xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision 979a8902)
1 /*
2  * QEMU emulation of an Intel IOMMU (VT-d)
3  *   (DMA Remapping device)
4  *
5  * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com>
6  * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12 
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17 
18  * You should have received a copy of the GNU General Public License along
19  * with this program; if not, see <http://www.gnu.org/licenses/>.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/error-report.h"
24 #include "qemu/main-loop.h"
25 #include "qapi/error.h"
26 #include "hw/sysbus.h"
27 #include "exec/address-spaces.h"
28 #include "intel_iommu_internal.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bus.h"
31 #include "hw/qdev-properties.h"
32 #include "hw/i386/pc.h"
33 #include "hw/i386/apic-msidef.h"
34 #include "hw/boards.h"
35 #include "hw/i386/x86-iommu.h"
36 #include "hw/pci-host/q35.h"
37 #include "sysemu/kvm.h"
38 #include "sysemu/sysemu.h"
39 #include "hw/i386/apic_internal.h"
40 #include "kvm_i386.h"
41 #include "migration/vmstate.h"
42 #include "trace.h"
43 
44 /* context entry operations */
45 #define VTD_CE_GET_RID2PASID(ce) \
46     ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK)
47 #define VTD_CE_GET_PASID_DIR_TABLE(ce) \
48     ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK)
49 
50 /* pe operations */
51 #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT)
52 #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW))
53 #define VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write) {\
54     if (ret_fr) {                                                             \
55         ret_fr = -ret_fr;                                                     \
56         if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) {                   \
57             trace_vtd_fault_disabled();                                       \
58         } else {                                                              \
59             vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);      \
60         }                                                                     \
61         goto error;                                                           \
62     }                                                                         \
63 }
64 
65 static void vtd_address_space_refresh_all(IntelIOMMUState *s);
66 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n);
67 
68 static void vtd_panic_require_caching_mode(void)
69 {
70     error_report("We need to set caching-mode=on for intel-iommu to enable "
71                  "device assignment with IOMMU protection.");
72     exit(1);
73 }
74 
75 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val,
76                             uint64_t wmask, uint64_t w1cmask)
77 {
78     stq_le_p(&s->csr[addr], val);
79     stq_le_p(&s->wmask[addr], wmask);
80     stq_le_p(&s->w1cmask[addr], w1cmask);
81 }
82 
83 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask)
84 {
85     stq_le_p(&s->womask[addr], mask);
86 }
87 
88 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val,
89                             uint32_t wmask, uint32_t w1cmask)
90 {
91     stl_le_p(&s->csr[addr], val);
92     stl_le_p(&s->wmask[addr], wmask);
93     stl_le_p(&s->w1cmask[addr], w1cmask);
94 }
95 
96 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask)
97 {
98     stl_le_p(&s->womask[addr], mask);
99 }
100 
101 /* "External" get/set operations */
102 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
103 {
104     uint64_t oldval = ldq_le_p(&s->csr[addr]);
105     uint64_t wmask = ldq_le_p(&s->wmask[addr]);
106     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
107     stq_le_p(&s->csr[addr],
108              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
109 }
110 
111 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
112 {
113     uint32_t oldval = ldl_le_p(&s->csr[addr]);
114     uint32_t wmask = ldl_le_p(&s->wmask[addr]);
115     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
116     stl_le_p(&s->csr[addr],
117              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
118 }
119 
120 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr)
121 {
122     uint64_t val = ldq_le_p(&s->csr[addr]);
123     uint64_t womask = ldq_le_p(&s->womask[addr]);
124     return val & ~womask;
125 }
126 
127 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr)
128 {
129     uint32_t val = ldl_le_p(&s->csr[addr]);
130     uint32_t womask = ldl_le_p(&s->womask[addr]);
131     return val & ~womask;
132 }
133 
134 /* "Internal" get/set operations */
135 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr)
136 {
137     return ldq_le_p(&s->csr[addr]);
138 }
139 
140 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr)
141 {
142     return ldl_le_p(&s->csr[addr]);
143 }
144 
145 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val)
146 {
147     stq_le_p(&s->csr[addr], val);
148 }
149 
150 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
151                                         uint32_t clear, uint32_t mask)
152 {
153     uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
154     stl_le_p(&s->csr[addr], new_val);
155     return new_val;
156 }
157 
158 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
159                                         uint64_t clear, uint64_t mask)
160 {
161     uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
162     stq_le_p(&s->csr[addr], new_val);
163     return new_val;
164 }
165 
166 static inline void vtd_iommu_lock(IntelIOMMUState *s)
167 {
168     qemu_mutex_lock(&s->iommu_lock);
169 }
170 
171 static inline void vtd_iommu_unlock(IntelIOMMUState *s)
172 {
173     qemu_mutex_unlock(&s->iommu_lock);
174 }
175 
176 static void vtd_update_scalable_state(IntelIOMMUState *s)
177 {
178     uint64_t val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
179 
180     if (s->scalable_mode) {
181         s->root_scalable = val & VTD_RTADDR_SMT;
182     }
183 }
184 
185 /* Whether the address space needs to notify new mappings */
186 static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as)
187 {
188     return as->notifier_flags & IOMMU_NOTIFIER_MAP;
189 }
190 
191 /* GHashTable functions */
192 static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2)
193 {
194     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
195 }
196 
197 static guint vtd_uint64_hash(gconstpointer v)
198 {
199     return (guint)*(const uint64_t *)v;
200 }
201 
202 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value,
203                                           gpointer user_data)
204 {
205     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
206     uint16_t domain_id = *(uint16_t *)user_data;
207     return entry->domain_id == domain_id;
208 }
209 
210 /* The shift of an addr for a certain level of paging structure */
211 static inline uint32_t vtd_slpt_level_shift(uint32_t level)
212 {
213     assert(level != 0);
214     return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
215 }
216 
217 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level)
218 {
219     return ~((1ULL << vtd_slpt_level_shift(level)) - 1);
220 }
221 
222 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value,
223                                         gpointer user_data)
224 {
225     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
226     VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data;
227     uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask;
228     uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K;
229     return (entry->domain_id == info->domain_id) &&
230             (((entry->gfn & info->mask) == gfn) ||
231              (entry->gfn == gfn_tlb));
232 }
233 
234 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
235  * IntelIOMMUState to 1.  Must be called with IOMMU lock held.
236  */
237 static void vtd_reset_context_cache_locked(IntelIOMMUState *s)
238 {
239     VTDAddressSpace *vtd_as;
240     VTDBus *vtd_bus;
241     GHashTableIter bus_it;
242     uint32_t devfn_it;
243 
244     trace_vtd_context_cache_reset();
245 
246     g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr);
247 
248     while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) {
249         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
250             vtd_as = vtd_bus->dev_as[devfn_it];
251             if (!vtd_as) {
252                 continue;
253             }
254             vtd_as->context_cache_entry.context_cache_gen = 0;
255         }
256     }
257     s->context_cache_gen = 1;
258 }
259 
260 /* Must be called with IOMMU lock held. */
261 static void vtd_reset_iotlb_locked(IntelIOMMUState *s)
262 {
263     assert(s->iotlb);
264     g_hash_table_remove_all(s->iotlb);
265 }
266 
267 static void vtd_reset_iotlb(IntelIOMMUState *s)
268 {
269     vtd_iommu_lock(s);
270     vtd_reset_iotlb_locked(s);
271     vtd_iommu_unlock(s);
272 }
273 
274 static void vtd_reset_caches(IntelIOMMUState *s)
275 {
276     vtd_iommu_lock(s);
277     vtd_reset_iotlb_locked(s);
278     vtd_reset_context_cache_locked(s);
279     vtd_iommu_unlock(s);
280 }
281 
282 static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id,
283                                   uint32_t level)
284 {
285     return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) |
286            ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT);
287 }
288 
289 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level)
290 {
291     return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K;
292 }
293 
294 /* Must be called with IOMMU lock held */
295 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id,
296                                        hwaddr addr)
297 {
298     VTDIOTLBEntry *entry;
299     uint64_t key;
300     int level;
301 
302     for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) {
303         key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level),
304                                 source_id, level);
305         entry = g_hash_table_lookup(s->iotlb, &key);
306         if (entry) {
307             goto out;
308         }
309     }
310 
311 out:
312     return entry;
313 }
314 
315 /* Must be with IOMMU lock held */
316 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id,
317                              uint16_t domain_id, hwaddr addr, uint64_t slpte,
318                              uint8_t access_flags, uint32_t level)
319 {
320     VTDIOTLBEntry *entry = g_malloc(sizeof(*entry));
321     uint64_t *key = g_malloc(sizeof(*key));
322     uint64_t gfn = vtd_get_iotlb_gfn(addr, level);
323 
324     trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id);
325     if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) {
326         trace_vtd_iotlb_reset("iotlb exceeds size limit");
327         vtd_reset_iotlb_locked(s);
328     }
329 
330     entry->gfn = gfn;
331     entry->domain_id = domain_id;
332     entry->slpte = slpte;
333     entry->access_flags = access_flags;
334     entry->mask = vtd_slpt_level_page_mask(level);
335     *key = vtd_get_iotlb_key(gfn, source_id, level);
336     g_hash_table_replace(s->iotlb, key, entry);
337 }
338 
339 /* Given the reg addr of both the message data and address, generate an
340  * interrupt via MSI.
341  */
342 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
343                                    hwaddr mesg_data_reg)
344 {
345     MSIMessage msi;
346 
347     assert(mesg_data_reg < DMAR_REG_SIZE);
348     assert(mesg_addr_reg < DMAR_REG_SIZE);
349 
350     msi.address = vtd_get_long_raw(s, mesg_addr_reg);
351     msi.data = vtd_get_long_raw(s, mesg_data_reg);
352 
353     trace_vtd_irq_generate(msi.address, msi.data);
354 
355     apic_get_class()->send_msi(&msi);
356 }
357 
358 /* Generate a fault event to software via MSI if conditions are met.
359  * Notice that the value of FSTS_REG being passed to it should be the one
360  * before any update.
361  */
362 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
363 {
364     if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
365         pre_fsts & VTD_FSTS_IQE) {
366         error_report_once("There are previous interrupt conditions "
367                           "to be serviced by software, fault event "
368                           "is not generated");
369         return;
370     }
371     vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
372     if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
373         error_report_once("Interrupt Mask set, irq is not generated");
374     } else {
375         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
376         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
377     }
378 }
379 
380 /* Check if the Fault (F) field of the Fault Recording Register referenced by
381  * @index is Set.
382  */
383 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index)
384 {
385     /* Each reg is 128-bit */
386     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
387     addr += 8; /* Access the high 64-bit half */
388 
389     assert(index < DMAR_FRCD_REG_NR);
390 
391     return vtd_get_quad_raw(s, addr) & VTD_FRCD_F;
392 }
393 
394 /* Update the PPF field of Fault Status Register.
395  * Should be called whenever change the F field of any fault recording
396  * registers.
397  */
398 static void vtd_update_fsts_ppf(IntelIOMMUState *s)
399 {
400     uint32_t i;
401     uint32_t ppf_mask = 0;
402 
403     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
404         if (vtd_is_frcd_set(s, i)) {
405             ppf_mask = VTD_FSTS_PPF;
406             break;
407         }
408     }
409     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
410     trace_vtd_fsts_ppf(!!ppf_mask);
411 }
412 
413 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index)
414 {
415     /* Each reg is 128-bit */
416     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
417     addr += 8; /* Access the high 64-bit half */
418 
419     assert(index < DMAR_FRCD_REG_NR);
420 
421     vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
422     vtd_update_fsts_ppf(s);
423 }
424 
425 /* Must not update F field now, should be done later */
426 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index,
427                             uint16_t source_id, hwaddr addr,
428                             VTDFaultReason fault, bool is_write)
429 {
430     uint64_t hi = 0, lo;
431     hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
432 
433     assert(index < DMAR_FRCD_REG_NR);
434 
435     lo = VTD_FRCD_FI(addr);
436     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
437     if (!is_write) {
438         hi |= VTD_FRCD_T;
439     }
440     vtd_set_quad_raw(s, frcd_reg_addr, lo);
441     vtd_set_quad_raw(s, frcd_reg_addr + 8, hi);
442 
443     trace_vtd_frr_new(index, hi, lo);
444 }
445 
446 /* Try to collapse multiple pending faults from the same requester */
447 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id)
448 {
449     uint32_t i;
450     uint64_t frcd_reg;
451     hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
452 
453     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
454         frcd_reg = vtd_get_quad_raw(s, addr);
455         if ((frcd_reg & VTD_FRCD_F) &&
456             ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
457             return true;
458         }
459         addr += 16; /* 128-bit for each */
460     }
461     return false;
462 }
463 
464 /* Log and report an DMAR (address translation) fault to software */
465 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
466                                   hwaddr addr, VTDFaultReason fault,
467                                   bool is_write)
468 {
469     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
470 
471     assert(fault < VTD_FR_MAX);
472 
473     if (fault == VTD_FR_RESERVED_ERR) {
474         /* This is not a normal fault reason case. Drop it. */
475         return;
476     }
477 
478     trace_vtd_dmar_fault(source_id, fault, addr, is_write);
479 
480     if (fsts_reg & VTD_FSTS_PFO) {
481         error_report_once("New fault is not recorded due to "
482                           "Primary Fault Overflow");
483         return;
484     }
485 
486     if (vtd_try_collapse_fault(s, source_id)) {
487         error_report_once("New fault is not recorded due to "
488                           "compression of faults");
489         return;
490     }
491 
492     if (vtd_is_frcd_set(s, s->next_frcd_reg)) {
493         error_report_once("Next Fault Recording Reg is used, "
494                           "new fault is not recorded, set PFO field");
495         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
496         return;
497     }
498 
499     vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write);
500 
501     if (fsts_reg & VTD_FSTS_PPF) {
502         error_report_once("There are pending faults already, "
503                           "fault event is not generated");
504         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg);
505         s->next_frcd_reg++;
506         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
507             s->next_frcd_reg = 0;
508         }
509     } else {
510         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
511                                 VTD_FSTS_FRI(s->next_frcd_reg));
512         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */
513         s->next_frcd_reg++;
514         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
515             s->next_frcd_reg = 0;
516         }
517         /* This case actually cause the PPF to be Set.
518          * So generate fault event (interrupt).
519          */
520          vtd_generate_fault_event(s, fsts_reg);
521     }
522 }
523 
524 /* Handle Invalidation Queue Errors of queued invalidation interface error
525  * conditions.
526  */
527 static void vtd_handle_inv_queue_error(IntelIOMMUState *s)
528 {
529     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
530 
531     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE);
532     vtd_generate_fault_event(s, fsts_reg);
533 }
534 
535 /* Set the IWC field and try to generate an invalidation completion interrupt */
536 static void vtd_generate_completion_event(IntelIOMMUState *s)
537 {
538     if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) {
539         trace_vtd_inv_desc_wait_irq("One pending, skip current");
540         return;
541     }
542     vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC);
543     vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP);
544     if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) {
545         trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
546                                     "new event not generated");
547         return;
548     } else {
549         /* Generate the interrupt event */
550         trace_vtd_inv_desc_wait_irq("Generating complete event");
551         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
552         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
553     }
554 }
555 
556 static inline bool vtd_root_entry_present(IntelIOMMUState *s,
557                                           VTDRootEntry *re,
558                                           uint8_t devfn)
559 {
560     if (s->root_scalable && devfn > UINT8_MAX / 2) {
561         return re->hi & VTD_ROOT_ENTRY_P;
562     }
563 
564     return re->lo & VTD_ROOT_ENTRY_P;
565 }
566 
567 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index,
568                               VTDRootEntry *re)
569 {
570     dma_addr_t addr;
571 
572     addr = s->root + index * sizeof(*re);
573     if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) {
574         re->lo = 0;
575         return -VTD_FR_ROOT_TABLE_INV;
576     }
577     re->lo = le64_to_cpu(re->lo);
578     re->hi = le64_to_cpu(re->hi);
579     return 0;
580 }
581 
582 static inline bool vtd_ce_present(VTDContextEntry *context)
583 {
584     return context->lo & VTD_CONTEXT_ENTRY_P;
585 }
586 
587 static int vtd_get_context_entry_from_root(IntelIOMMUState *s,
588                                            VTDRootEntry *re,
589                                            uint8_t index,
590                                            VTDContextEntry *ce)
591 {
592     dma_addr_t addr, ce_size;
593 
594     /* we have checked that root entry is present */
595     ce_size = s->root_scalable ? VTD_CTX_ENTRY_SCALABLE_SIZE :
596               VTD_CTX_ENTRY_LEGACY_SIZE;
597 
598     if (s->root_scalable && index > UINT8_MAX / 2) {
599         index = index & (~VTD_DEVFN_CHECK_MASK);
600         addr = re->hi & VTD_ROOT_ENTRY_CTP;
601     } else {
602         addr = re->lo & VTD_ROOT_ENTRY_CTP;
603     }
604 
605     addr = addr + index * ce_size;
606     if (dma_memory_read(&address_space_memory, addr, ce, ce_size)) {
607         return -VTD_FR_CONTEXT_TABLE_INV;
608     }
609 
610     ce->lo = le64_to_cpu(ce->lo);
611     ce->hi = le64_to_cpu(ce->hi);
612     if (ce_size == VTD_CTX_ENTRY_SCALABLE_SIZE) {
613         ce->val[2] = le64_to_cpu(ce->val[2]);
614         ce->val[3] = le64_to_cpu(ce->val[3]);
615     }
616     return 0;
617 }
618 
619 static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce)
620 {
621     return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
622 }
623 
624 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw)
625 {
626     return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw);
627 }
628 
629 /* Whether the pte indicates the address of the page frame */
630 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level)
631 {
632     return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
633 }
634 
635 /* Get the content of a spte located in @base_addr[@index] */
636 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index)
637 {
638     uint64_t slpte;
639 
640     assert(index < VTD_SL_PT_ENTRY_NR);
641 
642     if (dma_memory_read(&address_space_memory,
643                         base_addr + index * sizeof(slpte), &slpte,
644                         sizeof(slpte))) {
645         slpte = (uint64_t)-1;
646         return slpte;
647     }
648     slpte = le64_to_cpu(slpte);
649     return slpte;
650 }
651 
652 /* Given an iova and the level of paging structure, return the offset
653  * of current level.
654  */
655 static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level)
656 {
657     return (iova >> vtd_slpt_level_shift(level)) &
658             ((1ULL << VTD_SL_LEVEL_BITS) - 1);
659 }
660 
661 /* Check Capability Register to see if the @level of page-table is supported */
662 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level)
663 {
664     return VTD_CAP_SAGAW_MASK & s->cap &
665            (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
666 }
667 
668 /* Return true if check passed, otherwise false */
669 static inline bool vtd_pe_type_check(X86IOMMUState *x86_iommu,
670                                      VTDPASIDEntry *pe)
671 {
672     switch (VTD_PE_GET_TYPE(pe)) {
673     case VTD_SM_PASID_ENTRY_FLT:
674     case VTD_SM_PASID_ENTRY_SLT:
675     case VTD_SM_PASID_ENTRY_NESTED:
676         break;
677     case VTD_SM_PASID_ENTRY_PT:
678         if (!x86_iommu->pt_supported) {
679             return false;
680         }
681         break;
682     default:
683         /* Unknwon type */
684         return false;
685     }
686     return true;
687 }
688 
689 static inline bool vtd_pdire_present(VTDPASIDDirEntry *pdire)
690 {
691     return pdire->val & 1;
692 }
693 
694 /**
695  * Caller of this function should check present bit if wants
696  * to use pdir entry for futher usage except for fpd bit check.
697  */
698 static int vtd_get_pdire_from_pdir_table(dma_addr_t pasid_dir_base,
699                                          uint32_t pasid,
700                                          VTDPASIDDirEntry *pdire)
701 {
702     uint32_t index;
703     dma_addr_t addr, entry_size;
704 
705     index = VTD_PASID_DIR_INDEX(pasid);
706     entry_size = VTD_PASID_DIR_ENTRY_SIZE;
707     addr = pasid_dir_base + index * entry_size;
708     if (dma_memory_read(&address_space_memory, addr, pdire, entry_size)) {
709         return -VTD_FR_PASID_TABLE_INV;
710     }
711 
712     return 0;
713 }
714 
715 static inline bool vtd_pe_present(VTDPASIDEntry *pe)
716 {
717     return pe->val[0] & VTD_PASID_ENTRY_P;
718 }
719 
720 static int vtd_get_pe_in_pasid_leaf_table(IntelIOMMUState *s,
721                                           uint32_t pasid,
722                                           dma_addr_t addr,
723                                           VTDPASIDEntry *pe)
724 {
725     uint32_t index;
726     dma_addr_t entry_size;
727     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
728 
729     index = VTD_PASID_TABLE_INDEX(pasid);
730     entry_size = VTD_PASID_ENTRY_SIZE;
731     addr = addr + index * entry_size;
732     if (dma_memory_read(&address_space_memory, addr, pe, entry_size)) {
733         return -VTD_FR_PASID_TABLE_INV;
734     }
735 
736     /* Do translation type check */
737     if (!vtd_pe_type_check(x86_iommu, pe)) {
738         return -VTD_FR_PASID_TABLE_INV;
739     }
740 
741     if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) {
742         return -VTD_FR_PASID_TABLE_INV;
743     }
744 
745     return 0;
746 }
747 
748 /**
749  * Caller of this function should check present bit if wants
750  * to use pasid entry for futher usage except for fpd bit check.
751  */
752 static int vtd_get_pe_from_pdire(IntelIOMMUState *s,
753                                  uint32_t pasid,
754                                  VTDPASIDDirEntry *pdire,
755                                  VTDPASIDEntry *pe)
756 {
757     dma_addr_t addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK;
758 
759     return vtd_get_pe_in_pasid_leaf_table(s, pasid, addr, pe);
760 }
761 
762 /**
763  * This function gets a pasid entry from a specified pasid
764  * table (includes dir and leaf table) with a specified pasid.
765  * Sanity check should be done to ensure return a present
766  * pasid entry to caller.
767  */
768 static int vtd_get_pe_from_pasid_table(IntelIOMMUState *s,
769                                        dma_addr_t pasid_dir_base,
770                                        uint32_t pasid,
771                                        VTDPASIDEntry *pe)
772 {
773     int ret;
774     VTDPASIDDirEntry pdire;
775 
776     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base,
777                                         pasid, &pdire);
778     if (ret) {
779         return ret;
780     }
781 
782     if (!vtd_pdire_present(&pdire)) {
783         return -VTD_FR_PASID_TABLE_INV;
784     }
785 
786     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, pe);
787     if (ret) {
788         return ret;
789     }
790 
791     if (!vtd_pe_present(pe)) {
792         return -VTD_FR_PASID_TABLE_INV;
793     }
794 
795     return 0;
796 }
797 
798 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s,
799                                       VTDContextEntry *ce,
800                                       VTDPASIDEntry *pe)
801 {
802     uint32_t pasid;
803     dma_addr_t pasid_dir_base;
804     int ret = 0;
805 
806     pasid = VTD_CE_GET_RID2PASID(ce);
807     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
808     ret = vtd_get_pe_from_pasid_table(s, pasid_dir_base, pasid, pe);
809 
810     return ret;
811 }
812 
813 static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s,
814                                 VTDContextEntry *ce,
815                                 bool *pe_fpd_set)
816 {
817     int ret;
818     uint32_t pasid;
819     dma_addr_t pasid_dir_base;
820     VTDPASIDDirEntry pdire;
821     VTDPASIDEntry pe;
822 
823     pasid = VTD_CE_GET_RID2PASID(ce);
824     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
825 
826     /*
827      * No present bit check since fpd is meaningful even
828      * if the present bit is clear.
829      */
830     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base, pasid, &pdire);
831     if (ret) {
832         return ret;
833     }
834 
835     if (pdire.val & VTD_PASID_DIR_FPD) {
836         *pe_fpd_set = true;
837         return 0;
838     }
839 
840     if (!vtd_pdire_present(&pdire)) {
841         return -VTD_FR_PASID_TABLE_INV;
842     }
843 
844     /*
845      * No present bit check since fpd is meaningful even
846      * if the present bit is clear.
847      */
848     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, &pe);
849     if (ret) {
850         return ret;
851     }
852 
853     if (pe.val[0] & VTD_PASID_ENTRY_FPD) {
854         *pe_fpd_set = true;
855     }
856 
857     return 0;
858 }
859 
860 /* Get the page-table level that hardware should use for the second-level
861  * page-table walk from the Address Width field of context-entry.
862  */
863 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
864 {
865     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
866 }
867 
868 static uint32_t vtd_get_iova_level(IntelIOMMUState *s,
869                                    VTDContextEntry *ce)
870 {
871     VTDPASIDEntry pe;
872 
873     if (s->root_scalable) {
874         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
875         return VTD_PE_GET_LEVEL(&pe);
876     }
877 
878     return vtd_ce_get_level(ce);
879 }
880 
881 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
882 {
883     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
884 }
885 
886 static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s,
887                                   VTDContextEntry *ce)
888 {
889     VTDPASIDEntry pe;
890 
891     if (s->root_scalable) {
892         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
893         return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9;
894     }
895 
896     return vtd_ce_get_agaw(ce);
897 }
898 
899 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
900 {
901     return ce->lo & VTD_CONTEXT_ENTRY_TT;
902 }
903 
904 /* Only for Legacy Mode. Return true if check passed, otherwise false */
905 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
906                                      VTDContextEntry *ce)
907 {
908     switch (vtd_ce_get_type(ce)) {
909     case VTD_CONTEXT_TT_MULTI_LEVEL:
910         /* Always supported */
911         break;
912     case VTD_CONTEXT_TT_DEV_IOTLB:
913         if (!x86_iommu->dt_supported) {
914             error_report_once("%s: DT specified but not supported", __func__);
915             return false;
916         }
917         break;
918     case VTD_CONTEXT_TT_PASS_THROUGH:
919         if (!x86_iommu->pt_supported) {
920             error_report_once("%s: PT specified but not supported", __func__);
921             return false;
922         }
923         break;
924     default:
925         /* Unknown type */
926         error_report_once("%s: unknown ce type: %"PRIu32, __func__,
927                           vtd_ce_get_type(ce));
928         return false;
929     }
930     return true;
931 }
932 
933 static inline uint64_t vtd_iova_limit(IntelIOMMUState *s,
934                                       VTDContextEntry *ce, uint8_t aw)
935 {
936     uint32_t ce_agaw = vtd_get_iova_agaw(s, ce);
937     return 1ULL << MIN(ce_agaw, aw);
938 }
939 
940 /* Return true if IOVA passes range check, otherwise false. */
941 static inline bool vtd_iova_range_check(IntelIOMMUState *s,
942                                         uint64_t iova, VTDContextEntry *ce,
943                                         uint8_t aw)
944 {
945     /*
946      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
947      * in CAP_REG and AW in context-entry.
948      */
949     return !(iova & ~(vtd_iova_limit(s, ce, aw) - 1));
950 }
951 
952 static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s,
953                                           VTDContextEntry *ce)
954 {
955     VTDPASIDEntry pe;
956 
957     if (s->root_scalable) {
958         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
959         return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR;
960     }
961 
962     return vtd_ce_get_slpt_base(ce);
963 }
964 
965 /*
966  * Rsvd field masks for spte:
967  *     vtd_spte_rsvd 4k pages
968  *     vtd_spte_rsvd_large large pages
969  */
970 static uint64_t vtd_spte_rsvd[5];
971 static uint64_t vtd_spte_rsvd_large[5];
972 
973 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
974 {
975     uint64_t rsvd_mask = vtd_spte_rsvd[level];
976 
977     if ((level == VTD_SL_PD_LEVEL || level == VTD_SL_PDP_LEVEL) &&
978         (slpte & VTD_SL_PT_PAGE_SIZE_MASK)) {
979         /* large page */
980         rsvd_mask = vtd_spte_rsvd_large[level];
981     }
982 
983     return slpte & rsvd_mask;
984 }
985 
986 /* Find the VTD address space associated with a given bus number */
987 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num)
988 {
989     VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num];
990     if (!vtd_bus) {
991         /*
992          * Iterate over the registered buses to find the one which
993          * currently hold this bus number, and update the bus_num
994          * lookup table:
995          */
996         GHashTableIter iter;
997 
998         g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
999         while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1000             if (pci_bus_num(vtd_bus->bus) == bus_num) {
1001                 s->vtd_as_by_bus_num[bus_num] = vtd_bus;
1002                 return vtd_bus;
1003             }
1004         }
1005         vtd_bus = NULL;
1006     }
1007     return vtd_bus;
1008 }
1009 
1010 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
1011  * of the translation, can be used for deciding the size of large page.
1012  */
1013 static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce,
1014                              uint64_t iova, bool is_write,
1015                              uint64_t *slptep, uint32_t *slpte_level,
1016                              bool *reads, bool *writes, uint8_t aw_bits)
1017 {
1018     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1019     uint32_t level = vtd_get_iova_level(s, ce);
1020     uint32_t offset;
1021     uint64_t slpte;
1022     uint64_t access_right_check;
1023 
1024     if (!vtd_iova_range_check(s, iova, ce, aw_bits)) {
1025         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")",
1026                           __func__, iova);
1027         return -VTD_FR_ADDR_BEYOND_MGAW;
1028     }
1029 
1030     /* FIXME: what is the Atomics request here? */
1031     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
1032 
1033     while (true) {
1034         offset = vtd_iova_level_offset(iova, level);
1035         slpte = vtd_get_slpte(addr, offset);
1036 
1037         if (slpte == (uint64_t)-1) {
1038             error_report_once("%s: detected read error on DMAR slpte "
1039                               "(iova=0x%" PRIx64 ")", __func__, iova);
1040             if (level == vtd_get_iova_level(s, ce)) {
1041                 /* Invalid programming of context-entry */
1042                 return -VTD_FR_CONTEXT_ENTRY_INV;
1043             } else {
1044                 return -VTD_FR_PAGING_ENTRY_INV;
1045             }
1046         }
1047         *reads = (*reads) && (slpte & VTD_SL_R);
1048         *writes = (*writes) && (slpte & VTD_SL_W);
1049         if (!(slpte & access_right_check)) {
1050             error_report_once("%s: detected slpte permission error "
1051                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
1052                               "slpte=0x%" PRIx64 ", write=%d)", __func__,
1053                               iova, level, slpte, is_write);
1054             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
1055         }
1056         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1057             error_report_once("%s: detected splte reserve non-zero "
1058                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
1059                               "slpte=0x%" PRIx64 ")", __func__, iova,
1060                               level, slpte);
1061             return -VTD_FR_PAGING_ENTRY_RSVD;
1062         }
1063 
1064         if (vtd_is_last_slpte(slpte, level)) {
1065             *slptep = slpte;
1066             *slpte_level = level;
1067             return 0;
1068         }
1069         addr = vtd_get_slpte_addr(slpte, aw_bits);
1070         level--;
1071     }
1072 }
1073 
1074 typedef int (*vtd_page_walk_hook)(IOMMUTLBEntry *entry, void *private);
1075 
1076 /**
1077  * Constant information used during page walking
1078  *
1079  * @hook_fn: hook func to be called when detected page
1080  * @private: private data to be passed into hook func
1081  * @notify_unmap: whether we should notify invalid entries
1082  * @as: VT-d address space of the device
1083  * @aw: maximum address width
1084  * @domain: domain ID of the page walk
1085  */
1086 typedef struct {
1087     VTDAddressSpace *as;
1088     vtd_page_walk_hook hook_fn;
1089     void *private;
1090     bool notify_unmap;
1091     uint8_t aw;
1092     uint16_t domain_id;
1093 } vtd_page_walk_info;
1094 
1095 static int vtd_page_walk_one(IOMMUTLBEntry *entry, vtd_page_walk_info *info)
1096 {
1097     VTDAddressSpace *as = info->as;
1098     vtd_page_walk_hook hook_fn = info->hook_fn;
1099     void *private = info->private;
1100     DMAMap target = {
1101         .iova = entry->iova,
1102         .size = entry->addr_mask,
1103         .translated_addr = entry->translated_addr,
1104         .perm = entry->perm,
1105     };
1106     DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
1107 
1108     if (entry->perm == IOMMU_NONE && !info->notify_unmap) {
1109         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1110         return 0;
1111     }
1112 
1113     assert(hook_fn);
1114 
1115     /* Update local IOVA mapped ranges */
1116     if (entry->perm) {
1117         if (mapped) {
1118             /* If it's exactly the same translation, skip */
1119             if (!memcmp(mapped, &target, sizeof(target))) {
1120                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
1121                                                  entry->translated_addr);
1122                 return 0;
1123             } else {
1124                 /*
1125                  * Translation changed.  Normally this should not
1126                  * happen, but it can happen when with buggy guest
1127                  * OSes.  Note that there will be a small window that
1128                  * we don't have map at all.  But that's the best
1129                  * effort we can do.  The ideal way to emulate this is
1130                  * atomically modify the PTE to follow what has
1131                  * changed, but we can't.  One example is that vfio
1132                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1133                  * interface to modify a mapping (meanwhile it seems
1134                  * meaningless to even provide one).  Anyway, let's
1135                  * mark this as a TODO in case one day we'll have
1136                  * a better solution.
1137                  */
1138                 IOMMUAccessFlags cache_perm = entry->perm;
1139                 int ret;
1140 
1141                 /* Emulate an UNMAP */
1142                 entry->perm = IOMMU_NONE;
1143                 trace_vtd_page_walk_one(info->domain_id,
1144                                         entry->iova,
1145                                         entry->translated_addr,
1146                                         entry->addr_mask,
1147                                         entry->perm);
1148                 ret = hook_fn(entry, private);
1149                 if (ret) {
1150                     return ret;
1151                 }
1152                 /* Drop any existing mapping */
1153                 iova_tree_remove(as->iova_tree, &target);
1154                 /* Recover the correct permission */
1155                 entry->perm = cache_perm;
1156             }
1157         }
1158         iova_tree_insert(as->iova_tree, &target);
1159     } else {
1160         if (!mapped) {
1161             /* Skip since we didn't map this range at all */
1162             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1163             return 0;
1164         }
1165         iova_tree_remove(as->iova_tree, &target);
1166     }
1167 
1168     trace_vtd_page_walk_one(info->domain_id, entry->iova,
1169                             entry->translated_addr, entry->addr_mask,
1170                             entry->perm);
1171     return hook_fn(entry, private);
1172 }
1173 
1174 /**
1175  * vtd_page_walk_level - walk over specific level for IOVA range
1176  *
1177  * @addr: base GPA addr to start the walk
1178  * @start: IOVA range start address
1179  * @end: IOVA range end address (start <= addr < end)
1180  * @read: whether parent level has read permission
1181  * @write: whether parent level has write permission
1182  * @info: constant information for the page walk
1183  */
1184 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
1185                                uint64_t end, uint32_t level, bool read,
1186                                bool write, vtd_page_walk_info *info)
1187 {
1188     bool read_cur, write_cur, entry_valid;
1189     uint32_t offset;
1190     uint64_t slpte;
1191     uint64_t subpage_size, subpage_mask;
1192     IOMMUTLBEntry entry;
1193     uint64_t iova = start;
1194     uint64_t iova_next;
1195     int ret = 0;
1196 
1197     trace_vtd_page_walk_level(addr, level, start, end);
1198 
1199     subpage_size = 1ULL << vtd_slpt_level_shift(level);
1200     subpage_mask = vtd_slpt_level_page_mask(level);
1201 
1202     while (iova < end) {
1203         iova_next = (iova & subpage_mask) + subpage_size;
1204 
1205         offset = vtd_iova_level_offset(iova, level);
1206         slpte = vtd_get_slpte(addr, offset);
1207 
1208         if (slpte == (uint64_t)-1) {
1209             trace_vtd_page_walk_skip_read(iova, iova_next);
1210             goto next;
1211         }
1212 
1213         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1214             trace_vtd_page_walk_skip_reserve(iova, iova_next);
1215             goto next;
1216         }
1217 
1218         /* Permissions are stacked with parents' */
1219         read_cur = read && (slpte & VTD_SL_R);
1220         write_cur = write && (slpte & VTD_SL_W);
1221 
1222         /*
1223          * As long as we have either read/write permission, this is a
1224          * valid entry. The rule works for both page entries and page
1225          * table entries.
1226          */
1227         entry_valid = read_cur | write_cur;
1228 
1229         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
1230             /*
1231              * This is a valid PDE (or even bigger than PDE).  We need
1232              * to walk one further level.
1233              */
1234             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
1235                                       iova, MIN(iova_next, end), level - 1,
1236                                       read_cur, write_cur, info);
1237         } else {
1238             /*
1239              * This means we are either:
1240              *
1241              * (1) the real page entry (either 4K page, or huge page)
1242              * (2) the whole range is invalid
1243              *
1244              * In either case, we send an IOTLB notification down.
1245              */
1246             entry.target_as = &address_space_memory;
1247             entry.iova = iova & subpage_mask;
1248             entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
1249             entry.addr_mask = ~subpage_mask;
1250             /* NOTE: this is only meaningful if entry_valid == true */
1251             entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
1252             ret = vtd_page_walk_one(&entry, info);
1253         }
1254 
1255         if (ret < 0) {
1256             return ret;
1257         }
1258 
1259 next:
1260         iova = iova_next;
1261     }
1262 
1263     return 0;
1264 }
1265 
1266 /**
1267  * vtd_page_walk - walk specific IOVA range, and call the hook
1268  *
1269  * @s: intel iommu state
1270  * @ce: context entry to walk upon
1271  * @start: IOVA address to start the walk
1272  * @end: IOVA range end address (start <= addr < end)
1273  * @info: page walking information struct
1274  */
1275 static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce,
1276                          uint64_t start, uint64_t end,
1277                          vtd_page_walk_info *info)
1278 {
1279     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1280     uint32_t level = vtd_get_iova_level(s, ce);
1281 
1282     if (!vtd_iova_range_check(s, start, ce, info->aw)) {
1283         return -VTD_FR_ADDR_BEYOND_MGAW;
1284     }
1285 
1286     if (!vtd_iova_range_check(s, end, ce, info->aw)) {
1287         /* Fix end so that it reaches the maximum */
1288         end = vtd_iova_limit(s, ce, info->aw);
1289     }
1290 
1291     return vtd_page_walk_level(addr, start, end, level, true, true, info);
1292 }
1293 
1294 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s,
1295                                           VTDRootEntry *re)
1296 {
1297     /* Legacy Mode reserved bits check */
1298     if (!s->root_scalable &&
1299         (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1300         goto rsvd_err;
1301 
1302     /* Scalable Mode reserved bits check */
1303     if (s->root_scalable &&
1304         ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) ||
1305          (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1306         goto rsvd_err;
1307 
1308     return 0;
1309 
1310 rsvd_err:
1311     error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1312                       ", lo=0x%"PRIx64,
1313                       __func__, re->hi, re->lo);
1314     return -VTD_FR_ROOT_ENTRY_RSVD;
1315 }
1316 
1317 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s,
1318                                                     VTDContextEntry *ce)
1319 {
1320     if (!s->root_scalable &&
1321         (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI ||
1322          ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1323         error_report_once("%s: invalid context entry: hi=%"PRIx64
1324                           ", lo=%"PRIx64" (reserved nonzero)",
1325                           __func__, ce->hi, ce->lo);
1326         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1327     }
1328 
1329     if (s->root_scalable &&
1330         (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) ||
1331          ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 ||
1332          ce->val[2] ||
1333          ce->val[3])) {
1334         error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1335                           ", val[2]=%"PRIx64
1336                           ", val[1]=%"PRIx64
1337                           ", val[0]=%"PRIx64" (reserved nonzero)",
1338                           __func__, ce->val[3], ce->val[2],
1339                           ce->val[1], ce->val[0]);
1340         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1341     }
1342 
1343     return 0;
1344 }
1345 
1346 static int vtd_ce_rid2pasid_check(IntelIOMMUState *s,
1347                                   VTDContextEntry *ce)
1348 {
1349     VTDPASIDEntry pe;
1350 
1351     /*
1352      * Make sure in Scalable Mode, a present context entry
1353      * has valid rid2pasid setting, which includes valid
1354      * rid2pasid field and corresponding pasid entry setting
1355      */
1356     return vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1357 }
1358 
1359 /* Map a device to its corresponding domain (context-entry) */
1360 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
1361                                     uint8_t devfn, VTDContextEntry *ce)
1362 {
1363     VTDRootEntry re;
1364     int ret_fr;
1365     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
1366 
1367     ret_fr = vtd_get_root_entry(s, bus_num, &re);
1368     if (ret_fr) {
1369         return ret_fr;
1370     }
1371 
1372     if (!vtd_root_entry_present(s, &re, devfn)) {
1373         /* Not error - it's okay we don't have root entry. */
1374         trace_vtd_re_not_present(bus_num);
1375         return -VTD_FR_ROOT_ENTRY_P;
1376     }
1377 
1378     ret_fr = vtd_root_entry_rsvd_bits_check(s, &re);
1379     if (ret_fr) {
1380         return ret_fr;
1381     }
1382 
1383     ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce);
1384     if (ret_fr) {
1385         return ret_fr;
1386     }
1387 
1388     if (!vtd_ce_present(ce)) {
1389         /* Not error - it's okay we don't have context entry. */
1390         trace_vtd_ce_not_present(bus_num, devfn);
1391         return -VTD_FR_CONTEXT_ENTRY_P;
1392     }
1393 
1394     ret_fr = vtd_context_entry_rsvd_bits_check(s, ce);
1395     if (ret_fr) {
1396         return ret_fr;
1397     }
1398 
1399     /* Check if the programming of context-entry is valid */
1400     if (!s->root_scalable &&
1401         !vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1402         error_report_once("%s: invalid context entry: hi=%"PRIx64
1403                           ", lo=%"PRIx64" (level %d not supported)",
1404                           __func__, ce->hi, ce->lo,
1405                           vtd_ce_get_level(ce));
1406         return -VTD_FR_CONTEXT_ENTRY_INV;
1407     }
1408 
1409     if (!s->root_scalable) {
1410         /* Do translation type check */
1411         if (!vtd_ce_type_check(x86_iommu, ce)) {
1412             /* Errors dumped in vtd_ce_type_check() */
1413             return -VTD_FR_CONTEXT_ENTRY_INV;
1414         }
1415     } else {
1416         /*
1417          * Check if the programming of context-entry.rid2pasid
1418          * and corresponding pasid setting is valid, and thus
1419          * avoids to check pasid entry fetching result in future
1420          * helper function calling.
1421          */
1422         ret_fr = vtd_ce_rid2pasid_check(s, ce);
1423         if (ret_fr) {
1424             return ret_fr;
1425         }
1426     }
1427 
1428     return 0;
1429 }
1430 
1431 static int vtd_sync_shadow_page_hook(IOMMUTLBEntry *entry,
1432                                      void *private)
1433 {
1434     memory_region_notify_iommu((IOMMUMemoryRegion *)private, 0, *entry);
1435     return 0;
1436 }
1437 
1438 static uint16_t vtd_get_domain_id(IntelIOMMUState *s,
1439                                   VTDContextEntry *ce)
1440 {
1441     VTDPASIDEntry pe;
1442 
1443     if (s->root_scalable) {
1444         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1445         return VTD_SM_PASID_ENTRY_DID(pe.val[1]);
1446     }
1447 
1448     return VTD_CONTEXT_ENTRY_DID(ce->hi);
1449 }
1450 
1451 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1452                                             VTDContextEntry *ce,
1453                                             hwaddr addr, hwaddr size)
1454 {
1455     IntelIOMMUState *s = vtd_as->iommu_state;
1456     vtd_page_walk_info info = {
1457         .hook_fn = vtd_sync_shadow_page_hook,
1458         .private = (void *)&vtd_as->iommu,
1459         .notify_unmap = true,
1460         .aw = s->aw_bits,
1461         .as = vtd_as,
1462         .domain_id = vtd_get_domain_id(s, ce),
1463     };
1464 
1465     return vtd_page_walk(s, ce, addr, addr + size, &info);
1466 }
1467 
1468 static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
1469 {
1470     int ret;
1471     VTDContextEntry ce;
1472     IOMMUNotifier *n;
1473 
1474     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1475                                    pci_bus_num(vtd_as->bus),
1476                                    vtd_as->devfn, &ce);
1477     if (ret) {
1478         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1479             /*
1480              * It's a valid scenario to have a context entry that is
1481              * not present.  For example, when a device is removed
1482              * from an existing domain then the context entry will be
1483              * zeroed by the guest before it was put into another
1484              * domain.  When this happens, instead of synchronizing
1485              * the shadow pages we should invalidate all existing
1486              * mappings and notify the backends.
1487              */
1488             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1489                 vtd_address_space_unmap(vtd_as, n);
1490             }
1491             ret = 0;
1492         }
1493         return ret;
1494     }
1495 
1496     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1497 }
1498 
1499 /*
1500  * Check if specific device is configed to bypass address
1501  * translation for DMA requests. In Scalable Mode, bypass
1502  * 1st-level translation or 2nd-level translation, it depends
1503  * on PGTT setting.
1504  */
1505 static bool vtd_dev_pt_enabled(VTDAddressSpace *as)
1506 {
1507     IntelIOMMUState *s;
1508     VTDContextEntry ce;
1509     VTDPASIDEntry pe;
1510     int ret;
1511 
1512     assert(as);
1513 
1514     s = as->iommu_state;
1515     ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus),
1516                                    as->devfn, &ce);
1517     if (ret) {
1518         /*
1519          * Possibly failed to parse the context entry for some reason
1520          * (e.g., during init, or any guest configuration errors on
1521          * context entries). We should assume PT not enabled for
1522          * safety.
1523          */
1524         return false;
1525     }
1526 
1527     if (s->root_scalable) {
1528         ret = vtd_ce_get_rid2pasid_entry(s, &ce, &pe);
1529         if (ret) {
1530             error_report_once("%s: vtd_ce_get_rid2pasid_entry error: %"PRId32,
1531                               __func__, ret);
1532             return false;
1533         }
1534         return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT);
1535     }
1536 
1537     return (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH);
1538 }
1539 
1540 /* Return whether the device is using IOMMU translation. */
1541 static bool vtd_switch_address_space(VTDAddressSpace *as)
1542 {
1543     bool use_iommu;
1544     /* Whether we need to take the BQL on our own */
1545     bool take_bql = !qemu_mutex_iothread_locked();
1546 
1547     assert(as);
1548 
1549     use_iommu = as->iommu_state->dmar_enabled && !vtd_dev_pt_enabled(as);
1550 
1551     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1552                                    VTD_PCI_SLOT(as->devfn),
1553                                    VTD_PCI_FUNC(as->devfn),
1554                                    use_iommu);
1555 
1556     /*
1557      * It's possible that we reach here without BQL, e.g., when called
1558      * from vtd_pt_enable_fast_path(). However the memory APIs need
1559      * it. We'd better make sure we have had it already, or, take it.
1560      */
1561     if (take_bql) {
1562         qemu_mutex_lock_iothread();
1563     }
1564 
1565     /* Turn off first then on the other */
1566     if (use_iommu) {
1567         memory_region_set_enabled(&as->nodmar, false);
1568         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1569     } else {
1570         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1571         memory_region_set_enabled(&as->nodmar, true);
1572     }
1573 
1574     if (take_bql) {
1575         qemu_mutex_unlock_iothread();
1576     }
1577 
1578     return use_iommu;
1579 }
1580 
1581 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1582 {
1583     GHashTableIter iter;
1584     VTDBus *vtd_bus;
1585     int i;
1586 
1587     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1588     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1589         for (i = 0; i < PCI_DEVFN_MAX; i++) {
1590             if (!vtd_bus->dev_as[i]) {
1591                 continue;
1592             }
1593             vtd_switch_address_space(vtd_bus->dev_as[i]);
1594         }
1595     }
1596 }
1597 
1598 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn)
1599 {
1600     return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
1601 }
1602 
1603 static const bool vtd_qualified_faults[] = {
1604     [VTD_FR_RESERVED] = false,
1605     [VTD_FR_ROOT_ENTRY_P] = false,
1606     [VTD_FR_CONTEXT_ENTRY_P] = true,
1607     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1608     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1609     [VTD_FR_WRITE] = true,
1610     [VTD_FR_READ] = true,
1611     [VTD_FR_PAGING_ENTRY_INV] = true,
1612     [VTD_FR_ROOT_TABLE_INV] = false,
1613     [VTD_FR_CONTEXT_TABLE_INV] = false,
1614     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1615     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1616     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1617     [VTD_FR_PASID_TABLE_INV] = false,
1618     [VTD_FR_RESERVED_ERR] = false,
1619     [VTD_FR_MAX] = false,
1620 };
1621 
1622 /* To see if a fault condition is "qualified", which is reported to software
1623  * only if the FPD field in the context-entry used to process the faulting
1624  * request is 0.
1625  */
1626 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1627 {
1628     return vtd_qualified_faults[fault];
1629 }
1630 
1631 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1632 {
1633     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1634 }
1635 
1636 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1637 {
1638     VTDBus *vtd_bus;
1639     VTDAddressSpace *vtd_as;
1640     bool success = false;
1641 
1642     vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id));
1643     if (!vtd_bus) {
1644         goto out;
1645     }
1646 
1647     vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)];
1648     if (!vtd_as) {
1649         goto out;
1650     }
1651 
1652     if (vtd_switch_address_space(vtd_as) == false) {
1653         /* We switched off IOMMU region successfully. */
1654         success = true;
1655     }
1656 
1657 out:
1658     trace_vtd_pt_enable_fast_path(source_id, success);
1659 }
1660 
1661 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1662  * translation.
1663  *
1664  * Called from RCU critical section.
1665  *
1666  * @bus_num: The bus number
1667  * @devfn: The devfn, which is the  combined of device and function number
1668  * @is_write: The access is a write operation
1669  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1670  *
1671  * Returns true if translation is successful, otherwise false.
1672  */
1673 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1674                                    uint8_t devfn, hwaddr addr, bool is_write,
1675                                    IOMMUTLBEntry *entry)
1676 {
1677     IntelIOMMUState *s = vtd_as->iommu_state;
1678     VTDContextEntry ce;
1679     uint8_t bus_num = pci_bus_num(bus);
1680     VTDContextCacheEntry *cc_entry;
1681     uint64_t slpte, page_mask;
1682     uint32_t level;
1683     uint16_t source_id = vtd_make_source_id(bus_num, devfn);
1684     int ret_fr;
1685     bool is_fpd_set = false;
1686     bool reads = true;
1687     bool writes = true;
1688     uint8_t access_flags;
1689     VTDIOTLBEntry *iotlb_entry;
1690 
1691     /*
1692      * We have standalone memory region for interrupt addresses, we
1693      * should never receive translation requests in this region.
1694      */
1695     assert(!vtd_is_interrupt_addr(addr));
1696 
1697     vtd_iommu_lock(s);
1698 
1699     cc_entry = &vtd_as->context_cache_entry;
1700 
1701     /* Try to fetch slpte form IOTLB */
1702     iotlb_entry = vtd_lookup_iotlb(s, source_id, addr);
1703     if (iotlb_entry) {
1704         trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1705                                  iotlb_entry->domain_id);
1706         slpte = iotlb_entry->slpte;
1707         access_flags = iotlb_entry->access_flags;
1708         page_mask = iotlb_entry->mask;
1709         goto out;
1710     }
1711 
1712     /* Try to fetch context-entry from cache first */
1713     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1714         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1715                                cc_entry->context_entry.lo,
1716                                cc_entry->context_cache_gen);
1717         ce = cc_entry->context_entry;
1718         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1719         if (!is_fpd_set && s->root_scalable) {
1720             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1721             VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1722         }
1723     } else {
1724         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1725         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1726         if (!ret_fr && !is_fpd_set && s->root_scalable) {
1727             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1728         }
1729         VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1730         /* Update context-cache */
1731         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1732                                   cc_entry->context_cache_gen,
1733                                   s->context_cache_gen);
1734         cc_entry->context_entry = ce;
1735         cc_entry->context_cache_gen = s->context_cache_gen;
1736     }
1737 
1738     /*
1739      * We don't need to translate for pass-through context entries.
1740      * Also, let's ignore IOTLB caching as well for PT devices.
1741      */
1742     if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) {
1743         entry->iova = addr & VTD_PAGE_MASK_4K;
1744         entry->translated_addr = entry->iova;
1745         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1746         entry->perm = IOMMU_RW;
1747         trace_vtd_translate_pt(source_id, entry->iova);
1748 
1749         /*
1750          * When this happens, it means firstly caching-mode is not
1751          * enabled, and this is the first passthrough translation for
1752          * the device. Let's enable the fast path for passthrough.
1753          *
1754          * When passthrough is disabled again for the device, we can
1755          * capture it via the context entry invalidation, then the
1756          * IOMMU region can be swapped back.
1757          */
1758         vtd_pt_enable_fast_path(s, source_id);
1759         vtd_iommu_unlock(s);
1760         return true;
1761     }
1762 
1763     ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level,
1764                                &reads, &writes, s->aw_bits);
1765     VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1766 
1767     page_mask = vtd_slpt_level_page_mask(level);
1768     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1769     vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce), addr, slpte,
1770                      access_flags, level);
1771 out:
1772     vtd_iommu_unlock(s);
1773     entry->iova = addr & page_mask;
1774     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1775     entry->addr_mask = ~page_mask;
1776     entry->perm = access_flags;
1777     return true;
1778 
1779 error:
1780     vtd_iommu_unlock(s);
1781     entry->iova = 0;
1782     entry->translated_addr = 0;
1783     entry->addr_mask = 0;
1784     entry->perm = IOMMU_NONE;
1785     return false;
1786 }
1787 
1788 static void vtd_root_table_setup(IntelIOMMUState *s)
1789 {
1790     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1791     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
1792 
1793     vtd_update_scalable_state(s);
1794 
1795     trace_vtd_reg_dmar_root(s->root, s->root_scalable);
1796 }
1797 
1798 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
1799                                uint32_t index, uint32_t mask)
1800 {
1801     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
1802 }
1803 
1804 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
1805 {
1806     uint64_t value = 0;
1807     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
1808     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
1809     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
1810     s->intr_eime = value & VTD_IRTA_EIME;
1811 
1812     /* Notify global invalidation */
1813     vtd_iec_notify_all(s, true, 0, 0);
1814 
1815     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
1816 }
1817 
1818 static void vtd_iommu_replay_all(IntelIOMMUState *s)
1819 {
1820     VTDAddressSpace *vtd_as;
1821 
1822     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1823         vtd_sync_shadow_page_table(vtd_as);
1824     }
1825 }
1826 
1827 static void vtd_context_global_invalidate(IntelIOMMUState *s)
1828 {
1829     trace_vtd_inv_desc_cc_global();
1830     /* Protects context cache */
1831     vtd_iommu_lock(s);
1832     s->context_cache_gen++;
1833     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
1834         vtd_reset_context_cache_locked(s);
1835     }
1836     vtd_iommu_unlock(s);
1837     vtd_address_space_refresh_all(s);
1838     /*
1839      * From VT-d spec 6.5.2.1, a global context entry invalidation
1840      * should be followed by a IOTLB global invalidation, so we should
1841      * be safe even without this. Hoewever, let's replay the region as
1842      * well to be safer, and go back here when we need finer tunes for
1843      * VT-d emulation codes.
1844      */
1845     vtd_iommu_replay_all(s);
1846 }
1847 
1848 /* Do a context-cache device-selective invalidation.
1849  * @func_mask: FM field after shifting
1850  */
1851 static void vtd_context_device_invalidate(IntelIOMMUState *s,
1852                                           uint16_t source_id,
1853                                           uint16_t func_mask)
1854 {
1855     uint16_t mask;
1856     VTDBus *vtd_bus;
1857     VTDAddressSpace *vtd_as;
1858     uint8_t bus_n, devfn;
1859     uint16_t devfn_it;
1860 
1861     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
1862 
1863     switch (func_mask & 3) {
1864     case 0:
1865         mask = 0;   /* No bits in the SID field masked */
1866         break;
1867     case 1:
1868         mask = 4;   /* Mask bit 2 in the SID field */
1869         break;
1870     case 2:
1871         mask = 6;   /* Mask bit 2:1 in the SID field */
1872         break;
1873     case 3:
1874         mask = 7;   /* Mask bit 2:0 in the SID field */
1875         break;
1876     }
1877     mask = ~mask;
1878 
1879     bus_n = VTD_SID_TO_BUS(source_id);
1880     vtd_bus = vtd_find_as_from_bus_num(s, bus_n);
1881     if (vtd_bus) {
1882         devfn = VTD_SID_TO_DEVFN(source_id);
1883         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
1884             vtd_as = vtd_bus->dev_as[devfn_it];
1885             if (vtd_as && ((devfn_it & mask) == (devfn & mask))) {
1886                 trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it),
1887                                              VTD_PCI_FUNC(devfn_it));
1888                 vtd_iommu_lock(s);
1889                 vtd_as->context_cache_entry.context_cache_gen = 0;
1890                 vtd_iommu_unlock(s);
1891                 /*
1892                  * Do switch address space when needed, in case if the
1893                  * device passthrough bit is switched.
1894                  */
1895                 vtd_switch_address_space(vtd_as);
1896                 /*
1897                  * So a device is moving out of (or moving into) a
1898                  * domain, resync the shadow page table.
1899                  * This won't bring bad even if we have no such
1900                  * notifier registered - the IOMMU notification
1901                  * framework will skip MAP notifications if that
1902                  * happened.
1903                  */
1904                 vtd_sync_shadow_page_table(vtd_as);
1905             }
1906         }
1907     }
1908 }
1909 
1910 /* Context-cache invalidation
1911  * Returns the Context Actual Invalidation Granularity.
1912  * @val: the content of the CCMD_REG
1913  */
1914 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
1915 {
1916     uint64_t caig;
1917     uint64_t type = val & VTD_CCMD_CIRG_MASK;
1918 
1919     switch (type) {
1920     case VTD_CCMD_DOMAIN_INVL:
1921         /* Fall through */
1922     case VTD_CCMD_GLOBAL_INVL:
1923         caig = VTD_CCMD_GLOBAL_INVL_A;
1924         vtd_context_global_invalidate(s);
1925         break;
1926 
1927     case VTD_CCMD_DEVICE_INVL:
1928         caig = VTD_CCMD_DEVICE_INVL_A;
1929         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
1930         break;
1931 
1932     default:
1933         error_report_once("%s: invalid context: 0x%" PRIx64,
1934                           __func__, val);
1935         caig = 0;
1936     }
1937     return caig;
1938 }
1939 
1940 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
1941 {
1942     trace_vtd_inv_desc_iotlb_global();
1943     vtd_reset_iotlb(s);
1944     vtd_iommu_replay_all(s);
1945 }
1946 
1947 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
1948 {
1949     VTDContextEntry ce;
1950     VTDAddressSpace *vtd_as;
1951 
1952     trace_vtd_inv_desc_iotlb_domain(domain_id);
1953 
1954     vtd_iommu_lock(s);
1955     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
1956                                 &domain_id);
1957     vtd_iommu_unlock(s);
1958 
1959     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1960         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1961                                       vtd_as->devfn, &ce) &&
1962             domain_id == vtd_get_domain_id(s, &ce)) {
1963             vtd_sync_shadow_page_table(vtd_as);
1964         }
1965     }
1966 }
1967 
1968 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
1969                                            uint16_t domain_id, hwaddr addr,
1970                                            uint8_t am)
1971 {
1972     VTDAddressSpace *vtd_as;
1973     VTDContextEntry ce;
1974     int ret;
1975     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
1976 
1977     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
1978         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1979                                        vtd_as->devfn, &ce);
1980         if (!ret && domain_id == vtd_get_domain_id(s, &ce)) {
1981             if (vtd_as_has_map_notifier(vtd_as)) {
1982                 /*
1983                  * As long as we have MAP notifications registered in
1984                  * any of our IOMMU notifiers, we need to sync the
1985                  * shadow page table.
1986                  */
1987                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
1988             } else {
1989                 /*
1990                  * For UNMAP-only notifiers, we don't need to walk the
1991                  * page tables.  We just deliver the PSI down to
1992                  * invalidate caches.
1993                  */
1994                 IOMMUTLBEntry entry = {
1995                     .target_as = &address_space_memory,
1996                     .iova = addr,
1997                     .translated_addr = 0,
1998                     .addr_mask = size - 1,
1999                     .perm = IOMMU_NONE,
2000                 };
2001                 memory_region_notify_iommu(&vtd_as->iommu, 0, entry);
2002             }
2003         }
2004     }
2005 }
2006 
2007 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
2008                                       hwaddr addr, uint8_t am)
2009 {
2010     VTDIOTLBPageInvInfo info;
2011 
2012     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
2013 
2014     assert(am <= VTD_MAMV);
2015     info.domain_id = domain_id;
2016     info.addr = addr;
2017     info.mask = ~((1 << am) - 1);
2018     vtd_iommu_lock(s);
2019     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
2020     vtd_iommu_unlock(s);
2021     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am);
2022 }
2023 
2024 /* Flush IOTLB
2025  * Returns the IOTLB Actual Invalidation Granularity.
2026  * @val: the content of the IOTLB_REG
2027  */
2028 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
2029 {
2030     uint64_t iaig;
2031     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
2032     uint16_t domain_id;
2033     hwaddr addr;
2034     uint8_t am;
2035 
2036     switch (type) {
2037     case VTD_TLB_GLOBAL_FLUSH:
2038         iaig = VTD_TLB_GLOBAL_FLUSH_A;
2039         vtd_iotlb_global_invalidate(s);
2040         break;
2041 
2042     case VTD_TLB_DSI_FLUSH:
2043         domain_id = VTD_TLB_DID(val);
2044         iaig = VTD_TLB_DSI_FLUSH_A;
2045         vtd_iotlb_domain_invalidate(s, domain_id);
2046         break;
2047 
2048     case VTD_TLB_PSI_FLUSH:
2049         domain_id = VTD_TLB_DID(val);
2050         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
2051         am = VTD_IVA_AM(addr);
2052         addr = VTD_IVA_ADDR(addr);
2053         if (am > VTD_MAMV) {
2054             error_report_once("%s: address mask overflow: 0x%" PRIx64,
2055                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
2056             iaig = 0;
2057             break;
2058         }
2059         iaig = VTD_TLB_PSI_FLUSH_A;
2060         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2061         break;
2062 
2063     default:
2064         error_report_once("%s: invalid granularity: 0x%" PRIx64,
2065                           __func__, val);
2066         iaig = 0;
2067     }
2068     return iaig;
2069 }
2070 
2071 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
2072 
2073 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
2074 {
2075     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
2076            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
2077 }
2078 
2079 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
2080 {
2081     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
2082 
2083     trace_vtd_inv_qi_enable(en);
2084 
2085     if (en) {
2086         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
2087         /* 2^(x+8) entries */
2088         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0));
2089         s->qi_enabled = true;
2090         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
2091         /* Ok - report back to driver */
2092         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
2093 
2094         if (s->iq_tail != 0) {
2095             /*
2096              * This is a spec violation but Windows guests are known to set up
2097              * Queued Invalidation this way so we allow the write and process
2098              * Invalidation Descriptors right away.
2099              */
2100             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
2101             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2102                 vtd_fetch_inv_desc(s);
2103             }
2104         }
2105     } else {
2106         if (vtd_queued_inv_disable_check(s)) {
2107             /* disable Queued Invalidation */
2108             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
2109             s->iq_head = 0;
2110             s->qi_enabled = false;
2111             /* Ok - report back to driver */
2112             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
2113         } else {
2114             error_report_once("%s: detected improper state when disable QI "
2115                               "(head=0x%x, tail=0x%x, last_type=%d)",
2116                               __func__,
2117                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
2118         }
2119     }
2120 }
2121 
2122 /* Set Root Table Pointer */
2123 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
2124 {
2125     vtd_root_table_setup(s);
2126     /* Ok - report back to driver */
2127     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
2128     vtd_reset_caches(s);
2129     vtd_address_space_refresh_all(s);
2130 }
2131 
2132 /* Set Interrupt Remap Table Pointer */
2133 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
2134 {
2135     vtd_interrupt_remap_table_setup(s);
2136     /* Ok - report back to driver */
2137     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
2138 }
2139 
2140 /* Handle Translation Enable/Disable */
2141 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
2142 {
2143     if (s->dmar_enabled == en) {
2144         return;
2145     }
2146 
2147     trace_vtd_dmar_enable(en);
2148 
2149     if (en) {
2150         s->dmar_enabled = true;
2151         /* Ok - report back to driver */
2152         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
2153     } else {
2154         s->dmar_enabled = false;
2155 
2156         /* Clear the index of Fault Recording Register */
2157         s->next_frcd_reg = 0;
2158         /* Ok - report back to driver */
2159         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
2160     }
2161 
2162     vtd_reset_caches(s);
2163     vtd_address_space_refresh_all(s);
2164 }
2165 
2166 /* Handle Interrupt Remap Enable/Disable */
2167 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
2168 {
2169     trace_vtd_ir_enable(en);
2170 
2171     if (en) {
2172         s->intr_enabled = true;
2173         /* Ok - report back to driver */
2174         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
2175     } else {
2176         s->intr_enabled = false;
2177         /* Ok - report back to driver */
2178         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
2179     }
2180 }
2181 
2182 /* Handle write to Global Command Register */
2183 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
2184 {
2185     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
2186     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
2187     uint32_t changed = status ^ val;
2188 
2189     trace_vtd_reg_write_gcmd(status, val);
2190     if (changed & VTD_GCMD_TE) {
2191         /* Translation enable/disable */
2192         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
2193     }
2194     if (val & VTD_GCMD_SRTP) {
2195         /* Set/update the root-table pointer */
2196         vtd_handle_gcmd_srtp(s);
2197     }
2198     if (changed & VTD_GCMD_QIE) {
2199         /* Queued Invalidation Enable */
2200         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
2201     }
2202     if (val & VTD_GCMD_SIRTP) {
2203         /* Set/update the interrupt remapping root-table pointer */
2204         vtd_handle_gcmd_sirtp(s);
2205     }
2206     if (changed & VTD_GCMD_IRE) {
2207         /* Interrupt remap enable/disable */
2208         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
2209     }
2210 }
2211 
2212 /* Handle write to Context Command Register */
2213 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
2214 {
2215     uint64_t ret;
2216     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
2217 
2218     /* Context-cache invalidation request */
2219     if (val & VTD_CCMD_ICC) {
2220         if (s->qi_enabled) {
2221             error_report_once("Queued Invalidation enabled, "
2222                               "should not use register-based invalidation");
2223             return;
2224         }
2225         ret = vtd_context_cache_invalidate(s, val);
2226         /* Invalidation completed. Change something to show */
2227         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
2228         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
2229                                       ret);
2230     }
2231 }
2232 
2233 /* Handle write to IOTLB Invalidation Register */
2234 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
2235 {
2236     uint64_t ret;
2237     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
2238 
2239     /* IOTLB invalidation request */
2240     if (val & VTD_TLB_IVT) {
2241         if (s->qi_enabled) {
2242             error_report_once("Queued Invalidation enabled, "
2243                               "should not use register-based invalidation");
2244             return;
2245         }
2246         ret = vtd_iotlb_flush(s, val);
2247         /* Invalidation completed. Change something to show */
2248         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
2249         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
2250                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
2251     }
2252 }
2253 
2254 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2255 static bool vtd_get_inv_desc(IntelIOMMUState *s,
2256                              VTDInvDesc *inv_desc)
2257 {
2258     dma_addr_t base_addr = s->iq;
2259     uint32_t offset = s->iq_head;
2260     uint32_t dw = s->iq_dw ? 32 : 16;
2261     dma_addr_t addr = base_addr + offset * dw;
2262 
2263     if (dma_memory_read(&address_space_memory, addr, inv_desc, dw)) {
2264         error_report_once("Read INV DESC failed.");
2265         return false;
2266     }
2267     inv_desc->lo = le64_to_cpu(inv_desc->lo);
2268     inv_desc->hi = le64_to_cpu(inv_desc->hi);
2269     if (dw == 32) {
2270         inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]);
2271         inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]);
2272     }
2273     return true;
2274 }
2275 
2276 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2277 {
2278     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
2279         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
2280         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2281                           " (reserved nonzero)", __func__, inv_desc->hi,
2282                           inv_desc->lo);
2283         return false;
2284     }
2285     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
2286         /* Status Write */
2287         uint32_t status_data = (uint32_t)(inv_desc->lo >>
2288                                VTD_INV_DESC_WAIT_DATA_SHIFT);
2289 
2290         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
2291 
2292         /* FIXME: need to be masked with HAW? */
2293         dma_addr_t status_addr = inv_desc->hi;
2294         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
2295         status_data = cpu_to_le32(status_data);
2296         if (dma_memory_write(&address_space_memory, status_addr, &status_data,
2297                              sizeof(status_data))) {
2298             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
2299             return false;
2300         }
2301     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
2302         /* Interrupt flag */
2303         vtd_generate_completion_event(s);
2304     } else {
2305         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2306                           " (unknown type)", __func__, inv_desc->hi,
2307                           inv_desc->lo);
2308         return false;
2309     }
2310     return true;
2311 }
2312 
2313 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
2314                                            VTDInvDesc *inv_desc)
2315 {
2316     uint16_t sid, fmask;
2317 
2318     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
2319         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2320                           " (reserved nonzero)", __func__, inv_desc->hi,
2321                           inv_desc->lo);
2322         return false;
2323     }
2324     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
2325     case VTD_INV_DESC_CC_DOMAIN:
2326         trace_vtd_inv_desc_cc_domain(
2327             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
2328         /* Fall through */
2329     case VTD_INV_DESC_CC_GLOBAL:
2330         vtd_context_global_invalidate(s);
2331         break;
2332 
2333     case VTD_INV_DESC_CC_DEVICE:
2334         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
2335         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
2336         vtd_context_device_invalidate(s, sid, fmask);
2337         break;
2338 
2339     default:
2340         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2341                           " (invalid type)", __func__, inv_desc->hi,
2342                           inv_desc->lo);
2343         return false;
2344     }
2345     return true;
2346 }
2347 
2348 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2349 {
2350     uint16_t domain_id;
2351     uint8_t am;
2352     hwaddr addr;
2353 
2354     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
2355         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
2356         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2357                           ", lo=0x%"PRIx64" (reserved bits unzero)\n",
2358                           __func__, inv_desc->hi, inv_desc->lo);
2359         return false;
2360     }
2361 
2362     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
2363     case VTD_INV_DESC_IOTLB_GLOBAL:
2364         vtd_iotlb_global_invalidate(s);
2365         break;
2366 
2367     case VTD_INV_DESC_IOTLB_DOMAIN:
2368         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2369         vtd_iotlb_domain_invalidate(s, domain_id);
2370         break;
2371 
2372     case VTD_INV_DESC_IOTLB_PAGE:
2373         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2374         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
2375         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
2376         if (am > VTD_MAMV) {
2377             error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2378                               ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)\n",
2379                               __func__, inv_desc->hi, inv_desc->lo,
2380                               am, (unsigned)VTD_MAMV);
2381             return false;
2382         }
2383         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2384         break;
2385 
2386     default:
2387         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2388                           ", lo=0x%"PRIx64" (type mismatch: 0x%llx)\n",
2389                           __func__, inv_desc->hi, inv_desc->lo,
2390                           inv_desc->lo & VTD_INV_DESC_IOTLB_G);
2391         return false;
2392     }
2393     return true;
2394 }
2395 
2396 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
2397                                      VTDInvDesc *inv_desc)
2398 {
2399     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
2400                            inv_desc->iec.index,
2401                            inv_desc->iec.index_mask);
2402 
2403     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
2404                        inv_desc->iec.index,
2405                        inv_desc->iec.index_mask);
2406     return true;
2407 }
2408 
2409 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
2410                                           VTDInvDesc *inv_desc)
2411 {
2412     VTDAddressSpace *vtd_dev_as;
2413     IOMMUTLBEntry entry;
2414     struct VTDBus *vtd_bus;
2415     hwaddr addr;
2416     uint64_t sz;
2417     uint16_t sid;
2418     uint8_t devfn;
2419     bool size;
2420     uint8_t bus_num;
2421 
2422     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2423     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2424     devfn = sid & 0xff;
2425     bus_num = sid >> 8;
2426     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2427 
2428     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2429         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2430         error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2431                           ", lo=%"PRIx64" (reserved nonzero)", __func__,
2432                           inv_desc->hi, inv_desc->lo);
2433         return false;
2434     }
2435 
2436     vtd_bus = vtd_find_as_from_bus_num(s, bus_num);
2437     if (!vtd_bus) {
2438         goto done;
2439     }
2440 
2441     vtd_dev_as = vtd_bus->dev_as[devfn];
2442     if (!vtd_dev_as) {
2443         goto done;
2444     }
2445 
2446     /* According to ATS spec table 2.4:
2447      * S = 0, bits 15:12 = xxxx     range size: 4K
2448      * S = 1, bits 15:12 = xxx0     range size: 8K
2449      * S = 1, bits 15:12 = xx01     range size: 16K
2450      * S = 1, bits 15:12 = x011     range size: 32K
2451      * S = 1, bits 15:12 = 0111     range size: 64K
2452      * ...
2453      */
2454     if (size) {
2455         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2456         addr &= ~(sz - 1);
2457     } else {
2458         sz = VTD_PAGE_SIZE;
2459     }
2460 
2461     entry.target_as = &vtd_dev_as->as;
2462     entry.addr_mask = sz - 1;
2463     entry.iova = addr;
2464     entry.perm = IOMMU_NONE;
2465     entry.translated_addr = 0;
2466     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry);
2467 
2468 done:
2469     return true;
2470 }
2471 
2472 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2473 {
2474     VTDInvDesc inv_desc;
2475     uint8_t desc_type;
2476 
2477     trace_vtd_inv_qi_head(s->iq_head);
2478     if (!vtd_get_inv_desc(s, &inv_desc)) {
2479         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2480         return false;
2481     }
2482 
2483     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2484     /* FIXME: should update at first or at last? */
2485     s->iq_last_desc_type = desc_type;
2486 
2487     switch (desc_type) {
2488     case VTD_INV_DESC_CC:
2489         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2490         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2491             return false;
2492         }
2493         break;
2494 
2495     case VTD_INV_DESC_IOTLB:
2496         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2497         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2498             return false;
2499         }
2500         break;
2501 
2502     /*
2503      * TODO: the entity of below two cases will be implemented in future series.
2504      * To make guest (which integrates scalable mode support patch set in
2505      * iommu driver) work, just return true is enough so far.
2506      */
2507     case VTD_INV_DESC_PC:
2508         break;
2509 
2510     case VTD_INV_DESC_PIOTLB:
2511         break;
2512 
2513     case VTD_INV_DESC_WAIT:
2514         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2515         if (!vtd_process_wait_desc(s, &inv_desc)) {
2516             return false;
2517         }
2518         break;
2519 
2520     case VTD_INV_DESC_IEC:
2521         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2522         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2523             return false;
2524         }
2525         break;
2526 
2527     case VTD_INV_DESC_DEVICE:
2528         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2529         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2530             return false;
2531         }
2532         break;
2533 
2534     default:
2535         error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64
2536                           " (unknown type)", __func__, inv_desc.hi,
2537                           inv_desc.lo);
2538         return false;
2539     }
2540     s->iq_head++;
2541     if (s->iq_head == s->iq_size) {
2542         s->iq_head = 0;
2543     }
2544     return true;
2545 }
2546 
2547 /* Try to fetch and process more Invalidation Descriptors */
2548 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2549 {
2550     trace_vtd_inv_qi_fetch();
2551 
2552     if (s->iq_tail >= s->iq_size) {
2553         /* Detects an invalid Tail pointer */
2554         error_report_once("%s: detected invalid QI tail "
2555                           "(tail=0x%x, size=0x%x)",
2556                           __func__, s->iq_tail, s->iq_size);
2557         vtd_handle_inv_queue_error(s);
2558         return;
2559     }
2560     while (s->iq_head != s->iq_tail) {
2561         if (!vtd_process_inv_desc(s)) {
2562             /* Invalidation Queue Errors */
2563             vtd_handle_inv_queue_error(s);
2564             break;
2565         }
2566         /* Must update the IQH_REG in time */
2567         vtd_set_quad_raw(s, DMAR_IQH_REG,
2568                          (((uint64_t)(s->iq_head)) << VTD_IQH_QH_SHIFT) &
2569                          VTD_IQH_QH_MASK);
2570     }
2571 }
2572 
2573 /* Handle write to Invalidation Queue Tail Register */
2574 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2575 {
2576     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2577 
2578     if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) {
2579         error_report_once("%s: RSV bit is set: val=0x%"PRIx64,
2580                           __func__, val);
2581         return;
2582     }
2583     s->iq_tail = VTD_IQT_QT(s->iq_dw, val);
2584     trace_vtd_inv_qi_tail(s->iq_tail);
2585 
2586     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2587         /* Process Invalidation Queue here */
2588         vtd_fetch_inv_desc(s);
2589     }
2590 }
2591 
2592 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2593 {
2594     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2595     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2596     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2597 
2598     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2599         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2600         trace_vtd_fsts_clear_ip();
2601     }
2602     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2603      * Descriptors if there are any when Queued Invalidation is enabled?
2604      */
2605 }
2606 
2607 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2608 {
2609     uint32_t fectl_reg;
2610     /* FIXME: when software clears the IM field, check the IP field. But do we
2611      * need to compare the old value and the new value to conclude that
2612      * software clears the IM field? Or just check if the IM field is zero?
2613      */
2614     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2615 
2616     trace_vtd_reg_write_fectl(fectl_reg);
2617 
2618     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2619         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2620         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2621     }
2622 }
2623 
2624 static void vtd_handle_ics_write(IntelIOMMUState *s)
2625 {
2626     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2627     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2628 
2629     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2630         trace_vtd_reg_ics_clear_ip();
2631         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2632     }
2633 }
2634 
2635 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2636 {
2637     uint32_t iectl_reg;
2638     /* FIXME: when software clears the IM field, check the IP field. But do we
2639      * need to compare the old value and the new value to conclude that
2640      * software clears the IM field? Or just check if the IM field is zero?
2641      */
2642     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2643 
2644     trace_vtd_reg_write_iectl(iectl_reg);
2645 
2646     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2647         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2648         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2649     }
2650 }
2651 
2652 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2653 {
2654     IntelIOMMUState *s = opaque;
2655     uint64_t val;
2656 
2657     trace_vtd_reg_read(addr, size);
2658 
2659     if (addr + size > DMAR_REG_SIZE) {
2660         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2661                           " size=0x%u", __func__, addr, size);
2662         return (uint64_t)-1;
2663     }
2664 
2665     switch (addr) {
2666     /* Root Table Address Register, 64-bit */
2667     case DMAR_RTADDR_REG:
2668         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
2669         if (size == 4) {
2670             val = val & ((1ULL << 32) - 1);
2671         }
2672         break;
2673 
2674     case DMAR_RTADDR_REG_HI:
2675         assert(size == 4);
2676         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG) >> 32;
2677         break;
2678 
2679     /* Invalidation Queue Address Register, 64-bit */
2680     case DMAR_IQA_REG:
2681         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2682         if (size == 4) {
2683             val = val & ((1ULL << 32) - 1);
2684         }
2685         break;
2686 
2687     case DMAR_IQA_REG_HI:
2688         assert(size == 4);
2689         val = s->iq >> 32;
2690         break;
2691 
2692     default:
2693         if (size == 4) {
2694             val = vtd_get_long(s, addr);
2695         } else {
2696             val = vtd_get_quad(s, addr);
2697         }
2698     }
2699 
2700     return val;
2701 }
2702 
2703 static void vtd_mem_write(void *opaque, hwaddr addr,
2704                           uint64_t val, unsigned size)
2705 {
2706     IntelIOMMUState *s = opaque;
2707 
2708     trace_vtd_reg_write(addr, size, val);
2709 
2710     if (addr + size > DMAR_REG_SIZE) {
2711         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2712                           " size=0x%u", __func__, addr, size);
2713         return;
2714     }
2715 
2716     switch (addr) {
2717     /* Global Command Register, 32-bit */
2718     case DMAR_GCMD_REG:
2719         vtd_set_long(s, addr, val);
2720         vtd_handle_gcmd_write(s);
2721         break;
2722 
2723     /* Context Command Register, 64-bit */
2724     case DMAR_CCMD_REG:
2725         if (size == 4) {
2726             vtd_set_long(s, addr, val);
2727         } else {
2728             vtd_set_quad(s, addr, val);
2729             vtd_handle_ccmd_write(s);
2730         }
2731         break;
2732 
2733     case DMAR_CCMD_REG_HI:
2734         assert(size == 4);
2735         vtd_set_long(s, addr, val);
2736         vtd_handle_ccmd_write(s);
2737         break;
2738 
2739     /* IOTLB Invalidation Register, 64-bit */
2740     case DMAR_IOTLB_REG:
2741         if (size == 4) {
2742             vtd_set_long(s, addr, val);
2743         } else {
2744             vtd_set_quad(s, addr, val);
2745             vtd_handle_iotlb_write(s);
2746         }
2747         break;
2748 
2749     case DMAR_IOTLB_REG_HI:
2750         assert(size == 4);
2751         vtd_set_long(s, addr, val);
2752         vtd_handle_iotlb_write(s);
2753         break;
2754 
2755     /* Invalidate Address Register, 64-bit */
2756     case DMAR_IVA_REG:
2757         if (size == 4) {
2758             vtd_set_long(s, addr, val);
2759         } else {
2760             vtd_set_quad(s, addr, val);
2761         }
2762         break;
2763 
2764     case DMAR_IVA_REG_HI:
2765         assert(size == 4);
2766         vtd_set_long(s, addr, val);
2767         break;
2768 
2769     /* Fault Status Register, 32-bit */
2770     case DMAR_FSTS_REG:
2771         assert(size == 4);
2772         vtd_set_long(s, addr, val);
2773         vtd_handle_fsts_write(s);
2774         break;
2775 
2776     /* Fault Event Control Register, 32-bit */
2777     case DMAR_FECTL_REG:
2778         assert(size == 4);
2779         vtd_set_long(s, addr, val);
2780         vtd_handle_fectl_write(s);
2781         break;
2782 
2783     /* Fault Event Data Register, 32-bit */
2784     case DMAR_FEDATA_REG:
2785         assert(size == 4);
2786         vtd_set_long(s, addr, val);
2787         break;
2788 
2789     /* Fault Event Address Register, 32-bit */
2790     case DMAR_FEADDR_REG:
2791         if (size == 4) {
2792             vtd_set_long(s, addr, val);
2793         } else {
2794             /*
2795              * While the register is 32-bit only, some guests (Xen...) write to
2796              * it with 64-bit.
2797              */
2798             vtd_set_quad(s, addr, val);
2799         }
2800         break;
2801 
2802     /* Fault Event Upper Address Register, 32-bit */
2803     case DMAR_FEUADDR_REG:
2804         assert(size == 4);
2805         vtd_set_long(s, addr, val);
2806         break;
2807 
2808     /* Protected Memory Enable Register, 32-bit */
2809     case DMAR_PMEN_REG:
2810         assert(size == 4);
2811         vtd_set_long(s, addr, val);
2812         break;
2813 
2814     /* Root Table Address Register, 64-bit */
2815     case DMAR_RTADDR_REG:
2816         if (size == 4) {
2817             vtd_set_long(s, addr, val);
2818         } else {
2819             vtd_set_quad(s, addr, val);
2820         }
2821         break;
2822 
2823     case DMAR_RTADDR_REG_HI:
2824         assert(size == 4);
2825         vtd_set_long(s, addr, val);
2826         break;
2827 
2828     /* Invalidation Queue Tail Register, 64-bit */
2829     case DMAR_IQT_REG:
2830         if (size == 4) {
2831             vtd_set_long(s, addr, val);
2832         } else {
2833             vtd_set_quad(s, addr, val);
2834         }
2835         vtd_handle_iqt_write(s);
2836         break;
2837 
2838     case DMAR_IQT_REG_HI:
2839         assert(size == 4);
2840         vtd_set_long(s, addr, val);
2841         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
2842         break;
2843 
2844     /* Invalidation Queue Address Register, 64-bit */
2845     case DMAR_IQA_REG:
2846         if (size == 4) {
2847             vtd_set_long(s, addr, val);
2848         } else {
2849             vtd_set_quad(s, addr, val);
2850         }
2851         if (s->ecap & VTD_ECAP_SMTS &&
2852             val & VTD_IQA_DW_MASK) {
2853             s->iq_dw = true;
2854         } else {
2855             s->iq_dw = false;
2856         }
2857         break;
2858 
2859     case DMAR_IQA_REG_HI:
2860         assert(size == 4);
2861         vtd_set_long(s, addr, val);
2862         break;
2863 
2864     /* Invalidation Completion Status Register, 32-bit */
2865     case DMAR_ICS_REG:
2866         assert(size == 4);
2867         vtd_set_long(s, addr, val);
2868         vtd_handle_ics_write(s);
2869         break;
2870 
2871     /* Invalidation Event Control Register, 32-bit */
2872     case DMAR_IECTL_REG:
2873         assert(size == 4);
2874         vtd_set_long(s, addr, val);
2875         vtd_handle_iectl_write(s);
2876         break;
2877 
2878     /* Invalidation Event Data Register, 32-bit */
2879     case DMAR_IEDATA_REG:
2880         assert(size == 4);
2881         vtd_set_long(s, addr, val);
2882         break;
2883 
2884     /* Invalidation Event Address Register, 32-bit */
2885     case DMAR_IEADDR_REG:
2886         assert(size == 4);
2887         vtd_set_long(s, addr, val);
2888         break;
2889 
2890     /* Invalidation Event Upper Address Register, 32-bit */
2891     case DMAR_IEUADDR_REG:
2892         assert(size == 4);
2893         vtd_set_long(s, addr, val);
2894         break;
2895 
2896     /* Fault Recording Registers, 128-bit */
2897     case DMAR_FRCD_REG_0_0:
2898         if (size == 4) {
2899             vtd_set_long(s, addr, val);
2900         } else {
2901             vtd_set_quad(s, addr, val);
2902         }
2903         break;
2904 
2905     case DMAR_FRCD_REG_0_1:
2906         assert(size == 4);
2907         vtd_set_long(s, addr, val);
2908         break;
2909 
2910     case DMAR_FRCD_REG_0_2:
2911         if (size == 4) {
2912             vtd_set_long(s, addr, val);
2913         } else {
2914             vtd_set_quad(s, addr, val);
2915             /* May clear bit 127 (Fault), update PPF */
2916             vtd_update_fsts_ppf(s);
2917         }
2918         break;
2919 
2920     case DMAR_FRCD_REG_0_3:
2921         assert(size == 4);
2922         vtd_set_long(s, addr, val);
2923         /* May clear bit 127 (Fault), update PPF */
2924         vtd_update_fsts_ppf(s);
2925         break;
2926 
2927     case DMAR_IRTA_REG:
2928         if (size == 4) {
2929             vtd_set_long(s, addr, val);
2930         } else {
2931             vtd_set_quad(s, addr, val);
2932         }
2933         break;
2934 
2935     case DMAR_IRTA_REG_HI:
2936         assert(size == 4);
2937         vtd_set_long(s, addr, val);
2938         break;
2939 
2940     default:
2941         if (size == 4) {
2942             vtd_set_long(s, addr, val);
2943         } else {
2944             vtd_set_quad(s, addr, val);
2945         }
2946     }
2947 }
2948 
2949 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
2950                                          IOMMUAccessFlags flag, int iommu_idx)
2951 {
2952     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2953     IntelIOMMUState *s = vtd_as->iommu_state;
2954     IOMMUTLBEntry iotlb = {
2955         /* We'll fill in the rest later. */
2956         .target_as = &address_space_memory,
2957     };
2958     bool success;
2959 
2960     if (likely(s->dmar_enabled)) {
2961         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
2962                                          addr, flag & IOMMU_WO, &iotlb);
2963     } else {
2964         /* DMAR disabled, passthrough, use 4k-page*/
2965         iotlb.iova = addr & VTD_PAGE_MASK_4K;
2966         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
2967         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
2968         iotlb.perm = IOMMU_RW;
2969         success = true;
2970     }
2971 
2972     if (likely(success)) {
2973         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
2974                                  VTD_PCI_SLOT(vtd_as->devfn),
2975                                  VTD_PCI_FUNC(vtd_as->devfn),
2976                                  iotlb.iova, iotlb.translated_addr,
2977                                  iotlb.addr_mask);
2978     } else {
2979         error_report_once("%s: detected translation failure "
2980                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
2981                           __func__, pci_bus_num(vtd_as->bus),
2982                           VTD_PCI_SLOT(vtd_as->devfn),
2983                           VTD_PCI_FUNC(vtd_as->devfn),
2984                           addr);
2985     }
2986 
2987     return iotlb;
2988 }
2989 
2990 static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
2991                                          IOMMUNotifierFlag old,
2992                                          IOMMUNotifierFlag new,
2993                                          Error **errp)
2994 {
2995     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2996     IntelIOMMUState *s = vtd_as->iommu_state;
2997 
2998     /* Update per-address-space notifier flags */
2999     vtd_as->notifier_flags = new;
3000 
3001     if (old == IOMMU_NOTIFIER_NONE) {
3002         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
3003     } else if (new == IOMMU_NOTIFIER_NONE) {
3004         QLIST_REMOVE(vtd_as, next);
3005     }
3006     return 0;
3007 }
3008 
3009 static int vtd_post_load(void *opaque, int version_id)
3010 {
3011     IntelIOMMUState *iommu = opaque;
3012 
3013     /*
3014      * Memory regions are dynamically turned on/off depending on
3015      * context entry configurations from the guest. After migration,
3016      * we need to make sure the memory regions are still correct.
3017      */
3018     vtd_switch_address_space_all(iommu);
3019 
3020     /*
3021      * We don't need to migrate the root_scalable because we can
3022      * simply do the calculation after the loading is complete.  We
3023      * can actually do similar things with root, dmar_enabled, etc.
3024      * however since we've had them already so we'd better keep them
3025      * for compatibility of migration.
3026      */
3027     vtd_update_scalable_state(iommu);
3028 
3029     return 0;
3030 }
3031 
3032 static const VMStateDescription vtd_vmstate = {
3033     .name = "iommu-intel",
3034     .version_id = 1,
3035     .minimum_version_id = 1,
3036     .priority = MIG_PRI_IOMMU,
3037     .post_load = vtd_post_load,
3038     .fields = (VMStateField[]) {
3039         VMSTATE_UINT64(root, IntelIOMMUState),
3040         VMSTATE_UINT64(intr_root, IntelIOMMUState),
3041         VMSTATE_UINT64(iq, IntelIOMMUState),
3042         VMSTATE_UINT32(intr_size, IntelIOMMUState),
3043         VMSTATE_UINT16(iq_head, IntelIOMMUState),
3044         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
3045         VMSTATE_UINT16(iq_size, IntelIOMMUState),
3046         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
3047         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
3048         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
3049         VMSTATE_UNUSED(1),      /* bool root_extended is obsolete by VT-d */
3050         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
3051         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
3052         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
3053         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
3054         VMSTATE_END_OF_LIST()
3055     }
3056 };
3057 
3058 static const MemoryRegionOps vtd_mem_ops = {
3059     .read = vtd_mem_read,
3060     .write = vtd_mem_write,
3061     .endianness = DEVICE_LITTLE_ENDIAN,
3062     .impl = {
3063         .min_access_size = 4,
3064         .max_access_size = 8,
3065     },
3066     .valid = {
3067         .min_access_size = 4,
3068         .max_access_size = 8,
3069     },
3070 };
3071 
3072 static Property vtd_properties[] = {
3073     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
3074     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
3075                             ON_OFF_AUTO_AUTO),
3076     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
3077     DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits,
3078                       VTD_HOST_ADDRESS_WIDTH),
3079     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
3080     DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE),
3081     DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true),
3082     DEFINE_PROP_END_OF_LIST(),
3083 };
3084 
3085 /* Read IRTE entry with specific index */
3086 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
3087                         VTD_IR_TableEntry *entry, uint16_t sid)
3088 {
3089     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
3090         {0xffff, 0xfffb, 0xfff9, 0xfff8};
3091     dma_addr_t addr = 0x00;
3092     uint16_t mask, source_id;
3093     uint8_t bus, bus_max, bus_min;
3094 
3095     addr = iommu->intr_root + index * sizeof(*entry);
3096     if (dma_memory_read(&address_space_memory, addr, entry,
3097                         sizeof(*entry))) {
3098         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
3099                           __func__, index, addr);
3100         return -VTD_FR_IR_ROOT_INVAL;
3101     }
3102 
3103     trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]),
3104                           le64_to_cpu(entry->data[0]));
3105 
3106     if (!entry->irte.present) {
3107         error_report_once("%s: detected non-present IRTE "
3108                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3109                           __func__, index, le64_to_cpu(entry->data[1]),
3110                           le64_to_cpu(entry->data[0]));
3111         return -VTD_FR_IR_ENTRY_P;
3112     }
3113 
3114     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
3115         entry->irte.__reserved_2) {
3116         error_report_once("%s: detected non-zero reserved IRTE "
3117                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3118                           __func__, index, le64_to_cpu(entry->data[1]),
3119                           le64_to_cpu(entry->data[0]));
3120         return -VTD_FR_IR_IRTE_RSVD;
3121     }
3122 
3123     if (sid != X86_IOMMU_SID_INVALID) {
3124         /* Validate IRTE SID */
3125         source_id = le32_to_cpu(entry->irte.source_id);
3126         switch (entry->irte.sid_vtype) {
3127         case VTD_SVT_NONE:
3128             break;
3129 
3130         case VTD_SVT_ALL:
3131             mask = vtd_svt_mask[entry->irte.sid_q];
3132             if ((source_id & mask) != (sid & mask)) {
3133                 error_report_once("%s: invalid IRTE SID "
3134                                   "(index=%u, sid=%u, source_id=%u)",
3135                                   __func__, index, sid, source_id);
3136                 return -VTD_FR_IR_SID_ERR;
3137             }
3138             break;
3139 
3140         case VTD_SVT_BUS:
3141             bus_max = source_id >> 8;
3142             bus_min = source_id & 0xff;
3143             bus = sid >> 8;
3144             if (bus > bus_max || bus < bus_min) {
3145                 error_report_once("%s: invalid SVT_BUS "
3146                                   "(index=%u, bus=%u, min=%u, max=%u)",
3147                                   __func__, index, bus, bus_min, bus_max);
3148                 return -VTD_FR_IR_SID_ERR;
3149             }
3150             break;
3151 
3152         default:
3153             error_report_once("%s: detected invalid IRTE SVT "
3154                               "(index=%u, type=%d)", __func__,
3155                               index, entry->irte.sid_vtype);
3156             /* Take this as verification failure. */
3157             return -VTD_FR_IR_SID_ERR;
3158             break;
3159         }
3160     }
3161 
3162     return 0;
3163 }
3164 
3165 /* Fetch IRQ information of specific IR index */
3166 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
3167                              X86IOMMUIrq *irq, uint16_t sid)
3168 {
3169     VTD_IR_TableEntry irte = {};
3170     int ret = 0;
3171 
3172     ret = vtd_irte_get(iommu, index, &irte, sid);
3173     if (ret) {
3174         return ret;
3175     }
3176 
3177     irq->trigger_mode = irte.irte.trigger_mode;
3178     irq->vector = irte.irte.vector;
3179     irq->delivery_mode = irte.irte.delivery_mode;
3180     irq->dest = le32_to_cpu(irte.irte.dest_id);
3181     if (!iommu->intr_eime) {
3182 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
3183 #define  VTD_IR_APIC_DEST_SHIFT        (8)
3184         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
3185             VTD_IR_APIC_DEST_SHIFT;
3186     }
3187     irq->dest_mode = irte.irte.dest_mode;
3188     irq->redir_hint = irte.irte.redir_hint;
3189 
3190     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
3191                        irq->delivery_mode, irq->dest, irq->dest_mode);
3192 
3193     return 0;
3194 }
3195 
3196 /* Interrupt remapping for MSI/MSI-X entry */
3197 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
3198                                    MSIMessage *origin,
3199                                    MSIMessage *translated,
3200                                    uint16_t sid)
3201 {
3202     int ret = 0;
3203     VTD_IR_MSIAddress addr;
3204     uint16_t index;
3205     X86IOMMUIrq irq = {};
3206 
3207     assert(origin && translated);
3208 
3209     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
3210 
3211     if (!iommu || !iommu->intr_enabled) {
3212         memcpy(translated, origin, sizeof(*origin));
3213         goto out;
3214     }
3215 
3216     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
3217         error_report_once("%s: MSI address high 32 bits non-zero detected: "
3218                           "address=0x%" PRIx64, __func__, origin->address);
3219         return -VTD_FR_IR_REQ_RSVD;
3220     }
3221 
3222     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
3223     if (addr.addr.__head != 0xfee) {
3224         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
3225                           __func__, addr.data);
3226         return -VTD_FR_IR_REQ_RSVD;
3227     }
3228 
3229     /* This is compatible mode. */
3230     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
3231         memcpy(translated, origin, sizeof(*origin));
3232         goto out;
3233     }
3234 
3235     index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l);
3236 
3237 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
3238 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
3239 
3240     if (addr.addr.sub_valid) {
3241         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3242         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
3243     }
3244 
3245     ret = vtd_remap_irq_get(iommu, index, &irq, sid);
3246     if (ret) {
3247         return ret;
3248     }
3249 
3250     if (addr.addr.sub_valid) {
3251         trace_vtd_ir_remap_type("MSI");
3252         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
3253             error_report_once("%s: invalid IR MSI "
3254                               "(sid=%u, address=0x%" PRIx64
3255                               ", data=0x%" PRIx32 ")",
3256                               __func__, sid, origin->address, origin->data);
3257             return -VTD_FR_IR_REQ_RSVD;
3258         }
3259     } else {
3260         uint8_t vector = origin->data & 0xff;
3261         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
3262 
3263         trace_vtd_ir_remap_type("IOAPIC");
3264         /* IOAPIC entry vector should be aligned with IRTE vector
3265          * (see vt-d spec 5.1.5.1). */
3266         if (vector != irq.vector) {
3267             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
3268         }
3269 
3270         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3271          * (see vt-d spec 5.1.5.1). */
3272         if (trigger_mode != irq.trigger_mode) {
3273             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
3274                                       irq.trigger_mode);
3275         }
3276     }
3277 
3278     /*
3279      * We'd better keep the last two bits, assuming that guest OS
3280      * might modify it. Keep it does not hurt after all.
3281      */
3282     irq.msi_addr_last_bits = addr.addr.__not_care;
3283 
3284     /* Translate X86IOMMUIrq to MSI message */
3285     x86_iommu_irq_to_msi_message(&irq, translated);
3286 
3287 out:
3288     trace_vtd_ir_remap_msi(origin->address, origin->data,
3289                            translated->address, translated->data);
3290     return 0;
3291 }
3292 
3293 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
3294                          MSIMessage *dst, uint16_t sid)
3295 {
3296     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
3297                                    src, dst, sid);
3298 }
3299 
3300 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
3301                                    uint64_t *data, unsigned size,
3302                                    MemTxAttrs attrs)
3303 {
3304     return MEMTX_OK;
3305 }
3306 
3307 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
3308                                     uint64_t value, unsigned size,
3309                                     MemTxAttrs attrs)
3310 {
3311     int ret = 0;
3312     MSIMessage from = {}, to = {};
3313     uint16_t sid = X86_IOMMU_SID_INVALID;
3314 
3315     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
3316     from.data = (uint32_t) value;
3317 
3318     if (!attrs.unspecified) {
3319         /* We have explicit Source ID */
3320         sid = attrs.requester_id;
3321     }
3322 
3323     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid);
3324     if (ret) {
3325         /* TODO: report error */
3326         /* Drop this interrupt */
3327         return MEMTX_ERROR;
3328     }
3329 
3330     apic_get_class()->send_msi(&to);
3331 
3332     return MEMTX_OK;
3333 }
3334 
3335 static const MemoryRegionOps vtd_mem_ir_ops = {
3336     .read_with_attrs = vtd_mem_ir_read,
3337     .write_with_attrs = vtd_mem_ir_write,
3338     .endianness = DEVICE_LITTLE_ENDIAN,
3339     .impl = {
3340         .min_access_size = 4,
3341         .max_access_size = 4,
3342     },
3343     .valid = {
3344         .min_access_size = 4,
3345         .max_access_size = 4,
3346     },
3347 };
3348 
3349 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
3350 {
3351     uintptr_t key = (uintptr_t)bus;
3352     VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
3353     VTDAddressSpace *vtd_dev_as;
3354     char name[128];
3355 
3356     if (!vtd_bus) {
3357         uintptr_t *new_key = g_malloc(sizeof(*new_key));
3358         *new_key = (uintptr_t)bus;
3359         /* No corresponding free() */
3360         vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
3361                             PCI_DEVFN_MAX);
3362         vtd_bus->bus = bus;
3363         g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
3364     }
3365 
3366     vtd_dev_as = vtd_bus->dev_as[devfn];
3367 
3368     if (!vtd_dev_as) {
3369         snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn),
3370                  PCI_FUNC(devfn));
3371         vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace));
3372 
3373         vtd_dev_as->bus = bus;
3374         vtd_dev_as->devfn = (uint8_t)devfn;
3375         vtd_dev_as->iommu_state = s;
3376         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
3377         vtd_dev_as->iova_tree = iova_tree_new();
3378 
3379         memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX);
3380         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root");
3381 
3382         /*
3383          * Build the DMAR-disabled container with aliases to the
3384          * shared MRs.  Note that aliasing to a shared memory region
3385          * could help the memory API to detect same FlatViews so we
3386          * can have devices to share the same FlatView when DMAR is
3387          * disabled (either by not providing "intel_iommu=on" or with
3388          * "iommu=pt").  It will greatly reduce the total number of
3389          * FlatViews of the system hence VM runs faster.
3390          */
3391         memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s),
3392                                  "vtd-nodmar", &s->mr_nodmar, 0,
3393                                  memory_region_size(&s->mr_nodmar));
3394 
3395         /*
3396          * Build the per-device DMAR-enabled container.
3397          *
3398          * TODO: currently we have per-device IOMMU memory region only
3399          * because we have per-device IOMMU notifiers for devices.  If
3400          * one day we can abstract the IOMMU notifiers out of the
3401          * memory regions then we can also share the same memory
3402          * region here just like what we've done above with the nodmar
3403          * region.
3404          */
3405         strcat(name, "-dmar");
3406         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
3407                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
3408                                  name, UINT64_MAX);
3409         memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir",
3410                                  &s->mr_ir, 0, memory_region_size(&s->mr_ir));
3411         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
3412                                             VTD_INTERRUPT_ADDR_FIRST,
3413                                             &vtd_dev_as->iommu_ir, 1);
3414 
3415         /*
3416          * Hook both the containers under the root container, we
3417          * switch between DMAR & noDMAR by enable/disable
3418          * corresponding sub-containers
3419          */
3420         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3421                                             MEMORY_REGION(&vtd_dev_as->iommu),
3422                                             0);
3423         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3424                                             &vtd_dev_as->nodmar, 0);
3425 
3426         vtd_switch_address_space(vtd_dev_as);
3427     }
3428     return vtd_dev_as;
3429 }
3430 
3431 static uint64_t get_naturally_aligned_size(uint64_t start,
3432                                            uint64_t size, int gaw)
3433 {
3434     uint64_t max_mask = 1ULL << gaw;
3435     uint64_t alignment = start ? start & -start : max_mask;
3436 
3437     alignment = MIN(alignment, max_mask);
3438     size = MIN(size, max_mask);
3439 
3440     if (alignment <= size) {
3441         /* Increase the alignment of start */
3442         return alignment;
3443     } else {
3444         /* Find the largest page mask from size */
3445         return 1ULL << (63 - clz64(size));
3446     }
3447 }
3448 
3449 /* Unmap the whole range in the notifier's scope. */
3450 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
3451 {
3452     hwaddr size, remain;
3453     hwaddr start = n->start;
3454     hwaddr end = n->end;
3455     IntelIOMMUState *s = as->iommu_state;
3456     DMAMap map;
3457 
3458     /*
3459      * Note: all the codes in this function has a assumption that IOVA
3460      * bits are no more than VTD_MGAW bits (which is restricted by
3461      * VT-d spec), otherwise we need to consider overflow of 64 bits.
3462      */
3463 
3464     if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) {
3465         /*
3466          * Don't need to unmap regions that is bigger than the whole
3467          * VT-d supported address space size
3468          */
3469         end = VTD_ADDRESS_SIZE(s->aw_bits) - 1;
3470     }
3471 
3472     assert(start <= end);
3473     size = remain = end - start + 1;
3474 
3475     while (remain >= VTD_PAGE_SIZE) {
3476         IOMMUTLBEntry entry;
3477         uint64_t mask = get_naturally_aligned_size(start, remain, s->aw_bits);
3478 
3479         assert(mask);
3480 
3481         entry.iova = start;
3482         entry.addr_mask = mask - 1;
3483         entry.target_as = &address_space_memory;
3484         entry.perm = IOMMU_NONE;
3485         /* This field is meaningless for unmap */
3486         entry.translated_addr = 0;
3487 
3488         memory_region_notify_one(n, &entry);
3489 
3490         start += mask;
3491         remain -= mask;
3492     }
3493 
3494     assert(!remain);
3495 
3496     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3497                              VTD_PCI_SLOT(as->devfn),
3498                              VTD_PCI_FUNC(as->devfn),
3499                              n->start, size);
3500 
3501     map.iova = n->start;
3502     map.size = size;
3503     iova_tree_remove(as->iova_tree, &map);
3504 }
3505 
3506 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3507 {
3508     VTDAddressSpace *vtd_as;
3509     IOMMUNotifier *n;
3510 
3511     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3512         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3513             vtd_address_space_unmap(vtd_as, n);
3514         }
3515     }
3516 }
3517 
3518 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3519 {
3520     vtd_address_space_unmap_all(s);
3521     vtd_switch_address_space_all(s);
3522 }
3523 
3524 static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private)
3525 {
3526     memory_region_notify_one((IOMMUNotifier *)private, entry);
3527     return 0;
3528 }
3529 
3530 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3531 {
3532     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3533     IntelIOMMUState *s = vtd_as->iommu_state;
3534     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3535     VTDContextEntry ce;
3536 
3537     /*
3538      * The replay can be triggered by either a invalidation or a newly
3539      * created entry. No matter what, we release existing mappings
3540      * (it means flushing caches for UNMAP-only registers).
3541      */
3542     vtd_address_space_unmap(vtd_as, n);
3543 
3544     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3545         trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" :
3546                                   "legacy mode",
3547                                   bus_n, PCI_SLOT(vtd_as->devfn),
3548                                   PCI_FUNC(vtd_as->devfn),
3549                                   vtd_get_domain_id(s, &ce),
3550                                   ce.hi, ce.lo);
3551         if (vtd_as_has_map_notifier(vtd_as)) {
3552             /* This is required only for MAP typed notifiers */
3553             vtd_page_walk_info info = {
3554                 .hook_fn = vtd_replay_hook,
3555                 .private = (void *)n,
3556                 .notify_unmap = false,
3557                 .aw = s->aw_bits,
3558                 .as = vtd_as,
3559                 .domain_id = vtd_get_domain_id(s, &ce),
3560             };
3561 
3562             vtd_page_walk(s, &ce, 0, ~0ULL, &info);
3563         }
3564     } else {
3565         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3566                                     PCI_FUNC(vtd_as->devfn));
3567     }
3568 
3569     return;
3570 }
3571 
3572 /* Do the initialization. It will also be called when reset, so pay
3573  * attention when adding new initialization stuff.
3574  */
3575 static void vtd_init(IntelIOMMUState *s)
3576 {
3577     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3578 
3579     memset(s->csr, 0, DMAR_REG_SIZE);
3580     memset(s->wmask, 0, DMAR_REG_SIZE);
3581     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3582     memset(s->womask, 0, DMAR_REG_SIZE);
3583 
3584     s->root = 0;
3585     s->root_scalable = false;
3586     s->dmar_enabled = false;
3587     s->intr_enabled = false;
3588     s->iq_head = 0;
3589     s->iq_tail = 0;
3590     s->iq = 0;
3591     s->iq_size = 0;
3592     s->qi_enabled = false;
3593     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3594     s->iq_dw = false;
3595     s->next_frcd_reg = 0;
3596     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3597              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3598              VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits);
3599     if (s->dma_drain) {
3600         s->cap |= VTD_CAP_DRAIN;
3601     }
3602     if (s->aw_bits == VTD_HOST_AW_48BIT) {
3603         s->cap |= VTD_CAP_SAGAW_48bit;
3604     }
3605     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3606 
3607     /*
3608      * Rsvd field masks for spte
3609      */
3610     vtd_spte_rsvd[0] = ~0ULL;
3611     vtd_spte_rsvd[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits,
3612                                                   x86_iommu->dt_supported);
3613     vtd_spte_rsvd[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3614     vtd_spte_rsvd[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3615     vtd_spte_rsvd[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3616 
3617     vtd_spte_rsvd_large[2] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits,
3618                                                          x86_iommu->dt_supported);
3619     vtd_spte_rsvd_large[3] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits,
3620                                                          x86_iommu->dt_supported);
3621 
3622     if (x86_iommu_ir_supported(x86_iommu)) {
3623         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3624         if (s->intr_eim == ON_OFF_AUTO_ON) {
3625             s->ecap |= VTD_ECAP_EIM;
3626         }
3627         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3628     }
3629 
3630     if (x86_iommu->dt_supported) {
3631         s->ecap |= VTD_ECAP_DT;
3632     }
3633 
3634     if (x86_iommu->pt_supported) {
3635         s->ecap |= VTD_ECAP_PT;
3636     }
3637 
3638     if (s->caching_mode) {
3639         s->cap |= VTD_CAP_CM;
3640     }
3641 
3642     /* TODO: read cap/ecap from host to decide which cap to be exposed. */
3643     if (s->scalable_mode) {
3644         s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS;
3645     }
3646 
3647     vtd_reset_caches(s);
3648 
3649     /* Define registers with default values and bit semantics */
3650     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
3651     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
3652     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
3653     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
3654     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
3655     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
3656     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0);
3657     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
3658     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
3659 
3660     /* Advanced Fault Logging not supported */
3661     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
3662     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3663     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
3664     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
3665 
3666     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3667      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3668      */
3669     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
3670 
3671     /* Treated as RO for implementations that PLMR and PHMR fields reported
3672      * as Clear in the CAP_REG.
3673      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3674      */
3675     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
3676 
3677     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
3678     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
3679     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0);
3680     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
3681     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3682     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
3683     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
3684     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3685     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
3686 
3687     /* IOTLB registers */
3688     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
3689     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
3690     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
3691 
3692     /* Fault Recording Registers, 128-bit */
3693     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
3694     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
3695 
3696     /*
3697      * Interrupt remapping registers.
3698      */
3699     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
3700 }
3701 
3702 /* Should not reset address_spaces when reset because devices will still use
3703  * the address space they got at first (won't ask the bus again).
3704  */
3705 static void vtd_reset(DeviceState *dev)
3706 {
3707     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3708 
3709     vtd_init(s);
3710     vtd_address_space_refresh_all(s);
3711 }
3712 
3713 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
3714 {
3715     IntelIOMMUState *s = opaque;
3716     VTDAddressSpace *vtd_as;
3717 
3718     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
3719 
3720     vtd_as = vtd_find_add_as(s, bus, devfn);
3721     return &vtd_as->as;
3722 }
3723 
3724 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
3725 {
3726     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3727 
3728     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) {
3729         error_setg(errp, "eim=on cannot be selected without intremap=on");
3730         return false;
3731     }
3732 
3733     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
3734         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
3735                       && x86_iommu_ir_supported(x86_iommu) ?
3736                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
3737     }
3738     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
3739         if (!kvm_irqchip_in_kernel()) {
3740             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
3741             return false;
3742         }
3743         if (!kvm_enable_x2apic()) {
3744             error_setg(errp, "eim=on requires support on the KVM side"
3745                              "(X2APIC_API, first shipped in v4.7)");
3746             return false;
3747         }
3748     }
3749 
3750     /* Currently only address widths supported are 39 and 48 bits */
3751     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
3752         (s->aw_bits != VTD_HOST_AW_48BIT)) {
3753         error_setg(errp, "Supported values for x-aw-bits are: %d, %d",
3754                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
3755         return false;
3756     }
3757 
3758     if (s->scalable_mode && !s->dma_drain) {
3759         error_setg(errp, "Need to set dma_drain for scalable mode");
3760         return false;
3761     }
3762 
3763     return true;
3764 }
3765 
3766 static int vtd_machine_done_notify_one(Object *child, void *unused)
3767 {
3768     IntelIOMMUState *iommu = INTEL_IOMMU_DEVICE(x86_iommu_get_default());
3769 
3770     /*
3771      * We hard-coded here because vfio-pci is the only special case
3772      * here.  Let's be more elegant in the future when we can, but so
3773      * far there seems to be no better way.
3774      */
3775     if (object_dynamic_cast(child, "vfio-pci") && !iommu->caching_mode) {
3776         vtd_panic_require_caching_mode();
3777     }
3778 
3779     return 0;
3780 }
3781 
3782 static void vtd_machine_done_hook(Notifier *notifier, void *unused)
3783 {
3784     object_child_foreach_recursive(object_get_root(),
3785                                    vtd_machine_done_notify_one, NULL);
3786 }
3787 
3788 static Notifier vtd_machine_done_notify = {
3789     .notify = vtd_machine_done_hook,
3790 };
3791 
3792 static void vtd_realize(DeviceState *dev, Error **errp)
3793 {
3794     MachineState *ms = MACHINE(qdev_get_machine());
3795     PCMachineState *pcms = PC_MACHINE(ms);
3796     X86MachineState *x86ms = X86_MACHINE(ms);
3797     PCIBus *bus = pcms->bus;
3798     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3799     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
3800 
3801     x86_iommu->type = TYPE_INTEL;
3802 
3803     if (!vtd_decide_config(s, errp)) {
3804         return;
3805     }
3806 
3807     QLIST_INIT(&s->vtd_as_with_notifiers);
3808     qemu_mutex_init(&s->iommu_lock);
3809     memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num));
3810     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
3811                           "intel_iommu", DMAR_REG_SIZE);
3812 
3813     /* Create the shared memory regions by all devices */
3814     memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar",
3815                        UINT64_MAX);
3816     memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops,
3817                           s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE);
3818     memory_region_init_alias(&s->mr_sys_alias, OBJECT(s),
3819                              "vtd-sys-alias", get_system_memory(), 0,
3820                              memory_region_size(get_system_memory()));
3821     memory_region_add_subregion_overlap(&s->mr_nodmar, 0,
3822                                         &s->mr_sys_alias, 0);
3823     memory_region_add_subregion_overlap(&s->mr_nodmar,
3824                                         VTD_INTERRUPT_ADDR_FIRST,
3825                                         &s->mr_ir, 1);
3826 
3827     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
3828     /* No corresponding destroy */
3829     s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3830                                      g_free, g_free);
3831     s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3832                                               g_free, g_free);
3833     vtd_init(s);
3834     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
3835     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
3836     /* Pseudo address space under root PCI bus. */
3837     x86ms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
3838     qemu_add_machine_init_done_notifier(&vtd_machine_done_notify);
3839 }
3840 
3841 static void vtd_class_init(ObjectClass *klass, void *data)
3842 {
3843     DeviceClass *dc = DEVICE_CLASS(klass);
3844     X86IOMMUClass *x86_class = X86_IOMMU_CLASS(klass);
3845 
3846     dc->reset = vtd_reset;
3847     dc->vmsd = &vtd_vmstate;
3848     device_class_set_props(dc, vtd_properties);
3849     dc->hotpluggable = false;
3850     x86_class->realize = vtd_realize;
3851     x86_class->int_remap = vtd_int_remap;
3852     /* Supported by the pc-q35-* machine types */
3853     dc->user_creatable = true;
3854     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
3855     dc->desc = "Intel IOMMU (VT-d) DMA Remapping device";
3856 }
3857 
3858 static const TypeInfo vtd_info = {
3859     .name          = TYPE_INTEL_IOMMU_DEVICE,
3860     .parent        = TYPE_X86_IOMMU_DEVICE,
3861     .instance_size = sizeof(IntelIOMMUState),
3862     .class_init    = vtd_class_init,
3863 };
3864 
3865 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
3866                                                      void *data)
3867 {
3868     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
3869 
3870     imrc->translate = vtd_iommu_translate;
3871     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
3872     imrc->replay = vtd_iommu_replay;
3873 }
3874 
3875 static const TypeInfo vtd_iommu_memory_region_info = {
3876     .parent = TYPE_IOMMU_MEMORY_REGION,
3877     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
3878     .class_init = vtd_iommu_memory_region_class_init,
3879 };
3880 
3881 static void vtd_register_types(void)
3882 {
3883     type_register_static(&vtd_info);
3884     type_register_static(&vtd_iommu_memory_region_info);
3885 }
3886 
3887 type_init(vtd_register_types)
3888