xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision 75c5bb0b)
1 /*
2  * QEMU emulation of an Intel IOMMU (VT-d)
3  *   (DMA Remapping device)
4  *
5  * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com>
6  * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12 
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17 
18  * You should have received a copy of the GNU General Public License along
19  * with this program; if not, see <http://www.gnu.org/licenses/>.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/error-report.h"
24 #include "qemu/main-loop.h"
25 #include "qapi/error.h"
26 #include "hw/sysbus.h"
27 #include "exec/address-spaces.h"
28 #include "intel_iommu_internal.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bus.h"
31 #include "hw/qdev-properties.h"
32 #include "hw/i386/pc.h"
33 #include "hw/i386/apic-msidef.h"
34 #include "hw/boards.h"
35 #include "hw/i386/x86-iommu.h"
36 #include "hw/pci-host/q35.h"
37 #include "sysemu/kvm.h"
38 #include "sysemu/sysemu.h"
39 #include "hw/i386/apic_internal.h"
40 #include "kvm_i386.h"
41 #include "migration/vmstate.h"
42 #include "trace.h"
43 
44 /* context entry operations */
45 #define VTD_CE_GET_RID2PASID(ce) \
46     ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK)
47 #define VTD_CE_GET_PASID_DIR_TABLE(ce) \
48     ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK)
49 
50 /* pe operations */
51 #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT)
52 #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW))
53 #define VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write) {\
54     if (ret_fr) {                                                             \
55         ret_fr = -ret_fr;                                                     \
56         if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) {                   \
57             trace_vtd_fault_disabled();                                       \
58         } else {                                                              \
59             vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);      \
60         }                                                                     \
61         goto error;                                                           \
62     }                                                                         \
63 }
64 
65 static void vtd_address_space_refresh_all(IntelIOMMUState *s);
66 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n);
67 
68 static void vtd_panic_require_caching_mode(void)
69 {
70     error_report("We need to set caching-mode=on for intel-iommu to enable "
71                  "device assignment with IOMMU protection.");
72     exit(1);
73 }
74 
75 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val,
76                             uint64_t wmask, uint64_t w1cmask)
77 {
78     stq_le_p(&s->csr[addr], val);
79     stq_le_p(&s->wmask[addr], wmask);
80     stq_le_p(&s->w1cmask[addr], w1cmask);
81 }
82 
83 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask)
84 {
85     stq_le_p(&s->womask[addr], mask);
86 }
87 
88 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val,
89                             uint32_t wmask, uint32_t w1cmask)
90 {
91     stl_le_p(&s->csr[addr], val);
92     stl_le_p(&s->wmask[addr], wmask);
93     stl_le_p(&s->w1cmask[addr], w1cmask);
94 }
95 
96 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask)
97 {
98     stl_le_p(&s->womask[addr], mask);
99 }
100 
101 /* "External" get/set operations */
102 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
103 {
104     uint64_t oldval = ldq_le_p(&s->csr[addr]);
105     uint64_t wmask = ldq_le_p(&s->wmask[addr]);
106     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
107     stq_le_p(&s->csr[addr],
108              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
109 }
110 
111 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
112 {
113     uint32_t oldval = ldl_le_p(&s->csr[addr]);
114     uint32_t wmask = ldl_le_p(&s->wmask[addr]);
115     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
116     stl_le_p(&s->csr[addr],
117              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
118 }
119 
120 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr)
121 {
122     uint64_t val = ldq_le_p(&s->csr[addr]);
123     uint64_t womask = ldq_le_p(&s->womask[addr]);
124     return val & ~womask;
125 }
126 
127 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr)
128 {
129     uint32_t val = ldl_le_p(&s->csr[addr]);
130     uint32_t womask = ldl_le_p(&s->womask[addr]);
131     return val & ~womask;
132 }
133 
134 /* "Internal" get/set operations */
135 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr)
136 {
137     return ldq_le_p(&s->csr[addr]);
138 }
139 
140 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr)
141 {
142     return ldl_le_p(&s->csr[addr]);
143 }
144 
145 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val)
146 {
147     stq_le_p(&s->csr[addr], val);
148 }
149 
150 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
151                                         uint32_t clear, uint32_t mask)
152 {
153     uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
154     stl_le_p(&s->csr[addr], new_val);
155     return new_val;
156 }
157 
158 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
159                                         uint64_t clear, uint64_t mask)
160 {
161     uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
162     stq_le_p(&s->csr[addr], new_val);
163     return new_val;
164 }
165 
166 static inline void vtd_iommu_lock(IntelIOMMUState *s)
167 {
168     qemu_mutex_lock(&s->iommu_lock);
169 }
170 
171 static inline void vtd_iommu_unlock(IntelIOMMUState *s)
172 {
173     qemu_mutex_unlock(&s->iommu_lock);
174 }
175 
176 static void vtd_update_scalable_state(IntelIOMMUState *s)
177 {
178     uint64_t val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
179 
180     if (s->scalable_mode) {
181         s->root_scalable = val & VTD_RTADDR_SMT;
182     }
183 }
184 
185 /* Whether the address space needs to notify new mappings */
186 static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as)
187 {
188     return as->notifier_flags & IOMMU_NOTIFIER_MAP;
189 }
190 
191 /* GHashTable functions */
192 static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2)
193 {
194     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
195 }
196 
197 static guint vtd_uint64_hash(gconstpointer v)
198 {
199     return (guint)*(const uint64_t *)v;
200 }
201 
202 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value,
203                                           gpointer user_data)
204 {
205     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
206     uint16_t domain_id = *(uint16_t *)user_data;
207     return entry->domain_id == domain_id;
208 }
209 
210 /* The shift of an addr for a certain level of paging structure */
211 static inline uint32_t vtd_slpt_level_shift(uint32_t level)
212 {
213     assert(level != 0);
214     return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
215 }
216 
217 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level)
218 {
219     return ~((1ULL << vtd_slpt_level_shift(level)) - 1);
220 }
221 
222 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value,
223                                         gpointer user_data)
224 {
225     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
226     VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data;
227     uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask;
228     uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K;
229     return (entry->domain_id == info->domain_id) &&
230             (((entry->gfn & info->mask) == gfn) ||
231              (entry->gfn == gfn_tlb));
232 }
233 
234 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
235  * IntelIOMMUState to 1.  Must be called with IOMMU lock held.
236  */
237 static void vtd_reset_context_cache_locked(IntelIOMMUState *s)
238 {
239     VTDAddressSpace *vtd_as;
240     VTDBus *vtd_bus;
241     GHashTableIter bus_it;
242     uint32_t devfn_it;
243 
244     trace_vtd_context_cache_reset();
245 
246     g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr);
247 
248     while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) {
249         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
250             vtd_as = vtd_bus->dev_as[devfn_it];
251             if (!vtd_as) {
252                 continue;
253             }
254             vtd_as->context_cache_entry.context_cache_gen = 0;
255         }
256     }
257     s->context_cache_gen = 1;
258 }
259 
260 /* Must be called with IOMMU lock held. */
261 static void vtd_reset_iotlb_locked(IntelIOMMUState *s)
262 {
263     assert(s->iotlb);
264     g_hash_table_remove_all(s->iotlb);
265 }
266 
267 static void vtd_reset_iotlb(IntelIOMMUState *s)
268 {
269     vtd_iommu_lock(s);
270     vtd_reset_iotlb_locked(s);
271     vtd_iommu_unlock(s);
272 }
273 
274 static void vtd_reset_caches(IntelIOMMUState *s)
275 {
276     vtd_iommu_lock(s);
277     vtd_reset_iotlb_locked(s);
278     vtd_reset_context_cache_locked(s);
279     vtd_iommu_unlock(s);
280 }
281 
282 static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id,
283                                   uint32_t level)
284 {
285     return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) |
286            ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT);
287 }
288 
289 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level)
290 {
291     return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K;
292 }
293 
294 /* Must be called with IOMMU lock held */
295 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id,
296                                        hwaddr addr)
297 {
298     VTDIOTLBEntry *entry;
299     uint64_t key;
300     int level;
301 
302     for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) {
303         key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level),
304                                 source_id, level);
305         entry = g_hash_table_lookup(s->iotlb, &key);
306         if (entry) {
307             goto out;
308         }
309     }
310 
311 out:
312     return entry;
313 }
314 
315 /* Must be with IOMMU lock held */
316 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id,
317                              uint16_t domain_id, hwaddr addr, uint64_t slpte,
318                              uint8_t access_flags, uint32_t level)
319 {
320     VTDIOTLBEntry *entry = g_malloc(sizeof(*entry));
321     uint64_t *key = g_malloc(sizeof(*key));
322     uint64_t gfn = vtd_get_iotlb_gfn(addr, level);
323 
324     trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id);
325     if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) {
326         trace_vtd_iotlb_reset("iotlb exceeds size limit");
327         vtd_reset_iotlb_locked(s);
328     }
329 
330     entry->gfn = gfn;
331     entry->domain_id = domain_id;
332     entry->slpte = slpte;
333     entry->access_flags = access_flags;
334     entry->mask = vtd_slpt_level_page_mask(level);
335     *key = vtd_get_iotlb_key(gfn, source_id, level);
336     g_hash_table_replace(s->iotlb, key, entry);
337 }
338 
339 /* Given the reg addr of both the message data and address, generate an
340  * interrupt via MSI.
341  */
342 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
343                                    hwaddr mesg_data_reg)
344 {
345     MSIMessage msi;
346 
347     assert(mesg_data_reg < DMAR_REG_SIZE);
348     assert(mesg_addr_reg < DMAR_REG_SIZE);
349 
350     msi.address = vtd_get_long_raw(s, mesg_addr_reg);
351     msi.data = vtd_get_long_raw(s, mesg_data_reg);
352 
353     trace_vtd_irq_generate(msi.address, msi.data);
354 
355     apic_get_class()->send_msi(&msi);
356 }
357 
358 /* Generate a fault event to software via MSI if conditions are met.
359  * Notice that the value of FSTS_REG being passed to it should be the one
360  * before any update.
361  */
362 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
363 {
364     if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
365         pre_fsts & VTD_FSTS_IQE) {
366         error_report_once("There are previous interrupt conditions "
367                           "to be serviced by software, fault event "
368                           "is not generated");
369         return;
370     }
371     vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
372     if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
373         error_report_once("Interrupt Mask set, irq is not generated");
374     } else {
375         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
376         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
377     }
378 }
379 
380 /* Check if the Fault (F) field of the Fault Recording Register referenced by
381  * @index is Set.
382  */
383 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index)
384 {
385     /* Each reg is 128-bit */
386     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
387     addr += 8; /* Access the high 64-bit half */
388 
389     assert(index < DMAR_FRCD_REG_NR);
390 
391     return vtd_get_quad_raw(s, addr) & VTD_FRCD_F;
392 }
393 
394 /* Update the PPF field of Fault Status Register.
395  * Should be called whenever change the F field of any fault recording
396  * registers.
397  */
398 static void vtd_update_fsts_ppf(IntelIOMMUState *s)
399 {
400     uint32_t i;
401     uint32_t ppf_mask = 0;
402 
403     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
404         if (vtd_is_frcd_set(s, i)) {
405             ppf_mask = VTD_FSTS_PPF;
406             break;
407         }
408     }
409     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
410     trace_vtd_fsts_ppf(!!ppf_mask);
411 }
412 
413 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index)
414 {
415     /* Each reg is 128-bit */
416     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
417     addr += 8; /* Access the high 64-bit half */
418 
419     assert(index < DMAR_FRCD_REG_NR);
420 
421     vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
422     vtd_update_fsts_ppf(s);
423 }
424 
425 /* Must not update F field now, should be done later */
426 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index,
427                             uint16_t source_id, hwaddr addr,
428                             VTDFaultReason fault, bool is_write)
429 {
430     uint64_t hi = 0, lo;
431     hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
432 
433     assert(index < DMAR_FRCD_REG_NR);
434 
435     lo = VTD_FRCD_FI(addr);
436     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
437     if (!is_write) {
438         hi |= VTD_FRCD_T;
439     }
440     vtd_set_quad_raw(s, frcd_reg_addr, lo);
441     vtd_set_quad_raw(s, frcd_reg_addr + 8, hi);
442 
443     trace_vtd_frr_new(index, hi, lo);
444 }
445 
446 /* Try to collapse multiple pending faults from the same requester */
447 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id)
448 {
449     uint32_t i;
450     uint64_t frcd_reg;
451     hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
452 
453     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
454         frcd_reg = vtd_get_quad_raw(s, addr);
455         if ((frcd_reg & VTD_FRCD_F) &&
456             ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
457             return true;
458         }
459         addr += 16; /* 128-bit for each */
460     }
461     return false;
462 }
463 
464 /* Log and report an DMAR (address translation) fault to software */
465 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
466                                   hwaddr addr, VTDFaultReason fault,
467                                   bool is_write)
468 {
469     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
470 
471     assert(fault < VTD_FR_MAX);
472 
473     if (fault == VTD_FR_RESERVED_ERR) {
474         /* This is not a normal fault reason case. Drop it. */
475         return;
476     }
477 
478     trace_vtd_dmar_fault(source_id, fault, addr, is_write);
479 
480     if (fsts_reg & VTD_FSTS_PFO) {
481         error_report_once("New fault is not recorded due to "
482                           "Primary Fault Overflow");
483         return;
484     }
485 
486     if (vtd_try_collapse_fault(s, source_id)) {
487         error_report_once("New fault is not recorded due to "
488                           "compression of faults");
489         return;
490     }
491 
492     if (vtd_is_frcd_set(s, s->next_frcd_reg)) {
493         error_report_once("Next Fault Recording Reg is used, "
494                           "new fault is not recorded, set PFO field");
495         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
496         return;
497     }
498 
499     vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write);
500 
501     if (fsts_reg & VTD_FSTS_PPF) {
502         error_report_once("There are pending faults already, "
503                           "fault event is not generated");
504         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg);
505         s->next_frcd_reg++;
506         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
507             s->next_frcd_reg = 0;
508         }
509     } else {
510         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
511                                 VTD_FSTS_FRI(s->next_frcd_reg));
512         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */
513         s->next_frcd_reg++;
514         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
515             s->next_frcd_reg = 0;
516         }
517         /* This case actually cause the PPF to be Set.
518          * So generate fault event (interrupt).
519          */
520          vtd_generate_fault_event(s, fsts_reg);
521     }
522 }
523 
524 /* Handle Invalidation Queue Errors of queued invalidation interface error
525  * conditions.
526  */
527 static void vtd_handle_inv_queue_error(IntelIOMMUState *s)
528 {
529     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
530 
531     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE);
532     vtd_generate_fault_event(s, fsts_reg);
533 }
534 
535 /* Set the IWC field and try to generate an invalidation completion interrupt */
536 static void vtd_generate_completion_event(IntelIOMMUState *s)
537 {
538     if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) {
539         trace_vtd_inv_desc_wait_irq("One pending, skip current");
540         return;
541     }
542     vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC);
543     vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP);
544     if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) {
545         trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
546                                     "new event not generated");
547         return;
548     } else {
549         /* Generate the interrupt event */
550         trace_vtd_inv_desc_wait_irq("Generating complete event");
551         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
552         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
553     }
554 }
555 
556 static inline bool vtd_root_entry_present(IntelIOMMUState *s,
557                                           VTDRootEntry *re,
558                                           uint8_t devfn)
559 {
560     if (s->root_scalable && devfn > UINT8_MAX / 2) {
561         return re->hi & VTD_ROOT_ENTRY_P;
562     }
563 
564     return re->lo & VTD_ROOT_ENTRY_P;
565 }
566 
567 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index,
568                               VTDRootEntry *re)
569 {
570     dma_addr_t addr;
571 
572     addr = s->root + index * sizeof(*re);
573     if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) {
574         re->lo = 0;
575         return -VTD_FR_ROOT_TABLE_INV;
576     }
577     re->lo = le64_to_cpu(re->lo);
578     re->hi = le64_to_cpu(re->hi);
579     return 0;
580 }
581 
582 static inline bool vtd_ce_present(VTDContextEntry *context)
583 {
584     return context->lo & VTD_CONTEXT_ENTRY_P;
585 }
586 
587 static int vtd_get_context_entry_from_root(IntelIOMMUState *s,
588                                            VTDRootEntry *re,
589                                            uint8_t index,
590                                            VTDContextEntry *ce)
591 {
592     dma_addr_t addr, ce_size;
593 
594     /* we have checked that root entry is present */
595     ce_size = s->root_scalable ? VTD_CTX_ENTRY_SCALABLE_SIZE :
596               VTD_CTX_ENTRY_LEGACY_SIZE;
597 
598     if (s->root_scalable && index > UINT8_MAX / 2) {
599         index = index & (~VTD_DEVFN_CHECK_MASK);
600         addr = re->hi & VTD_ROOT_ENTRY_CTP;
601     } else {
602         addr = re->lo & VTD_ROOT_ENTRY_CTP;
603     }
604 
605     addr = addr + index * ce_size;
606     if (dma_memory_read(&address_space_memory, addr, ce, ce_size)) {
607         return -VTD_FR_CONTEXT_TABLE_INV;
608     }
609 
610     ce->lo = le64_to_cpu(ce->lo);
611     ce->hi = le64_to_cpu(ce->hi);
612     if (ce_size == VTD_CTX_ENTRY_SCALABLE_SIZE) {
613         ce->val[2] = le64_to_cpu(ce->val[2]);
614         ce->val[3] = le64_to_cpu(ce->val[3]);
615     }
616     return 0;
617 }
618 
619 static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce)
620 {
621     return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
622 }
623 
624 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw)
625 {
626     return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw);
627 }
628 
629 /* Whether the pte indicates the address of the page frame */
630 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level)
631 {
632     return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
633 }
634 
635 /* Get the content of a spte located in @base_addr[@index] */
636 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index)
637 {
638     uint64_t slpte;
639 
640     assert(index < VTD_SL_PT_ENTRY_NR);
641 
642     if (dma_memory_read(&address_space_memory,
643                         base_addr + index * sizeof(slpte), &slpte,
644                         sizeof(slpte))) {
645         slpte = (uint64_t)-1;
646         return slpte;
647     }
648     slpte = le64_to_cpu(slpte);
649     return slpte;
650 }
651 
652 /* Given an iova and the level of paging structure, return the offset
653  * of current level.
654  */
655 static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level)
656 {
657     return (iova >> vtd_slpt_level_shift(level)) &
658             ((1ULL << VTD_SL_LEVEL_BITS) - 1);
659 }
660 
661 /* Check Capability Register to see if the @level of page-table is supported */
662 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level)
663 {
664     return VTD_CAP_SAGAW_MASK & s->cap &
665            (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
666 }
667 
668 /* Return true if check passed, otherwise false */
669 static inline bool vtd_pe_type_check(X86IOMMUState *x86_iommu,
670                                      VTDPASIDEntry *pe)
671 {
672     switch (VTD_PE_GET_TYPE(pe)) {
673     case VTD_SM_PASID_ENTRY_FLT:
674     case VTD_SM_PASID_ENTRY_SLT:
675     case VTD_SM_PASID_ENTRY_NESTED:
676         break;
677     case VTD_SM_PASID_ENTRY_PT:
678         if (!x86_iommu->pt_supported) {
679             return false;
680         }
681         break;
682     default:
683         /* Unknwon type */
684         return false;
685     }
686     return true;
687 }
688 
689 static int vtd_get_pasid_dire(dma_addr_t pasid_dir_base,
690                               uint32_t pasid,
691                               VTDPASIDDirEntry *pdire)
692 {
693     uint32_t index;
694     dma_addr_t addr, entry_size;
695 
696     index = VTD_PASID_DIR_INDEX(pasid);
697     entry_size = VTD_PASID_DIR_ENTRY_SIZE;
698     addr = pasid_dir_base + index * entry_size;
699     if (dma_memory_read(&address_space_memory, addr, pdire, entry_size)) {
700         return -VTD_FR_PASID_TABLE_INV;
701     }
702 
703     return 0;
704 }
705 
706 static int vtd_get_pasid_entry(IntelIOMMUState *s,
707                                uint32_t pasid,
708                                VTDPASIDDirEntry *pdire,
709                                VTDPASIDEntry *pe)
710 {
711     uint32_t index;
712     dma_addr_t addr, entry_size;
713     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
714 
715     index = VTD_PASID_TABLE_INDEX(pasid);
716     entry_size = VTD_PASID_ENTRY_SIZE;
717     addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK;
718     addr = addr + index * entry_size;
719     if (dma_memory_read(&address_space_memory, addr, pe, entry_size)) {
720         return -VTD_FR_PASID_TABLE_INV;
721     }
722 
723     /* Do translation type check */
724     if (!vtd_pe_type_check(x86_iommu, pe)) {
725         return -VTD_FR_PASID_TABLE_INV;
726     }
727 
728     if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) {
729         return -VTD_FR_PASID_TABLE_INV;
730     }
731 
732     return 0;
733 }
734 
735 static int vtd_get_pasid_entry_from_pasid(IntelIOMMUState *s,
736                                           dma_addr_t pasid_dir_base,
737                                           uint32_t pasid,
738                                           VTDPASIDEntry *pe)
739 {
740     int ret;
741     VTDPASIDDirEntry pdire;
742 
743     ret = vtd_get_pasid_dire(pasid_dir_base, pasid, &pdire);
744     if (ret) {
745         return ret;
746     }
747 
748     ret = vtd_get_pasid_entry(s, pasid, &pdire, pe);
749     if (ret) {
750         return ret;
751     }
752 
753     return ret;
754 }
755 
756 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s,
757                                       VTDContextEntry *ce,
758                                       VTDPASIDEntry *pe)
759 {
760     uint32_t pasid;
761     dma_addr_t pasid_dir_base;
762     int ret = 0;
763 
764     pasid = VTD_CE_GET_RID2PASID(ce);
765     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
766     ret = vtd_get_pasid_entry_from_pasid(s, pasid_dir_base, pasid, pe);
767 
768     return ret;
769 }
770 
771 static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s,
772                                 VTDContextEntry *ce,
773                                 bool *pe_fpd_set)
774 {
775     int ret;
776     uint32_t pasid;
777     dma_addr_t pasid_dir_base;
778     VTDPASIDDirEntry pdire;
779     VTDPASIDEntry pe;
780 
781     pasid = VTD_CE_GET_RID2PASID(ce);
782     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
783 
784     ret = vtd_get_pasid_dire(pasid_dir_base, pasid, &pdire);
785     if (ret) {
786         return ret;
787     }
788 
789     if (pdire.val & VTD_PASID_DIR_FPD) {
790         *pe_fpd_set = true;
791         return 0;
792     }
793 
794     ret = vtd_get_pasid_entry(s, pasid, &pdire, &pe);
795     if (ret) {
796         return ret;
797     }
798 
799     if (pe.val[0] & VTD_PASID_ENTRY_FPD) {
800         *pe_fpd_set = true;
801     }
802 
803     return 0;
804 }
805 
806 /* Get the page-table level that hardware should use for the second-level
807  * page-table walk from the Address Width field of context-entry.
808  */
809 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
810 {
811     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
812 }
813 
814 static uint32_t vtd_get_iova_level(IntelIOMMUState *s,
815                                    VTDContextEntry *ce)
816 {
817     VTDPASIDEntry pe;
818 
819     if (s->root_scalable) {
820         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
821         return VTD_PE_GET_LEVEL(&pe);
822     }
823 
824     return vtd_ce_get_level(ce);
825 }
826 
827 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
828 {
829     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
830 }
831 
832 static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s,
833                                   VTDContextEntry *ce)
834 {
835     VTDPASIDEntry pe;
836 
837     if (s->root_scalable) {
838         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
839         return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9;
840     }
841 
842     return vtd_ce_get_agaw(ce);
843 }
844 
845 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
846 {
847     return ce->lo & VTD_CONTEXT_ENTRY_TT;
848 }
849 
850 /* Only for Legacy Mode. Return true if check passed, otherwise false */
851 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
852                                      VTDContextEntry *ce)
853 {
854     switch (vtd_ce_get_type(ce)) {
855     case VTD_CONTEXT_TT_MULTI_LEVEL:
856         /* Always supported */
857         break;
858     case VTD_CONTEXT_TT_DEV_IOTLB:
859         if (!x86_iommu->dt_supported) {
860             error_report_once("%s: DT specified but not supported", __func__);
861             return false;
862         }
863         break;
864     case VTD_CONTEXT_TT_PASS_THROUGH:
865         if (!x86_iommu->pt_supported) {
866             error_report_once("%s: PT specified but not supported", __func__);
867             return false;
868         }
869         break;
870     default:
871         /* Unknown type */
872         error_report_once("%s: unknown ce type: %"PRIu32, __func__,
873                           vtd_ce_get_type(ce));
874         return false;
875     }
876     return true;
877 }
878 
879 static inline uint64_t vtd_iova_limit(IntelIOMMUState *s,
880                                       VTDContextEntry *ce, uint8_t aw)
881 {
882     uint32_t ce_agaw = vtd_get_iova_agaw(s, ce);
883     return 1ULL << MIN(ce_agaw, aw);
884 }
885 
886 /* Return true if IOVA passes range check, otherwise false. */
887 static inline bool vtd_iova_range_check(IntelIOMMUState *s,
888                                         uint64_t iova, VTDContextEntry *ce,
889                                         uint8_t aw)
890 {
891     /*
892      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
893      * in CAP_REG and AW in context-entry.
894      */
895     return !(iova & ~(vtd_iova_limit(s, ce, aw) - 1));
896 }
897 
898 static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s,
899                                           VTDContextEntry *ce)
900 {
901     VTDPASIDEntry pe;
902 
903     if (s->root_scalable) {
904         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
905         return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR;
906     }
907 
908     return vtd_ce_get_slpt_base(ce);
909 }
910 
911 /*
912  * Rsvd field masks for spte:
913  *     vtd_spte_rsvd 4k pages
914  *     vtd_spte_rsvd_large large pages
915  */
916 static uint64_t vtd_spte_rsvd[5];
917 static uint64_t vtd_spte_rsvd_large[5];
918 
919 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
920 {
921     uint64_t rsvd_mask = vtd_spte_rsvd[level];
922 
923     if ((level == VTD_SL_PD_LEVEL || level == VTD_SL_PDP_LEVEL) &&
924         (slpte & VTD_SL_PT_PAGE_SIZE_MASK)) {
925         /* large page */
926         rsvd_mask = vtd_spte_rsvd_large[level];
927     }
928 
929     return slpte & rsvd_mask;
930 }
931 
932 /* Find the VTD address space associated with a given bus number */
933 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num)
934 {
935     VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num];
936     if (!vtd_bus) {
937         /*
938          * Iterate over the registered buses to find the one which
939          * currently hold this bus number, and update the bus_num
940          * lookup table:
941          */
942         GHashTableIter iter;
943 
944         g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
945         while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
946             if (pci_bus_num(vtd_bus->bus) == bus_num) {
947                 s->vtd_as_by_bus_num[bus_num] = vtd_bus;
948                 return vtd_bus;
949             }
950         }
951     }
952     return vtd_bus;
953 }
954 
955 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
956  * of the translation, can be used for deciding the size of large page.
957  */
958 static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce,
959                              uint64_t iova, bool is_write,
960                              uint64_t *slptep, uint32_t *slpte_level,
961                              bool *reads, bool *writes, uint8_t aw_bits)
962 {
963     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
964     uint32_t level = vtd_get_iova_level(s, ce);
965     uint32_t offset;
966     uint64_t slpte;
967     uint64_t access_right_check;
968 
969     if (!vtd_iova_range_check(s, iova, ce, aw_bits)) {
970         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")",
971                           __func__, iova);
972         return -VTD_FR_ADDR_BEYOND_MGAW;
973     }
974 
975     /* FIXME: what is the Atomics request here? */
976     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
977 
978     while (true) {
979         offset = vtd_iova_level_offset(iova, level);
980         slpte = vtd_get_slpte(addr, offset);
981 
982         if (slpte == (uint64_t)-1) {
983             error_report_once("%s: detected read error on DMAR slpte "
984                               "(iova=0x%" PRIx64 ")", __func__, iova);
985             if (level == vtd_get_iova_level(s, ce)) {
986                 /* Invalid programming of context-entry */
987                 return -VTD_FR_CONTEXT_ENTRY_INV;
988             } else {
989                 return -VTD_FR_PAGING_ENTRY_INV;
990             }
991         }
992         *reads = (*reads) && (slpte & VTD_SL_R);
993         *writes = (*writes) && (slpte & VTD_SL_W);
994         if (!(slpte & access_right_check)) {
995             error_report_once("%s: detected slpte permission error "
996                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
997                               "slpte=0x%" PRIx64 ", write=%d)", __func__,
998                               iova, level, slpte, is_write);
999             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
1000         }
1001         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1002             error_report_once("%s: detected splte reserve non-zero "
1003                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
1004                               "slpte=0x%" PRIx64 ")", __func__, iova,
1005                               level, slpte);
1006             return -VTD_FR_PAGING_ENTRY_RSVD;
1007         }
1008 
1009         if (vtd_is_last_slpte(slpte, level)) {
1010             *slptep = slpte;
1011             *slpte_level = level;
1012             return 0;
1013         }
1014         addr = vtd_get_slpte_addr(slpte, aw_bits);
1015         level--;
1016     }
1017 }
1018 
1019 typedef int (*vtd_page_walk_hook)(IOMMUTLBEntry *entry, void *private);
1020 
1021 /**
1022  * Constant information used during page walking
1023  *
1024  * @hook_fn: hook func to be called when detected page
1025  * @private: private data to be passed into hook func
1026  * @notify_unmap: whether we should notify invalid entries
1027  * @as: VT-d address space of the device
1028  * @aw: maximum address width
1029  * @domain: domain ID of the page walk
1030  */
1031 typedef struct {
1032     VTDAddressSpace *as;
1033     vtd_page_walk_hook hook_fn;
1034     void *private;
1035     bool notify_unmap;
1036     uint8_t aw;
1037     uint16_t domain_id;
1038 } vtd_page_walk_info;
1039 
1040 static int vtd_page_walk_one(IOMMUTLBEntry *entry, vtd_page_walk_info *info)
1041 {
1042     VTDAddressSpace *as = info->as;
1043     vtd_page_walk_hook hook_fn = info->hook_fn;
1044     void *private = info->private;
1045     DMAMap target = {
1046         .iova = entry->iova,
1047         .size = entry->addr_mask,
1048         .translated_addr = entry->translated_addr,
1049         .perm = entry->perm,
1050     };
1051     DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
1052 
1053     if (entry->perm == IOMMU_NONE && !info->notify_unmap) {
1054         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1055         return 0;
1056     }
1057 
1058     assert(hook_fn);
1059 
1060     /* Update local IOVA mapped ranges */
1061     if (entry->perm) {
1062         if (mapped) {
1063             /* If it's exactly the same translation, skip */
1064             if (!memcmp(mapped, &target, sizeof(target))) {
1065                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
1066                                                  entry->translated_addr);
1067                 return 0;
1068             } else {
1069                 /*
1070                  * Translation changed.  Normally this should not
1071                  * happen, but it can happen when with buggy guest
1072                  * OSes.  Note that there will be a small window that
1073                  * we don't have map at all.  But that's the best
1074                  * effort we can do.  The ideal way to emulate this is
1075                  * atomically modify the PTE to follow what has
1076                  * changed, but we can't.  One example is that vfio
1077                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1078                  * interface to modify a mapping (meanwhile it seems
1079                  * meaningless to even provide one).  Anyway, let's
1080                  * mark this as a TODO in case one day we'll have
1081                  * a better solution.
1082                  */
1083                 IOMMUAccessFlags cache_perm = entry->perm;
1084                 int ret;
1085 
1086                 /* Emulate an UNMAP */
1087                 entry->perm = IOMMU_NONE;
1088                 trace_vtd_page_walk_one(info->domain_id,
1089                                         entry->iova,
1090                                         entry->translated_addr,
1091                                         entry->addr_mask,
1092                                         entry->perm);
1093                 ret = hook_fn(entry, private);
1094                 if (ret) {
1095                     return ret;
1096                 }
1097                 /* Drop any existing mapping */
1098                 iova_tree_remove(as->iova_tree, &target);
1099                 /* Recover the correct permission */
1100                 entry->perm = cache_perm;
1101             }
1102         }
1103         iova_tree_insert(as->iova_tree, &target);
1104     } else {
1105         if (!mapped) {
1106             /* Skip since we didn't map this range at all */
1107             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1108             return 0;
1109         }
1110         iova_tree_remove(as->iova_tree, &target);
1111     }
1112 
1113     trace_vtd_page_walk_one(info->domain_id, entry->iova,
1114                             entry->translated_addr, entry->addr_mask,
1115                             entry->perm);
1116     return hook_fn(entry, private);
1117 }
1118 
1119 /**
1120  * vtd_page_walk_level - walk over specific level for IOVA range
1121  *
1122  * @addr: base GPA addr to start the walk
1123  * @start: IOVA range start address
1124  * @end: IOVA range end address (start <= addr < end)
1125  * @read: whether parent level has read permission
1126  * @write: whether parent level has write permission
1127  * @info: constant information for the page walk
1128  */
1129 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
1130                                uint64_t end, uint32_t level, bool read,
1131                                bool write, vtd_page_walk_info *info)
1132 {
1133     bool read_cur, write_cur, entry_valid;
1134     uint32_t offset;
1135     uint64_t slpte;
1136     uint64_t subpage_size, subpage_mask;
1137     IOMMUTLBEntry entry;
1138     uint64_t iova = start;
1139     uint64_t iova_next;
1140     int ret = 0;
1141 
1142     trace_vtd_page_walk_level(addr, level, start, end);
1143 
1144     subpage_size = 1ULL << vtd_slpt_level_shift(level);
1145     subpage_mask = vtd_slpt_level_page_mask(level);
1146 
1147     while (iova < end) {
1148         iova_next = (iova & subpage_mask) + subpage_size;
1149 
1150         offset = vtd_iova_level_offset(iova, level);
1151         slpte = vtd_get_slpte(addr, offset);
1152 
1153         if (slpte == (uint64_t)-1) {
1154             trace_vtd_page_walk_skip_read(iova, iova_next);
1155             goto next;
1156         }
1157 
1158         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1159             trace_vtd_page_walk_skip_reserve(iova, iova_next);
1160             goto next;
1161         }
1162 
1163         /* Permissions are stacked with parents' */
1164         read_cur = read && (slpte & VTD_SL_R);
1165         write_cur = write && (slpte & VTD_SL_W);
1166 
1167         /*
1168          * As long as we have either read/write permission, this is a
1169          * valid entry. The rule works for both page entries and page
1170          * table entries.
1171          */
1172         entry_valid = read_cur | write_cur;
1173 
1174         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
1175             /*
1176              * This is a valid PDE (or even bigger than PDE).  We need
1177              * to walk one further level.
1178              */
1179             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
1180                                       iova, MIN(iova_next, end), level - 1,
1181                                       read_cur, write_cur, info);
1182         } else {
1183             /*
1184              * This means we are either:
1185              *
1186              * (1) the real page entry (either 4K page, or huge page)
1187              * (2) the whole range is invalid
1188              *
1189              * In either case, we send an IOTLB notification down.
1190              */
1191             entry.target_as = &address_space_memory;
1192             entry.iova = iova & subpage_mask;
1193             entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
1194             entry.addr_mask = ~subpage_mask;
1195             /* NOTE: this is only meaningful if entry_valid == true */
1196             entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
1197             ret = vtd_page_walk_one(&entry, info);
1198         }
1199 
1200         if (ret < 0) {
1201             return ret;
1202         }
1203 
1204 next:
1205         iova = iova_next;
1206     }
1207 
1208     return 0;
1209 }
1210 
1211 /**
1212  * vtd_page_walk - walk specific IOVA range, and call the hook
1213  *
1214  * @s: intel iommu state
1215  * @ce: context entry to walk upon
1216  * @start: IOVA address to start the walk
1217  * @end: IOVA range end address (start <= addr < end)
1218  * @info: page walking information struct
1219  */
1220 static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce,
1221                          uint64_t start, uint64_t end,
1222                          vtd_page_walk_info *info)
1223 {
1224     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1225     uint32_t level = vtd_get_iova_level(s, ce);
1226 
1227     if (!vtd_iova_range_check(s, start, ce, info->aw)) {
1228         return -VTD_FR_ADDR_BEYOND_MGAW;
1229     }
1230 
1231     if (!vtd_iova_range_check(s, end, ce, info->aw)) {
1232         /* Fix end so that it reaches the maximum */
1233         end = vtd_iova_limit(s, ce, info->aw);
1234     }
1235 
1236     return vtd_page_walk_level(addr, start, end, level, true, true, info);
1237 }
1238 
1239 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s,
1240                                           VTDRootEntry *re)
1241 {
1242     /* Legacy Mode reserved bits check */
1243     if (!s->root_scalable &&
1244         (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1245         goto rsvd_err;
1246 
1247     /* Scalable Mode reserved bits check */
1248     if (s->root_scalable &&
1249         ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) ||
1250          (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1251         goto rsvd_err;
1252 
1253     return 0;
1254 
1255 rsvd_err:
1256     error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1257                       ", lo=0x%"PRIx64,
1258                       __func__, re->hi, re->lo);
1259     return -VTD_FR_ROOT_ENTRY_RSVD;
1260 }
1261 
1262 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s,
1263                                                     VTDContextEntry *ce)
1264 {
1265     if (!s->root_scalable &&
1266         (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI ||
1267          ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1268         error_report_once("%s: invalid context entry: hi=%"PRIx64
1269                           ", lo=%"PRIx64" (reserved nonzero)",
1270                           __func__, ce->hi, ce->lo);
1271         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1272     }
1273 
1274     if (s->root_scalable &&
1275         (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) ||
1276          ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 ||
1277          ce->val[2] ||
1278          ce->val[3])) {
1279         error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1280                           ", val[2]=%"PRIx64
1281                           ", val[1]=%"PRIx64
1282                           ", val[0]=%"PRIx64" (reserved nonzero)",
1283                           __func__, ce->val[3], ce->val[2],
1284                           ce->val[1], ce->val[0]);
1285         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1286     }
1287 
1288     return 0;
1289 }
1290 
1291 static int vtd_ce_rid2pasid_check(IntelIOMMUState *s,
1292                                   VTDContextEntry *ce)
1293 {
1294     VTDPASIDEntry pe;
1295 
1296     /*
1297      * Make sure in Scalable Mode, a present context entry
1298      * has valid rid2pasid setting, which includes valid
1299      * rid2pasid field and corresponding pasid entry setting
1300      */
1301     return vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1302 }
1303 
1304 /* Map a device to its corresponding domain (context-entry) */
1305 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
1306                                     uint8_t devfn, VTDContextEntry *ce)
1307 {
1308     VTDRootEntry re;
1309     int ret_fr;
1310     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
1311 
1312     ret_fr = vtd_get_root_entry(s, bus_num, &re);
1313     if (ret_fr) {
1314         return ret_fr;
1315     }
1316 
1317     if (!vtd_root_entry_present(s, &re, devfn)) {
1318         /* Not error - it's okay we don't have root entry. */
1319         trace_vtd_re_not_present(bus_num);
1320         return -VTD_FR_ROOT_ENTRY_P;
1321     }
1322 
1323     ret_fr = vtd_root_entry_rsvd_bits_check(s, &re);
1324     if (ret_fr) {
1325         return ret_fr;
1326     }
1327 
1328     ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce);
1329     if (ret_fr) {
1330         return ret_fr;
1331     }
1332 
1333     if (!vtd_ce_present(ce)) {
1334         /* Not error - it's okay we don't have context entry. */
1335         trace_vtd_ce_not_present(bus_num, devfn);
1336         return -VTD_FR_CONTEXT_ENTRY_P;
1337     }
1338 
1339     ret_fr = vtd_context_entry_rsvd_bits_check(s, ce);
1340     if (ret_fr) {
1341         return ret_fr;
1342     }
1343 
1344     /* Check if the programming of context-entry is valid */
1345     if (!s->root_scalable &&
1346         !vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1347         error_report_once("%s: invalid context entry: hi=%"PRIx64
1348                           ", lo=%"PRIx64" (level %d not supported)",
1349                           __func__, ce->hi, ce->lo,
1350                           vtd_ce_get_level(ce));
1351         return -VTD_FR_CONTEXT_ENTRY_INV;
1352     }
1353 
1354     if (!s->root_scalable) {
1355         /* Do translation type check */
1356         if (!vtd_ce_type_check(x86_iommu, ce)) {
1357             /* Errors dumped in vtd_ce_type_check() */
1358             return -VTD_FR_CONTEXT_ENTRY_INV;
1359         }
1360     } else {
1361         /*
1362          * Check if the programming of context-entry.rid2pasid
1363          * and corresponding pasid setting is valid, and thus
1364          * avoids to check pasid entry fetching result in future
1365          * helper function calling.
1366          */
1367         ret_fr = vtd_ce_rid2pasid_check(s, ce);
1368         if (ret_fr) {
1369             return ret_fr;
1370         }
1371     }
1372 
1373     return 0;
1374 }
1375 
1376 static int vtd_sync_shadow_page_hook(IOMMUTLBEntry *entry,
1377                                      void *private)
1378 {
1379     memory_region_notify_iommu((IOMMUMemoryRegion *)private, 0, *entry);
1380     return 0;
1381 }
1382 
1383 static uint16_t vtd_get_domain_id(IntelIOMMUState *s,
1384                                   VTDContextEntry *ce)
1385 {
1386     VTDPASIDEntry pe;
1387 
1388     if (s->root_scalable) {
1389         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1390         return VTD_SM_PASID_ENTRY_DID(pe.val[1]);
1391     }
1392 
1393     return VTD_CONTEXT_ENTRY_DID(ce->hi);
1394 }
1395 
1396 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1397                                             VTDContextEntry *ce,
1398                                             hwaddr addr, hwaddr size)
1399 {
1400     IntelIOMMUState *s = vtd_as->iommu_state;
1401     vtd_page_walk_info info = {
1402         .hook_fn = vtd_sync_shadow_page_hook,
1403         .private = (void *)&vtd_as->iommu,
1404         .notify_unmap = true,
1405         .aw = s->aw_bits,
1406         .as = vtd_as,
1407         .domain_id = vtd_get_domain_id(s, ce),
1408     };
1409 
1410     return vtd_page_walk(s, ce, addr, addr + size, &info);
1411 }
1412 
1413 static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
1414 {
1415     int ret;
1416     VTDContextEntry ce;
1417     IOMMUNotifier *n;
1418 
1419     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1420                                    pci_bus_num(vtd_as->bus),
1421                                    vtd_as->devfn, &ce);
1422     if (ret) {
1423         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1424             /*
1425              * It's a valid scenario to have a context entry that is
1426              * not present.  For example, when a device is removed
1427              * from an existing domain then the context entry will be
1428              * zeroed by the guest before it was put into another
1429              * domain.  When this happens, instead of synchronizing
1430              * the shadow pages we should invalidate all existing
1431              * mappings and notify the backends.
1432              */
1433             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1434                 vtd_address_space_unmap(vtd_as, n);
1435             }
1436             ret = 0;
1437         }
1438         return ret;
1439     }
1440 
1441     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1442 }
1443 
1444 /*
1445  * Check if specific device is configed to bypass address
1446  * translation for DMA requests. In Scalable Mode, bypass
1447  * 1st-level translation or 2nd-level translation, it depends
1448  * on PGTT setting.
1449  */
1450 static bool vtd_dev_pt_enabled(VTDAddressSpace *as)
1451 {
1452     IntelIOMMUState *s;
1453     VTDContextEntry ce;
1454     VTDPASIDEntry pe;
1455     int ret;
1456 
1457     assert(as);
1458 
1459     s = as->iommu_state;
1460     ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus),
1461                                    as->devfn, &ce);
1462     if (ret) {
1463         /*
1464          * Possibly failed to parse the context entry for some reason
1465          * (e.g., during init, or any guest configuration errors on
1466          * context entries). We should assume PT not enabled for
1467          * safety.
1468          */
1469         return false;
1470     }
1471 
1472     if (s->root_scalable) {
1473         ret = vtd_ce_get_rid2pasid_entry(s, &ce, &pe);
1474         if (ret) {
1475             error_report_once("%s: vtd_ce_get_rid2pasid_entry error: %"PRId32,
1476                               __func__, ret);
1477             return false;
1478         }
1479         return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT);
1480     }
1481 
1482     return (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH);
1483 }
1484 
1485 /* Return whether the device is using IOMMU translation. */
1486 static bool vtd_switch_address_space(VTDAddressSpace *as)
1487 {
1488     bool use_iommu;
1489     /* Whether we need to take the BQL on our own */
1490     bool take_bql = !qemu_mutex_iothread_locked();
1491 
1492     assert(as);
1493 
1494     use_iommu = as->iommu_state->dmar_enabled && !vtd_dev_pt_enabled(as);
1495 
1496     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1497                                    VTD_PCI_SLOT(as->devfn),
1498                                    VTD_PCI_FUNC(as->devfn),
1499                                    use_iommu);
1500 
1501     /*
1502      * It's possible that we reach here without BQL, e.g., when called
1503      * from vtd_pt_enable_fast_path(). However the memory APIs need
1504      * it. We'd better make sure we have had it already, or, take it.
1505      */
1506     if (take_bql) {
1507         qemu_mutex_lock_iothread();
1508     }
1509 
1510     /* Turn off first then on the other */
1511     if (use_iommu) {
1512         memory_region_set_enabled(&as->nodmar, false);
1513         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1514     } else {
1515         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1516         memory_region_set_enabled(&as->nodmar, true);
1517     }
1518 
1519     if (take_bql) {
1520         qemu_mutex_unlock_iothread();
1521     }
1522 
1523     return use_iommu;
1524 }
1525 
1526 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1527 {
1528     GHashTableIter iter;
1529     VTDBus *vtd_bus;
1530     int i;
1531 
1532     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1533     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1534         for (i = 0; i < PCI_DEVFN_MAX; i++) {
1535             if (!vtd_bus->dev_as[i]) {
1536                 continue;
1537             }
1538             vtd_switch_address_space(vtd_bus->dev_as[i]);
1539         }
1540     }
1541 }
1542 
1543 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn)
1544 {
1545     return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
1546 }
1547 
1548 static const bool vtd_qualified_faults[] = {
1549     [VTD_FR_RESERVED] = false,
1550     [VTD_FR_ROOT_ENTRY_P] = false,
1551     [VTD_FR_CONTEXT_ENTRY_P] = true,
1552     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1553     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1554     [VTD_FR_WRITE] = true,
1555     [VTD_FR_READ] = true,
1556     [VTD_FR_PAGING_ENTRY_INV] = true,
1557     [VTD_FR_ROOT_TABLE_INV] = false,
1558     [VTD_FR_CONTEXT_TABLE_INV] = false,
1559     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1560     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1561     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1562     [VTD_FR_PASID_TABLE_INV] = false,
1563     [VTD_FR_RESERVED_ERR] = false,
1564     [VTD_FR_MAX] = false,
1565 };
1566 
1567 /* To see if a fault condition is "qualified", which is reported to software
1568  * only if the FPD field in the context-entry used to process the faulting
1569  * request is 0.
1570  */
1571 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1572 {
1573     return vtd_qualified_faults[fault];
1574 }
1575 
1576 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1577 {
1578     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1579 }
1580 
1581 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1582 {
1583     VTDBus *vtd_bus;
1584     VTDAddressSpace *vtd_as;
1585     bool success = false;
1586 
1587     vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id));
1588     if (!vtd_bus) {
1589         goto out;
1590     }
1591 
1592     vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)];
1593     if (!vtd_as) {
1594         goto out;
1595     }
1596 
1597     if (vtd_switch_address_space(vtd_as) == false) {
1598         /* We switched off IOMMU region successfully. */
1599         success = true;
1600     }
1601 
1602 out:
1603     trace_vtd_pt_enable_fast_path(source_id, success);
1604 }
1605 
1606 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1607  * translation.
1608  *
1609  * Called from RCU critical section.
1610  *
1611  * @bus_num: The bus number
1612  * @devfn: The devfn, which is the  combined of device and function number
1613  * @is_write: The access is a write operation
1614  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1615  *
1616  * Returns true if translation is successful, otherwise false.
1617  */
1618 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1619                                    uint8_t devfn, hwaddr addr, bool is_write,
1620                                    IOMMUTLBEntry *entry)
1621 {
1622     IntelIOMMUState *s = vtd_as->iommu_state;
1623     VTDContextEntry ce;
1624     uint8_t bus_num = pci_bus_num(bus);
1625     VTDContextCacheEntry *cc_entry;
1626     uint64_t slpte, page_mask;
1627     uint32_t level;
1628     uint16_t source_id = vtd_make_source_id(bus_num, devfn);
1629     int ret_fr;
1630     bool is_fpd_set = false;
1631     bool reads = true;
1632     bool writes = true;
1633     uint8_t access_flags;
1634     VTDIOTLBEntry *iotlb_entry;
1635 
1636     /*
1637      * We have standalone memory region for interrupt addresses, we
1638      * should never receive translation requests in this region.
1639      */
1640     assert(!vtd_is_interrupt_addr(addr));
1641 
1642     vtd_iommu_lock(s);
1643 
1644     cc_entry = &vtd_as->context_cache_entry;
1645 
1646     /* Try to fetch slpte form IOTLB */
1647     iotlb_entry = vtd_lookup_iotlb(s, source_id, addr);
1648     if (iotlb_entry) {
1649         trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1650                                  iotlb_entry->domain_id);
1651         slpte = iotlb_entry->slpte;
1652         access_flags = iotlb_entry->access_flags;
1653         page_mask = iotlb_entry->mask;
1654         goto out;
1655     }
1656 
1657     /* Try to fetch context-entry from cache first */
1658     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1659         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1660                                cc_entry->context_entry.lo,
1661                                cc_entry->context_cache_gen);
1662         ce = cc_entry->context_entry;
1663         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1664         if (!is_fpd_set && s->root_scalable) {
1665             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1666             VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1667         }
1668     } else {
1669         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1670         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1671         if (!ret_fr && !is_fpd_set && s->root_scalable) {
1672             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1673         }
1674         VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1675         /* Update context-cache */
1676         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1677                                   cc_entry->context_cache_gen,
1678                                   s->context_cache_gen);
1679         cc_entry->context_entry = ce;
1680         cc_entry->context_cache_gen = s->context_cache_gen;
1681     }
1682 
1683     /*
1684      * We don't need to translate for pass-through context entries.
1685      * Also, let's ignore IOTLB caching as well for PT devices.
1686      */
1687     if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) {
1688         entry->iova = addr & VTD_PAGE_MASK_4K;
1689         entry->translated_addr = entry->iova;
1690         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1691         entry->perm = IOMMU_RW;
1692         trace_vtd_translate_pt(source_id, entry->iova);
1693 
1694         /*
1695          * When this happens, it means firstly caching-mode is not
1696          * enabled, and this is the first passthrough translation for
1697          * the device. Let's enable the fast path for passthrough.
1698          *
1699          * When passthrough is disabled again for the device, we can
1700          * capture it via the context entry invalidation, then the
1701          * IOMMU region can be swapped back.
1702          */
1703         vtd_pt_enable_fast_path(s, source_id);
1704         vtd_iommu_unlock(s);
1705         return true;
1706     }
1707 
1708     ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level,
1709                                &reads, &writes, s->aw_bits);
1710     VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1711 
1712     page_mask = vtd_slpt_level_page_mask(level);
1713     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1714     vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce), addr, slpte,
1715                      access_flags, level);
1716 out:
1717     vtd_iommu_unlock(s);
1718     entry->iova = addr & page_mask;
1719     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1720     entry->addr_mask = ~page_mask;
1721     entry->perm = access_flags;
1722     return true;
1723 
1724 error:
1725     vtd_iommu_unlock(s);
1726     entry->iova = 0;
1727     entry->translated_addr = 0;
1728     entry->addr_mask = 0;
1729     entry->perm = IOMMU_NONE;
1730     return false;
1731 }
1732 
1733 static void vtd_root_table_setup(IntelIOMMUState *s)
1734 {
1735     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1736     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
1737 
1738     vtd_update_scalable_state(s);
1739 
1740     trace_vtd_reg_dmar_root(s->root, s->root_scalable);
1741 }
1742 
1743 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
1744                                uint32_t index, uint32_t mask)
1745 {
1746     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
1747 }
1748 
1749 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
1750 {
1751     uint64_t value = 0;
1752     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
1753     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
1754     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
1755     s->intr_eime = value & VTD_IRTA_EIME;
1756 
1757     /* Notify global invalidation */
1758     vtd_iec_notify_all(s, true, 0, 0);
1759 
1760     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
1761 }
1762 
1763 static void vtd_iommu_replay_all(IntelIOMMUState *s)
1764 {
1765     VTDAddressSpace *vtd_as;
1766 
1767     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1768         vtd_sync_shadow_page_table(vtd_as);
1769     }
1770 }
1771 
1772 static void vtd_context_global_invalidate(IntelIOMMUState *s)
1773 {
1774     trace_vtd_inv_desc_cc_global();
1775     /* Protects context cache */
1776     vtd_iommu_lock(s);
1777     s->context_cache_gen++;
1778     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
1779         vtd_reset_context_cache_locked(s);
1780     }
1781     vtd_iommu_unlock(s);
1782     vtd_address_space_refresh_all(s);
1783     /*
1784      * From VT-d spec 6.5.2.1, a global context entry invalidation
1785      * should be followed by a IOTLB global invalidation, so we should
1786      * be safe even without this. Hoewever, let's replay the region as
1787      * well to be safer, and go back here when we need finer tunes for
1788      * VT-d emulation codes.
1789      */
1790     vtd_iommu_replay_all(s);
1791 }
1792 
1793 /* Do a context-cache device-selective invalidation.
1794  * @func_mask: FM field after shifting
1795  */
1796 static void vtd_context_device_invalidate(IntelIOMMUState *s,
1797                                           uint16_t source_id,
1798                                           uint16_t func_mask)
1799 {
1800     uint16_t mask;
1801     VTDBus *vtd_bus;
1802     VTDAddressSpace *vtd_as;
1803     uint8_t bus_n, devfn;
1804     uint16_t devfn_it;
1805 
1806     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
1807 
1808     switch (func_mask & 3) {
1809     case 0:
1810         mask = 0;   /* No bits in the SID field masked */
1811         break;
1812     case 1:
1813         mask = 4;   /* Mask bit 2 in the SID field */
1814         break;
1815     case 2:
1816         mask = 6;   /* Mask bit 2:1 in the SID field */
1817         break;
1818     case 3:
1819         mask = 7;   /* Mask bit 2:0 in the SID field */
1820         break;
1821     }
1822     mask = ~mask;
1823 
1824     bus_n = VTD_SID_TO_BUS(source_id);
1825     vtd_bus = vtd_find_as_from_bus_num(s, bus_n);
1826     if (vtd_bus) {
1827         devfn = VTD_SID_TO_DEVFN(source_id);
1828         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
1829             vtd_as = vtd_bus->dev_as[devfn_it];
1830             if (vtd_as && ((devfn_it & mask) == (devfn & mask))) {
1831                 trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it),
1832                                              VTD_PCI_FUNC(devfn_it));
1833                 vtd_iommu_lock(s);
1834                 vtd_as->context_cache_entry.context_cache_gen = 0;
1835                 vtd_iommu_unlock(s);
1836                 /*
1837                  * Do switch address space when needed, in case if the
1838                  * device passthrough bit is switched.
1839                  */
1840                 vtd_switch_address_space(vtd_as);
1841                 /*
1842                  * So a device is moving out of (or moving into) a
1843                  * domain, resync the shadow page table.
1844                  * This won't bring bad even if we have no such
1845                  * notifier registered - the IOMMU notification
1846                  * framework will skip MAP notifications if that
1847                  * happened.
1848                  */
1849                 vtd_sync_shadow_page_table(vtd_as);
1850             }
1851         }
1852     }
1853 }
1854 
1855 /* Context-cache invalidation
1856  * Returns the Context Actual Invalidation Granularity.
1857  * @val: the content of the CCMD_REG
1858  */
1859 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
1860 {
1861     uint64_t caig;
1862     uint64_t type = val & VTD_CCMD_CIRG_MASK;
1863 
1864     switch (type) {
1865     case VTD_CCMD_DOMAIN_INVL:
1866         /* Fall through */
1867     case VTD_CCMD_GLOBAL_INVL:
1868         caig = VTD_CCMD_GLOBAL_INVL_A;
1869         vtd_context_global_invalidate(s);
1870         break;
1871 
1872     case VTD_CCMD_DEVICE_INVL:
1873         caig = VTD_CCMD_DEVICE_INVL_A;
1874         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
1875         break;
1876 
1877     default:
1878         error_report_once("%s: invalid context: 0x%" PRIx64,
1879                           __func__, val);
1880         caig = 0;
1881     }
1882     return caig;
1883 }
1884 
1885 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
1886 {
1887     trace_vtd_inv_desc_iotlb_global();
1888     vtd_reset_iotlb(s);
1889     vtd_iommu_replay_all(s);
1890 }
1891 
1892 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
1893 {
1894     VTDContextEntry ce;
1895     VTDAddressSpace *vtd_as;
1896 
1897     trace_vtd_inv_desc_iotlb_domain(domain_id);
1898 
1899     vtd_iommu_lock(s);
1900     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
1901                                 &domain_id);
1902     vtd_iommu_unlock(s);
1903 
1904     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1905         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1906                                       vtd_as->devfn, &ce) &&
1907             domain_id == vtd_get_domain_id(s, &ce)) {
1908             vtd_sync_shadow_page_table(vtd_as);
1909         }
1910     }
1911 }
1912 
1913 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
1914                                            uint16_t domain_id, hwaddr addr,
1915                                            uint8_t am)
1916 {
1917     VTDAddressSpace *vtd_as;
1918     VTDContextEntry ce;
1919     int ret;
1920     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
1921 
1922     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
1923         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1924                                        vtd_as->devfn, &ce);
1925         if (!ret && domain_id == vtd_get_domain_id(s, &ce)) {
1926             if (vtd_as_has_map_notifier(vtd_as)) {
1927                 /*
1928                  * As long as we have MAP notifications registered in
1929                  * any of our IOMMU notifiers, we need to sync the
1930                  * shadow page table.
1931                  */
1932                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
1933             } else {
1934                 /*
1935                  * For UNMAP-only notifiers, we don't need to walk the
1936                  * page tables.  We just deliver the PSI down to
1937                  * invalidate caches.
1938                  */
1939                 IOMMUTLBEntry entry = {
1940                     .target_as = &address_space_memory,
1941                     .iova = addr,
1942                     .translated_addr = 0,
1943                     .addr_mask = size - 1,
1944                     .perm = IOMMU_NONE,
1945                 };
1946                 memory_region_notify_iommu(&vtd_as->iommu, 0, entry);
1947             }
1948         }
1949     }
1950 }
1951 
1952 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
1953                                       hwaddr addr, uint8_t am)
1954 {
1955     VTDIOTLBPageInvInfo info;
1956 
1957     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
1958 
1959     assert(am <= VTD_MAMV);
1960     info.domain_id = domain_id;
1961     info.addr = addr;
1962     info.mask = ~((1 << am) - 1);
1963     vtd_iommu_lock(s);
1964     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
1965     vtd_iommu_unlock(s);
1966     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am);
1967 }
1968 
1969 /* Flush IOTLB
1970  * Returns the IOTLB Actual Invalidation Granularity.
1971  * @val: the content of the IOTLB_REG
1972  */
1973 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
1974 {
1975     uint64_t iaig;
1976     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
1977     uint16_t domain_id;
1978     hwaddr addr;
1979     uint8_t am;
1980 
1981     switch (type) {
1982     case VTD_TLB_GLOBAL_FLUSH:
1983         iaig = VTD_TLB_GLOBAL_FLUSH_A;
1984         vtd_iotlb_global_invalidate(s);
1985         break;
1986 
1987     case VTD_TLB_DSI_FLUSH:
1988         domain_id = VTD_TLB_DID(val);
1989         iaig = VTD_TLB_DSI_FLUSH_A;
1990         vtd_iotlb_domain_invalidate(s, domain_id);
1991         break;
1992 
1993     case VTD_TLB_PSI_FLUSH:
1994         domain_id = VTD_TLB_DID(val);
1995         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
1996         am = VTD_IVA_AM(addr);
1997         addr = VTD_IVA_ADDR(addr);
1998         if (am > VTD_MAMV) {
1999             error_report_once("%s: address mask overflow: 0x%" PRIx64,
2000                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
2001             iaig = 0;
2002             break;
2003         }
2004         iaig = VTD_TLB_PSI_FLUSH_A;
2005         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2006         break;
2007 
2008     default:
2009         error_report_once("%s: invalid granularity: 0x%" PRIx64,
2010                           __func__, val);
2011         iaig = 0;
2012     }
2013     return iaig;
2014 }
2015 
2016 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
2017 
2018 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
2019 {
2020     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
2021            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
2022 }
2023 
2024 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
2025 {
2026     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
2027 
2028     trace_vtd_inv_qi_enable(en);
2029 
2030     if (en) {
2031         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
2032         /* 2^(x+8) entries */
2033         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0));
2034         s->qi_enabled = true;
2035         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
2036         /* Ok - report back to driver */
2037         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
2038 
2039         if (s->iq_tail != 0) {
2040             /*
2041              * This is a spec violation but Windows guests are known to set up
2042              * Queued Invalidation this way so we allow the write and process
2043              * Invalidation Descriptors right away.
2044              */
2045             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
2046             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2047                 vtd_fetch_inv_desc(s);
2048             }
2049         }
2050     } else {
2051         if (vtd_queued_inv_disable_check(s)) {
2052             /* disable Queued Invalidation */
2053             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
2054             s->iq_head = 0;
2055             s->qi_enabled = false;
2056             /* Ok - report back to driver */
2057             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
2058         } else {
2059             error_report_once("%s: detected improper state when disable QI "
2060                               "(head=0x%x, tail=0x%x, last_type=%d)",
2061                               __func__,
2062                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
2063         }
2064     }
2065 }
2066 
2067 /* Set Root Table Pointer */
2068 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
2069 {
2070     vtd_root_table_setup(s);
2071     /* Ok - report back to driver */
2072     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
2073     vtd_reset_caches(s);
2074     vtd_address_space_refresh_all(s);
2075 }
2076 
2077 /* Set Interrupt Remap Table Pointer */
2078 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
2079 {
2080     vtd_interrupt_remap_table_setup(s);
2081     /* Ok - report back to driver */
2082     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
2083 }
2084 
2085 /* Handle Translation Enable/Disable */
2086 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
2087 {
2088     if (s->dmar_enabled == en) {
2089         return;
2090     }
2091 
2092     trace_vtd_dmar_enable(en);
2093 
2094     if (en) {
2095         s->dmar_enabled = true;
2096         /* Ok - report back to driver */
2097         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
2098     } else {
2099         s->dmar_enabled = false;
2100 
2101         /* Clear the index of Fault Recording Register */
2102         s->next_frcd_reg = 0;
2103         /* Ok - report back to driver */
2104         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
2105     }
2106 
2107     vtd_reset_caches(s);
2108     vtd_address_space_refresh_all(s);
2109 }
2110 
2111 /* Handle Interrupt Remap Enable/Disable */
2112 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
2113 {
2114     trace_vtd_ir_enable(en);
2115 
2116     if (en) {
2117         s->intr_enabled = true;
2118         /* Ok - report back to driver */
2119         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
2120     } else {
2121         s->intr_enabled = false;
2122         /* Ok - report back to driver */
2123         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
2124     }
2125 }
2126 
2127 /* Handle write to Global Command Register */
2128 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
2129 {
2130     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
2131     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
2132     uint32_t changed = status ^ val;
2133 
2134     trace_vtd_reg_write_gcmd(status, val);
2135     if (changed & VTD_GCMD_TE) {
2136         /* Translation enable/disable */
2137         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
2138     }
2139     if (val & VTD_GCMD_SRTP) {
2140         /* Set/update the root-table pointer */
2141         vtd_handle_gcmd_srtp(s);
2142     }
2143     if (changed & VTD_GCMD_QIE) {
2144         /* Queued Invalidation Enable */
2145         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
2146     }
2147     if (val & VTD_GCMD_SIRTP) {
2148         /* Set/update the interrupt remapping root-table pointer */
2149         vtd_handle_gcmd_sirtp(s);
2150     }
2151     if (changed & VTD_GCMD_IRE) {
2152         /* Interrupt remap enable/disable */
2153         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
2154     }
2155 }
2156 
2157 /* Handle write to Context Command Register */
2158 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
2159 {
2160     uint64_t ret;
2161     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
2162 
2163     /* Context-cache invalidation request */
2164     if (val & VTD_CCMD_ICC) {
2165         if (s->qi_enabled) {
2166             error_report_once("Queued Invalidation enabled, "
2167                               "should not use register-based invalidation");
2168             return;
2169         }
2170         ret = vtd_context_cache_invalidate(s, val);
2171         /* Invalidation completed. Change something to show */
2172         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
2173         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
2174                                       ret);
2175     }
2176 }
2177 
2178 /* Handle write to IOTLB Invalidation Register */
2179 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
2180 {
2181     uint64_t ret;
2182     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
2183 
2184     /* IOTLB invalidation request */
2185     if (val & VTD_TLB_IVT) {
2186         if (s->qi_enabled) {
2187             error_report_once("Queued Invalidation enabled, "
2188                               "should not use register-based invalidation");
2189             return;
2190         }
2191         ret = vtd_iotlb_flush(s, val);
2192         /* Invalidation completed. Change something to show */
2193         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
2194         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
2195                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
2196     }
2197 }
2198 
2199 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2200 static bool vtd_get_inv_desc(IntelIOMMUState *s,
2201                              VTDInvDesc *inv_desc)
2202 {
2203     dma_addr_t base_addr = s->iq;
2204     uint32_t offset = s->iq_head;
2205     uint32_t dw = s->iq_dw ? 32 : 16;
2206     dma_addr_t addr = base_addr + offset * dw;
2207 
2208     if (dma_memory_read(&address_space_memory, addr, inv_desc, dw)) {
2209         error_report_once("Read INV DESC failed.");
2210         return false;
2211     }
2212     inv_desc->lo = le64_to_cpu(inv_desc->lo);
2213     inv_desc->hi = le64_to_cpu(inv_desc->hi);
2214     if (dw == 32) {
2215         inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]);
2216         inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]);
2217     }
2218     return true;
2219 }
2220 
2221 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2222 {
2223     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
2224         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
2225         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2226                           " (reserved nonzero)", __func__, inv_desc->hi,
2227                           inv_desc->lo);
2228         return false;
2229     }
2230     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
2231         /* Status Write */
2232         uint32_t status_data = (uint32_t)(inv_desc->lo >>
2233                                VTD_INV_DESC_WAIT_DATA_SHIFT);
2234 
2235         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
2236 
2237         /* FIXME: need to be masked with HAW? */
2238         dma_addr_t status_addr = inv_desc->hi;
2239         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
2240         status_data = cpu_to_le32(status_data);
2241         if (dma_memory_write(&address_space_memory, status_addr, &status_data,
2242                              sizeof(status_data))) {
2243             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
2244             return false;
2245         }
2246     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
2247         /* Interrupt flag */
2248         vtd_generate_completion_event(s);
2249     } else {
2250         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2251                           " (unknown type)", __func__, inv_desc->hi,
2252                           inv_desc->lo);
2253         return false;
2254     }
2255     return true;
2256 }
2257 
2258 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
2259                                            VTDInvDesc *inv_desc)
2260 {
2261     uint16_t sid, fmask;
2262 
2263     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
2264         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2265                           " (reserved nonzero)", __func__, inv_desc->hi,
2266                           inv_desc->lo);
2267         return false;
2268     }
2269     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
2270     case VTD_INV_DESC_CC_DOMAIN:
2271         trace_vtd_inv_desc_cc_domain(
2272             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
2273         /* Fall through */
2274     case VTD_INV_DESC_CC_GLOBAL:
2275         vtd_context_global_invalidate(s);
2276         break;
2277 
2278     case VTD_INV_DESC_CC_DEVICE:
2279         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
2280         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
2281         vtd_context_device_invalidate(s, sid, fmask);
2282         break;
2283 
2284     default:
2285         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2286                           " (invalid type)", __func__, inv_desc->hi,
2287                           inv_desc->lo);
2288         return false;
2289     }
2290     return true;
2291 }
2292 
2293 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2294 {
2295     uint16_t domain_id;
2296     uint8_t am;
2297     hwaddr addr;
2298 
2299     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
2300         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
2301         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2302                           ", lo=0x%"PRIx64" (reserved bits unzero)\n",
2303                           __func__, inv_desc->hi, inv_desc->lo);
2304         return false;
2305     }
2306 
2307     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
2308     case VTD_INV_DESC_IOTLB_GLOBAL:
2309         vtd_iotlb_global_invalidate(s);
2310         break;
2311 
2312     case VTD_INV_DESC_IOTLB_DOMAIN:
2313         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2314         vtd_iotlb_domain_invalidate(s, domain_id);
2315         break;
2316 
2317     case VTD_INV_DESC_IOTLB_PAGE:
2318         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2319         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
2320         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
2321         if (am > VTD_MAMV) {
2322             error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2323                               ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)\n",
2324                               __func__, inv_desc->hi, inv_desc->lo,
2325                               am, (unsigned)VTD_MAMV);
2326             return false;
2327         }
2328         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2329         break;
2330 
2331     default:
2332         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2333                           ", lo=0x%"PRIx64" (type mismatch: 0x%llx)\n",
2334                           __func__, inv_desc->hi, inv_desc->lo,
2335                           inv_desc->lo & VTD_INV_DESC_IOTLB_G);
2336         return false;
2337     }
2338     return true;
2339 }
2340 
2341 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
2342                                      VTDInvDesc *inv_desc)
2343 {
2344     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
2345                            inv_desc->iec.index,
2346                            inv_desc->iec.index_mask);
2347 
2348     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
2349                        inv_desc->iec.index,
2350                        inv_desc->iec.index_mask);
2351     return true;
2352 }
2353 
2354 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
2355                                           VTDInvDesc *inv_desc)
2356 {
2357     VTDAddressSpace *vtd_dev_as;
2358     IOMMUTLBEntry entry;
2359     struct VTDBus *vtd_bus;
2360     hwaddr addr;
2361     uint64_t sz;
2362     uint16_t sid;
2363     uint8_t devfn;
2364     bool size;
2365     uint8_t bus_num;
2366 
2367     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2368     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2369     devfn = sid & 0xff;
2370     bus_num = sid >> 8;
2371     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2372 
2373     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2374         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2375         error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2376                           ", lo=%"PRIx64" (reserved nonzero)", __func__,
2377                           inv_desc->hi, inv_desc->lo);
2378         return false;
2379     }
2380 
2381     vtd_bus = vtd_find_as_from_bus_num(s, bus_num);
2382     if (!vtd_bus) {
2383         goto done;
2384     }
2385 
2386     vtd_dev_as = vtd_bus->dev_as[devfn];
2387     if (!vtd_dev_as) {
2388         goto done;
2389     }
2390 
2391     /* According to ATS spec table 2.4:
2392      * S = 0, bits 15:12 = xxxx     range size: 4K
2393      * S = 1, bits 15:12 = xxx0     range size: 8K
2394      * S = 1, bits 15:12 = xx01     range size: 16K
2395      * S = 1, bits 15:12 = x011     range size: 32K
2396      * S = 1, bits 15:12 = 0111     range size: 64K
2397      * ...
2398      */
2399     if (size) {
2400         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2401         addr &= ~(sz - 1);
2402     } else {
2403         sz = VTD_PAGE_SIZE;
2404     }
2405 
2406     entry.target_as = &vtd_dev_as->as;
2407     entry.addr_mask = sz - 1;
2408     entry.iova = addr;
2409     entry.perm = IOMMU_NONE;
2410     entry.translated_addr = 0;
2411     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry);
2412 
2413 done:
2414     return true;
2415 }
2416 
2417 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2418 {
2419     VTDInvDesc inv_desc;
2420     uint8_t desc_type;
2421 
2422     trace_vtd_inv_qi_head(s->iq_head);
2423     if (!vtd_get_inv_desc(s, &inv_desc)) {
2424         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2425         return false;
2426     }
2427 
2428     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2429     /* FIXME: should update at first or at last? */
2430     s->iq_last_desc_type = desc_type;
2431 
2432     switch (desc_type) {
2433     case VTD_INV_DESC_CC:
2434         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2435         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2436             return false;
2437         }
2438         break;
2439 
2440     case VTD_INV_DESC_IOTLB:
2441         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2442         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2443             return false;
2444         }
2445         break;
2446 
2447     /*
2448      * TODO: the entity of below two cases will be implemented in future series.
2449      * To make guest (which integrates scalable mode support patch set in
2450      * iommu driver) work, just return true is enough so far.
2451      */
2452     case VTD_INV_DESC_PC:
2453         break;
2454 
2455     case VTD_INV_DESC_PIOTLB:
2456         break;
2457 
2458     case VTD_INV_DESC_WAIT:
2459         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2460         if (!vtd_process_wait_desc(s, &inv_desc)) {
2461             return false;
2462         }
2463         break;
2464 
2465     case VTD_INV_DESC_IEC:
2466         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2467         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2468             return false;
2469         }
2470         break;
2471 
2472     case VTD_INV_DESC_DEVICE:
2473         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2474         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2475             return false;
2476         }
2477         break;
2478 
2479     default:
2480         error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64
2481                           " (unknown type)", __func__, inv_desc.hi,
2482                           inv_desc.lo);
2483         return false;
2484     }
2485     s->iq_head++;
2486     if (s->iq_head == s->iq_size) {
2487         s->iq_head = 0;
2488     }
2489     return true;
2490 }
2491 
2492 /* Try to fetch and process more Invalidation Descriptors */
2493 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2494 {
2495     trace_vtd_inv_qi_fetch();
2496 
2497     if (s->iq_tail >= s->iq_size) {
2498         /* Detects an invalid Tail pointer */
2499         error_report_once("%s: detected invalid QI tail "
2500                           "(tail=0x%x, size=0x%x)",
2501                           __func__, s->iq_tail, s->iq_size);
2502         vtd_handle_inv_queue_error(s);
2503         return;
2504     }
2505     while (s->iq_head != s->iq_tail) {
2506         if (!vtd_process_inv_desc(s)) {
2507             /* Invalidation Queue Errors */
2508             vtd_handle_inv_queue_error(s);
2509             break;
2510         }
2511         /* Must update the IQH_REG in time */
2512         vtd_set_quad_raw(s, DMAR_IQH_REG,
2513                          (((uint64_t)(s->iq_head)) << VTD_IQH_QH_SHIFT) &
2514                          VTD_IQH_QH_MASK);
2515     }
2516 }
2517 
2518 /* Handle write to Invalidation Queue Tail Register */
2519 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2520 {
2521     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2522 
2523     if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) {
2524         error_report_once("%s: RSV bit is set: val=0x%"PRIx64,
2525                           __func__, val);
2526         return;
2527     }
2528     s->iq_tail = VTD_IQT_QT(s->iq_dw, val);
2529     trace_vtd_inv_qi_tail(s->iq_tail);
2530 
2531     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2532         /* Process Invalidation Queue here */
2533         vtd_fetch_inv_desc(s);
2534     }
2535 }
2536 
2537 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2538 {
2539     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2540     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2541     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2542 
2543     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2544         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2545         trace_vtd_fsts_clear_ip();
2546     }
2547     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2548      * Descriptors if there are any when Queued Invalidation is enabled?
2549      */
2550 }
2551 
2552 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2553 {
2554     uint32_t fectl_reg;
2555     /* FIXME: when software clears the IM field, check the IP field. But do we
2556      * need to compare the old value and the new value to conclude that
2557      * software clears the IM field? Or just check if the IM field is zero?
2558      */
2559     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2560 
2561     trace_vtd_reg_write_fectl(fectl_reg);
2562 
2563     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2564         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2565         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2566     }
2567 }
2568 
2569 static void vtd_handle_ics_write(IntelIOMMUState *s)
2570 {
2571     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2572     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2573 
2574     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2575         trace_vtd_reg_ics_clear_ip();
2576         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2577     }
2578 }
2579 
2580 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2581 {
2582     uint32_t iectl_reg;
2583     /* FIXME: when software clears the IM field, check the IP field. But do we
2584      * need to compare the old value and the new value to conclude that
2585      * software clears the IM field? Or just check if the IM field is zero?
2586      */
2587     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2588 
2589     trace_vtd_reg_write_iectl(iectl_reg);
2590 
2591     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2592         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2593         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2594     }
2595 }
2596 
2597 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2598 {
2599     IntelIOMMUState *s = opaque;
2600     uint64_t val;
2601 
2602     trace_vtd_reg_read(addr, size);
2603 
2604     if (addr + size > DMAR_REG_SIZE) {
2605         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2606                           " size=0x%u", __func__, addr, size);
2607         return (uint64_t)-1;
2608     }
2609 
2610     switch (addr) {
2611     /* Root Table Address Register, 64-bit */
2612     case DMAR_RTADDR_REG:
2613         if (size == 4) {
2614             val = s->root & ((1ULL << 32) - 1);
2615         } else {
2616             val = s->root;
2617         }
2618         break;
2619 
2620     case DMAR_RTADDR_REG_HI:
2621         assert(size == 4);
2622         val = s->root >> 32;
2623         break;
2624 
2625     /* Invalidation Queue Address Register, 64-bit */
2626     case DMAR_IQA_REG:
2627         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2628         if (size == 4) {
2629             val = val & ((1ULL << 32) - 1);
2630         }
2631         break;
2632 
2633     case DMAR_IQA_REG_HI:
2634         assert(size == 4);
2635         val = s->iq >> 32;
2636         break;
2637 
2638     default:
2639         if (size == 4) {
2640             val = vtd_get_long(s, addr);
2641         } else {
2642             val = vtd_get_quad(s, addr);
2643         }
2644     }
2645 
2646     return val;
2647 }
2648 
2649 static void vtd_mem_write(void *opaque, hwaddr addr,
2650                           uint64_t val, unsigned size)
2651 {
2652     IntelIOMMUState *s = opaque;
2653 
2654     trace_vtd_reg_write(addr, size, val);
2655 
2656     if (addr + size > DMAR_REG_SIZE) {
2657         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2658                           " size=0x%u", __func__, addr, size);
2659         return;
2660     }
2661 
2662     switch (addr) {
2663     /* Global Command Register, 32-bit */
2664     case DMAR_GCMD_REG:
2665         vtd_set_long(s, addr, val);
2666         vtd_handle_gcmd_write(s);
2667         break;
2668 
2669     /* Context Command Register, 64-bit */
2670     case DMAR_CCMD_REG:
2671         if (size == 4) {
2672             vtd_set_long(s, addr, val);
2673         } else {
2674             vtd_set_quad(s, addr, val);
2675             vtd_handle_ccmd_write(s);
2676         }
2677         break;
2678 
2679     case DMAR_CCMD_REG_HI:
2680         assert(size == 4);
2681         vtd_set_long(s, addr, val);
2682         vtd_handle_ccmd_write(s);
2683         break;
2684 
2685     /* IOTLB Invalidation Register, 64-bit */
2686     case DMAR_IOTLB_REG:
2687         if (size == 4) {
2688             vtd_set_long(s, addr, val);
2689         } else {
2690             vtd_set_quad(s, addr, val);
2691             vtd_handle_iotlb_write(s);
2692         }
2693         break;
2694 
2695     case DMAR_IOTLB_REG_HI:
2696         assert(size == 4);
2697         vtd_set_long(s, addr, val);
2698         vtd_handle_iotlb_write(s);
2699         break;
2700 
2701     /* Invalidate Address Register, 64-bit */
2702     case DMAR_IVA_REG:
2703         if (size == 4) {
2704             vtd_set_long(s, addr, val);
2705         } else {
2706             vtd_set_quad(s, addr, val);
2707         }
2708         break;
2709 
2710     case DMAR_IVA_REG_HI:
2711         assert(size == 4);
2712         vtd_set_long(s, addr, val);
2713         break;
2714 
2715     /* Fault Status Register, 32-bit */
2716     case DMAR_FSTS_REG:
2717         assert(size == 4);
2718         vtd_set_long(s, addr, val);
2719         vtd_handle_fsts_write(s);
2720         break;
2721 
2722     /* Fault Event Control Register, 32-bit */
2723     case DMAR_FECTL_REG:
2724         assert(size == 4);
2725         vtd_set_long(s, addr, val);
2726         vtd_handle_fectl_write(s);
2727         break;
2728 
2729     /* Fault Event Data Register, 32-bit */
2730     case DMAR_FEDATA_REG:
2731         assert(size == 4);
2732         vtd_set_long(s, addr, val);
2733         break;
2734 
2735     /* Fault Event Address Register, 32-bit */
2736     case DMAR_FEADDR_REG:
2737         if (size == 4) {
2738             vtd_set_long(s, addr, val);
2739         } else {
2740             /*
2741              * While the register is 32-bit only, some guests (Xen...) write to
2742              * it with 64-bit.
2743              */
2744             vtd_set_quad(s, addr, val);
2745         }
2746         break;
2747 
2748     /* Fault Event Upper Address Register, 32-bit */
2749     case DMAR_FEUADDR_REG:
2750         assert(size == 4);
2751         vtd_set_long(s, addr, val);
2752         break;
2753 
2754     /* Protected Memory Enable Register, 32-bit */
2755     case DMAR_PMEN_REG:
2756         assert(size == 4);
2757         vtd_set_long(s, addr, val);
2758         break;
2759 
2760     /* Root Table Address Register, 64-bit */
2761     case DMAR_RTADDR_REG:
2762         if (size == 4) {
2763             vtd_set_long(s, addr, val);
2764         } else {
2765             vtd_set_quad(s, addr, val);
2766         }
2767         break;
2768 
2769     case DMAR_RTADDR_REG_HI:
2770         assert(size == 4);
2771         vtd_set_long(s, addr, val);
2772         break;
2773 
2774     /* Invalidation Queue Tail Register, 64-bit */
2775     case DMAR_IQT_REG:
2776         if (size == 4) {
2777             vtd_set_long(s, addr, val);
2778         } else {
2779             vtd_set_quad(s, addr, val);
2780         }
2781         vtd_handle_iqt_write(s);
2782         break;
2783 
2784     case DMAR_IQT_REG_HI:
2785         assert(size == 4);
2786         vtd_set_long(s, addr, val);
2787         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
2788         break;
2789 
2790     /* Invalidation Queue Address Register, 64-bit */
2791     case DMAR_IQA_REG:
2792         if (size == 4) {
2793             vtd_set_long(s, addr, val);
2794         } else {
2795             vtd_set_quad(s, addr, val);
2796         }
2797         if (s->ecap & VTD_ECAP_SMTS &&
2798             val & VTD_IQA_DW_MASK) {
2799             s->iq_dw = true;
2800         } else {
2801             s->iq_dw = false;
2802         }
2803         break;
2804 
2805     case DMAR_IQA_REG_HI:
2806         assert(size == 4);
2807         vtd_set_long(s, addr, val);
2808         break;
2809 
2810     /* Invalidation Completion Status Register, 32-bit */
2811     case DMAR_ICS_REG:
2812         assert(size == 4);
2813         vtd_set_long(s, addr, val);
2814         vtd_handle_ics_write(s);
2815         break;
2816 
2817     /* Invalidation Event Control Register, 32-bit */
2818     case DMAR_IECTL_REG:
2819         assert(size == 4);
2820         vtd_set_long(s, addr, val);
2821         vtd_handle_iectl_write(s);
2822         break;
2823 
2824     /* Invalidation Event Data Register, 32-bit */
2825     case DMAR_IEDATA_REG:
2826         assert(size == 4);
2827         vtd_set_long(s, addr, val);
2828         break;
2829 
2830     /* Invalidation Event Address Register, 32-bit */
2831     case DMAR_IEADDR_REG:
2832         assert(size == 4);
2833         vtd_set_long(s, addr, val);
2834         break;
2835 
2836     /* Invalidation Event Upper Address Register, 32-bit */
2837     case DMAR_IEUADDR_REG:
2838         assert(size == 4);
2839         vtd_set_long(s, addr, val);
2840         break;
2841 
2842     /* Fault Recording Registers, 128-bit */
2843     case DMAR_FRCD_REG_0_0:
2844         if (size == 4) {
2845             vtd_set_long(s, addr, val);
2846         } else {
2847             vtd_set_quad(s, addr, val);
2848         }
2849         break;
2850 
2851     case DMAR_FRCD_REG_0_1:
2852         assert(size == 4);
2853         vtd_set_long(s, addr, val);
2854         break;
2855 
2856     case DMAR_FRCD_REG_0_2:
2857         if (size == 4) {
2858             vtd_set_long(s, addr, val);
2859         } else {
2860             vtd_set_quad(s, addr, val);
2861             /* May clear bit 127 (Fault), update PPF */
2862             vtd_update_fsts_ppf(s);
2863         }
2864         break;
2865 
2866     case DMAR_FRCD_REG_0_3:
2867         assert(size == 4);
2868         vtd_set_long(s, addr, val);
2869         /* May clear bit 127 (Fault), update PPF */
2870         vtd_update_fsts_ppf(s);
2871         break;
2872 
2873     case DMAR_IRTA_REG:
2874         if (size == 4) {
2875             vtd_set_long(s, addr, val);
2876         } else {
2877             vtd_set_quad(s, addr, val);
2878         }
2879         break;
2880 
2881     case DMAR_IRTA_REG_HI:
2882         assert(size == 4);
2883         vtd_set_long(s, addr, val);
2884         break;
2885 
2886     default:
2887         if (size == 4) {
2888             vtd_set_long(s, addr, val);
2889         } else {
2890             vtd_set_quad(s, addr, val);
2891         }
2892     }
2893 }
2894 
2895 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
2896                                          IOMMUAccessFlags flag, int iommu_idx)
2897 {
2898     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2899     IntelIOMMUState *s = vtd_as->iommu_state;
2900     IOMMUTLBEntry iotlb = {
2901         /* We'll fill in the rest later. */
2902         .target_as = &address_space_memory,
2903     };
2904     bool success;
2905 
2906     if (likely(s->dmar_enabled)) {
2907         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
2908                                          addr, flag & IOMMU_WO, &iotlb);
2909     } else {
2910         /* DMAR disabled, passthrough, use 4k-page*/
2911         iotlb.iova = addr & VTD_PAGE_MASK_4K;
2912         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
2913         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
2914         iotlb.perm = IOMMU_RW;
2915         success = true;
2916     }
2917 
2918     if (likely(success)) {
2919         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
2920                                  VTD_PCI_SLOT(vtd_as->devfn),
2921                                  VTD_PCI_FUNC(vtd_as->devfn),
2922                                  iotlb.iova, iotlb.translated_addr,
2923                                  iotlb.addr_mask);
2924     } else {
2925         error_report_once("%s: detected translation failure "
2926                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
2927                           __func__, pci_bus_num(vtd_as->bus),
2928                           VTD_PCI_SLOT(vtd_as->devfn),
2929                           VTD_PCI_FUNC(vtd_as->devfn),
2930                           addr);
2931     }
2932 
2933     return iotlb;
2934 }
2935 
2936 static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
2937                                          IOMMUNotifierFlag old,
2938                                          IOMMUNotifierFlag new,
2939                                          Error **errp)
2940 {
2941     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2942     IntelIOMMUState *s = vtd_as->iommu_state;
2943 
2944     /* Update per-address-space notifier flags */
2945     vtd_as->notifier_flags = new;
2946 
2947     if (old == IOMMU_NOTIFIER_NONE) {
2948         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
2949     } else if (new == IOMMU_NOTIFIER_NONE) {
2950         QLIST_REMOVE(vtd_as, next);
2951     }
2952     return 0;
2953 }
2954 
2955 static int vtd_post_load(void *opaque, int version_id)
2956 {
2957     IntelIOMMUState *iommu = opaque;
2958 
2959     /*
2960      * Memory regions are dynamically turned on/off depending on
2961      * context entry configurations from the guest. After migration,
2962      * we need to make sure the memory regions are still correct.
2963      */
2964     vtd_switch_address_space_all(iommu);
2965 
2966     /*
2967      * We don't need to migrate the root_scalable because we can
2968      * simply do the calculation after the loading is complete.  We
2969      * can actually do similar things with root, dmar_enabled, etc.
2970      * however since we've had them already so we'd better keep them
2971      * for compatibility of migration.
2972      */
2973     vtd_update_scalable_state(iommu);
2974 
2975     return 0;
2976 }
2977 
2978 static const VMStateDescription vtd_vmstate = {
2979     .name = "iommu-intel",
2980     .version_id = 1,
2981     .minimum_version_id = 1,
2982     .priority = MIG_PRI_IOMMU,
2983     .post_load = vtd_post_load,
2984     .fields = (VMStateField[]) {
2985         VMSTATE_UINT64(root, IntelIOMMUState),
2986         VMSTATE_UINT64(intr_root, IntelIOMMUState),
2987         VMSTATE_UINT64(iq, IntelIOMMUState),
2988         VMSTATE_UINT32(intr_size, IntelIOMMUState),
2989         VMSTATE_UINT16(iq_head, IntelIOMMUState),
2990         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
2991         VMSTATE_UINT16(iq_size, IntelIOMMUState),
2992         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
2993         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
2994         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
2995         VMSTATE_UNUSED(1),      /* bool root_extended is obsolete by VT-d */
2996         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
2997         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
2998         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
2999         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
3000         VMSTATE_END_OF_LIST()
3001     }
3002 };
3003 
3004 static const MemoryRegionOps vtd_mem_ops = {
3005     .read = vtd_mem_read,
3006     .write = vtd_mem_write,
3007     .endianness = DEVICE_LITTLE_ENDIAN,
3008     .impl = {
3009         .min_access_size = 4,
3010         .max_access_size = 8,
3011     },
3012     .valid = {
3013         .min_access_size = 4,
3014         .max_access_size = 8,
3015     },
3016 };
3017 
3018 static Property vtd_properties[] = {
3019     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
3020     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
3021                             ON_OFF_AUTO_AUTO),
3022     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
3023     DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits,
3024                       VTD_HOST_ADDRESS_WIDTH),
3025     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
3026     DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE),
3027     DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true),
3028     DEFINE_PROP_END_OF_LIST(),
3029 };
3030 
3031 /* Read IRTE entry with specific index */
3032 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
3033                         VTD_IR_TableEntry *entry, uint16_t sid)
3034 {
3035     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
3036         {0xffff, 0xfffb, 0xfff9, 0xfff8};
3037     dma_addr_t addr = 0x00;
3038     uint16_t mask, source_id;
3039     uint8_t bus, bus_max, bus_min;
3040 
3041     addr = iommu->intr_root + index * sizeof(*entry);
3042     if (dma_memory_read(&address_space_memory, addr, entry,
3043                         sizeof(*entry))) {
3044         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
3045                           __func__, index, addr);
3046         return -VTD_FR_IR_ROOT_INVAL;
3047     }
3048 
3049     trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]),
3050                           le64_to_cpu(entry->data[0]));
3051 
3052     if (!entry->irte.present) {
3053         error_report_once("%s: detected non-present IRTE "
3054                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3055                           __func__, index, le64_to_cpu(entry->data[1]),
3056                           le64_to_cpu(entry->data[0]));
3057         return -VTD_FR_IR_ENTRY_P;
3058     }
3059 
3060     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
3061         entry->irte.__reserved_2) {
3062         error_report_once("%s: detected non-zero reserved IRTE "
3063                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3064                           __func__, index, le64_to_cpu(entry->data[1]),
3065                           le64_to_cpu(entry->data[0]));
3066         return -VTD_FR_IR_IRTE_RSVD;
3067     }
3068 
3069     if (sid != X86_IOMMU_SID_INVALID) {
3070         /* Validate IRTE SID */
3071         source_id = le32_to_cpu(entry->irte.source_id);
3072         switch (entry->irte.sid_vtype) {
3073         case VTD_SVT_NONE:
3074             break;
3075 
3076         case VTD_SVT_ALL:
3077             mask = vtd_svt_mask[entry->irte.sid_q];
3078             if ((source_id & mask) != (sid & mask)) {
3079                 error_report_once("%s: invalid IRTE SID "
3080                                   "(index=%u, sid=%u, source_id=%u)",
3081                                   __func__, index, sid, source_id);
3082                 return -VTD_FR_IR_SID_ERR;
3083             }
3084             break;
3085 
3086         case VTD_SVT_BUS:
3087             bus_max = source_id >> 8;
3088             bus_min = source_id & 0xff;
3089             bus = sid >> 8;
3090             if (bus > bus_max || bus < bus_min) {
3091                 error_report_once("%s: invalid SVT_BUS "
3092                                   "(index=%u, bus=%u, min=%u, max=%u)",
3093                                   __func__, index, bus, bus_min, bus_max);
3094                 return -VTD_FR_IR_SID_ERR;
3095             }
3096             break;
3097 
3098         default:
3099             error_report_once("%s: detected invalid IRTE SVT "
3100                               "(index=%u, type=%d)", __func__,
3101                               index, entry->irte.sid_vtype);
3102             /* Take this as verification failure. */
3103             return -VTD_FR_IR_SID_ERR;
3104             break;
3105         }
3106     }
3107 
3108     return 0;
3109 }
3110 
3111 /* Fetch IRQ information of specific IR index */
3112 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
3113                              X86IOMMUIrq *irq, uint16_t sid)
3114 {
3115     VTD_IR_TableEntry irte = {};
3116     int ret = 0;
3117 
3118     ret = vtd_irte_get(iommu, index, &irte, sid);
3119     if (ret) {
3120         return ret;
3121     }
3122 
3123     irq->trigger_mode = irte.irte.trigger_mode;
3124     irq->vector = irte.irte.vector;
3125     irq->delivery_mode = irte.irte.delivery_mode;
3126     irq->dest = le32_to_cpu(irte.irte.dest_id);
3127     if (!iommu->intr_eime) {
3128 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
3129 #define  VTD_IR_APIC_DEST_SHIFT        (8)
3130         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
3131             VTD_IR_APIC_DEST_SHIFT;
3132     }
3133     irq->dest_mode = irte.irte.dest_mode;
3134     irq->redir_hint = irte.irte.redir_hint;
3135 
3136     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
3137                        irq->delivery_mode, irq->dest, irq->dest_mode);
3138 
3139     return 0;
3140 }
3141 
3142 /* Interrupt remapping for MSI/MSI-X entry */
3143 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
3144                                    MSIMessage *origin,
3145                                    MSIMessage *translated,
3146                                    uint16_t sid)
3147 {
3148     int ret = 0;
3149     VTD_IR_MSIAddress addr;
3150     uint16_t index;
3151     X86IOMMUIrq irq = {};
3152 
3153     assert(origin && translated);
3154 
3155     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
3156 
3157     if (!iommu || !iommu->intr_enabled) {
3158         memcpy(translated, origin, sizeof(*origin));
3159         goto out;
3160     }
3161 
3162     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
3163         error_report_once("%s: MSI address high 32 bits non-zero detected: "
3164                           "address=0x%" PRIx64, __func__, origin->address);
3165         return -VTD_FR_IR_REQ_RSVD;
3166     }
3167 
3168     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
3169     if (addr.addr.__head != 0xfee) {
3170         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
3171                           __func__, addr.data);
3172         return -VTD_FR_IR_REQ_RSVD;
3173     }
3174 
3175     /* This is compatible mode. */
3176     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
3177         memcpy(translated, origin, sizeof(*origin));
3178         goto out;
3179     }
3180 
3181     index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l);
3182 
3183 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
3184 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
3185 
3186     if (addr.addr.sub_valid) {
3187         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3188         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
3189     }
3190 
3191     ret = vtd_remap_irq_get(iommu, index, &irq, sid);
3192     if (ret) {
3193         return ret;
3194     }
3195 
3196     if (addr.addr.sub_valid) {
3197         trace_vtd_ir_remap_type("MSI");
3198         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
3199             error_report_once("%s: invalid IR MSI "
3200                               "(sid=%u, address=0x%" PRIx64
3201                               ", data=0x%" PRIx32 ")",
3202                               __func__, sid, origin->address, origin->data);
3203             return -VTD_FR_IR_REQ_RSVD;
3204         }
3205     } else {
3206         uint8_t vector = origin->data & 0xff;
3207         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
3208 
3209         trace_vtd_ir_remap_type("IOAPIC");
3210         /* IOAPIC entry vector should be aligned with IRTE vector
3211          * (see vt-d spec 5.1.5.1). */
3212         if (vector != irq.vector) {
3213             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
3214         }
3215 
3216         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3217          * (see vt-d spec 5.1.5.1). */
3218         if (trigger_mode != irq.trigger_mode) {
3219             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
3220                                       irq.trigger_mode);
3221         }
3222     }
3223 
3224     /*
3225      * We'd better keep the last two bits, assuming that guest OS
3226      * might modify it. Keep it does not hurt after all.
3227      */
3228     irq.msi_addr_last_bits = addr.addr.__not_care;
3229 
3230     /* Translate X86IOMMUIrq to MSI message */
3231     x86_iommu_irq_to_msi_message(&irq, translated);
3232 
3233 out:
3234     trace_vtd_ir_remap_msi(origin->address, origin->data,
3235                            translated->address, translated->data);
3236     return 0;
3237 }
3238 
3239 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
3240                          MSIMessage *dst, uint16_t sid)
3241 {
3242     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
3243                                    src, dst, sid);
3244 }
3245 
3246 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
3247                                    uint64_t *data, unsigned size,
3248                                    MemTxAttrs attrs)
3249 {
3250     return MEMTX_OK;
3251 }
3252 
3253 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
3254                                     uint64_t value, unsigned size,
3255                                     MemTxAttrs attrs)
3256 {
3257     int ret = 0;
3258     MSIMessage from = {}, to = {};
3259     uint16_t sid = X86_IOMMU_SID_INVALID;
3260 
3261     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
3262     from.data = (uint32_t) value;
3263 
3264     if (!attrs.unspecified) {
3265         /* We have explicit Source ID */
3266         sid = attrs.requester_id;
3267     }
3268 
3269     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid);
3270     if (ret) {
3271         /* TODO: report error */
3272         /* Drop this interrupt */
3273         return MEMTX_ERROR;
3274     }
3275 
3276     apic_get_class()->send_msi(&to);
3277 
3278     return MEMTX_OK;
3279 }
3280 
3281 static const MemoryRegionOps vtd_mem_ir_ops = {
3282     .read_with_attrs = vtd_mem_ir_read,
3283     .write_with_attrs = vtd_mem_ir_write,
3284     .endianness = DEVICE_LITTLE_ENDIAN,
3285     .impl = {
3286         .min_access_size = 4,
3287         .max_access_size = 4,
3288     },
3289     .valid = {
3290         .min_access_size = 4,
3291         .max_access_size = 4,
3292     },
3293 };
3294 
3295 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
3296 {
3297     uintptr_t key = (uintptr_t)bus;
3298     VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
3299     VTDAddressSpace *vtd_dev_as;
3300     char name[128];
3301 
3302     if (!vtd_bus) {
3303         uintptr_t *new_key = g_malloc(sizeof(*new_key));
3304         *new_key = (uintptr_t)bus;
3305         /* No corresponding free() */
3306         vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
3307                             PCI_DEVFN_MAX);
3308         vtd_bus->bus = bus;
3309         g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
3310     }
3311 
3312     vtd_dev_as = vtd_bus->dev_as[devfn];
3313 
3314     if (!vtd_dev_as) {
3315         snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn),
3316                  PCI_FUNC(devfn));
3317         vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace));
3318 
3319         vtd_dev_as->bus = bus;
3320         vtd_dev_as->devfn = (uint8_t)devfn;
3321         vtd_dev_as->iommu_state = s;
3322         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
3323         vtd_dev_as->iova_tree = iova_tree_new();
3324 
3325         memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX);
3326         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root");
3327 
3328         /*
3329          * Build the DMAR-disabled container with aliases to the
3330          * shared MRs.  Note that aliasing to a shared memory region
3331          * could help the memory API to detect same FlatViews so we
3332          * can have devices to share the same FlatView when DMAR is
3333          * disabled (either by not providing "intel_iommu=on" or with
3334          * "iommu=pt").  It will greatly reduce the total number of
3335          * FlatViews of the system hence VM runs faster.
3336          */
3337         memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s),
3338                                  "vtd-nodmar", &s->mr_nodmar, 0,
3339                                  memory_region_size(&s->mr_nodmar));
3340 
3341         /*
3342          * Build the per-device DMAR-enabled container.
3343          *
3344          * TODO: currently we have per-device IOMMU memory region only
3345          * because we have per-device IOMMU notifiers for devices.  If
3346          * one day we can abstract the IOMMU notifiers out of the
3347          * memory regions then we can also share the same memory
3348          * region here just like what we've done above with the nodmar
3349          * region.
3350          */
3351         strcat(name, "-dmar");
3352         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
3353                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
3354                                  name, UINT64_MAX);
3355         memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir",
3356                                  &s->mr_ir, 0, memory_region_size(&s->mr_ir));
3357         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
3358                                             VTD_INTERRUPT_ADDR_FIRST,
3359                                             &vtd_dev_as->iommu_ir, 1);
3360 
3361         /*
3362          * Hook both the containers under the root container, we
3363          * switch between DMAR & noDMAR by enable/disable
3364          * corresponding sub-containers
3365          */
3366         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3367                                             MEMORY_REGION(&vtd_dev_as->iommu),
3368                                             0);
3369         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3370                                             &vtd_dev_as->nodmar, 0);
3371 
3372         vtd_switch_address_space(vtd_dev_as);
3373     }
3374     return vtd_dev_as;
3375 }
3376 
3377 static uint64_t get_naturally_aligned_size(uint64_t start,
3378                                            uint64_t size, int gaw)
3379 {
3380     uint64_t max_mask = 1ULL << gaw;
3381     uint64_t alignment = start ? start & -start : max_mask;
3382 
3383     alignment = MIN(alignment, max_mask);
3384     size = MIN(size, max_mask);
3385 
3386     if (alignment <= size) {
3387         /* Increase the alignment of start */
3388         return alignment;
3389     } else {
3390         /* Find the largest page mask from size */
3391         return 1ULL << (63 - clz64(size));
3392     }
3393 }
3394 
3395 /* Unmap the whole range in the notifier's scope. */
3396 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
3397 {
3398     hwaddr size, remain;
3399     hwaddr start = n->start;
3400     hwaddr end = n->end;
3401     IntelIOMMUState *s = as->iommu_state;
3402     DMAMap map;
3403 
3404     /*
3405      * Note: all the codes in this function has a assumption that IOVA
3406      * bits are no more than VTD_MGAW bits (which is restricted by
3407      * VT-d spec), otherwise we need to consider overflow of 64 bits.
3408      */
3409 
3410     if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) {
3411         /*
3412          * Don't need to unmap regions that is bigger than the whole
3413          * VT-d supported address space size
3414          */
3415         end = VTD_ADDRESS_SIZE(s->aw_bits) - 1;
3416     }
3417 
3418     assert(start <= end);
3419     size = remain = end - start + 1;
3420 
3421     while (remain >= VTD_PAGE_SIZE) {
3422         IOMMUTLBEntry entry;
3423         uint64_t mask = get_naturally_aligned_size(start, remain, s->aw_bits);
3424 
3425         assert(mask);
3426 
3427         entry.iova = start;
3428         entry.addr_mask = mask - 1;
3429         entry.target_as = &address_space_memory;
3430         entry.perm = IOMMU_NONE;
3431         /* This field is meaningless for unmap */
3432         entry.translated_addr = 0;
3433 
3434         memory_region_notify_one(n, &entry);
3435 
3436         start += mask;
3437         remain -= mask;
3438     }
3439 
3440     assert(!remain);
3441 
3442     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3443                              VTD_PCI_SLOT(as->devfn),
3444                              VTD_PCI_FUNC(as->devfn),
3445                              n->start, size);
3446 
3447     map.iova = n->start;
3448     map.size = size;
3449     iova_tree_remove(as->iova_tree, &map);
3450 }
3451 
3452 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3453 {
3454     VTDAddressSpace *vtd_as;
3455     IOMMUNotifier *n;
3456 
3457     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3458         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3459             vtd_address_space_unmap(vtd_as, n);
3460         }
3461     }
3462 }
3463 
3464 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3465 {
3466     vtd_address_space_unmap_all(s);
3467     vtd_switch_address_space_all(s);
3468 }
3469 
3470 static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private)
3471 {
3472     memory_region_notify_one((IOMMUNotifier *)private, entry);
3473     return 0;
3474 }
3475 
3476 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3477 {
3478     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3479     IntelIOMMUState *s = vtd_as->iommu_state;
3480     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3481     VTDContextEntry ce;
3482 
3483     /*
3484      * The replay can be triggered by either a invalidation or a newly
3485      * created entry. No matter what, we release existing mappings
3486      * (it means flushing caches for UNMAP-only registers).
3487      */
3488     vtd_address_space_unmap(vtd_as, n);
3489 
3490     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3491         trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" :
3492                                   "legacy mode",
3493                                   bus_n, PCI_SLOT(vtd_as->devfn),
3494                                   PCI_FUNC(vtd_as->devfn),
3495                                   vtd_get_domain_id(s, &ce),
3496                                   ce.hi, ce.lo);
3497         if (vtd_as_has_map_notifier(vtd_as)) {
3498             /* This is required only for MAP typed notifiers */
3499             vtd_page_walk_info info = {
3500                 .hook_fn = vtd_replay_hook,
3501                 .private = (void *)n,
3502                 .notify_unmap = false,
3503                 .aw = s->aw_bits,
3504                 .as = vtd_as,
3505                 .domain_id = vtd_get_domain_id(s, &ce),
3506             };
3507 
3508             vtd_page_walk(s, &ce, 0, ~0ULL, &info);
3509         }
3510     } else {
3511         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3512                                     PCI_FUNC(vtd_as->devfn));
3513     }
3514 
3515     return;
3516 }
3517 
3518 /* Do the initialization. It will also be called when reset, so pay
3519  * attention when adding new initialization stuff.
3520  */
3521 static void vtd_init(IntelIOMMUState *s)
3522 {
3523     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3524 
3525     memset(s->csr, 0, DMAR_REG_SIZE);
3526     memset(s->wmask, 0, DMAR_REG_SIZE);
3527     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3528     memset(s->womask, 0, DMAR_REG_SIZE);
3529 
3530     s->root = 0;
3531     s->root_scalable = false;
3532     s->dmar_enabled = false;
3533     s->intr_enabled = false;
3534     s->iq_head = 0;
3535     s->iq_tail = 0;
3536     s->iq = 0;
3537     s->iq_size = 0;
3538     s->qi_enabled = false;
3539     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3540     s->iq_dw = false;
3541     s->next_frcd_reg = 0;
3542     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3543              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3544              VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits);
3545     if (s->dma_drain) {
3546         s->cap |= VTD_CAP_DRAIN;
3547     }
3548     if (s->aw_bits == VTD_HOST_AW_48BIT) {
3549         s->cap |= VTD_CAP_SAGAW_48bit;
3550     }
3551     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3552 
3553     /*
3554      * Rsvd field masks for spte
3555      */
3556     vtd_spte_rsvd[0] = ~0ULL;
3557     vtd_spte_rsvd[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits,
3558                                                   x86_iommu->dt_supported);
3559     vtd_spte_rsvd[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3560     vtd_spte_rsvd[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3561     vtd_spte_rsvd[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3562 
3563     vtd_spte_rsvd_large[2] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits,
3564                                                          x86_iommu->dt_supported);
3565     vtd_spte_rsvd_large[3] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits,
3566                                                          x86_iommu->dt_supported);
3567 
3568     if (x86_iommu_ir_supported(x86_iommu)) {
3569         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3570         if (s->intr_eim == ON_OFF_AUTO_ON) {
3571             s->ecap |= VTD_ECAP_EIM;
3572         }
3573         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3574     }
3575 
3576     if (x86_iommu->dt_supported) {
3577         s->ecap |= VTD_ECAP_DT;
3578     }
3579 
3580     if (x86_iommu->pt_supported) {
3581         s->ecap |= VTD_ECAP_PT;
3582     }
3583 
3584     if (s->caching_mode) {
3585         s->cap |= VTD_CAP_CM;
3586     }
3587 
3588     /* TODO: read cap/ecap from host to decide which cap to be exposed. */
3589     if (s->scalable_mode) {
3590         s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS;
3591     }
3592 
3593     vtd_reset_caches(s);
3594 
3595     /* Define registers with default values and bit semantics */
3596     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
3597     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
3598     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
3599     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
3600     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
3601     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
3602     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0);
3603     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
3604     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
3605 
3606     /* Advanced Fault Logging not supported */
3607     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
3608     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3609     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
3610     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
3611 
3612     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3613      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3614      */
3615     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
3616 
3617     /* Treated as RO for implementations that PLMR and PHMR fields reported
3618      * as Clear in the CAP_REG.
3619      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3620      */
3621     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
3622 
3623     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
3624     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
3625     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0);
3626     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
3627     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3628     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
3629     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
3630     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3631     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
3632 
3633     /* IOTLB registers */
3634     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
3635     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
3636     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
3637 
3638     /* Fault Recording Registers, 128-bit */
3639     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
3640     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
3641 
3642     /*
3643      * Interrupt remapping registers.
3644      */
3645     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
3646 }
3647 
3648 /* Should not reset address_spaces when reset because devices will still use
3649  * the address space they got at first (won't ask the bus again).
3650  */
3651 static void vtd_reset(DeviceState *dev)
3652 {
3653     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3654 
3655     vtd_init(s);
3656     vtd_address_space_refresh_all(s);
3657 }
3658 
3659 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
3660 {
3661     IntelIOMMUState *s = opaque;
3662     VTDAddressSpace *vtd_as;
3663 
3664     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
3665 
3666     vtd_as = vtd_find_add_as(s, bus, devfn);
3667     return &vtd_as->as;
3668 }
3669 
3670 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
3671 {
3672     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3673 
3674     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) {
3675         error_setg(errp, "eim=on cannot be selected without intremap=on");
3676         return false;
3677     }
3678 
3679     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
3680         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
3681                       && x86_iommu_ir_supported(x86_iommu) ?
3682                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
3683     }
3684     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
3685         if (!kvm_irqchip_in_kernel()) {
3686             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
3687             return false;
3688         }
3689         if (!kvm_enable_x2apic()) {
3690             error_setg(errp, "eim=on requires support on the KVM side"
3691                              "(X2APIC_API, first shipped in v4.7)");
3692             return false;
3693         }
3694     }
3695 
3696     /* Currently only address widths supported are 39 and 48 bits */
3697     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
3698         (s->aw_bits != VTD_HOST_AW_48BIT)) {
3699         error_setg(errp, "Supported values for x-aw-bits are: %d, %d",
3700                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
3701         return false;
3702     }
3703 
3704     if (s->scalable_mode && !s->dma_drain) {
3705         error_setg(errp, "Need to set dma_drain for scalable mode");
3706         return false;
3707     }
3708 
3709     return true;
3710 }
3711 
3712 static int vtd_machine_done_notify_one(Object *child, void *unused)
3713 {
3714     IntelIOMMUState *iommu = INTEL_IOMMU_DEVICE(x86_iommu_get_default());
3715 
3716     /*
3717      * We hard-coded here because vfio-pci is the only special case
3718      * here.  Let's be more elegant in the future when we can, but so
3719      * far there seems to be no better way.
3720      */
3721     if (object_dynamic_cast(child, "vfio-pci") && !iommu->caching_mode) {
3722         vtd_panic_require_caching_mode();
3723     }
3724 
3725     return 0;
3726 }
3727 
3728 static void vtd_machine_done_hook(Notifier *notifier, void *unused)
3729 {
3730     object_child_foreach_recursive(object_get_root(),
3731                                    vtd_machine_done_notify_one, NULL);
3732 }
3733 
3734 static Notifier vtd_machine_done_notify = {
3735     .notify = vtd_machine_done_hook,
3736 };
3737 
3738 static void vtd_realize(DeviceState *dev, Error **errp)
3739 {
3740     MachineState *ms = MACHINE(qdev_get_machine());
3741     PCMachineState *pcms = PC_MACHINE(ms);
3742     X86MachineState *x86ms = X86_MACHINE(ms);
3743     PCIBus *bus = pcms->bus;
3744     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3745     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
3746 
3747     x86_iommu->type = TYPE_INTEL;
3748 
3749     if (!vtd_decide_config(s, errp)) {
3750         return;
3751     }
3752 
3753     QLIST_INIT(&s->vtd_as_with_notifiers);
3754     qemu_mutex_init(&s->iommu_lock);
3755     memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num));
3756     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
3757                           "intel_iommu", DMAR_REG_SIZE);
3758 
3759     /* Create the shared memory regions by all devices */
3760     memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar",
3761                        UINT64_MAX);
3762     memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops,
3763                           s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE);
3764     memory_region_init_alias(&s->mr_sys_alias, OBJECT(s),
3765                              "vtd-sys-alias", get_system_memory(), 0,
3766                              memory_region_size(get_system_memory()));
3767     memory_region_add_subregion_overlap(&s->mr_nodmar, 0,
3768                                         &s->mr_sys_alias, 0);
3769     memory_region_add_subregion_overlap(&s->mr_nodmar,
3770                                         VTD_INTERRUPT_ADDR_FIRST,
3771                                         &s->mr_ir, 1);
3772 
3773     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
3774     /* No corresponding destroy */
3775     s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3776                                      g_free, g_free);
3777     s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3778                                               g_free, g_free);
3779     vtd_init(s);
3780     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
3781     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
3782     /* Pseudo address space under root PCI bus. */
3783     x86ms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
3784     qemu_add_machine_init_done_notifier(&vtd_machine_done_notify);
3785 }
3786 
3787 static void vtd_class_init(ObjectClass *klass, void *data)
3788 {
3789     DeviceClass *dc = DEVICE_CLASS(klass);
3790     X86IOMMUClass *x86_class = X86_IOMMU_CLASS(klass);
3791 
3792     dc->reset = vtd_reset;
3793     dc->vmsd = &vtd_vmstate;
3794     dc->props = vtd_properties;
3795     dc->hotpluggable = false;
3796     x86_class->realize = vtd_realize;
3797     x86_class->int_remap = vtd_int_remap;
3798     /* Supported by the pc-q35-* machine types */
3799     dc->user_creatable = true;
3800     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
3801     dc->desc = "Intel IOMMU (VT-d) DMA Remapping device";
3802 }
3803 
3804 static const TypeInfo vtd_info = {
3805     .name          = TYPE_INTEL_IOMMU_DEVICE,
3806     .parent        = TYPE_X86_IOMMU_DEVICE,
3807     .instance_size = sizeof(IntelIOMMUState),
3808     .class_init    = vtd_class_init,
3809 };
3810 
3811 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
3812                                                      void *data)
3813 {
3814     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
3815 
3816     imrc->translate = vtd_iommu_translate;
3817     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
3818     imrc->replay = vtd_iommu_replay;
3819 }
3820 
3821 static const TypeInfo vtd_iommu_memory_region_info = {
3822     .parent = TYPE_IOMMU_MEMORY_REGION,
3823     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
3824     .class_init = vtd_iommu_memory_region_class_init,
3825 };
3826 
3827 static void vtd_register_types(void)
3828 {
3829     type_register_static(&vtd_info);
3830     type_register_static(&vtd_iommu_memory_region_info);
3831 }
3832 
3833 type_init(vtd_register_types)
3834