
1359     }
1360 
1361     return 0;
1362 }
1363 
1364 static int vtd_sync_shadow_page_hook(IOMMUTLBEntry *entry,
1365                                      void *private)
1366 {
1367     memory_region_notify_iommu((IOMMUMemoryRegion *)private, 0, *entry);
1368     return 0;
1369 }
1370 
1371 static uint16_t vtd_get_domain_id(IntelIOMMUState *s,
1372                                   VTDContextEntry *ce)
1373 {
1374     VTDPASIDEntry pe;
1375 
1376     if (s->root_scalable) {
1377         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1378         return VTD_SM_PASID_ENTRY_DID(pe.val[1]);
1379     }
1380 
1381     return VTD_CONTEXT_ENTRY_DID(ce->hi);
1382 }
1383 
1384 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1385                                             VTDContextEntry *ce,
1386                                             hwaddr addr, hwaddr size)
1387 {
1388     IntelIOMMUState *s = vtd_as->iommu_state;
1389     vtd_page_walk_info info = {
1390         .hook_fn = vtd_sync_shadow_page_hook,
1391         .private = (void *)&vtd_as->iommu,
1392         .notify_unmap = true,
1393         .aw = s->aw_bits,
1394         .as = vtd_as,
1395         .domain_id = vtd_get_domain_id(s, ce),
1396     };
1397 
1398     return vtd_page_walk(s, ce, addr, addr + size, &info);
1399 }
1400 
1401 static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
1402 {
1403     int ret;
1404     VTDContextEntry ce;
1405     IOMMUNotifier *n;
1406 
1407     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1408                                    pci_bus_num(vtd_as->bus),
1409                                    vtd_as->devfn, &ce);
1410     if (ret) {
1411         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1412             /*
1413              * It's a valid scenario to have a context entry that is
1414              * not present.  For example, when a device is removed
1415              * from an existing domain then the context entry will be
1416              * zeroed by the guest before it was put into another
1417              * domain.  When this happens, instead of synchronizing
1418              * the shadow pages we should invalidate all existing
1419              * mappings and notify the backends.
1420              */
1421             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1422                 vtd_address_space_unmap(vtd_as, n);
1423             }
1424             ret = 0;
1425         }
1426         return ret;
1427     }
1428 
1429     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1430 }
1431 
1432 /*
1433  * Check if specific device is configed to bypass address
1434  * translation for DMA requests. In Scalable Mode, bypass
1435  * 1st-level translation or 2nd-level translation, it depends
1436  * on PGTT setting.
1437  */
1438 static bool vtd_dev_pt_enabled(VTDAddressSpace *as)
1439 {
1440     IntelIOMMUState *s;
1441     VTDContextEntry ce;
1442     VTDPASIDEntry pe;
1443     int ret;
1444 
1445     assert(as);
1446 
1447     s = as->iommu_state;
1448     ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus),
1449                                    as->devfn, &ce);
1450     if (ret) {
1451         /*
1452          * Possibly failed to parse the context entry for some reason
1453          * (e.g., during init, or any guest configuration errors on
1454          * context entries). We should assume PT not enabled for
1455          * safety.
1456          */
1457         return false;
1458     }
1459 
1460     if (s->root_scalable) {
1461         ret = vtd_ce_get_rid2pasid_entry(s, &ce, &pe);
1462         if (ret) {
1463             error_report_once("%s: vtd_ce_get_rid2pasid_entry error: %"PRId32,
1464                               __func__, ret);
1465             return false;
1466         }
1467         return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT);
1468     }
1469 
1470     return (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH);
1471 }
1472 
1473 /* Return whether the device is using IOMMU translation. */
1474 static bool vtd_switch_address_space(VTDAddressSpace *as)
1475 {
1476     bool use_iommu;
1477     /* Whether we need to take the BQL on our own */
1478     bool take_bql = !qemu_mutex_iothread_locked();
1479 
1480     assert(as);
1481 
1482     use_iommu = as->iommu_state->dmar_enabled && !vtd_dev_pt_enabled(as);
1483 
1484     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1485                                    VTD_PCI_SLOT(as->devfn),
1486                                    VTD_PCI_FUNC(as->devfn),
1487                                    use_iommu);
1488 
1489     /*
1490      * It's possible that we reach here without BQL, e.g., when called
1491      * from vtd_pt_enable_fast_path(). However the memory APIs need
1492      * it. We'd better make sure we have had it already, or, take it.
1493      */
1494     if (take_bql) {
1495         qemu_mutex_lock_iothread();
1496     }
1497 
1498     /* Turn off first then on the other */
1499     if (use_iommu) {
1500         memory_region_set_enabled(&as->nodmar, false);
1501         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1502     } else {
1503         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1504         memory_region_set_enabled(&as->nodmar, true);
1505     }
1506 
1507     if (take_bql) {
1508         qemu_mutex_unlock_iothread();
1509     }
1510 
1511     return use_iommu;
1512 }
1513 
1514 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1515 {
1516     GHashTableIter iter;
1517     VTDBus *vtd_bus;
1518     int i;
1519 
1520     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1521     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1522         for (i = 0; i < PCI_DEVFN_MAX; i++) {
1523             if (!vtd_bus->dev_as[i]) {
1524                 continue;
1525             }
1526             vtd_switch_address_space(vtd_bus->dev_as[i]);
1527         }
1528     }
1529 }
1530 
1531 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn)
1532 {
1533     return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
1534 }
1535 
1536 static const bool vtd_qualified_faults[] = {
1537     [VTD_FR_RESERVED] = false,
1538     [VTD_FR_ROOT_ENTRY_P] = false,
1539     [VTD_FR_CONTEXT_ENTRY_P] = true,
1540     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1541     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1542     [VTD_FR_WRITE] = true,
1543     [VTD_FR_READ] = true,
1544     [VTD_FR_PAGING_ENTRY_INV] = true,
1545     [VTD_FR_ROOT_TABLE_INV] = false,
1546     [VTD_FR_CONTEXT_TABLE_INV] = false,
1547     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1548     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1549     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1550     [VTD_FR_PASID_TABLE_INV] = false,
1551     [VTD_FR_RESERVED_ERR] = false,
1552     [VTD_FR_MAX] = false,
1553 };
1554 
1555 /* To see if a fault condition is "qualified", which is reported to software
1556  * only if the FPD field in the context-entry used to process the faulting
1557  * request is 0.
1558  */
1559 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1560 {
1561     return vtd_qualified_faults[fault];
1562 }
1563 
1564 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1565 {
1566     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1567 }
1568 
1569 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1570 {
1571     VTDBus *vtd_bus;
1572     VTDAddressSpace *vtd_as;
1573     bool success = false;
1574 
1575     vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id));
1576     if (!vtd_bus) {
1577         goto out;
1578     }
1579 
1580     vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)];
1581     if (!vtd_as) {
1582         goto out;
1583     }
1584 
1585     if (vtd_switch_address_space(vtd_as) == false) {
1586         /* We switched off IOMMU region successfully. */
1587         success = true;
1588     }
1589 
1590 out:
1591     trace_vtd_pt_enable_fast_path(source_id, success);
1592 }
1593 
1594 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1595  * translation.
1596  *
1597  * Called from RCU critical section.
1598  *
1599  * @bus_num: The bus number
1600  * @devfn: The devfn, which is the  combined of device and function number
1601  * @is_write: The access is a write operation
1602  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1603  *
1604  * Returns true if translation is successful, otherwise false.
1605  */
1606 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1607                                    uint8_t devfn, hwaddr addr, bool is_write,
1608                                    IOMMUTLBEntry *entry)
1609 {
1610     IntelIOMMUState *s = vtd_as->iommu_state;
1611     VTDContextEntry ce;
1612     uint8_t bus_num = pci_bus_num(bus);
1613     VTDContextCacheEntry *cc_entry;
1614     uint64_t slpte, page_mask;
1615     uint32_t level;
1616     uint16_t source_id = vtd_make_source_id(bus_num, devfn);
1617     int ret_fr;
1618     bool is_fpd_set = false;
1619     bool reads = true;
1620     bool writes = true;
1621     uint8_t access_flags;
1622     VTDIOTLBEntry *iotlb_entry;
1623 
1624     /*
1625      * We have standalone memory region for interrupt addresses, we
1626      * should never receive translation requests in this region.
1627      */
1628     assert(!vtd_is_interrupt_addr(addr));
1629 
1630     vtd_iommu_lock(s);
1631 
1632     cc_entry = &vtd_as->context_cache_entry;
1633 
1634     /* Try to fetch slpte form IOTLB */
1635     iotlb_entry = vtd_lookup_iotlb(s, source_id, addr);
1636     if (iotlb_entry) {
1637         trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1638                                  iotlb_entry->domain_id);
1639         slpte = iotlb_entry->slpte;
1640         access_flags = iotlb_entry->access_flags;
1641         page_mask = iotlb_entry->mask;
1642         goto out;
1643     }
1644 
1645     /* Try to fetch context-entry from cache first */
1646     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1647         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1648                                cc_entry->context_entry.lo,
1649                                cc_entry->context_cache_gen);
1650         ce = cc_entry->context_entry;
1651         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1652         if (!is_fpd_set && s->root_scalable) {
1653             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1654             VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1655         }
1656     } else {
1657         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1658         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1659         if (!ret_fr && !is_fpd_set && s->root_scalable) {
1660             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1661         }
1662         VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1663         /* Update context-cache */
1664         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1665                                   cc_entry->context_cache_gen,
1666                                   s->context_cache_gen);
1667         cc_entry->context_entry = ce;
1668         cc_entry->context_cache_gen = s->context_cache_gen;
1669     }
1670 
1671     /*
1672      * We don't need to translate for pass-through context entries.
1673      * Also, let's ignore IOTLB caching as well for PT devices.
1674      */
1675     if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) {
1676         entry->iova = addr & VTD_PAGE_MASK_4K;
1677         entry->translated_addr = entry->iova;
1678         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1679         entry->perm = IOMMU_RW;
1680         trace_vtd_translate_pt(source_id, entry->iova);
1681 
1682         /*
1683          * When this happens, it means firstly caching-mode is not
1684          * enabled, and this is the first passthrough translation for
1685          * the device. Let's enable the fast path for passthrough.
1686          *
1687          * When passthrough is disabled again for the device, we can
1688          * capture it via the context entry invalidation, then the
1689          * IOMMU region can be swapped back.
1690          */
1691         vtd_pt_enable_fast_path(s, source_id);
1692         vtd_iommu_unlock(s);
1693         return true;
1694     }
1695 
1696     ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level,
1697                                &reads, &writes, s->aw_bits);
1698     VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1699 
1700     page_mask = vtd_slpt_level_page_mask(level);
1701     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1702     vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce), addr, slpte,
1703                      access_flags, level);
1704 out:
1705     vtd_iommu_unlock(s);
1706     entry->iova = addr & page_mask;
1707     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1708     entry->addr_mask = ~page_mask;
1709     entry->perm = access_flags;
1710     return true;
1711 
1712 error:
1713     vtd_iommu_unlock(s);
1714     entry->iova = 0;
1715     entry->translated_addr = 0;
1716     entry->addr_mask = 0;
1717     entry->perm = IOMMU_NONE;
1718     return false;
1719 }
1720 
1721 static void vtd_root_table_setup(IntelIOMMUState *s)
1722 {
1723     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1724     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
1725 
1726     vtd_update_scalable_state(s);
1727 
1728     trace_vtd_reg_dmar_root(s->root, s->root_scalable);
1729 }
1730 
1731 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
1732                                uint32_t index, uint32_t mask)
1733 {
1734     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
1735 }
1736 
1737 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
1738 {
1739     uint64_t value = 0;
1740     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
1741     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
1742     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
1743     s->intr_eime = value & VTD_IRTA_EIME;
1744 
1745     /* Notify global invalidation */
1746     vtd_iec_notify_all(s, true, 0, 0);
1747 
1748     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
1749 }
1750 
1751 static void vtd_iommu_replay_all(IntelIOMMUState *s)
1752 {
1753     VTDAddressSpace *vtd_as;
1754 
1755     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1756         vtd_sync_shadow_page_table(vtd_as);
1757     }
1758 }
1759 
1760 static void vtd_context_global_invalidate(IntelIOMMUState *s)
1761 {
1762     trace_vtd_inv_desc_cc_global();
1763     /* Protects context cache */
1764     vtd_iommu_lock(s);
1765     s->context_cache_gen++;
1766     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
1767         vtd_reset_context_cache_locked(s);
1768     }
1769     vtd_iommu_unlock(s);
1770     vtd_address_space_refresh_all(s);
1771     /*
1772      * From VT-d spec 6.5.2.1, a global context entry invalidation
1773      * should be followed by a IOTLB global invalidation, so we should
1774      * be safe even without this. Hoewever, let's replay the region as
1775      * well to be safer, and go back here when we need finer tunes for
1776      * VT-d emulation codes.
1777      */
1778     vtd_iommu_replay_all(s);
1779 }
1780 
1781 /* Do a context-cache device-selective invalidation.
1782  * @func_mask: FM field after shifting
1783  */
1784 static void vtd_context_device_invalidate(IntelIOMMUState *s,
1785                                           uint16_t source_id,
1786                                           uint16_t func_mask)
1787 {
1788     uint16_t mask;
1789     VTDBus *vtd_bus;
1790     VTDAddressSpace *vtd_as;
1791     uint8_t bus_n, devfn;
1792     uint16_t devfn_it;
1793 
1794     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
1795 
1796     switch (func_mask & 3) {
1797     case 0:
1798         mask = 0;   /* No bits in the SID field masked */
1799         break;
1800     case 1:
1801         mask = 4;   /* Mask bit 2 in the SID field */
1802         break;
1803     case 2:
1804         mask = 6;   /* Mask bit 2:1 in the SID field */
1805         break;
1806     case 3:
1807         mask = 7;   /* Mask bit 2:0 in the SID field */
1808         break;
1809     }
1810     mask = ~mask;
1811 
1812     bus_n = VTD_SID_TO_BUS(source_id);
1813     vtd_bus = vtd_find_as_from_bus_num(s, bus_n);
1814     if (vtd_bus) {
1815         devfn = VTD_SID_TO_DEVFN(source_id);
1816         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
1817             vtd_as = vtd_bus->dev_as[devfn_it];
1818             if (vtd_as && ((devfn_it & mask) == (devfn & mask))) {
1819                 trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it),
1820                                              VTD_PCI_FUNC(devfn_it));
1821                 vtd_iommu_lock(s);
1822                 vtd_as->context_cache_entry.context_cache_gen = 0;
1823                 vtd_iommu_unlock(s);
1824                 /*
1825                  * Do switch address space when needed, in case if the
1826                  * device passthrough bit is switched.
1827                  */
1828                 vtd_switch_address_space(vtd_as);
1829                 /*
1830                  * So a device is moving out of (or moving into) a
1831                  * domain, resync the shadow page table.
1832                  * This won't bring bad even if we have no such
1833                  * notifier registered - the IOMMU notification
1834                  * framework will skip MAP notifications if that
1835                  * happened.
1836                  */
1837                 vtd_sync_shadow_page_table(vtd_as);
1838             }
1839         }
1840     }
1841 }
1842 
1843 /* Context-cache invalidation
1844  * Returns the Context Actual Invalidation Granularity.
1845  * @val: the content of the CCMD_REG
1846  */
1847 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
1848 {
1849     uint64_t caig;
1850     uint64_t type = val & VTD_CCMD_CIRG_MASK;
1851 
1852     switch (type) {
1853     case VTD_CCMD_DOMAIN_INVL:
1854         /* Fall through */
1855     case VTD_CCMD_GLOBAL_INVL:
1856         caig = VTD_CCMD_GLOBAL_INVL_A;
1857         vtd_context_global_invalidate(s);
1858         break;
1859 
1860     case VTD_CCMD_DEVICE_INVL:
1861         caig = VTD_CCMD_DEVICE_INVL_A;
1862         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
1863         break;
1864 
1865     default:
1866         error_report_once("%s: invalid context: 0x%" PRIx64,
1867                           __func__, val);
1868         caig = 0;
1869     }
1870     return caig;
1871 }
1872 
1873 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
1874 {
1875     trace_vtd_inv_desc_iotlb_global();
1876     vtd_reset_iotlb(s);
1877     vtd_iommu_replay_all(s);
1878 }
1879 
1880 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
1881 {
1882     VTDContextEntry ce;
1883     VTDAddressSpace *vtd_as;
1884 
1885     trace_vtd_inv_desc_iotlb_domain(domain_id);
1886 
1887     vtd_iommu_lock(s);
1888     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
1889                                 &domain_id);
1890     vtd_iommu_unlock(s);
1891 
1892     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1893         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1894                                       vtd_as->devfn, &ce) &&
1895             domain_id == vtd_get_domain_id(s, &ce)) {
1896             vtd_sync_shadow_page_table(vtd_as);
1897         }
1898     }
1899 }
1900 
1901 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
1902                                            uint16_t domain_id, hwaddr addr,
1903                                            uint8_t am)
1904 {
1905     VTDAddressSpace *vtd_as;
1906     VTDContextEntry ce;
1907     int ret;
1908     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
1909 
1910     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
1911         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1912                                        vtd_as->devfn, &ce);
1913         if (!ret && domain_id == vtd_get_domain_id(s, &ce)) {
1914             if (vtd_as_has_map_notifier(vtd_as)) {
1915                 /*
1916                  * As long as we have MAP notifications registered in
1917                  * any of our IOMMU notifiers, we need to sync the
1918                  * shadow page table.
1919                  */
1920                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
1921             } else {
1922                 /*
1923                  * For UNMAP-only notifiers, we don't need to walk the
1924                  * page tables.  We just deliver the PSI down to
1925                  * invalidate caches.
1926                  */
1927                 IOMMUTLBEntry entry = {
1928                     .target_as = &address_space_memory,
1929                     .iova = addr,
1930                     .translated_addr = 0,
1931                     .addr_mask = size - 1,
1932                     .perm = IOMMU_NONE,
1933                 };
1934                 memory_region_notify_iommu(&vtd_as->iommu, 0, entry);
1935             }
1936         }
1937     }
1938 }
1939 
1940 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
1941                                       hwaddr addr, uint8_t am)
1942 {
1943     VTDIOTLBPageInvInfo info;
1944 
1945     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
1946 
1947     assert(am <= VTD_MAMV);
1948     info.domain_id = domain_id;
1949     info.addr = addr;
1950     info.mask = ~((1 << am) - 1);
1951     vtd_iommu_lock(s);
1952     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
1953     vtd_iommu_unlock(s);
1954     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am);
1955 }
1956 
1957 /* Flush IOTLB
1958  * Returns the IOTLB Actual Invalidation Granularity.
1959  * @val: the content of the IOTLB_REG
1960  */
1961 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
1962 {
1963     uint64_t iaig;
1964     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
1965     uint16_t domain_id;
1966     hwaddr addr;
1967     uint8_t am;
1968 
1969     switch (type) {
1970     case VTD_TLB_GLOBAL_FLUSH:
1971         iaig = VTD_TLB_GLOBAL_FLUSH_A;
1972         vtd_iotlb_global_invalidate(s);
1973         break;
1974 
1975     case VTD_TLB_DSI_FLUSH:
1976         domain_id = VTD_TLB_DID(val);
1977         iaig = VTD_TLB_DSI_FLUSH_A;
1978         vtd_iotlb_domain_invalidate(s, domain_id);
1979         break;
1980 
1981     case VTD_TLB_PSI_FLUSH:
1982         domain_id = VTD_TLB_DID(val);
1983         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
1984         am = VTD_IVA_AM(addr);
1985         addr = VTD_IVA_ADDR(addr);
1986         if (am > VTD_MAMV) {
1987             error_report_once("%s: address mask overflow: 0x%" PRIx64,
1988                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
1989             iaig = 0;
1990             break;
1991         }
1992         iaig = VTD_TLB_PSI_FLUSH_A;
1993         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
1994         break;
1995 
1996     default:
1997         error_report_once("%s: invalid granularity: 0x%" PRIx64,
1998                           __func__, val);
1999         iaig = 0;
2000     }
2001     return iaig;
2002 }
2003 
2004 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
2005 
2006 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
2007 {
2008     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
2009            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
2010 }
2011 
2012 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
2013 {
2014     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
2015 
2016     trace_vtd_inv_qi_enable(en);
2017 
2018     if (en) {
2019         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
2020         /* 2^(x+8) entries */
2021         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0));
2022         s->qi_enabled = true;
2023         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
2024         /* Ok - report back to driver */
2025         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
2026 
2027         if (s->iq_tail != 0) {
2028             /*
2029              * This is a spec violation but Windows guests are known to set up
2030              * Queued Invalidation this way so we allow the write and process
2031              * Invalidation Descriptors right away.
2032              */
2033             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
2034             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2035                 vtd_fetch_inv_desc(s);
2036             }
2037         }
2038     } else {
2039         if (vtd_queued_inv_disable_check(s)) {
2040             /* disable Queued Invalidation */
2041             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
2042             s->iq_head = 0;
2043             s->qi_enabled = false;
2044             /* Ok - report back to driver */
2045             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
2046         } else {
2047             error_report_once("%s: detected improper state when disable QI "
2048                               "(head=0x%x, tail=0x%x, last_type=%d)",
2049                               __func__,
2050                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
2051         }
2052     }
2053 }
2054 
2055 /* Set Root Table Pointer */
2056 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
2057 {
2058     vtd_root_table_setup(s);
2059     /* Ok - report back to driver */
2060     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
2061     vtd_reset_caches(s);
2062     vtd_address_space_refresh_all(s);
2063 }
2064 
2065 /* Set Interrupt Remap Table Pointer */
2066 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
2067 {
2068     vtd_interrupt_remap_table_setup(s);
2069     /* Ok - report back to driver */
2070     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
2071 }
2072 
2073 /* Handle Translation Enable/Disable */
2074 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
2075 {
2076     if (s->dmar_enabled == en) {
2077         return;
2078     }
2079 
2080     trace_vtd_dmar_enable(en);
2081 
2082     if (en) {
2083         s->dmar_enabled = true;
2084         /* Ok - report back to driver */
2085         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
2086     } else {
2087         s->dmar_enabled = false;
2088 
2089         /* Clear the index of Fault Recording Register */
2090         s->next_frcd_reg = 0;
2091         /* Ok - report back to driver */
2092         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
2093     }
2094 
2095     vtd_reset_caches(s);
2096     vtd_address_space_refresh_all(s);
2097 }
2098 
2099 /* Handle Interrupt Remap Enable/Disable */
2100 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
2101 {
2102     trace_vtd_ir_enable(en);
2103 
2104     if (en) {
2105         s->intr_enabled = true;
2106         /* Ok - report back to driver */
2107         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
2108     } else {
2109         s->intr_enabled = false;
2110         /* Ok - report back to driver */
2111         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
2112     }
2113 }
2114 
2115 /* Handle write to Global Command Register */
2116 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
2117 {
2118     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
2119     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
2120     uint32_t changed = status ^ val;
2121 
2122     trace_vtd_reg_write_gcmd(status, val);
2123     if (changed & VTD_GCMD_TE) {
2124         /* Translation enable/disable */
2125         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
2126     }
2127     if (val & VTD_GCMD_SRTP) {
2128         /* Set/update the root-table pointer */
2129         vtd_handle_gcmd_srtp(s);
2130     }
2131     if (changed & VTD_GCMD_QIE) {
2132         /* Queued Invalidation Enable */
2133         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
2134     }
2135     if (val & VTD_GCMD_SIRTP) {
2136         /* Set/update the interrupt remapping root-table pointer */
2137         vtd_handle_gcmd_sirtp(s);
2138     }
2139     if (changed & VTD_GCMD_IRE) {
2140         /* Interrupt remap enable/disable */
2141         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
2142     }
2143 }
2144 
2145 /* Handle write to Context Command Register */
2146 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
2147 {
2148     uint64_t ret;
2149     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
2150 
2151     /* Context-cache invalidation request */
2152     if (val & VTD_CCMD_ICC) {
2153         if (s->qi_enabled) {
2154             error_report_once("Queued Invalidation enabled, "
2155                               "should not use register-based invalidation");
2156             return;
2157         }
2158         ret = vtd_context_cache_invalidate(s, val);
2159         /* Invalidation completed. Change something to show */
2160         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
2161         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
2162                                       ret);
2163     }
2164 }
2165 
2166 /* Handle write to IOTLB Invalidation Register */
2167 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
2168 {
2169     uint64_t ret;
2170     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
2171 
2172     /* IOTLB invalidation request */
2173     if (val & VTD_TLB_IVT) {
2174         if (s->qi_enabled) {
2175             error_report_once("Queued Invalidation enabled, "
2176                               "should not use register-based invalidation");
2177             return;
2178         }
2179         ret = vtd_iotlb_flush(s, val);
2180         /* Invalidation completed. Change something to show */
2181         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
2182         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
2183                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
2184     }
2185 }
2186 
2187 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2188 static bool vtd_get_inv_desc(IntelIOMMUState *s,
2189                              VTDInvDesc *inv_desc)
2190 {
2191     dma_addr_t base_addr = s->iq;
2192     uint32_t offset = s->iq_head;
2193     uint32_t dw = s->iq_dw ? 32 : 16;
2194     dma_addr_t addr = base_addr + offset * dw;
2195 
2196     if (dma_memory_read(&address_space_memory, addr, inv_desc, dw)) {
2197         error_report_once("Read INV DESC failed.");
2198         return false;
2199     }
2200     inv_desc->lo = le64_to_cpu(inv_desc->lo);
2201     inv_desc->hi = le64_to_cpu(inv_desc->hi);
2202     if (dw == 32) {
2203         inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]);
2204         inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]);
2205     }
2206     return true;
2207 }
2208 
2209 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2210 {
2211     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
2212         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
2213         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2214                           " (reserved nonzero)", __func__, inv_desc->hi,
2215                           inv_desc->lo);
2216         return false;
2217     }
2218     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
2219         /* Status Write */
2220         uint32_t status_data = (uint32_t)(inv_desc->lo >>
2221                                VTD_INV_DESC_WAIT_DATA_SHIFT);
2222 
2223         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
2224 
2225         /* FIXME: need to be masked with HAW? */
2226         dma_addr_t status_addr = inv_desc->hi;
2227         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
2228         status_data = cpu_to_le32(status_data);
2229         if (dma_memory_write(&address_space_memory, status_addr, &status_data,
2230                              sizeof(status_data))) {
2231             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
2232             return false;
2233         }
2234     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
2235         /* Interrupt flag */
2236         vtd_generate_completion_event(s);
2237     } else {
2238         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2239                           " (unknown type)", __func__, inv_desc->hi,
2240                           inv_desc->lo);
2241         return false;
2242     }
2243     return true;
2244 }
2245 
2246 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
2247                                            VTDInvDesc *inv_desc)
2248 {
2249     uint16_t sid, fmask;
2250 
2251     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
2252         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2253                           " (reserved nonzero)", __func__, inv_desc->hi,
2254                           inv_desc->lo);
2255         return false;
2256     }
2257     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
2258     case VTD_INV_DESC_CC_DOMAIN:
2259         trace_vtd_inv_desc_cc_domain(
2260             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
2261         /* Fall through */
2262     case VTD_INV_DESC_CC_GLOBAL:
2263         vtd_context_global_invalidate(s);
2264         break;
2265 
2266     case VTD_INV_DESC_CC_DEVICE:
2267         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
2268         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
2269         vtd_context_device_invalidate(s, sid, fmask);
2270         break;
2271 
2272     default:
2273         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2274                           " (invalid type)", __func__, inv_desc->hi,
2275                           inv_desc->lo);
2276         return false;
2277     }
2278     return true;
2279 }
2280 
2281 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2282 {
2283     uint16_t domain_id;
2284     uint8_t am;
2285     hwaddr addr;
2286 
2287     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
2288         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
2289         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2290                           ", lo=0x%"PRIx64" (reserved bits unzero)\n",
2291                           __func__, inv_desc->hi, inv_desc->lo);
2292         return false;
2293     }
2294 
2295     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
2296     case VTD_INV_DESC_IOTLB_GLOBAL:
2297         vtd_iotlb_global_invalidate(s);
2298         break;
2299 
2300     case VTD_INV_DESC_IOTLB_DOMAIN:
2301         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2302         vtd_iotlb_domain_invalidate(s, domain_id);
2303         break;
2304 
2305     case VTD_INV_DESC_IOTLB_PAGE:
2306         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2307         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
2308         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
2309         if (am > VTD_MAMV) {
2310             error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2311                               ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)\n",
2312                               __func__, inv_desc->hi, inv_desc->lo,
2313                               am, (unsigned)VTD_MAMV);
2314             return false;
2315         }
2316         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2317         break;
2318 
2319     default:
2320         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2321                           ", lo=0x%"PRIx64" (type mismatch: 0x%llx)\n",
2322                           __func__, inv_desc->hi, inv_desc->lo,
2323                           inv_desc->lo & VTD_INV_DESC_IOTLB_G);
2324         return false;
2325     }
2326     return true;
2327 }
2328 
2329 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
2330                                      VTDInvDesc *inv_desc)
2331 {
2332     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
2333                            inv_desc->iec.index,
2334                            inv_desc->iec.index_mask);
2335 
2336     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
2337                        inv_desc->iec.index,
2338                        inv_desc->iec.index_mask);
2339     return true;
2340 }
2341 
2342 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
2343                                           VTDInvDesc *inv_desc)
2344 {
2345     VTDAddressSpace *vtd_dev_as;
2346     IOMMUTLBEntry entry;
2347     struct VTDBus *vtd_bus;
2348     hwaddr addr;
2349     uint64_t sz;
2350     uint16_t sid;
2351     uint8_t devfn;
2352     bool size;
2353     uint8_t bus_num;
2354 
2355     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2356     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2357     devfn = sid & 0xff;
2358     bus_num = sid >> 8;
2359     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2360 
2361     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2362         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2363         error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2364                           ", lo=%"PRIx64" (reserved nonzero)", __func__,
2365                           inv_desc->hi, inv_desc->lo);
2366         return false;
2367     }
2368 
2369     vtd_bus = vtd_find_as_from_bus_num(s, bus_num);
2370     if (!vtd_bus) {
2371         goto done;
2372     }
2373 
2374     vtd_dev_as = vtd_bus->dev_as[devfn];
2375     if (!vtd_dev_as) {
2376         goto done;
2377     }
2378 
2379     /* According to ATS spec table 2.4:
2380      * S = 0, bits 15:12 = xxxx     range size: 4K
2381      * S = 1, bits 15:12 = xxx0     range size: 8K
2382      * S = 1, bits 15:12 = xx01     range size: 16K
2383      * S = 1, bits 15:12 = x011     range size: 32K
2384      * S = 1, bits 15:12 = 0111     range size: 64K
2385      * ...
2386      */
2387     if (size) {
2388         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2389         addr &= ~(sz - 1);
2390     } else {
2391         sz = VTD_PAGE_SIZE;
2392     }
2393 
2394     entry.target_as = &vtd_dev_as->as;
2395     entry.addr_mask = sz - 1;
2396     entry.iova = addr;
2397     entry.perm = IOMMU_NONE;
2398     entry.translated_addr = 0;
2399     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry);
2400 
2401 done:
2402     return true;
2403 }
2404 
2405 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2406 {
2407     VTDInvDesc inv_desc;
2408     uint8_t desc_type;
2409 
2410     trace_vtd_inv_qi_head(s->iq_head);
2411     if (!vtd_get_inv_desc(s, &inv_desc)) {
2412         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2413         return false;
2414     }
2415 
2416     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2417     /* FIXME: should update at first or at last? */
2418     s->iq_last_desc_type = desc_type;
2419 
2420     switch (desc_type) {
2421     case VTD_INV_DESC_CC:
2422         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2423         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2424             return false;
2425         }
2426         break;
2427 
2428     case VTD_INV_DESC_IOTLB:
2429         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2430         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2431             return false;
2432         }
2433         break;
2434 
2435     /*
2436      * TODO: the entity of below two cases will be implemented in future series.
2437      * To make guest (which integrates scalable mode support patch set in
2438      * iommu driver) work, just return true is enough so far.
2439      */
2440     case VTD_INV_DESC_PC:
2441         break;
2442 
2443     case VTD_INV_DESC_PIOTLB:
2444         break;
2445 
2446     case VTD_INV_DESC_WAIT:
2447         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2448         if (!vtd_process_wait_desc(s, &inv_desc)) {
2449             return false;
2450         }
2451         break;
2452 
2453     case VTD_INV_DESC_IEC:
2454         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2455         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2456             return false;
2457         }
2458         break;
2459 
2460     case VTD_INV_DESC_DEVICE:
2461         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2462         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2463             return false;
2464         }
2465         break;
2466 
2467     default:
2468         error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64
2469                           " (unknown type)", __func__, inv_desc.hi,
2470                           inv_desc.lo);
2471         return false;
2472     }
2473     s->iq_head++;
2474     if (s->iq_head == s->iq_size) {
2475         s->iq_head = 0;
2476     }
2477     return true;
2478 }
2479 
2480 /* Try to fetch and process more Invalidation Descriptors */
2481 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2482 {
2483     trace_vtd_inv_qi_fetch();
2484 
2485     if (s->iq_tail >= s->iq_size) {
2486         /* Detects an invalid Tail pointer */
2487         error_report_once("%s: detected invalid QI tail "
2488                           "(tail=0x%x, size=0x%x)",
2489                           __func__, s->iq_tail, s->iq_size);
2490         vtd_handle_inv_queue_error(s);
2491         return;
2492     }
2493     while (s->iq_head != s->iq_tail) {
2494         if (!vtd_process_inv_desc(s)) {
2495             /* Invalidation Queue Errors */
2496             vtd_handle_inv_queue_error(s);
2497             break;
2498         }
2499         /* Must update the IQH_REG in time */
2500         vtd_set_quad_raw(s, DMAR_IQH_REG,
2501                          (((uint64_t)(s->iq_head)) << VTD_IQH_QH_SHIFT) &
2502                          VTD_IQH_QH_MASK);
2503     }
2504 }
2505 
2506 /* Handle write to Invalidation Queue Tail Register */
2507 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2508 {
2509     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2510 
2511     if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) {
2512         error_report_once("%s: RSV bit is set: val=0x%"PRIx64,
2513                           __func__, val);
2514         return;
2515     }
2516     s->iq_tail = VTD_IQT_QT(s->iq_dw, val);
2517     trace_vtd_inv_qi_tail(s->iq_tail);
2518 
2519     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2520         /* Process Invalidation Queue here */
2521         vtd_fetch_inv_desc(s);
2522     }
2523 }
2524 
2525 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2526 {
2527     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2528     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2529     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2530 
2531     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2532         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2533         trace_vtd_fsts_clear_ip();
2534     }
2535     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2536      * Descriptors if there are any when Queued Invalidation is enabled?
2537      */
2538 }
2539 
2540 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2541 {
2542     uint32_t fectl_reg;
2543     /* FIXME: when software clears the IM field, check the IP field. But do we
2544      * need to compare the old value and the new value to conclude that
2545      * software clears the IM field? Or just check if the IM field is zero?
2546      */
2547     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2548 
2549     trace_vtd_reg_write_fectl(fectl_reg);
2550 
2551     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2552         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2553         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2554     }
2555 }
2556 
2557 static void vtd_handle_ics_write(IntelIOMMUState *s)
2558 {
2559     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2560     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2561 
2562     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2563         trace_vtd_reg_ics_clear_ip();
2564         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2565     }
2566 }
2567 
2568 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2569 {
2570     uint32_t iectl_reg;
2571     /* FIXME: when software clears the IM field, check the IP field. But do we
2572      * need to compare the old value and the new value to conclude that
2573      * software clears the IM field? Or just check if the IM field is zero?
2574      */
2575     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2576 
2577     trace_vtd_reg_write_iectl(iectl_reg);
2578 
2579     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2580         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2581         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2582     }
2583 }
2584 
2585 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2586 {
2587     IntelIOMMUState *s = opaque;
2588     uint64_t val;
2589 
2590     trace_vtd_reg_read(addr, size);
2591 
2592     if (addr + size > DMAR_REG_SIZE) {
2593         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2594                           " size=0x%u", __func__, addr, size);
2595         return (uint64_t)-1;
2596     }
2597 
2598     switch (addr) {
2599     /* Root Table Address Register, 64-bit */
2600     case DMAR_RTADDR_REG:
2601         if (size == 4) {
2602             val = s->root & ((1ULL << 32) - 1);
2603         } else {
2604             val = s->root;
2605         }
2606         break;
2607 
2608     case DMAR_RTADDR_REG_HI:
2609         assert(size == 4);
2610         val = s->root >> 32;
2611         break;
2612 
2613     /* Invalidation Queue Address Register, 64-bit */
2614     case DMAR_IQA_REG:
2615         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2616         if (size == 4) {
2617             val = val & ((1ULL << 32) - 1);
2618         }
2619         break;
2620 
2621     case DMAR_IQA_REG_HI:
2622         assert(size == 4);
2623         val = s->iq >> 32;
2624         break;
2625 
2626     default:
2627         if (size == 4) {
2628             val = vtd_get_long(s, addr);
2629         } else {
2630             val = vtd_get_quad(s, addr);
2631         }
2632     }
2633 
2634     return val;
2635 }
2636 
2637 static void vtd_mem_write(void *opaque, hwaddr addr,
2638                           uint64_t val, unsigned size)
2639 {
2640     IntelIOMMUState *s = opaque;
2641 
2642     trace_vtd_reg_write(addr, size, val);
2643 
2644     if (addr + size > DMAR_REG_SIZE) {
2645         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2646                           " size=0x%u", __func__, addr, size);
2647         return;
2648     }
2649 
2650     switch (addr) {
2651     /* Global Command Register, 32-bit */
2652     case DMAR_GCMD_REG:
2653         vtd_set_long(s, addr, val);
2654         vtd_handle_gcmd_write(s);
2655         break;
2656 
2657     /* Context Command Register, 64-bit */
2658     case DMAR_CCMD_REG:
2659         if (size == 4) {
2660             vtd_set_long(s, addr, val);
2661         } else {
2662             vtd_set_quad(s, addr, val);
2663             vtd_handle_ccmd_write(s);
2664         }
2665         break;
2666 
2667     case DMAR_CCMD_REG_HI:
2668         assert(size == 4);
2669         vtd_set_long(s, addr, val);
2670         vtd_handle_ccmd_write(s);
2671         break;
2672 
2673     /* IOTLB Invalidation Register, 64-bit */
2674     case DMAR_IOTLB_REG:
2675         if (size == 4) {
2676             vtd_set_long(s, addr, val);
2677         } else {
2678             vtd_set_quad(s, addr, val);
2679             vtd_handle_iotlb_write(s);
2680         }
2681         break;
2682 
2683     case DMAR_IOTLB_REG_HI:
2684         assert(size == 4);
2685         vtd_set_long(s, addr, val);
2686         vtd_handle_iotlb_write(s);
2687         break;
2688 
2689     /* Invalidate Address Register, 64-bit */
2690     case DMAR_IVA_REG:
2691         if (size == 4) {
2692             vtd_set_long(s, addr, val);
2693         } else {
2694             vtd_set_quad(s, addr, val);
2695         }
2696         break;
2697 
2698     case DMAR_IVA_REG_HI:
2699         assert(size == 4);
2700         vtd_set_long(s, addr, val);
2701         break;
2702 
2703     /* Fault Status Register, 32-bit */
2704     case DMAR_FSTS_REG:
2705         assert(size == 4);
2706         vtd_set_long(s, addr, val);
2707         vtd_handle_fsts_write(s);
2708         break;
2709 
2710     /* Fault Event Control Register, 32-bit */
2711     case DMAR_FECTL_REG:
2712         assert(size == 4);
2713         vtd_set_long(s, addr, val);
2714         vtd_handle_fectl_write(s);
2715         break;
2716 
2717     /* Fault Event Data Register, 32-bit */
2718     case DMAR_FEDATA_REG:
2719         assert(size == 4);
2720         vtd_set_long(s, addr, val);
2721         break;
2722 
2723     /* Fault Event Address Register, 32-bit */
2724     case DMAR_FEADDR_REG:
2725         if (size == 4) {
2726             vtd_set_long(s, addr, val);
2727         } else {
2728             /*
2729              * While the register is 32-bit only, some guests (Xen...) write to
2730              * it with 64-bit.
2731              */
2732             vtd_set_quad(s, addr, val);
2733         }
2734         break;
2735 
2736     /* Fault Event Upper Address Register, 32-bit */
2737     case DMAR_FEUADDR_REG:
2738         assert(size == 4);
2739         vtd_set_long(s, addr, val);
2740         break;
2741 
2742     /* Protected Memory Enable Register, 32-bit */
2743     case DMAR_PMEN_REG:
2744         assert(size == 4);
2745         vtd_set_long(s, addr, val);
2746         break;
2747 
2748     /* Root Table Address Register, 64-bit */
2749     case DMAR_RTADDR_REG:
2750         if (size == 4) {
2751             vtd_set_long(s, addr, val);
2752         } else {
2753             vtd_set_quad(s, addr, val);
2754         }
2755         break;
2756 
2757     case DMAR_RTADDR_REG_HI:
2758         assert(size == 4);
2759         vtd_set_long(s, addr, val);
2760         break;
2761 
2762     /* Invalidation Queue Tail Register, 64-bit */
2763     case DMAR_IQT_REG:
2764         if (size == 4) {
2765             vtd_set_long(s, addr, val);
2766         } else {
2767             vtd_set_quad(s, addr, val);
2768         }
2769         vtd_handle_iqt_write(s);
2770         break;
2771 
2772     case DMAR_IQT_REG_HI:
2773         assert(size == 4);
2774         vtd_set_long(s, addr, val);
2775         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
2776         break;
2777 
2778     /* Invalidation Queue Address Register, 64-bit */
2779     case DMAR_IQA_REG:
2780         if (size == 4) {
2781             vtd_set_long(s, addr, val);
2782         } else {
2783             vtd_set_quad(s, addr, val);
2784         }
2785         if (s->ecap & VTD_ECAP_SMTS &&
2786             val & VTD_IQA_DW_MASK) {
2787             s->iq_dw = true;
2788         } else {
2789             s->iq_dw = false;
2790         }
2791         break;
2792 
2793     case DMAR_IQA_REG_HI:
2794         assert(size == 4);
2795         vtd_set_long(s, addr, val);
2796         break;
2797 
2798     /* Invalidation Completion Status Register, 32-bit */
2799     case DMAR_ICS_REG:
2800         assert(size == 4);
2801         vtd_set_long(s, addr, val);
2802         vtd_handle_ics_write(s);
2803         break;
2804 
2805     /* Invalidation Event Control Register, 32-bit */
2806     case DMAR_IECTL_REG:
2807         assert(size == 4);
2808         vtd_set_long(s, addr, val);
2809         vtd_handle_iectl_write(s);
2810         break;
2811 
2812     /* Invalidation Event Data Register, 32-bit */
2813     case DMAR_IEDATA_REG:
2814         assert(size == 4);
2815         vtd_set_long(s, addr, val);
2816         break;
2817 
2818     /* Invalidation Event Address Register, 32-bit */
2819     case DMAR_IEADDR_REG:
2820         assert(size == 4);
2821         vtd_set_long(s, addr, val);
2822         break;
2823 
2824     /* Invalidation Event Upper Address Register, 32-bit */
2825     case DMAR_IEUADDR_REG:
2826         assert(size == 4);
2827         vtd_set_long(s, addr, val);
2828         break;
2829 
2830     /* Fault Recording Registers, 128-bit */
2831     case DMAR_FRCD_REG_0_0:
2832         if (size == 4) {
2833             vtd_set_long(s, addr, val);
2834         } else {
2835             vtd_set_quad(s, addr, val);
2836         }
2837         break;
2838 
2839     case DMAR_FRCD_REG_0_1:
2840         assert(size == 4);
2841         vtd_set_long(s, addr, val);
2842         break;
2843 
2844     case DMAR_FRCD_REG_0_2:
2845         if (size == 4) {
2846             vtd_set_long(s, addr, val);
2847         } else {
2848             vtd_set_quad(s, addr, val);
2849             /* May clear bit 127 (Fault), update PPF */
2850             vtd_update_fsts_ppf(s);
2851         }
2852         break;
2853 
2854     case DMAR_FRCD_REG_0_3:
2855         assert(size == 4);
2856         vtd_set_long(s, addr, val);
2857         /* May clear bit 127 (Fault), update PPF */
2858         vtd_update_fsts_ppf(s);
2859         break;
2860 
2861     case DMAR_IRTA_REG:
2862         if (size == 4) {
2863             vtd_set_long(s, addr, val);
2864         } else {
2865             vtd_set_quad(s, addr, val);
2866         }
2867         break;
2868 
2869     case DMAR_IRTA_REG_HI:
2870         assert(size == 4);
2871         vtd_set_long(s, addr, val);
2872         break;
2873 
2874     default:
2875         if (size == 4) {
2876             vtd_set_long(s, addr, val);
2877         } else {
2878             vtd_set_quad(s, addr, val);
2879         }
2880     }
2881 }
2882 
2883 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
2884                                          IOMMUAccessFlags flag, int iommu_idx)
2885 {
2886     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2887     IntelIOMMUState *s = vtd_as->iommu_state;
2888     IOMMUTLBEntry iotlb = {
2889         /* We'll fill in the rest later. */
2890         .target_as = &address_space_memory,
2891     };
2892     bool success;
2893 
2894     if (likely(s->dmar_enabled)) {
2895         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
2896                                          addr, flag & IOMMU_WO, &iotlb);
2897     } else {
2898         /* DMAR disabled, passthrough, use 4k-page*/
2899         iotlb.iova = addr & VTD_PAGE_MASK_4K;
2900         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
2901         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
2902         iotlb.perm = IOMMU_RW;
2903         success = true;
2904     }
2905 
2906     if (likely(success)) {
2907         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
2908                                  VTD_PCI_SLOT(vtd_as->devfn),
2909                                  VTD_PCI_FUNC(vtd_as->devfn),
2910                                  iotlb.iova, iotlb.translated_addr,
2911                                  iotlb.addr_mask);
2912     } else {
2913         error_report_once("%s: detected translation failure "
2914                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
2915                           __func__, pci_bus_num(vtd_as->bus),
2916                           VTD_PCI_SLOT(vtd_as->devfn),
2917                           VTD_PCI_FUNC(vtd_as->devfn),
2918                           addr);
2919     }
2920 
2921     return iotlb;
2922 }
2923 
2924 static void vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
2925                                           IOMMUNotifierFlag old,
2926                                           IOMMUNotifierFlag new)
2927 {
2928     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2929     IntelIOMMUState *s = vtd_as->iommu_state;
2930 
2931     if (!s->caching_mode && new & IOMMU_NOTIFIER_MAP) {
2932         error_report("We need to set caching-mode=on for intel-iommu to enable "
2933                      "device assignment with IOMMU protection.");
2934         exit(1);
2935     }
2936 
2937     /* Update per-address-space notifier flags */
2938     vtd_as->notifier_flags = new;
2939 
2940     if (old == IOMMU_NOTIFIER_NONE) {
2941         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
2942     } else if (new == IOMMU_NOTIFIER_NONE) {
2943         QLIST_REMOVE(vtd_as, next);
2944     }
2945 }
2946 
2947 static int vtd_post_load(void *opaque, int version_id)
2948 {
2949     IntelIOMMUState *iommu = opaque;
2950 
2951     /*
2952      * Memory regions are dynamically turned on/off depending on
2953      * context entry configurations from the guest. After migration,
2954      * we need to make sure the memory regions are still correct.
2955      */
2956     vtd_switch_address_space_all(iommu);
2957 
2958     /*
2959      * We don't need to migrate the root_scalable because we can
2960      * simply do the calculation after the loading is complete.  We
2961      * can actually do similar things with root, dmar_enabled, etc.
2962      * however since we've had them already so we'd better keep them
2963      * for compatibility of migration.
2964      */
2965     vtd_update_scalable_state(iommu);
2966 
2967     return 0;
2968 }
2969 
2970 static const VMStateDescription vtd_vmstate = {
2971     .name = "iommu-intel",
2972     .version_id = 1,
2973     .minimum_version_id = 1,
2974     .priority = MIG_PRI_IOMMU,
2975     .post_load = vtd_post_load,
2976     .fields = (VMStateField[]) {
2977         VMSTATE_UINT64(root, IntelIOMMUState),
2978         VMSTATE_UINT64(intr_root, IntelIOMMUState),
2979         VMSTATE_UINT64(iq, IntelIOMMUState),
2980         VMSTATE_UINT32(intr_size, IntelIOMMUState),
2981         VMSTATE_UINT16(iq_head, IntelIOMMUState),
2982         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
2983         VMSTATE_UINT16(iq_size, IntelIOMMUState),
2984         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
2985         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
2986         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
2987         VMSTATE_UNUSED(1),      /* bool root_extended is obsolete by VT-d */
2988         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
2989         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
2990         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
2991         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
2992         VMSTATE_END_OF_LIST()
2993     }
2994 };
2995 
2996 static const MemoryRegionOps vtd_mem_ops = {
2997     .read = vtd_mem_read,
2998     .write = vtd_mem_write,
2999     .endianness = DEVICE_LITTLE_ENDIAN,
3000     .impl = {
3001         .min_access_size = 4,
3002         .max_access_size = 8,
3003     },
3004     .valid = {
3005         .min_access_size = 4,
3006         .max_access_size = 8,
3007     },
3008 };
3009 
3010 static Property vtd_properties[] = {
3011     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
3012     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
3013                             ON_OFF_AUTO_AUTO),
3014     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
3015     DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits,
3016                       VTD_HOST_ADDRESS_WIDTH),
3017     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
3018     DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE),
3019     DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true),
3020     DEFINE_PROP_END_OF_LIST(),
3021 };
3022 
3023 /* Read IRTE entry with specific index */
3024 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
3025                         VTD_IR_TableEntry *entry, uint16_t sid)
3026 {
3027     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
3028         {0xffff, 0xfffb, 0xfff9, 0xfff8};
3029     dma_addr_t addr = 0x00;
3030     uint16_t mask, source_id;
3031     uint8_t bus, bus_max, bus_min;
3032 
3033     addr = iommu->intr_root + index * sizeof(*entry);
3034     if (dma_memory_read(&address_space_memory, addr, entry,
3035                         sizeof(*entry))) {
3036         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
3037                           __func__, index, addr);
3038         return -VTD_FR_IR_ROOT_INVAL;
3039     }
3040 
3041     trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]),
3042                           le64_to_cpu(entry->data[0]));
3043 
3044     if (!entry->irte.present) {
3045         error_report_once("%s: detected non-present IRTE "
3046                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3047                           __func__, index, le64_to_cpu(entry->data[1]),
3048                           le64_to_cpu(entry->data[0]));
3049         return -VTD_FR_IR_ENTRY_P;
3050     }
3051 
3052     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
3053         entry->irte.__reserved_2) {
3054         error_report_once("%s: detected non-zero reserved IRTE "
3055                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3056                           __func__, index, le64_to_cpu(entry->data[1]),
3057                           le64_to_cpu(entry->data[0]));
3058         return -VTD_FR_IR_IRTE_RSVD;
3059     }
3060 
3061     if (sid != X86_IOMMU_SID_INVALID) {
3062         /* Validate IRTE SID */
3063         source_id = le32_to_cpu(entry->irte.source_id);
3064         switch (entry->irte.sid_vtype) {
3065         case VTD_SVT_NONE:
3066             break;
3067 
3068         case VTD_SVT_ALL:
3069             mask = vtd_svt_mask[entry->irte.sid_q];
3070             if ((source_id & mask) != (sid & mask)) {
3071                 error_report_once("%s: invalid IRTE SID "
3072                                   "(index=%u, sid=%u, source_id=%u)",
3073                                   __func__, index, sid, source_id);
3074                 return -VTD_FR_IR_SID_ERR;
3075             }
3076             break;
3077 
3078         case VTD_SVT_BUS:
3079             bus_max = source_id >> 8;
3080             bus_min = source_id & 0xff;
3081             bus = sid >> 8;
3082             if (bus > bus_max || bus < bus_min) {
3083                 error_report_once("%s: invalid SVT_BUS "
3084                                   "(index=%u, bus=%u, min=%u, max=%u)",
3085                                   __func__, index, bus, bus_min, bus_max);
3086                 return -VTD_FR_IR_SID_ERR;
3087             }
3088             break;
3089 
3090         default:
3091             error_report_once("%s: detected invalid IRTE SVT "
3092                               "(index=%u, type=%d)", __func__,
3093                               index, entry->irte.sid_vtype);
3094             /* Take this as verification failure. */
3095             return -VTD_FR_IR_SID_ERR;
3096             break;
3097         }
3098     }
3099 
3100     return 0;
3101 }
3102 
3103 /* Fetch IRQ information of specific IR index */
3104 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
3105                              X86IOMMUIrq *irq, uint16_t sid)
3106 {
3107     VTD_IR_TableEntry irte = {};
3108     int ret = 0;
3109 
3110     ret = vtd_irte_get(iommu, index, &irte, sid);
3111     if (ret) {
3112         return ret;
3113     }
3114 
3115     irq->trigger_mode = irte.irte.trigger_mode;
3116     irq->vector = irte.irte.vector;
3117     irq->delivery_mode = irte.irte.delivery_mode;
3118     irq->dest = le32_to_cpu(irte.irte.dest_id);
3119     if (!iommu->intr_eime) {
3120 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
3121 #define  VTD_IR_APIC_DEST_SHIFT        (8)
3122         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
3123             VTD_IR_APIC_DEST_SHIFT;
3124     }
3125     irq->dest_mode = irte.irte.dest_mode;
3126     irq->redir_hint = irte.irte.redir_hint;
3127 
3128     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
3129                        irq->delivery_mode, irq->dest, irq->dest_mode);
3130 
3131     return 0;
3132 }
3133 
3134 /* Interrupt remapping for MSI/MSI-X entry */
3135 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
3136                                    MSIMessage *origin,
3137                                    MSIMessage *translated,
3138                                    uint16_t sid)
3139 {
3140     int ret = 0;
3141     VTD_IR_MSIAddress addr;
3142     uint16_t index;
3143     X86IOMMUIrq irq = {};
3144 
3145     assert(origin && translated);
3146 
3147     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
3148 
3149     if (!iommu || !iommu->intr_enabled) {
3150         memcpy(translated, origin, sizeof(*origin));
3151         goto out;
3152     }
3153 
3154     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
3155         error_report_once("%s: MSI address high 32 bits non-zero detected: "
3156                           "address=0x%" PRIx64, __func__, origin->address);
3157         return -VTD_FR_IR_REQ_RSVD;
3158     }
3159 
3160     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
3161     if (addr.addr.__head != 0xfee) {
3162         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
3163                           __func__, addr.data);
3164         return -VTD_FR_IR_REQ_RSVD;
3165     }
3166 
3167     /* This is compatible mode. */
3168     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
3169         memcpy(translated, origin, sizeof(*origin));
3170         goto out;
3171     }
3172 
3173     index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l);
3174 
3175 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
3176 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
3177 
3178     if (addr.addr.sub_valid) {
3179         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3180         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
3181     }
3182 
3183     ret = vtd_remap_irq_get(iommu, index, &irq, sid);
3184     if (ret) {
3185         return ret;
3186     }
3187 
3188     if (addr.addr.sub_valid) {
3189         trace_vtd_ir_remap_type("MSI");
3190         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
3191             error_report_once("%s: invalid IR MSI "
3192                               "(sid=%u, address=0x%" PRIx64
3193                               ", data=0x%" PRIx32 ")",
3194                               __func__, sid, origin->address, origin->data);
3195             return -VTD_FR_IR_REQ_RSVD;
3196         }
3197     } else {
3198         uint8_t vector = origin->data & 0xff;
3199         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
3200 
3201         trace_vtd_ir_remap_type("IOAPIC");
3202         /* IOAPIC entry vector should be aligned with IRTE vector
3203          * (see vt-d spec 5.1.5.1). */
3204         if (vector != irq.vector) {
3205             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
3206         }
3207 
3208         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3209          * (see vt-d spec 5.1.5.1). */
3210         if (trigger_mode != irq.trigger_mode) {
3211             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
3212                                       irq.trigger_mode);
3213         }
3214     }
3215 
3216     /*
3217      * We'd better keep the last two bits, assuming that guest OS
3218      * might modify it. Keep it does not hurt after all.
3219      */
3220     irq.msi_addr_last_bits = addr.addr.__not_care;
3221 
3222     /* Translate X86IOMMUIrq to MSI message */
3223     x86_iommu_irq_to_msi_message(&irq, translated);
3224 
3225 out:
3226     trace_vtd_ir_remap_msi(origin->address, origin->data,
3227                            translated->address, translated->data);
3228     return 0;
3229 }
3230 
3231 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
3232                          MSIMessage *dst, uint16_t sid)
3233 {
3234     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
3235                                    src, dst, sid);
3236 }
3237 
3238 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
3239                                    uint64_t *data, unsigned size,
3240                                    MemTxAttrs attrs)
3241 {
3242     return MEMTX_OK;
3243 }
3244 
3245 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
3246                                     uint64_t value, unsigned size,
3247                                     MemTxAttrs attrs)
3248 {
3249     int ret = 0;
3250     MSIMessage from = {}, to = {};
3251     uint16_t sid = X86_IOMMU_SID_INVALID;
3252 
3253     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
3254     from.data = (uint32_t) value;
3255 
3256     if (!attrs.unspecified) {
3257         /* We have explicit Source ID */
3258         sid = attrs.requester_id;
3259     }
3260 
3261     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid);
3262     if (ret) {
3263         /* TODO: report error */
3264         /* Drop this interrupt */
3265         return MEMTX_ERROR;
3266     }
3267 
3268     apic_get_class()->send_msi(&to);
3269 
3270     return MEMTX_OK;
3271 }
3272 
3273 static const MemoryRegionOps vtd_mem_ir_ops = {
3274     .read_with_attrs = vtd_mem_ir_read,
3275     .write_with_attrs = vtd_mem_ir_write,
3276     .endianness = DEVICE_LITTLE_ENDIAN,
3277     .impl = {
3278         .min_access_size = 4,
3279         .max_access_size = 4,
3280     },
3281     .valid = {
3282         .min_access_size = 4,
3283         .max_access_size = 4,
3284     },
3285 };
3286 
3287 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
3288 {
3289     uintptr_t key = (uintptr_t)bus;
3290     VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
3291     VTDAddressSpace *vtd_dev_as;
3292     char name[128];
3293 
3294     if (!vtd_bus) {
3295         uintptr_t *new_key = g_malloc(sizeof(*new_key));
3296         *new_key = (uintptr_t)bus;
3297         /* No corresponding free() */
3298         vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
3299                             PCI_DEVFN_MAX);
3300         vtd_bus->bus = bus;
3301         g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
3302     }
3303 
3304     vtd_dev_as = vtd_bus->dev_as[devfn];
3305 
3306     if (!vtd_dev_as) {
3307         snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn),
3308                  PCI_FUNC(devfn));
3309         vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace));
3310 
3311         vtd_dev_as->bus = bus;
3312         vtd_dev_as->devfn = (uint8_t)devfn;
3313         vtd_dev_as->iommu_state = s;
3314         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
3315         vtd_dev_as->iova_tree = iova_tree_new();
3316 
3317         memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX);
3318         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root");
3319 
3320         /*
3321          * Build the DMAR-disabled container with aliases to the
3322          * shared MRs.  Note that aliasing to a shared memory region
3323          * could help the memory API to detect same FlatViews so we
3324          * can have devices to share the same FlatView when DMAR is
3325          * disabled (either by not providing "intel_iommu=on" or with
3326          * "iommu=pt").  It will greatly reduce the total number of
3327          * FlatViews of the system hence VM runs faster.
3328          */
3329         memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s),
3330                                  "vtd-nodmar", &s->mr_nodmar, 0,
3331                                  memory_region_size(&s->mr_nodmar));
3332 
3333         /*
3334          * Build the per-device DMAR-enabled container.
3335          *
3336          * TODO: currently we have per-device IOMMU memory region only
3337          * because we have per-device IOMMU notifiers for devices.  If
3338          * one day we can abstract the IOMMU notifiers out of the
3339          * memory regions then we can also share the same memory
3340          * region here just like what we've done above with the nodmar
3341          * region.
3342          */
3343         strcat(name, "-dmar");
3344         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
3345                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
3346                                  name, UINT64_MAX);
3347         memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir",
3348                                  &s->mr_ir, 0, memory_region_size(&s->mr_ir));
3349         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
3350                                             VTD_INTERRUPT_ADDR_FIRST,
3351                                             &vtd_dev_as->iommu_ir, 1);
3352 
3353         /*
3354          * Hook both the containers under the root container, we
3355          * switch between DMAR & noDMAR by enable/disable
3356          * corresponding sub-containers
3357          */
3358         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3359                                             MEMORY_REGION(&vtd_dev_as->iommu),
3360                                             0);
3361         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3362                                             &vtd_dev_as->nodmar, 0);
3363 
3364         vtd_switch_address_space(vtd_dev_as);
3365     }
3366     return vtd_dev_as;
3367 }
3368 
3369 static uint64_t get_naturally_aligned_size(uint64_t start,
3370                                            uint64_t size, int gaw)
3371 {
3372     uint64_t max_mask = 1ULL << gaw;
3373     uint64_t alignment = start ? start & -start : max_mask;
3374 
3375     alignment = MIN(alignment, max_mask);
3376     size = MIN(size, max_mask);
3377 
3378     if (alignment <= size) {
3379         /* Increase the alignment of start */
3380         return alignment;
3381     } else {
3382         /* Find the largest page mask from size */
3383         return 1ULL << (63 - clz64(size));
3384     }
3385 }
3386 
3387 /* Unmap the whole range in the notifier's scope. */
3388 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
3389 {
3390     hwaddr size, remain;
3391     hwaddr start = n->start;
3392     hwaddr end = n->end;
3393     IntelIOMMUState *s = as->iommu_state;
3394     DMAMap map;
3395 
3396     /*
3397      * Note: all the codes in this function has a assumption that IOVA
3398      * bits are no more than VTD_MGAW bits (which is restricted by
3399      * VT-d spec), otherwise we need to consider overflow of 64 bits.
3400      */
3401 
3402     if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) {
3403         /*
3404          * Don't need to unmap regions that is bigger than the whole
3405          * VT-d supported address space size
3406          */
3407         end = VTD_ADDRESS_SIZE(s->aw_bits) - 1;
3408     }
3409 
3410     assert(start <= end);
3411     size = remain = end - start + 1;
3412 
3413     while (remain >= VTD_PAGE_SIZE) {
3414         IOMMUTLBEntry entry;
3415         uint64_t mask = get_naturally_aligned_size(start, remain, s->aw_bits);
3416 
3417         assert(mask);
3418 
3419         entry.iova = start;
3420         entry.addr_mask = mask - 1;
3421         entry.target_as = &address_space_memory;
3422         entry.perm = IOMMU_NONE;
3423         /* This field is meaningless for unmap */
3424         entry.translated_addr = 0;
3425 
3426         memory_region_notify_one(n, &entry);
3427 
3428         start += mask;
3429         remain -= mask;
3430     }
3431 
3432     assert(!remain);
3433 
3434     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3435                              VTD_PCI_SLOT(as->devfn),
3436                              VTD_PCI_FUNC(as->devfn),
3437                              n->start, size);
3438 
3439     map.iova = n->start;
3440     map.size = size;
3441     iova_tree_remove(as->iova_tree, &map);
3442 }
3443 
3444 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3445 {
3446     VTDAddressSpace *vtd_as;
3447     IOMMUNotifier *n;
3448 
3449     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3450         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3451             vtd_address_space_unmap(vtd_as, n);
3452         }
3453     }
3454 }
3455 
3456 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3457 {
3458     vtd_address_space_unmap_all(s);
3459     vtd_switch_address_space_all(s);
3460 }
3461 
3462 static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private)
3463 {
3464     memory_region_notify_one((IOMMUNotifier *)private, entry);
3465     return 0;
3466 }
3467 
3468 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3469 {
3470     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3471     IntelIOMMUState *s = vtd_as->iommu_state;
3472     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3473     VTDContextEntry ce;
3474 
3475     /*
3476      * The replay can be triggered by either a invalidation or a newly
3477      * created entry. No matter what, we release existing mappings
3478      * (it means flushing caches for UNMAP-only registers).
3479      */
3480     vtd_address_space_unmap(vtd_as, n);
3481 
3482     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3483         trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" :
3484                                   "legacy mode",
3485                                   bus_n, PCI_SLOT(vtd_as->devfn),
3486                                   PCI_FUNC(vtd_as->devfn),
3487                                   vtd_get_domain_id(s, &ce),
3488                                   ce.hi, ce.lo);
3489         if (vtd_as_has_map_notifier(vtd_as)) {
3490             /* This is required only for MAP typed notifiers */
3491             vtd_page_walk_info info = {
3492                 .hook_fn = vtd_replay_hook,
3493                 .private = (void *)n,
3494                 .notify_unmap = false,
3495                 .aw = s->aw_bits,
3496                 .as = vtd_as,
3497                 .domain_id = vtd_get_domain_id(s, &ce),
3498             };
3499 
3500             vtd_page_walk(s, &ce, 0, ~0ULL, &info);
3501         }
3502     } else {
3503         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3504                                     PCI_FUNC(vtd_as->devfn));
3505     }
3506 
3507     return;
3508 }
3509 
3510 /* Do the initialization. It will also be called when reset, so pay
3511  * attention when adding new initialization stuff.
3512  */
3513 static void vtd_init(IntelIOMMUState *s)
3514 {
3515     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3516 
3517     memset(s->csr, 0, DMAR_REG_SIZE);
3518     memset(s->wmask, 0, DMAR_REG_SIZE);
3519     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3520     memset(s->womask, 0, DMAR_REG_SIZE);
3521 
3522     s->root = 0;
3523     s->root_scalable = false;
3524     s->dmar_enabled = false;
3525     s->intr_enabled = false;
3526     s->iq_head = 0;
3527     s->iq_tail = 0;
3528     s->iq = 0;
3529     s->iq_size = 0;
3530     s->qi_enabled = false;
3531     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3532     s->iq_dw = false;
3533     s->next_frcd_reg = 0;
3534     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3535              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3536              VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits);
3537     if (s->dma_drain) {
3538         s->cap |= VTD_CAP_DRAIN;
3539     }
3540     if (s->aw_bits == VTD_HOST_AW_48BIT) {
3541         s->cap |= VTD_CAP_SAGAW_48bit;
3542     }
3543     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3544 
3545     /*
3546      * Rsvd field masks for spte
3547      */
3548     vtd_paging_entry_rsvd_field[0] = ~0ULL;
3549     vtd_paging_entry_rsvd_field[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits);
3550     vtd_paging_entry_rsvd_field[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3551     vtd_paging_entry_rsvd_field[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3552     vtd_paging_entry_rsvd_field[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3553     vtd_paging_entry_rsvd_field[5] = VTD_SPTE_LPAGE_L1_RSVD_MASK(s->aw_bits);
3554     vtd_paging_entry_rsvd_field[6] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits);
3555     vtd_paging_entry_rsvd_field[7] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits);
3556     vtd_paging_entry_rsvd_field[8] = VTD_SPTE_LPAGE_L4_RSVD_MASK(s->aw_bits);
3557 
3558     if (x86_iommu_ir_supported(x86_iommu)) {
3559         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3560         if (s->intr_eim == ON_OFF_AUTO_ON) {
3561             s->ecap |= VTD_ECAP_EIM;
3562         }
3563         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3564     }
3565 
3566     if (x86_iommu->dt_supported) {
3567         s->ecap |= VTD_ECAP_DT;
3568     }
3569 
3570     if (x86_iommu->pt_supported) {
3571         s->ecap |= VTD_ECAP_PT;
3572     }
3573 
3574     if (s->caching_mode) {
3575         s->cap |= VTD_CAP_CM;
3576     }
3577 
3578     /* TODO: read cap/ecap from host to decide which cap to be exposed. */
3579     if (s->scalable_mode) {
3580         s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS;
3581     }
3582 
3583     vtd_reset_caches(s);
3584 
3585     /* Define registers with default values and bit semantics */
3586     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
3587     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
3588     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
3589     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
3590     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
3591     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
3592     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0);
3593     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
3594     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
3595 
3596     /* Advanced Fault Logging not supported */
3597     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
3598     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3599     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
3600     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
3601 
3602     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3603      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3604      */
3605     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
3606 
3607     /* Treated as RO for implementations that PLMR and PHMR fields reported
3608      * as Clear in the CAP_REG.
3609      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3610      */
3611     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
3612 
3613     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
3614     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
3615     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0);
3616     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
3617     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3618     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
3619     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
3620     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3621     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
3622 
3623     /* IOTLB registers */
3624     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
3625     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
3626     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
3627 
3628     /* Fault Recording Registers, 128-bit */
3629     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
3630     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
3631 
3632     /*
3633      * Interrupt remapping registers.
3634      */
3635     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
3636 }
3637 
3638 /* Should not reset address_spaces when reset because devices will still use
3639  * the address space they got at first (won't ask the bus again).
3640  */
3641 static void vtd_reset(DeviceState *dev)
3642 {
3643     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3644 
3645     vtd_init(s);
3646     vtd_address_space_refresh_all(s);
3647 }
3648 
3649 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
3650 {
3651     IntelIOMMUState *s = opaque;
3652     VTDAddressSpace *vtd_as;
3653 
3654     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
3655 
3656     vtd_as = vtd_find_add_as(s, bus, devfn);
3657     return &vtd_as->as;
3658 }
3659 
3660 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
3661 {
3662     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3663 
3664     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) {
3665         error_setg(errp, "eim=on cannot be selected without intremap=on");
3666         return false;
3667     }
3668 
3669     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
3670         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
3671                       && x86_iommu_ir_supported(x86_iommu) ?
3672                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
3673     }
3674     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
3675         if (!kvm_irqchip_in_kernel()) {
3676             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
3677             return false;
3678         }
3679         if (!kvm_enable_x2apic()) {
3680             error_setg(errp, "eim=on requires support on the KVM side"
3681                              "(X2APIC_API, first shipped in v4.7)");
3682             return false;
3683         }
3684     }
3685 
3686     /* Currently only address widths supported are 39 and 48 bits */
3687     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
3688         (s->aw_bits != VTD_HOST_AW_48BIT)) {
3689         error_setg(errp, "Supported values for x-aw-bits are: %d, %d",
3690                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
3691         return false;
3692     }
3693 
3694     if (s->scalable_mode && !s->dma_drain) {
3695         error_setg(errp, "Need to set dma_drain for scalable mode");
3696         return false;
3697     }
3698 
3699     return true;
3700 }
3701 
3702 static void vtd_realize(DeviceState *dev, Error **errp)
3703 {
3704     MachineState *ms = MACHINE(qdev_get_machine());
3705     PCMachineState *pcms = PC_MACHINE(ms);
3706     PCIBus *bus = pcms->bus;
3707     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3708     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
3709 
3710     x86_iommu->type = TYPE_INTEL;
3711 
3712     if (!vtd_decide_config(s, errp)) {
3713         return;
3714     }
3715 
3716     QLIST_INIT(&s->vtd_as_with_notifiers);
3717     qemu_mutex_init(&s->iommu_lock);
3718     memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num));
3719     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
3720                           "intel_iommu", DMAR_REG_SIZE);
3721 
3722     /* Create the shared memory regions by all devices */
3723     memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar",
3724                        UINT64_MAX);
3725     memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops,
3726                           s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE);
3727     memory_region_init_alias(&s->mr_sys_alias, OBJECT(s),
3728                              "vtd-sys-alias", get_system_memory(), 0,
3729                              memory_region_size(get_system_memory()));
3730     memory_region_add_subregion_overlap(&s->mr_nodmar, 0,
3731                                         &s->mr_sys_alias, 0);
3732     memory_region_add_subregion_overlap(&s->mr_nodmar,
3733                                         VTD_INTERRUPT_ADDR_FIRST,
3734                                         &s->mr_ir, 1);
3735 
3736     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
3737     /* No corresponding destroy */
3738     s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3739                                      g_free, g_free);
3740     s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3741                                               g_free, g_free);
3742     vtd_init(s);
3743     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
3744     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
3745     /* Pseudo address space under root PCI bus. */
3746     pcms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
3747 }
3748 
3749 static void vtd_class_init(ObjectClass *klass, void *data)
3750 {
3751     DeviceClass *dc = DEVICE_CLASS(klass);
3752     X86IOMMUClass *x86_class = X86_IOMMU_CLASS(klass);
3753 
3754     dc->reset = vtd_reset;
3755     dc->vmsd = &vtd_vmstate;
3756     dc->props = vtd_properties;
3757     dc->hotpluggable = false;
3758     x86_class->realize = vtd_realize;
3759     x86_class->int_remap = vtd_int_remap;
3760     /* Supported by the pc-q35-* machine types */
3761     dc->user_creatable = true;
3762     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
3763     dc->desc = "Intel IOMMU (VT-d) DMA Remapping device";
3764 }
3765 
3766 static const TypeInfo vtd_info = {
3767     .name          = TYPE_INTEL_IOMMU_DEVICE,
3768     .parent        = TYPE_X86_IOMMU_DEVICE,
3769     .instance_size = sizeof(IntelIOMMUState),
3770     .class_init    = vtd_class_init,
3771 };
3772 
3773 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
3774                                                      void *data)
3775 {
3776     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
3777 
3778     imrc->translate = vtd_iommu_translate;
3779     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
3780     imrc->replay = vtd_iommu_replay;
3781 }
3782 
3783 static const TypeInfo vtd_iommu_memory_region_info = {
3784     .parent = TYPE_IOMMU_MEMORY_REGION,
3785     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
3786     .class_init = vtd_iommu_memory_region_class_init,
3787 };
3788 
3789 static void vtd_register_types(void)
3790 {
3791     type_register_static(&vtd_info);
3792     type_register_static(&vtd_iommu_memory_region_info);
3793 }
3794 
3795 type_init(vtd_register_types)
3796