xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision 5b3e7517)
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 "intel_iommu_internal.h"
28 #include "hw/pci/pci.h"
29 #include "hw/pci/pci_bus.h"
30 #include "hw/qdev-properties.h"
31 #include "hw/i386/pc.h"
32 #include "hw/i386/apic-msidef.h"
33 #include "hw/i386/x86-iommu.h"
34 #include "hw/pci-host/q35.h"
35 #include "sysemu/kvm.h"
36 #include "sysemu/dma.h"
37 #include "sysemu/sysemu.h"
38 #include "hw/i386/apic_internal.h"
39 #include "kvm/kvm_i386.h"
40 #include "migration/vmstate.h"
41 #include "trace.h"
42 
43 /* context entry operations */
44 #define VTD_CE_GET_RID2PASID(ce) \
45     ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK)
46 #define VTD_CE_GET_PASID_DIR_TABLE(ce) \
47     ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK)
48 
49 /* pe operations */
50 #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT)
51 #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW))
52 #define VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write) {\
53     if (ret_fr) {                                                             \
54         ret_fr = -ret_fr;                                                     \
55         if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) {                   \
56             trace_vtd_fault_disabled();                                       \
57         } else {                                                              \
58             vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);      \
59         }                                                                     \
60         goto error;                                                           \
61     }                                                                         \
62 }
63 
64 static void vtd_address_space_refresh_all(IntelIOMMUState *s);
65 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n);
66 
67 static void vtd_panic_require_caching_mode(void)
68 {
69     error_report("We need to set caching-mode=on for intel-iommu to enable "
70                  "device assignment with IOMMU protection.");
71     exit(1);
72 }
73 
74 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val,
75                             uint64_t wmask, uint64_t w1cmask)
76 {
77     stq_le_p(&s->csr[addr], val);
78     stq_le_p(&s->wmask[addr], wmask);
79     stq_le_p(&s->w1cmask[addr], w1cmask);
80 }
81 
82 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask)
83 {
84     stq_le_p(&s->womask[addr], mask);
85 }
86 
87 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val,
88                             uint32_t wmask, uint32_t w1cmask)
89 {
90     stl_le_p(&s->csr[addr], val);
91     stl_le_p(&s->wmask[addr], wmask);
92     stl_le_p(&s->w1cmask[addr], w1cmask);
93 }
94 
95 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask)
96 {
97     stl_le_p(&s->womask[addr], mask);
98 }
99 
100 /* "External" get/set operations */
101 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
102 {
103     uint64_t oldval = ldq_le_p(&s->csr[addr]);
104     uint64_t wmask = ldq_le_p(&s->wmask[addr]);
105     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
106     stq_le_p(&s->csr[addr],
107              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
108 }
109 
110 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
111 {
112     uint32_t oldval = ldl_le_p(&s->csr[addr]);
113     uint32_t wmask = ldl_le_p(&s->wmask[addr]);
114     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
115     stl_le_p(&s->csr[addr],
116              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
117 }
118 
119 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr)
120 {
121     uint64_t val = ldq_le_p(&s->csr[addr]);
122     uint64_t womask = ldq_le_p(&s->womask[addr]);
123     return val & ~womask;
124 }
125 
126 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr)
127 {
128     uint32_t val = ldl_le_p(&s->csr[addr]);
129     uint32_t womask = ldl_le_p(&s->womask[addr]);
130     return val & ~womask;
131 }
132 
133 /* "Internal" get/set operations */
134 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr)
135 {
136     return ldq_le_p(&s->csr[addr]);
137 }
138 
139 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr)
140 {
141     return ldl_le_p(&s->csr[addr]);
142 }
143 
144 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val)
145 {
146     stq_le_p(&s->csr[addr], val);
147 }
148 
149 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
150                                         uint32_t clear, uint32_t mask)
151 {
152     uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
153     stl_le_p(&s->csr[addr], new_val);
154     return new_val;
155 }
156 
157 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
158                                         uint64_t clear, uint64_t mask)
159 {
160     uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
161     stq_le_p(&s->csr[addr], new_val);
162     return new_val;
163 }
164 
165 static inline void vtd_iommu_lock(IntelIOMMUState *s)
166 {
167     qemu_mutex_lock(&s->iommu_lock);
168 }
169 
170 static inline void vtd_iommu_unlock(IntelIOMMUState *s)
171 {
172     qemu_mutex_unlock(&s->iommu_lock);
173 }
174 
175 static void vtd_update_scalable_state(IntelIOMMUState *s)
176 {
177     uint64_t val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
178 
179     if (s->scalable_mode) {
180         s->root_scalable = val & VTD_RTADDR_SMT;
181     }
182 }
183 
184 /* Whether the address space needs to notify new mappings */
185 static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as)
186 {
187     return as->notifier_flags & IOMMU_NOTIFIER_MAP;
188 }
189 
190 /* GHashTable functions */
191 static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2)
192 {
193     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
194 }
195 
196 static guint vtd_uint64_hash(gconstpointer v)
197 {
198     return (guint)*(const uint64_t *)v;
199 }
200 
201 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value,
202                                           gpointer user_data)
203 {
204     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
205     uint16_t domain_id = *(uint16_t *)user_data;
206     return entry->domain_id == domain_id;
207 }
208 
209 /* The shift of an addr for a certain level of paging structure */
210 static inline uint32_t vtd_slpt_level_shift(uint32_t level)
211 {
212     assert(level != 0);
213     return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
214 }
215 
216 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level)
217 {
218     return ~((1ULL << vtd_slpt_level_shift(level)) - 1);
219 }
220 
221 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value,
222                                         gpointer user_data)
223 {
224     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
225     VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data;
226     uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask;
227     uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K;
228     return (entry->domain_id == info->domain_id) &&
229             (((entry->gfn & info->mask) == gfn) ||
230              (entry->gfn == gfn_tlb));
231 }
232 
233 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
234  * IntelIOMMUState to 1.  Must be called with IOMMU lock held.
235  */
236 static void vtd_reset_context_cache_locked(IntelIOMMUState *s)
237 {
238     VTDAddressSpace *vtd_as;
239     VTDBus *vtd_bus;
240     GHashTableIter bus_it;
241     uint32_t devfn_it;
242 
243     trace_vtd_context_cache_reset();
244 
245     g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr);
246 
247     while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) {
248         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
249             vtd_as = vtd_bus->dev_as[devfn_it];
250             if (!vtd_as) {
251                 continue;
252             }
253             vtd_as->context_cache_entry.context_cache_gen = 0;
254         }
255     }
256     s->context_cache_gen = 1;
257 }
258 
259 /* Must be called with IOMMU lock held. */
260 static void vtd_reset_iotlb_locked(IntelIOMMUState *s)
261 {
262     assert(s->iotlb);
263     g_hash_table_remove_all(s->iotlb);
264 }
265 
266 static void vtd_reset_iotlb(IntelIOMMUState *s)
267 {
268     vtd_iommu_lock(s);
269     vtd_reset_iotlb_locked(s);
270     vtd_iommu_unlock(s);
271 }
272 
273 static void vtd_reset_caches(IntelIOMMUState *s)
274 {
275     vtd_iommu_lock(s);
276     vtd_reset_iotlb_locked(s);
277     vtd_reset_context_cache_locked(s);
278     vtd_iommu_unlock(s);
279 }
280 
281 static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id,
282                                   uint32_t level)
283 {
284     return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) |
285            ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT);
286 }
287 
288 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level)
289 {
290     return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K;
291 }
292 
293 /* Must be called with IOMMU lock held */
294 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id,
295                                        hwaddr addr)
296 {
297     VTDIOTLBEntry *entry;
298     uint64_t key;
299     int level;
300 
301     for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) {
302         key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level),
303                                 source_id, level);
304         entry = g_hash_table_lookup(s->iotlb, &key);
305         if (entry) {
306             goto out;
307         }
308     }
309 
310 out:
311     return entry;
312 }
313 
314 /* Must be with IOMMU lock held */
315 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id,
316                              uint16_t domain_id, hwaddr addr, uint64_t slpte,
317                              uint8_t access_flags, uint32_t level)
318 {
319     VTDIOTLBEntry *entry = g_malloc(sizeof(*entry));
320     uint64_t *key = g_malloc(sizeof(*key));
321     uint64_t gfn = vtd_get_iotlb_gfn(addr, level);
322 
323     trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id);
324     if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) {
325         trace_vtd_iotlb_reset("iotlb exceeds size limit");
326         vtd_reset_iotlb_locked(s);
327     }
328 
329     entry->gfn = gfn;
330     entry->domain_id = domain_id;
331     entry->slpte = slpte;
332     entry->access_flags = access_flags;
333     entry->mask = vtd_slpt_level_page_mask(level);
334     *key = vtd_get_iotlb_key(gfn, source_id, level);
335     g_hash_table_replace(s->iotlb, key, entry);
336 }
337 
338 /* Given the reg addr of both the message data and address, generate an
339  * interrupt via MSI.
340  */
341 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
342                                    hwaddr mesg_data_reg)
343 {
344     MSIMessage msi;
345 
346     assert(mesg_data_reg < DMAR_REG_SIZE);
347     assert(mesg_addr_reg < DMAR_REG_SIZE);
348 
349     msi.address = vtd_get_long_raw(s, mesg_addr_reg);
350     msi.data = vtd_get_long_raw(s, mesg_data_reg);
351 
352     trace_vtd_irq_generate(msi.address, msi.data);
353 
354     apic_get_class()->send_msi(&msi);
355 }
356 
357 /* Generate a fault event to software via MSI if conditions are met.
358  * Notice that the value of FSTS_REG being passed to it should be the one
359  * before any update.
360  */
361 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
362 {
363     if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
364         pre_fsts & VTD_FSTS_IQE) {
365         error_report_once("There are previous interrupt conditions "
366                           "to be serviced by software, fault event "
367                           "is not generated");
368         return;
369     }
370     vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
371     if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
372         error_report_once("Interrupt Mask set, irq is not generated");
373     } else {
374         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
375         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
376     }
377 }
378 
379 /* Check if the Fault (F) field of the Fault Recording Register referenced by
380  * @index is Set.
381  */
382 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index)
383 {
384     /* Each reg is 128-bit */
385     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
386     addr += 8; /* Access the high 64-bit half */
387 
388     assert(index < DMAR_FRCD_REG_NR);
389 
390     return vtd_get_quad_raw(s, addr) & VTD_FRCD_F;
391 }
392 
393 /* Update the PPF field of Fault Status Register.
394  * Should be called whenever change the F field of any fault recording
395  * registers.
396  */
397 static void vtd_update_fsts_ppf(IntelIOMMUState *s)
398 {
399     uint32_t i;
400     uint32_t ppf_mask = 0;
401 
402     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
403         if (vtd_is_frcd_set(s, i)) {
404             ppf_mask = VTD_FSTS_PPF;
405             break;
406         }
407     }
408     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
409     trace_vtd_fsts_ppf(!!ppf_mask);
410 }
411 
412 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index)
413 {
414     /* Each reg is 128-bit */
415     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
416     addr += 8; /* Access the high 64-bit half */
417 
418     assert(index < DMAR_FRCD_REG_NR);
419 
420     vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
421     vtd_update_fsts_ppf(s);
422 }
423 
424 /* Must not update F field now, should be done later */
425 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index,
426                             uint16_t source_id, hwaddr addr,
427                             VTDFaultReason fault, bool is_write)
428 {
429     uint64_t hi = 0, lo;
430     hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
431 
432     assert(index < DMAR_FRCD_REG_NR);
433 
434     lo = VTD_FRCD_FI(addr);
435     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
436     if (!is_write) {
437         hi |= VTD_FRCD_T;
438     }
439     vtd_set_quad_raw(s, frcd_reg_addr, lo);
440     vtd_set_quad_raw(s, frcd_reg_addr + 8, hi);
441 
442     trace_vtd_frr_new(index, hi, lo);
443 }
444 
445 /* Try to collapse multiple pending faults from the same requester */
446 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id)
447 {
448     uint32_t i;
449     uint64_t frcd_reg;
450     hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
451 
452     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
453         frcd_reg = vtd_get_quad_raw(s, addr);
454         if ((frcd_reg & VTD_FRCD_F) &&
455             ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
456             return true;
457         }
458         addr += 16; /* 128-bit for each */
459     }
460     return false;
461 }
462 
463 /* Log and report an DMAR (address translation) fault to software */
464 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
465                                   hwaddr addr, VTDFaultReason fault,
466                                   bool is_write)
467 {
468     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
469 
470     assert(fault < VTD_FR_MAX);
471 
472     if (fault == VTD_FR_RESERVED_ERR) {
473         /* This is not a normal fault reason case. Drop it. */
474         return;
475     }
476 
477     trace_vtd_dmar_fault(source_id, fault, addr, is_write);
478 
479     if (fsts_reg & VTD_FSTS_PFO) {
480         error_report_once("New fault is not recorded due to "
481                           "Primary Fault Overflow");
482         return;
483     }
484 
485     if (vtd_try_collapse_fault(s, source_id)) {
486         error_report_once("New fault is not recorded due to "
487                           "compression of faults");
488         return;
489     }
490 
491     if (vtd_is_frcd_set(s, s->next_frcd_reg)) {
492         error_report_once("Next Fault Recording Reg is used, "
493                           "new fault is not recorded, set PFO field");
494         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
495         return;
496     }
497 
498     vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write);
499 
500     if (fsts_reg & VTD_FSTS_PPF) {
501         error_report_once("There are pending faults already, "
502                           "fault event is not generated");
503         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg);
504         s->next_frcd_reg++;
505         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
506             s->next_frcd_reg = 0;
507         }
508     } else {
509         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
510                                 VTD_FSTS_FRI(s->next_frcd_reg));
511         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */
512         s->next_frcd_reg++;
513         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
514             s->next_frcd_reg = 0;
515         }
516         /* This case actually cause the PPF to be Set.
517          * So generate fault event (interrupt).
518          */
519          vtd_generate_fault_event(s, fsts_reg);
520     }
521 }
522 
523 /* Handle Invalidation Queue Errors of queued invalidation interface error
524  * conditions.
525  */
526 static void vtd_handle_inv_queue_error(IntelIOMMUState *s)
527 {
528     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
529 
530     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE);
531     vtd_generate_fault_event(s, fsts_reg);
532 }
533 
534 /* Set the IWC field and try to generate an invalidation completion interrupt */
535 static void vtd_generate_completion_event(IntelIOMMUState *s)
536 {
537     if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) {
538         trace_vtd_inv_desc_wait_irq("One pending, skip current");
539         return;
540     }
541     vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC);
542     vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP);
543     if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) {
544         trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
545                                     "new event not generated");
546         return;
547     } else {
548         /* Generate the interrupt event */
549         trace_vtd_inv_desc_wait_irq("Generating complete event");
550         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
551         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
552     }
553 }
554 
555 static inline bool vtd_root_entry_present(IntelIOMMUState *s,
556                                           VTDRootEntry *re,
557                                           uint8_t devfn)
558 {
559     if (s->root_scalable && devfn > UINT8_MAX / 2) {
560         return re->hi & VTD_ROOT_ENTRY_P;
561     }
562 
563     return re->lo & VTD_ROOT_ENTRY_P;
564 }
565 
566 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index,
567                               VTDRootEntry *re)
568 {
569     dma_addr_t addr;
570 
571     addr = s->root + index * sizeof(*re);
572     if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) {
573         re->lo = 0;
574         return -VTD_FR_ROOT_TABLE_INV;
575     }
576     re->lo = le64_to_cpu(re->lo);
577     re->hi = le64_to_cpu(re->hi);
578     return 0;
579 }
580 
581 static inline bool vtd_ce_present(VTDContextEntry *context)
582 {
583     return context->lo & VTD_CONTEXT_ENTRY_P;
584 }
585 
586 static int vtd_get_context_entry_from_root(IntelIOMMUState *s,
587                                            VTDRootEntry *re,
588                                            uint8_t index,
589                                            VTDContextEntry *ce)
590 {
591     dma_addr_t addr, ce_size;
592 
593     /* we have checked that root entry is present */
594     ce_size = s->root_scalable ? VTD_CTX_ENTRY_SCALABLE_SIZE :
595               VTD_CTX_ENTRY_LEGACY_SIZE;
596 
597     if (s->root_scalable && index > UINT8_MAX / 2) {
598         index = index & (~VTD_DEVFN_CHECK_MASK);
599         addr = re->hi & VTD_ROOT_ENTRY_CTP;
600     } else {
601         addr = re->lo & VTD_ROOT_ENTRY_CTP;
602     }
603 
604     addr = addr + index * ce_size;
605     if (dma_memory_read(&address_space_memory, addr, ce, ce_size)) {
606         return -VTD_FR_CONTEXT_TABLE_INV;
607     }
608 
609     ce->lo = le64_to_cpu(ce->lo);
610     ce->hi = le64_to_cpu(ce->hi);
611     if (ce_size == VTD_CTX_ENTRY_SCALABLE_SIZE) {
612         ce->val[2] = le64_to_cpu(ce->val[2]);
613         ce->val[3] = le64_to_cpu(ce->val[3]);
614     }
615     return 0;
616 }
617 
618 static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce)
619 {
620     return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
621 }
622 
623 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw)
624 {
625     return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw);
626 }
627 
628 /* Whether the pte indicates the address of the page frame */
629 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level)
630 {
631     return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
632 }
633 
634 /* Get the content of a spte located in @base_addr[@index] */
635 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index)
636 {
637     uint64_t slpte;
638 
639     assert(index < VTD_SL_PT_ENTRY_NR);
640 
641     if (dma_memory_read(&address_space_memory,
642                         base_addr + index * sizeof(slpte), &slpte,
643                         sizeof(slpte))) {
644         slpte = (uint64_t)-1;
645         return slpte;
646     }
647     slpte = le64_to_cpu(slpte);
648     return slpte;
649 }
650 
651 /* Given an iova and the level of paging structure, return the offset
652  * of current level.
653  */
654 static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level)
655 {
656     return (iova >> vtd_slpt_level_shift(level)) &
657             ((1ULL << VTD_SL_LEVEL_BITS) - 1);
658 }
659 
660 /* Check Capability Register to see if the @level of page-table is supported */
661 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level)
662 {
663     return VTD_CAP_SAGAW_MASK & s->cap &
664            (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
665 }
666 
667 /* Return true if check passed, otherwise false */
668 static inline bool vtd_pe_type_check(X86IOMMUState *x86_iommu,
669                                      VTDPASIDEntry *pe)
670 {
671     switch (VTD_PE_GET_TYPE(pe)) {
672     case VTD_SM_PASID_ENTRY_FLT:
673     case VTD_SM_PASID_ENTRY_SLT:
674     case VTD_SM_PASID_ENTRY_NESTED:
675         break;
676     case VTD_SM_PASID_ENTRY_PT:
677         if (!x86_iommu->pt_supported) {
678             return false;
679         }
680         break;
681     default:
682         /* Unknown type */
683         return false;
684     }
685     return true;
686 }
687 
688 static inline bool vtd_pdire_present(VTDPASIDDirEntry *pdire)
689 {
690     return pdire->val & 1;
691 }
692 
693 /**
694  * Caller of this function should check present bit if wants
695  * to use pdir entry for further usage except for fpd bit check.
696  */
697 static int vtd_get_pdire_from_pdir_table(dma_addr_t pasid_dir_base,
698                                          uint32_t pasid,
699                                          VTDPASIDDirEntry *pdire)
700 {
701     uint32_t index;
702     dma_addr_t addr, entry_size;
703 
704     index = VTD_PASID_DIR_INDEX(pasid);
705     entry_size = VTD_PASID_DIR_ENTRY_SIZE;
706     addr = pasid_dir_base + index * entry_size;
707     if (dma_memory_read(&address_space_memory, addr, pdire, entry_size)) {
708         return -VTD_FR_PASID_TABLE_INV;
709     }
710 
711     return 0;
712 }
713 
714 static inline bool vtd_pe_present(VTDPASIDEntry *pe)
715 {
716     return pe->val[0] & VTD_PASID_ENTRY_P;
717 }
718 
719 static int vtd_get_pe_in_pasid_leaf_table(IntelIOMMUState *s,
720                                           uint32_t pasid,
721                                           dma_addr_t addr,
722                                           VTDPASIDEntry *pe)
723 {
724     uint32_t index;
725     dma_addr_t entry_size;
726     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
727 
728     index = VTD_PASID_TABLE_INDEX(pasid);
729     entry_size = VTD_PASID_ENTRY_SIZE;
730     addr = addr + index * entry_size;
731     if (dma_memory_read(&address_space_memory, addr, pe, entry_size)) {
732         return -VTD_FR_PASID_TABLE_INV;
733     }
734 
735     /* Do translation type check */
736     if (!vtd_pe_type_check(x86_iommu, pe)) {
737         return -VTD_FR_PASID_TABLE_INV;
738     }
739 
740     if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) {
741         return -VTD_FR_PASID_TABLE_INV;
742     }
743 
744     return 0;
745 }
746 
747 /**
748  * Caller of this function should check present bit if wants
749  * to use pasid entry for further usage except for fpd bit check.
750  */
751 static int vtd_get_pe_from_pdire(IntelIOMMUState *s,
752                                  uint32_t pasid,
753                                  VTDPASIDDirEntry *pdire,
754                                  VTDPASIDEntry *pe)
755 {
756     dma_addr_t addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK;
757 
758     return vtd_get_pe_in_pasid_leaf_table(s, pasid, addr, pe);
759 }
760 
761 /**
762  * This function gets a pasid entry from a specified pasid
763  * table (includes dir and leaf table) with a specified pasid.
764  * Sanity check should be done to ensure return a present
765  * pasid entry to caller.
766  */
767 static int vtd_get_pe_from_pasid_table(IntelIOMMUState *s,
768                                        dma_addr_t pasid_dir_base,
769                                        uint32_t pasid,
770                                        VTDPASIDEntry *pe)
771 {
772     int ret;
773     VTDPASIDDirEntry pdire;
774 
775     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base,
776                                         pasid, &pdire);
777     if (ret) {
778         return ret;
779     }
780 
781     if (!vtd_pdire_present(&pdire)) {
782         return -VTD_FR_PASID_TABLE_INV;
783     }
784 
785     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, pe);
786     if (ret) {
787         return ret;
788     }
789 
790     if (!vtd_pe_present(pe)) {
791         return -VTD_FR_PASID_TABLE_INV;
792     }
793 
794     return 0;
795 }
796 
797 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s,
798                                       VTDContextEntry *ce,
799                                       VTDPASIDEntry *pe)
800 {
801     uint32_t pasid;
802     dma_addr_t pasid_dir_base;
803     int ret = 0;
804 
805     pasid = VTD_CE_GET_RID2PASID(ce);
806     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
807     ret = vtd_get_pe_from_pasid_table(s, pasid_dir_base, pasid, pe);
808 
809     return ret;
810 }
811 
812 static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s,
813                                 VTDContextEntry *ce,
814                                 bool *pe_fpd_set)
815 {
816     int ret;
817     uint32_t pasid;
818     dma_addr_t pasid_dir_base;
819     VTDPASIDDirEntry pdire;
820     VTDPASIDEntry pe;
821 
822     pasid = VTD_CE_GET_RID2PASID(ce);
823     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
824 
825     /*
826      * No present bit check since fpd is meaningful even
827      * if the present bit is clear.
828      */
829     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base, pasid, &pdire);
830     if (ret) {
831         return ret;
832     }
833 
834     if (pdire.val & VTD_PASID_DIR_FPD) {
835         *pe_fpd_set = true;
836         return 0;
837     }
838 
839     if (!vtd_pdire_present(&pdire)) {
840         return -VTD_FR_PASID_TABLE_INV;
841     }
842 
843     /*
844      * No present bit check since fpd is meaningful even
845      * if the present bit is clear.
846      */
847     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, &pe);
848     if (ret) {
849         return ret;
850     }
851 
852     if (pe.val[0] & VTD_PASID_ENTRY_FPD) {
853         *pe_fpd_set = true;
854     }
855 
856     return 0;
857 }
858 
859 /* Get the page-table level that hardware should use for the second-level
860  * page-table walk from the Address Width field of context-entry.
861  */
862 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
863 {
864     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
865 }
866 
867 static uint32_t vtd_get_iova_level(IntelIOMMUState *s,
868                                    VTDContextEntry *ce)
869 {
870     VTDPASIDEntry pe;
871 
872     if (s->root_scalable) {
873         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
874         return VTD_PE_GET_LEVEL(&pe);
875     }
876 
877     return vtd_ce_get_level(ce);
878 }
879 
880 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
881 {
882     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
883 }
884 
885 static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s,
886                                   VTDContextEntry *ce)
887 {
888     VTDPASIDEntry pe;
889 
890     if (s->root_scalable) {
891         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
892         return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9;
893     }
894 
895     return vtd_ce_get_agaw(ce);
896 }
897 
898 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
899 {
900     return ce->lo & VTD_CONTEXT_ENTRY_TT;
901 }
902 
903 /* Only for Legacy Mode. Return true if check passed, otherwise false */
904 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
905                                      VTDContextEntry *ce)
906 {
907     switch (vtd_ce_get_type(ce)) {
908     case VTD_CONTEXT_TT_MULTI_LEVEL:
909         /* Always supported */
910         break;
911     case VTD_CONTEXT_TT_DEV_IOTLB:
912         if (!x86_iommu->dt_supported) {
913             error_report_once("%s: DT specified but not supported", __func__);
914             return false;
915         }
916         break;
917     case VTD_CONTEXT_TT_PASS_THROUGH:
918         if (!x86_iommu->pt_supported) {
919             error_report_once("%s: PT specified but not supported", __func__);
920             return false;
921         }
922         break;
923     default:
924         /* Unknown type */
925         error_report_once("%s: unknown ce type: %"PRIu32, __func__,
926                           vtd_ce_get_type(ce));
927         return false;
928     }
929     return true;
930 }
931 
932 static inline uint64_t vtd_iova_limit(IntelIOMMUState *s,
933                                       VTDContextEntry *ce, uint8_t aw)
934 {
935     uint32_t ce_agaw = vtd_get_iova_agaw(s, ce);
936     return 1ULL << MIN(ce_agaw, aw);
937 }
938 
939 /* Return true if IOVA passes range check, otherwise false. */
940 static inline bool vtd_iova_range_check(IntelIOMMUState *s,
941                                         uint64_t iova, VTDContextEntry *ce,
942                                         uint8_t aw)
943 {
944     /*
945      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
946      * in CAP_REG and AW in context-entry.
947      */
948     return !(iova & ~(vtd_iova_limit(s, ce, aw) - 1));
949 }
950 
951 static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s,
952                                           VTDContextEntry *ce)
953 {
954     VTDPASIDEntry pe;
955 
956     if (s->root_scalable) {
957         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
958         return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR;
959     }
960 
961     return vtd_ce_get_slpt_base(ce);
962 }
963 
964 /*
965  * Rsvd field masks for spte:
966  *     vtd_spte_rsvd 4k pages
967  *     vtd_spte_rsvd_large large pages
968  */
969 static uint64_t vtd_spte_rsvd[5];
970 static uint64_t vtd_spte_rsvd_large[5];
971 
972 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
973 {
974     uint64_t rsvd_mask = vtd_spte_rsvd[level];
975 
976     if ((level == VTD_SL_PD_LEVEL || level == VTD_SL_PDP_LEVEL) &&
977         (slpte & VTD_SL_PT_PAGE_SIZE_MASK)) {
978         /* large page */
979         rsvd_mask = vtd_spte_rsvd_large[level];
980     }
981 
982     return slpte & rsvd_mask;
983 }
984 
985 /* Find the VTD address space associated with a given bus number */
986 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num)
987 {
988     VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num];
989     GHashTableIter iter;
990 
991     if (vtd_bus) {
992         return vtd_bus;
993     }
994 
995     /*
996      * Iterate over the registered buses to find the one which
997      * currently holds this bus number and update the bus_num
998      * lookup table.
999      */
1000     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1001     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1002         if (pci_bus_num(vtd_bus->bus) == bus_num) {
1003             s->vtd_as_by_bus_num[bus_num] = vtd_bus;
1004             return vtd_bus;
1005         }
1006     }
1007 
1008     return NULL;
1009 }
1010 
1011 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
1012  * of the translation, can be used for deciding the size of large page.
1013  */
1014 static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce,
1015                              uint64_t iova, bool is_write,
1016                              uint64_t *slptep, uint32_t *slpte_level,
1017                              bool *reads, bool *writes, uint8_t aw_bits)
1018 {
1019     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1020     uint32_t level = vtd_get_iova_level(s, ce);
1021     uint32_t offset;
1022     uint64_t slpte;
1023     uint64_t access_right_check;
1024 
1025     if (!vtd_iova_range_check(s, iova, ce, aw_bits)) {
1026         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")",
1027                           __func__, iova);
1028         return -VTD_FR_ADDR_BEYOND_MGAW;
1029     }
1030 
1031     /* FIXME: what is the Atomics request here? */
1032     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
1033 
1034     while (true) {
1035         offset = vtd_iova_level_offset(iova, level);
1036         slpte = vtd_get_slpte(addr, offset);
1037 
1038         if (slpte == (uint64_t)-1) {
1039             error_report_once("%s: detected read error on DMAR slpte "
1040                               "(iova=0x%" PRIx64 ")", __func__, iova);
1041             if (level == vtd_get_iova_level(s, ce)) {
1042                 /* Invalid programming of context-entry */
1043                 return -VTD_FR_CONTEXT_ENTRY_INV;
1044             } else {
1045                 return -VTD_FR_PAGING_ENTRY_INV;
1046             }
1047         }
1048         *reads = (*reads) && (slpte & VTD_SL_R);
1049         *writes = (*writes) && (slpte & VTD_SL_W);
1050         if (!(slpte & access_right_check)) {
1051             error_report_once("%s: detected slpte permission error "
1052                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
1053                               "slpte=0x%" PRIx64 ", write=%d)", __func__,
1054                               iova, level, slpte, is_write);
1055             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
1056         }
1057         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1058             error_report_once("%s: detected splte reserve non-zero "
1059                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
1060                               "slpte=0x%" PRIx64 ")", __func__, iova,
1061                               level, slpte);
1062             return -VTD_FR_PAGING_ENTRY_RSVD;
1063         }
1064 
1065         if (vtd_is_last_slpte(slpte, level)) {
1066             *slptep = slpte;
1067             *slpte_level = level;
1068             return 0;
1069         }
1070         addr = vtd_get_slpte_addr(slpte, aw_bits);
1071         level--;
1072     }
1073 }
1074 
1075 typedef int (*vtd_page_walk_hook)(IOMMUTLBEvent *event, void *private);
1076 
1077 /**
1078  * Constant information used during page walking
1079  *
1080  * @hook_fn: hook func to be called when detected page
1081  * @private: private data to be passed into hook func
1082  * @notify_unmap: whether we should notify invalid entries
1083  * @as: VT-d address space of the device
1084  * @aw: maximum address width
1085  * @domain: domain ID of the page walk
1086  */
1087 typedef struct {
1088     VTDAddressSpace *as;
1089     vtd_page_walk_hook hook_fn;
1090     void *private;
1091     bool notify_unmap;
1092     uint8_t aw;
1093     uint16_t domain_id;
1094 } vtd_page_walk_info;
1095 
1096 static int vtd_page_walk_one(IOMMUTLBEvent *event, vtd_page_walk_info *info)
1097 {
1098     VTDAddressSpace *as = info->as;
1099     vtd_page_walk_hook hook_fn = info->hook_fn;
1100     void *private = info->private;
1101     IOMMUTLBEntry *entry = &event->entry;
1102     DMAMap target = {
1103         .iova = entry->iova,
1104         .size = entry->addr_mask,
1105         .translated_addr = entry->translated_addr,
1106         .perm = entry->perm,
1107     };
1108     DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
1109 
1110     if (event->type == IOMMU_NOTIFIER_UNMAP && !info->notify_unmap) {
1111         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1112         return 0;
1113     }
1114 
1115     assert(hook_fn);
1116 
1117     /* Update local IOVA mapped ranges */
1118     if (event->type == IOMMU_NOTIFIER_MAP) {
1119         if (mapped) {
1120             /* If it's exactly the same translation, skip */
1121             if (!memcmp(mapped, &target, sizeof(target))) {
1122                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
1123                                                  entry->translated_addr);
1124                 return 0;
1125             } else {
1126                 /*
1127                  * Translation changed.  Normally this should not
1128                  * happen, but it can happen when with buggy guest
1129                  * OSes.  Note that there will be a small window that
1130                  * we don't have map at all.  But that's the best
1131                  * effort we can do.  The ideal way to emulate this is
1132                  * atomically modify the PTE to follow what has
1133                  * changed, but we can't.  One example is that vfio
1134                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1135                  * interface to modify a mapping (meanwhile it seems
1136                  * meaningless to even provide one).  Anyway, let's
1137                  * mark this as a TODO in case one day we'll have
1138                  * a better solution.
1139                  */
1140                 IOMMUAccessFlags cache_perm = entry->perm;
1141                 int ret;
1142 
1143                 /* Emulate an UNMAP */
1144                 event->type = IOMMU_NOTIFIER_UNMAP;
1145                 entry->perm = IOMMU_NONE;
1146                 trace_vtd_page_walk_one(info->domain_id,
1147                                         entry->iova,
1148                                         entry->translated_addr,
1149                                         entry->addr_mask,
1150                                         entry->perm);
1151                 ret = hook_fn(event, private);
1152                 if (ret) {
1153                     return ret;
1154                 }
1155                 /* Drop any existing mapping */
1156                 iova_tree_remove(as->iova_tree, &target);
1157                 /* Recover the correct type */
1158                 event->type = IOMMU_NOTIFIER_MAP;
1159                 entry->perm = cache_perm;
1160             }
1161         }
1162         iova_tree_insert(as->iova_tree, &target);
1163     } else {
1164         if (!mapped) {
1165             /* Skip since we didn't map this range at all */
1166             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1167             return 0;
1168         }
1169         iova_tree_remove(as->iova_tree, &target);
1170     }
1171 
1172     trace_vtd_page_walk_one(info->domain_id, entry->iova,
1173                             entry->translated_addr, entry->addr_mask,
1174                             entry->perm);
1175     return hook_fn(event, private);
1176 }
1177 
1178 /**
1179  * vtd_page_walk_level - walk over specific level for IOVA range
1180  *
1181  * @addr: base GPA addr to start the walk
1182  * @start: IOVA range start address
1183  * @end: IOVA range end address (start <= addr < end)
1184  * @read: whether parent level has read permission
1185  * @write: whether parent level has write permission
1186  * @info: constant information for the page walk
1187  */
1188 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
1189                                uint64_t end, uint32_t level, bool read,
1190                                bool write, vtd_page_walk_info *info)
1191 {
1192     bool read_cur, write_cur, entry_valid;
1193     uint32_t offset;
1194     uint64_t slpte;
1195     uint64_t subpage_size, subpage_mask;
1196     IOMMUTLBEvent event;
1197     uint64_t iova = start;
1198     uint64_t iova_next;
1199     int ret = 0;
1200 
1201     trace_vtd_page_walk_level(addr, level, start, end);
1202 
1203     subpage_size = 1ULL << vtd_slpt_level_shift(level);
1204     subpage_mask = vtd_slpt_level_page_mask(level);
1205 
1206     while (iova < end) {
1207         iova_next = (iova & subpage_mask) + subpage_size;
1208 
1209         offset = vtd_iova_level_offset(iova, level);
1210         slpte = vtd_get_slpte(addr, offset);
1211 
1212         if (slpte == (uint64_t)-1) {
1213             trace_vtd_page_walk_skip_read(iova, iova_next);
1214             goto next;
1215         }
1216 
1217         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1218             trace_vtd_page_walk_skip_reserve(iova, iova_next);
1219             goto next;
1220         }
1221 
1222         /* Permissions are stacked with parents' */
1223         read_cur = read && (slpte & VTD_SL_R);
1224         write_cur = write && (slpte & VTD_SL_W);
1225 
1226         /*
1227          * As long as we have either read/write permission, this is a
1228          * valid entry. The rule works for both page entries and page
1229          * table entries.
1230          */
1231         entry_valid = read_cur | write_cur;
1232 
1233         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
1234             /*
1235              * This is a valid PDE (or even bigger than PDE).  We need
1236              * to walk one further level.
1237              */
1238             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
1239                                       iova, MIN(iova_next, end), level - 1,
1240                                       read_cur, write_cur, info);
1241         } else {
1242             /*
1243              * This means we are either:
1244              *
1245              * (1) the real page entry (either 4K page, or huge page)
1246              * (2) the whole range is invalid
1247              *
1248              * In either case, we send an IOTLB notification down.
1249              */
1250             event.entry.target_as = &address_space_memory;
1251             event.entry.iova = iova & subpage_mask;
1252             event.entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
1253             event.entry.addr_mask = ~subpage_mask;
1254             /* NOTE: this is only meaningful if entry_valid == true */
1255             event.entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
1256             event.type = event.entry.perm ? IOMMU_NOTIFIER_MAP :
1257                                             IOMMU_NOTIFIER_UNMAP;
1258             ret = vtd_page_walk_one(&event, info);
1259         }
1260 
1261         if (ret < 0) {
1262             return ret;
1263         }
1264 
1265 next:
1266         iova = iova_next;
1267     }
1268 
1269     return 0;
1270 }
1271 
1272 /**
1273  * vtd_page_walk - walk specific IOVA range, and call the hook
1274  *
1275  * @s: intel iommu state
1276  * @ce: context entry to walk upon
1277  * @start: IOVA address to start the walk
1278  * @end: IOVA range end address (start <= addr < end)
1279  * @info: page walking information struct
1280  */
1281 static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce,
1282                          uint64_t start, uint64_t end,
1283                          vtd_page_walk_info *info)
1284 {
1285     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1286     uint32_t level = vtd_get_iova_level(s, ce);
1287 
1288     if (!vtd_iova_range_check(s, start, ce, info->aw)) {
1289         return -VTD_FR_ADDR_BEYOND_MGAW;
1290     }
1291 
1292     if (!vtd_iova_range_check(s, end, ce, info->aw)) {
1293         /* Fix end so that it reaches the maximum */
1294         end = vtd_iova_limit(s, ce, info->aw);
1295     }
1296 
1297     return vtd_page_walk_level(addr, start, end, level, true, true, info);
1298 }
1299 
1300 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s,
1301                                           VTDRootEntry *re)
1302 {
1303     /* Legacy Mode reserved bits check */
1304     if (!s->root_scalable &&
1305         (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1306         goto rsvd_err;
1307 
1308     /* Scalable Mode reserved bits check */
1309     if (s->root_scalable &&
1310         ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) ||
1311          (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1312         goto rsvd_err;
1313 
1314     return 0;
1315 
1316 rsvd_err:
1317     error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1318                       ", lo=0x%"PRIx64,
1319                       __func__, re->hi, re->lo);
1320     return -VTD_FR_ROOT_ENTRY_RSVD;
1321 }
1322 
1323 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s,
1324                                                     VTDContextEntry *ce)
1325 {
1326     if (!s->root_scalable &&
1327         (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI ||
1328          ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1329         error_report_once("%s: invalid context entry: hi=%"PRIx64
1330                           ", lo=%"PRIx64" (reserved nonzero)",
1331                           __func__, ce->hi, ce->lo);
1332         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1333     }
1334 
1335     if (s->root_scalable &&
1336         (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) ||
1337          ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 ||
1338          ce->val[2] ||
1339          ce->val[3])) {
1340         error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1341                           ", val[2]=%"PRIx64
1342                           ", val[1]=%"PRIx64
1343                           ", val[0]=%"PRIx64" (reserved nonzero)",
1344                           __func__, ce->val[3], ce->val[2],
1345                           ce->val[1], ce->val[0]);
1346         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1347     }
1348 
1349     return 0;
1350 }
1351 
1352 static int vtd_ce_rid2pasid_check(IntelIOMMUState *s,
1353                                   VTDContextEntry *ce)
1354 {
1355     VTDPASIDEntry pe;
1356 
1357     /*
1358      * Make sure in Scalable Mode, a present context entry
1359      * has valid rid2pasid setting, which includes valid
1360      * rid2pasid field and corresponding pasid entry setting
1361      */
1362     return vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1363 }
1364 
1365 /* Map a device to its corresponding domain (context-entry) */
1366 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
1367                                     uint8_t devfn, VTDContextEntry *ce)
1368 {
1369     VTDRootEntry re;
1370     int ret_fr;
1371     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
1372 
1373     ret_fr = vtd_get_root_entry(s, bus_num, &re);
1374     if (ret_fr) {
1375         return ret_fr;
1376     }
1377 
1378     if (!vtd_root_entry_present(s, &re, devfn)) {
1379         /* Not error - it's okay we don't have root entry. */
1380         trace_vtd_re_not_present(bus_num);
1381         return -VTD_FR_ROOT_ENTRY_P;
1382     }
1383 
1384     ret_fr = vtd_root_entry_rsvd_bits_check(s, &re);
1385     if (ret_fr) {
1386         return ret_fr;
1387     }
1388 
1389     ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce);
1390     if (ret_fr) {
1391         return ret_fr;
1392     }
1393 
1394     if (!vtd_ce_present(ce)) {
1395         /* Not error - it's okay we don't have context entry. */
1396         trace_vtd_ce_not_present(bus_num, devfn);
1397         return -VTD_FR_CONTEXT_ENTRY_P;
1398     }
1399 
1400     ret_fr = vtd_context_entry_rsvd_bits_check(s, ce);
1401     if (ret_fr) {
1402         return ret_fr;
1403     }
1404 
1405     /* Check if the programming of context-entry is valid */
1406     if (!s->root_scalable &&
1407         !vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1408         error_report_once("%s: invalid context entry: hi=%"PRIx64
1409                           ", lo=%"PRIx64" (level %d not supported)",
1410                           __func__, ce->hi, ce->lo,
1411                           vtd_ce_get_level(ce));
1412         return -VTD_FR_CONTEXT_ENTRY_INV;
1413     }
1414 
1415     if (!s->root_scalable) {
1416         /* Do translation type check */
1417         if (!vtd_ce_type_check(x86_iommu, ce)) {
1418             /* Errors dumped in vtd_ce_type_check() */
1419             return -VTD_FR_CONTEXT_ENTRY_INV;
1420         }
1421     } else {
1422         /*
1423          * Check if the programming of context-entry.rid2pasid
1424          * and corresponding pasid setting is valid, and thus
1425          * avoids to check pasid entry fetching result in future
1426          * helper function calling.
1427          */
1428         ret_fr = vtd_ce_rid2pasid_check(s, ce);
1429         if (ret_fr) {
1430             return ret_fr;
1431         }
1432     }
1433 
1434     return 0;
1435 }
1436 
1437 static int vtd_sync_shadow_page_hook(IOMMUTLBEvent *event,
1438                                      void *private)
1439 {
1440     memory_region_notify_iommu(private, 0, *event);
1441     return 0;
1442 }
1443 
1444 static uint16_t vtd_get_domain_id(IntelIOMMUState *s,
1445                                   VTDContextEntry *ce)
1446 {
1447     VTDPASIDEntry pe;
1448 
1449     if (s->root_scalable) {
1450         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1451         return VTD_SM_PASID_ENTRY_DID(pe.val[1]);
1452     }
1453 
1454     return VTD_CONTEXT_ENTRY_DID(ce->hi);
1455 }
1456 
1457 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1458                                             VTDContextEntry *ce,
1459                                             hwaddr addr, hwaddr size)
1460 {
1461     IntelIOMMUState *s = vtd_as->iommu_state;
1462     vtd_page_walk_info info = {
1463         .hook_fn = vtd_sync_shadow_page_hook,
1464         .private = (void *)&vtd_as->iommu,
1465         .notify_unmap = true,
1466         .aw = s->aw_bits,
1467         .as = vtd_as,
1468         .domain_id = vtd_get_domain_id(s, ce),
1469     };
1470 
1471     return vtd_page_walk(s, ce, addr, addr + size, &info);
1472 }
1473 
1474 static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
1475 {
1476     int ret;
1477     VTDContextEntry ce;
1478     IOMMUNotifier *n;
1479 
1480     if (!(vtd_as->iommu.iommu_notify_flags & IOMMU_NOTIFIER_IOTLB_EVENTS)) {
1481         return 0;
1482     }
1483 
1484     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1485                                    pci_bus_num(vtd_as->bus),
1486                                    vtd_as->devfn, &ce);
1487     if (ret) {
1488         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1489             /*
1490              * It's a valid scenario to have a context entry that is
1491              * not present.  For example, when a device is removed
1492              * from an existing domain then the context entry will be
1493              * zeroed by the guest before it was put into another
1494              * domain.  When this happens, instead of synchronizing
1495              * the shadow pages we should invalidate all existing
1496              * mappings and notify the backends.
1497              */
1498             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1499                 vtd_address_space_unmap(vtd_as, n);
1500             }
1501             ret = 0;
1502         }
1503         return ret;
1504     }
1505 
1506     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1507 }
1508 
1509 /*
1510  * Check if specific device is configured to bypass address
1511  * translation for DMA requests. In Scalable Mode, bypass
1512  * 1st-level translation or 2nd-level translation, it depends
1513  * on PGTT setting.
1514  */
1515 static bool vtd_dev_pt_enabled(VTDAddressSpace *as)
1516 {
1517     IntelIOMMUState *s;
1518     VTDContextEntry ce;
1519     VTDPASIDEntry pe;
1520     int ret;
1521 
1522     assert(as);
1523 
1524     s = as->iommu_state;
1525     ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus),
1526                                    as->devfn, &ce);
1527     if (ret) {
1528         /*
1529          * Possibly failed to parse the context entry for some reason
1530          * (e.g., during init, or any guest configuration errors on
1531          * context entries). We should assume PT not enabled for
1532          * safety.
1533          */
1534         return false;
1535     }
1536 
1537     if (s->root_scalable) {
1538         ret = vtd_ce_get_rid2pasid_entry(s, &ce, &pe);
1539         if (ret) {
1540             error_report_once("%s: vtd_ce_get_rid2pasid_entry error: %"PRId32,
1541                               __func__, ret);
1542             return false;
1543         }
1544         return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT);
1545     }
1546 
1547     return (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH);
1548 }
1549 
1550 /* Return whether the device is using IOMMU translation. */
1551 static bool vtd_switch_address_space(VTDAddressSpace *as)
1552 {
1553     bool use_iommu;
1554     /* Whether we need to take the BQL on our own */
1555     bool take_bql = !qemu_mutex_iothread_locked();
1556 
1557     assert(as);
1558 
1559     use_iommu = as->iommu_state->dmar_enabled && !vtd_dev_pt_enabled(as);
1560 
1561     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1562                                    VTD_PCI_SLOT(as->devfn),
1563                                    VTD_PCI_FUNC(as->devfn),
1564                                    use_iommu);
1565 
1566     /*
1567      * It's possible that we reach here without BQL, e.g., when called
1568      * from vtd_pt_enable_fast_path(). However the memory APIs need
1569      * it. We'd better make sure we have had it already, or, take it.
1570      */
1571     if (take_bql) {
1572         qemu_mutex_lock_iothread();
1573     }
1574 
1575     /* Turn off first then on the other */
1576     if (use_iommu) {
1577         memory_region_set_enabled(&as->nodmar, false);
1578         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1579     } else {
1580         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1581         memory_region_set_enabled(&as->nodmar, true);
1582     }
1583 
1584     if (take_bql) {
1585         qemu_mutex_unlock_iothread();
1586     }
1587 
1588     return use_iommu;
1589 }
1590 
1591 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1592 {
1593     GHashTableIter iter;
1594     VTDBus *vtd_bus;
1595     int i;
1596 
1597     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1598     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1599         for (i = 0; i < PCI_DEVFN_MAX; i++) {
1600             if (!vtd_bus->dev_as[i]) {
1601                 continue;
1602             }
1603             vtd_switch_address_space(vtd_bus->dev_as[i]);
1604         }
1605     }
1606 }
1607 
1608 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn)
1609 {
1610     return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
1611 }
1612 
1613 static const bool vtd_qualified_faults[] = {
1614     [VTD_FR_RESERVED] = false,
1615     [VTD_FR_ROOT_ENTRY_P] = false,
1616     [VTD_FR_CONTEXT_ENTRY_P] = true,
1617     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1618     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1619     [VTD_FR_WRITE] = true,
1620     [VTD_FR_READ] = true,
1621     [VTD_FR_PAGING_ENTRY_INV] = true,
1622     [VTD_FR_ROOT_TABLE_INV] = false,
1623     [VTD_FR_CONTEXT_TABLE_INV] = false,
1624     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1625     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1626     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1627     [VTD_FR_PASID_TABLE_INV] = false,
1628     [VTD_FR_RESERVED_ERR] = false,
1629     [VTD_FR_MAX] = false,
1630 };
1631 
1632 /* To see if a fault condition is "qualified", which is reported to software
1633  * only if the FPD field in the context-entry used to process the faulting
1634  * request is 0.
1635  */
1636 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1637 {
1638     return vtd_qualified_faults[fault];
1639 }
1640 
1641 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1642 {
1643     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1644 }
1645 
1646 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1647 {
1648     VTDBus *vtd_bus;
1649     VTDAddressSpace *vtd_as;
1650     bool success = false;
1651 
1652     vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id));
1653     if (!vtd_bus) {
1654         goto out;
1655     }
1656 
1657     vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)];
1658     if (!vtd_as) {
1659         goto out;
1660     }
1661 
1662     if (vtd_switch_address_space(vtd_as) == false) {
1663         /* We switched off IOMMU region successfully. */
1664         success = true;
1665     }
1666 
1667 out:
1668     trace_vtd_pt_enable_fast_path(source_id, success);
1669 }
1670 
1671 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1672  * translation.
1673  *
1674  * Called from RCU critical section.
1675  *
1676  * @bus_num: The bus number
1677  * @devfn: The devfn, which is the  combined of device and function number
1678  * @is_write: The access is a write operation
1679  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1680  *
1681  * Returns true if translation is successful, otherwise false.
1682  */
1683 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1684                                    uint8_t devfn, hwaddr addr, bool is_write,
1685                                    IOMMUTLBEntry *entry)
1686 {
1687     IntelIOMMUState *s = vtd_as->iommu_state;
1688     VTDContextEntry ce;
1689     uint8_t bus_num = pci_bus_num(bus);
1690     VTDContextCacheEntry *cc_entry;
1691     uint64_t slpte, page_mask;
1692     uint32_t level;
1693     uint16_t source_id = vtd_make_source_id(bus_num, devfn);
1694     int ret_fr;
1695     bool is_fpd_set = false;
1696     bool reads = true;
1697     bool writes = true;
1698     uint8_t access_flags;
1699     VTDIOTLBEntry *iotlb_entry;
1700 
1701     /*
1702      * We have standalone memory region for interrupt addresses, we
1703      * should never receive translation requests in this region.
1704      */
1705     assert(!vtd_is_interrupt_addr(addr));
1706 
1707     vtd_iommu_lock(s);
1708 
1709     cc_entry = &vtd_as->context_cache_entry;
1710 
1711     /* Try to fetch slpte form IOTLB */
1712     iotlb_entry = vtd_lookup_iotlb(s, source_id, addr);
1713     if (iotlb_entry) {
1714         trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1715                                  iotlb_entry->domain_id);
1716         slpte = iotlb_entry->slpte;
1717         access_flags = iotlb_entry->access_flags;
1718         page_mask = iotlb_entry->mask;
1719         goto out;
1720     }
1721 
1722     /* Try to fetch context-entry from cache first */
1723     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1724         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1725                                cc_entry->context_entry.lo,
1726                                cc_entry->context_cache_gen);
1727         ce = cc_entry->context_entry;
1728         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1729         if (!is_fpd_set && s->root_scalable) {
1730             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1731             VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1732         }
1733     } else {
1734         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1735         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1736         if (!ret_fr && !is_fpd_set && s->root_scalable) {
1737             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1738         }
1739         VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1740         /* Update context-cache */
1741         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1742                                   cc_entry->context_cache_gen,
1743                                   s->context_cache_gen);
1744         cc_entry->context_entry = ce;
1745         cc_entry->context_cache_gen = s->context_cache_gen;
1746     }
1747 
1748     /*
1749      * We don't need to translate for pass-through context entries.
1750      * Also, let's ignore IOTLB caching as well for PT devices.
1751      */
1752     if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) {
1753         entry->iova = addr & VTD_PAGE_MASK_4K;
1754         entry->translated_addr = entry->iova;
1755         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1756         entry->perm = IOMMU_RW;
1757         trace_vtd_translate_pt(source_id, entry->iova);
1758 
1759         /*
1760          * When this happens, it means firstly caching-mode is not
1761          * enabled, and this is the first passthrough translation for
1762          * the device. Let's enable the fast path for passthrough.
1763          *
1764          * When passthrough is disabled again for the device, we can
1765          * capture it via the context entry invalidation, then the
1766          * IOMMU region can be swapped back.
1767          */
1768         vtd_pt_enable_fast_path(s, source_id);
1769         vtd_iommu_unlock(s);
1770         return true;
1771     }
1772 
1773     ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level,
1774                                &reads, &writes, s->aw_bits);
1775     VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1776 
1777     page_mask = vtd_slpt_level_page_mask(level);
1778     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1779     vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce), addr, slpte,
1780                      access_flags, level);
1781 out:
1782     vtd_iommu_unlock(s);
1783     entry->iova = addr & page_mask;
1784     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1785     entry->addr_mask = ~page_mask;
1786     entry->perm = access_flags;
1787     return true;
1788 
1789 error:
1790     vtd_iommu_unlock(s);
1791     entry->iova = 0;
1792     entry->translated_addr = 0;
1793     entry->addr_mask = 0;
1794     entry->perm = IOMMU_NONE;
1795     return false;
1796 }
1797 
1798 static void vtd_root_table_setup(IntelIOMMUState *s)
1799 {
1800     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1801     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
1802 
1803     vtd_update_scalable_state(s);
1804 
1805     trace_vtd_reg_dmar_root(s->root, s->root_scalable);
1806 }
1807 
1808 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
1809                                uint32_t index, uint32_t mask)
1810 {
1811     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
1812 }
1813 
1814 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
1815 {
1816     uint64_t value = 0;
1817     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
1818     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
1819     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
1820     s->intr_eime = value & VTD_IRTA_EIME;
1821 
1822     /* Notify global invalidation */
1823     vtd_iec_notify_all(s, true, 0, 0);
1824 
1825     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
1826 }
1827 
1828 static void vtd_iommu_replay_all(IntelIOMMUState *s)
1829 {
1830     VTDAddressSpace *vtd_as;
1831 
1832     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1833         vtd_sync_shadow_page_table(vtd_as);
1834     }
1835 }
1836 
1837 static void vtd_context_global_invalidate(IntelIOMMUState *s)
1838 {
1839     trace_vtd_inv_desc_cc_global();
1840     /* Protects context cache */
1841     vtd_iommu_lock(s);
1842     s->context_cache_gen++;
1843     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
1844         vtd_reset_context_cache_locked(s);
1845     }
1846     vtd_iommu_unlock(s);
1847     vtd_address_space_refresh_all(s);
1848     /*
1849      * From VT-d spec 6.5.2.1, a global context entry invalidation
1850      * should be followed by a IOTLB global invalidation, so we should
1851      * be safe even without this. Hoewever, let's replay the region as
1852      * well to be safer, and go back here when we need finer tunes for
1853      * VT-d emulation codes.
1854      */
1855     vtd_iommu_replay_all(s);
1856 }
1857 
1858 /* Do a context-cache device-selective invalidation.
1859  * @func_mask: FM field after shifting
1860  */
1861 static void vtd_context_device_invalidate(IntelIOMMUState *s,
1862                                           uint16_t source_id,
1863                                           uint16_t func_mask)
1864 {
1865     uint16_t mask;
1866     VTDBus *vtd_bus;
1867     VTDAddressSpace *vtd_as;
1868     uint8_t bus_n, devfn;
1869     uint16_t devfn_it;
1870 
1871     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
1872 
1873     switch (func_mask & 3) {
1874     case 0:
1875         mask = 0;   /* No bits in the SID field masked */
1876         break;
1877     case 1:
1878         mask = 4;   /* Mask bit 2 in the SID field */
1879         break;
1880     case 2:
1881         mask = 6;   /* Mask bit 2:1 in the SID field */
1882         break;
1883     case 3:
1884         mask = 7;   /* Mask bit 2:0 in the SID field */
1885         break;
1886     default:
1887         g_assert_not_reached();
1888     }
1889     mask = ~mask;
1890 
1891     bus_n = VTD_SID_TO_BUS(source_id);
1892     vtd_bus = vtd_find_as_from_bus_num(s, bus_n);
1893     if (vtd_bus) {
1894         devfn = VTD_SID_TO_DEVFN(source_id);
1895         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
1896             vtd_as = vtd_bus->dev_as[devfn_it];
1897             if (vtd_as && ((devfn_it & mask) == (devfn & mask))) {
1898                 trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it),
1899                                              VTD_PCI_FUNC(devfn_it));
1900                 vtd_iommu_lock(s);
1901                 vtd_as->context_cache_entry.context_cache_gen = 0;
1902                 vtd_iommu_unlock(s);
1903                 /*
1904                  * Do switch address space when needed, in case if the
1905                  * device passthrough bit is switched.
1906                  */
1907                 vtd_switch_address_space(vtd_as);
1908                 /*
1909                  * So a device is moving out of (or moving into) a
1910                  * domain, resync the shadow page table.
1911                  * This won't bring bad even if we have no such
1912                  * notifier registered - the IOMMU notification
1913                  * framework will skip MAP notifications if that
1914                  * happened.
1915                  */
1916                 vtd_sync_shadow_page_table(vtd_as);
1917             }
1918         }
1919     }
1920 }
1921 
1922 /* Context-cache invalidation
1923  * Returns the Context Actual Invalidation Granularity.
1924  * @val: the content of the CCMD_REG
1925  */
1926 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
1927 {
1928     uint64_t caig;
1929     uint64_t type = val & VTD_CCMD_CIRG_MASK;
1930 
1931     switch (type) {
1932     case VTD_CCMD_DOMAIN_INVL:
1933         /* Fall through */
1934     case VTD_CCMD_GLOBAL_INVL:
1935         caig = VTD_CCMD_GLOBAL_INVL_A;
1936         vtd_context_global_invalidate(s);
1937         break;
1938 
1939     case VTD_CCMD_DEVICE_INVL:
1940         caig = VTD_CCMD_DEVICE_INVL_A;
1941         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
1942         break;
1943 
1944     default:
1945         error_report_once("%s: invalid context: 0x%" PRIx64,
1946                           __func__, val);
1947         caig = 0;
1948     }
1949     return caig;
1950 }
1951 
1952 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
1953 {
1954     trace_vtd_inv_desc_iotlb_global();
1955     vtd_reset_iotlb(s);
1956     vtd_iommu_replay_all(s);
1957 }
1958 
1959 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
1960 {
1961     VTDContextEntry ce;
1962     VTDAddressSpace *vtd_as;
1963 
1964     trace_vtd_inv_desc_iotlb_domain(domain_id);
1965 
1966     vtd_iommu_lock(s);
1967     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
1968                                 &domain_id);
1969     vtd_iommu_unlock(s);
1970 
1971     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1972         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1973                                       vtd_as->devfn, &ce) &&
1974             domain_id == vtd_get_domain_id(s, &ce)) {
1975             vtd_sync_shadow_page_table(vtd_as);
1976         }
1977     }
1978 }
1979 
1980 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
1981                                            uint16_t domain_id, hwaddr addr,
1982                                            uint8_t am)
1983 {
1984     VTDAddressSpace *vtd_as;
1985     VTDContextEntry ce;
1986     int ret;
1987     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
1988 
1989     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
1990         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1991                                        vtd_as->devfn, &ce);
1992         if (!ret && domain_id == vtd_get_domain_id(s, &ce)) {
1993             if (vtd_as_has_map_notifier(vtd_as)) {
1994                 /*
1995                  * As long as we have MAP notifications registered in
1996                  * any of our IOMMU notifiers, we need to sync the
1997                  * shadow page table.
1998                  */
1999                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
2000             } else {
2001                 /*
2002                  * For UNMAP-only notifiers, we don't need to walk the
2003                  * page tables.  We just deliver the PSI down to
2004                  * invalidate caches.
2005                  */
2006                 IOMMUTLBEvent event = {
2007                     .type = IOMMU_NOTIFIER_UNMAP,
2008                     .entry = {
2009                         .target_as = &address_space_memory,
2010                         .iova = addr,
2011                         .translated_addr = 0,
2012                         .addr_mask = size - 1,
2013                         .perm = IOMMU_NONE,
2014                     },
2015                 };
2016                 memory_region_notify_iommu(&vtd_as->iommu, 0, event);
2017             }
2018         }
2019     }
2020 }
2021 
2022 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
2023                                       hwaddr addr, uint8_t am)
2024 {
2025     VTDIOTLBPageInvInfo info;
2026 
2027     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
2028 
2029     assert(am <= VTD_MAMV);
2030     info.domain_id = domain_id;
2031     info.addr = addr;
2032     info.mask = ~((1 << am) - 1);
2033     vtd_iommu_lock(s);
2034     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
2035     vtd_iommu_unlock(s);
2036     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am);
2037 }
2038 
2039 /* Flush IOTLB
2040  * Returns the IOTLB Actual Invalidation Granularity.
2041  * @val: the content of the IOTLB_REG
2042  */
2043 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
2044 {
2045     uint64_t iaig;
2046     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
2047     uint16_t domain_id;
2048     hwaddr addr;
2049     uint8_t am;
2050 
2051     switch (type) {
2052     case VTD_TLB_GLOBAL_FLUSH:
2053         iaig = VTD_TLB_GLOBAL_FLUSH_A;
2054         vtd_iotlb_global_invalidate(s);
2055         break;
2056 
2057     case VTD_TLB_DSI_FLUSH:
2058         domain_id = VTD_TLB_DID(val);
2059         iaig = VTD_TLB_DSI_FLUSH_A;
2060         vtd_iotlb_domain_invalidate(s, domain_id);
2061         break;
2062 
2063     case VTD_TLB_PSI_FLUSH:
2064         domain_id = VTD_TLB_DID(val);
2065         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
2066         am = VTD_IVA_AM(addr);
2067         addr = VTD_IVA_ADDR(addr);
2068         if (am > VTD_MAMV) {
2069             error_report_once("%s: address mask overflow: 0x%" PRIx64,
2070                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
2071             iaig = 0;
2072             break;
2073         }
2074         iaig = VTD_TLB_PSI_FLUSH_A;
2075         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2076         break;
2077 
2078     default:
2079         error_report_once("%s: invalid granularity: 0x%" PRIx64,
2080                           __func__, val);
2081         iaig = 0;
2082     }
2083     return iaig;
2084 }
2085 
2086 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
2087 
2088 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
2089 {
2090     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
2091            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
2092 }
2093 
2094 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
2095 {
2096     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
2097 
2098     trace_vtd_inv_qi_enable(en);
2099 
2100     if (en) {
2101         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
2102         /* 2^(x+8) entries */
2103         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0));
2104         s->qi_enabled = true;
2105         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
2106         /* Ok - report back to driver */
2107         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
2108 
2109         if (s->iq_tail != 0) {
2110             /*
2111              * This is a spec violation but Windows guests are known to set up
2112              * Queued Invalidation this way so we allow the write and process
2113              * Invalidation Descriptors right away.
2114              */
2115             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
2116             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2117                 vtd_fetch_inv_desc(s);
2118             }
2119         }
2120     } else {
2121         if (vtd_queued_inv_disable_check(s)) {
2122             /* disable Queued Invalidation */
2123             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
2124             s->iq_head = 0;
2125             s->qi_enabled = false;
2126             /* Ok - report back to driver */
2127             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
2128         } else {
2129             error_report_once("%s: detected improper state when disable QI "
2130                               "(head=0x%x, tail=0x%x, last_type=%d)",
2131                               __func__,
2132                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
2133         }
2134     }
2135 }
2136 
2137 /* Set Root Table Pointer */
2138 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
2139 {
2140     vtd_root_table_setup(s);
2141     /* Ok - report back to driver */
2142     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
2143     vtd_reset_caches(s);
2144     vtd_address_space_refresh_all(s);
2145 }
2146 
2147 /* Set Interrupt Remap Table Pointer */
2148 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
2149 {
2150     vtd_interrupt_remap_table_setup(s);
2151     /* Ok - report back to driver */
2152     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
2153 }
2154 
2155 /* Handle Translation Enable/Disable */
2156 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
2157 {
2158     if (s->dmar_enabled == en) {
2159         return;
2160     }
2161 
2162     trace_vtd_dmar_enable(en);
2163 
2164     if (en) {
2165         s->dmar_enabled = true;
2166         /* Ok - report back to driver */
2167         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
2168     } else {
2169         s->dmar_enabled = false;
2170 
2171         /* Clear the index of Fault Recording Register */
2172         s->next_frcd_reg = 0;
2173         /* Ok - report back to driver */
2174         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
2175     }
2176 
2177     vtd_reset_caches(s);
2178     vtd_address_space_refresh_all(s);
2179 }
2180 
2181 /* Handle Interrupt Remap Enable/Disable */
2182 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
2183 {
2184     trace_vtd_ir_enable(en);
2185 
2186     if (en) {
2187         s->intr_enabled = true;
2188         /* Ok - report back to driver */
2189         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
2190     } else {
2191         s->intr_enabled = false;
2192         /* Ok - report back to driver */
2193         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
2194     }
2195 }
2196 
2197 /* Handle write to Global Command Register */
2198 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
2199 {
2200     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
2201     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
2202     uint32_t changed = status ^ val;
2203 
2204     trace_vtd_reg_write_gcmd(status, val);
2205     if (changed & VTD_GCMD_TE) {
2206         /* Translation enable/disable */
2207         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
2208     }
2209     if (val & VTD_GCMD_SRTP) {
2210         /* Set/update the root-table pointer */
2211         vtd_handle_gcmd_srtp(s);
2212     }
2213     if (changed & VTD_GCMD_QIE) {
2214         /* Queued Invalidation Enable */
2215         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
2216     }
2217     if (val & VTD_GCMD_SIRTP) {
2218         /* Set/update the interrupt remapping root-table pointer */
2219         vtd_handle_gcmd_sirtp(s);
2220     }
2221     if (changed & VTD_GCMD_IRE) {
2222         /* Interrupt remap enable/disable */
2223         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
2224     }
2225 }
2226 
2227 /* Handle write to Context Command Register */
2228 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
2229 {
2230     uint64_t ret;
2231     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
2232 
2233     /* Context-cache invalidation request */
2234     if (val & VTD_CCMD_ICC) {
2235         if (s->qi_enabled) {
2236             error_report_once("Queued Invalidation enabled, "
2237                               "should not use register-based invalidation");
2238             return;
2239         }
2240         ret = vtd_context_cache_invalidate(s, val);
2241         /* Invalidation completed. Change something to show */
2242         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
2243         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
2244                                       ret);
2245     }
2246 }
2247 
2248 /* Handle write to IOTLB Invalidation Register */
2249 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
2250 {
2251     uint64_t ret;
2252     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
2253 
2254     /* IOTLB invalidation request */
2255     if (val & VTD_TLB_IVT) {
2256         if (s->qi_enabled) {
2257             error_report_once("Queued Invalidation enabled, "
2258                               "should not use register-based invalidation");
2259             return;
2260         }
2261         ret = vtd_iotlb_flush(s, val);
2262         /* Invalidation completed. Change something to show */
2263         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
2264         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
2265                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
2266     }
2267 }
2268 
2269 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2270 static bool vtd_get_inv_desc(IntelIOMMUState *s,
2271                              VTDInvDesc *inv_desc)
2272 {
2273     dma_addr_t base_addr = s->iq;
2274     uint32_t offset = s->iq_head;
2275     uint32_t dw = s->iq_dw ? 32 : 16;
2276     dma_addr_t addr = base_addr + offset * dw;
2277 
2278     if (dma_memory_read(&address_space_memory, addr, inv_desc, dw)) {
2279         error_report_once("Read INV DESC failed.");
2280         return false;
2281     }
2282     inv_desc->lo = le64_to_cpu(inv_desc->lo);
2283     inv_desc->hi = le64_to_cpu(inv_desc->hi);
2284     if (dw == 32) {
2285         inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]);
2286         inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]);
2287     }
2288     return true;
2289 }
2290 
2291 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2292 {
2293     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
2294         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
2295         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2296                           " (reserved nonzero)", __func__, inv_desc->hi,
2297                           inv_desc->lo);
2298         return false;
2299     }
2300     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
2301         /* Status Write */
2302         uint32_t status_data = (uint32_t)(inv_desc->lo >>
2303                                VTD_INV_DESC_WAIT_DATA_SHIFT);
2304 
2305         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
2306 
2307         /* FIXME: need to be masked with HAW? */
2308         dma_addr_t status_addr = inv_desc->hi;
2309         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
2310         status_data = cpu_to_le32(status_data);
2311         if (dma_memory_write(&address_space_memory, status_addr, &status_data,
2312                              sizeof(status_data))) {
2313             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
2314             return false;
2315         }
2316     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
2317         /* Interrupt flag */
2318         vtd_generate_completion_event(s);
2319     } else {
2320         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2321                           " (unknown type)", __func__, inv_desc->hi,
2322                           inv_desc->lo);
2323         return false;
2324     }
2325     return true;
2326 }
2327 
2328 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
2329                                            VTDInvDesc *inv_desc)
2330 {
2331     uint16_t sid, fmask;
2332 
2333     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
2334         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2335                           " (reserved nonzero)", __func__, inv_desc->hi,
2336                           inv_desc->lo);
2337         return false;
2338     }
2339     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
2340     case VTD_INV_DESC_CC_DOMAIN:
2341         trace_vtd_inv_desc_cc_domain(
2342             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
2343         /* Fall through */
2344     case VTD_INV_DESC_CC_GLOBAL:
2345         vtd_context_global_invalidate(s);
2346         break;
2347 
2348     case VTD_INV_DESC_CC_DEVICE:
2349         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
2350         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
2351         vtd_context_device_invalidate(s, sid, fmask);
2352         break;
2353 
2354     default:
2355         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2356                           " (invalid type)", __func__, inv_desc->hi,
2357                           inv_desc->lo);
2358         return false;
2359     }
2360     return true;
2361 }
2362 
2363 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2364 {
2365     uint16_t domain_id;
2366     uint8_t am;
2367     hwaddr addr;
2368 
2369     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
2370         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
2371         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2372                           ", lo=0x%"PRIx64" (reserved bits unzero)",
2373                           __func__, inv_desc->hi, inv_desc->lo);
2374         return false;
2375     }
2376 
2377     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
2378     case VTD_INV_DESC_IOTLB_GLOBAL:
2379         vtd_iotlb_global_invalidate(s);
2380         break;
2381 
2382     case VTD_INV_DESC_IOTLB_DOMAIN:
2383         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2384         vtd_iotlb_domain_invalidate(s, domain_id);
2385         break;
2386 
2387     case VTD_INV_DESC_IOTLB_PAGE:
2388         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2389         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
2390         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
2391         if (am > VTD_MAMV) {
2392             error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2393                               ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)",
2394                               __func__, inv_desc->hi, inv_desc->lo,
2395                               am, (unsigned)VTD_MAMV);
2396             return false;
2397         }
2398         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2399         break;
2400 
2401     default:
2402         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2403                           ", lo=0x%"PRIx64" (type mismatch: 0x%llx)",
2404                           __func__, inv_desc->hi, inv_desc->lo,
2405                           inv_desc->lo & VTD_INV_DESC_IOTLB_G);
2406         return false;
2407     }
2408     return true;
2409 }
2410 
2411 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
2412                                      VTDInvDesc *inv_desc)
2413 {
2414     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
2415                            inv_desc->iec.index,
2416                            inv_desc->iec.index_mask);
2417 
2418     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
2419                        inv_desc->iec.index,
2420                        inv_desc->iec.index_mask);
2421     return true;
2422 }
2423 
2424 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
2425                                           VTDInvDesc *inv_desc)
2426 {
2427     VTDAddressSpace *vtd_dev_as;
2428     IOMMUTLBEvent event;
2429     struct VTDBus *vtd_bus;
2430     hwaddr addr;
2431     uint64_t sz;
2432     uint16_t sid;
2433     uint8_t devfn;
2434     bool size;
2435     uint8_t bus_num;
2436 
2437     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2438     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2439     devfn = sid & 0xff;
2440     bus_num = sid >> 8;
2441     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2442 
2443     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2444         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2445         error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2446                           ", lo=%"PRIx64" (reserved nonzero)", __func__,
2447                           inv_desc->hi, inv_desc->lo);
2448         return false;
2449     }
2450 
2451     vtd_bus = vtd_find_as_from_bus_num(s, bus_num);
2452     if (!vtd_bus) {
2453         goto done;
2454     }
2455 
2456     vtd_dev_as = vtd_bus->dev_as[devfn];
2457     if (!vtd_dev_as) {
2458         goto done;
2459     }
2460 
2461     /* According to ATS spec table 2.4:
2462      * S = 0, bits 15:12 = xxxx     range size: 4K
2463      * S = 1, bits 15:12 = xxx0     range size: 8K
2464      * S = 1, bits 15:12 = xx01     range size: 16K
2465      * S = 1, bits 15:12 = x011     range size: 32K
2466      * S = 1, bits 15:12 = 0111     range size: 64K
2467      * ...
2468      */
2469     if (size) {
2470         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2471         addr &= ~(sz - 1);
2472     } else {
2473         sz = VTD_PAGE_SIZE;
2474     }
2475 
2476     event.type = IOMMU_NOTIFIER_DEVIOTLB_UNMAP;
2477     event.entry.target_as = &vtd_dev_as->as;
2478     event.entry.addr_mask = sz - 1;
2479     event.entry.iova = addr;
2480     event.entry.perm = IOMMU_NONE;
2481     event.entry.translated_addr = 0;
2482     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, event);
2483 
2484 done:
2485     return true;
2486 }
2487 
2488 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2489 {
2490     VTDInvDesc inv_desc;
2491     uint8_t desc_type;
2492 
2493     trace_vtd_inv_qi_head(s->iq_head);
2494     if (!vtd_get_inv_desc(s, &inv_desc)) {
2495         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2496         return false;
2497     }
2498 
2499     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2500     /* FIXME: should update at first or at last? */
2501     s->iq_last_desc_type = desc_type;
2502 
2503     switch (desc_type) {
2504     case VTD_INV_DESC_CC:
2505         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2506         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2507             return false;
2508         }
2509         break;
2510 
2511     case VTD_INV_DESC_IOTLB:
2512         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2513         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2514             return false;
2515         }
2516         break;
2517 
2518     /*
2519      * TODO: the entity of below two cases will be implemented in future series.
2520      * To make guest (which integrates scalable mode support patch set in
2521      * iommu driver) work, just return true is enough so far.
2522      */
2523     case VTD_INV_DESC_PC:
2524         break;
2525 
2526     case VTD_INV_DESC_PIOTLB:
2527         break;
2528 
2529     case VTD_INV_DESC_WAIT:
2530         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2531         if (!vtd_process_wait_desc(s, &inv_desc)) {
2532             return false;
2533         }
2534         break;
2535 
2536     case VTD_INV_DESC_IEC:
2537         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2538         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2539             return false;
2540         }
2541         break;
2542 
2543     case VTD_INV_DESC_DEVICE:
2544         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2545         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2546             return false;
2547         }
2548         break;
2549 
2550     default:
2551         error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64
2552                           " (unknown type)", __func__, inv_desc.hi,
2553                           inv_desc.lo);
2554         return false;
2555     }
2556     s->iq_head++;
2557     if (s->iq_head == s->iq_size) {
2558         s->iq_head = 0;
2559     }
2560     return true;
2561 }
2562 
2563 /* Try to fetch and process more Invalidation Descriptors */
2564 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2565 {
2566     int qi_shift;
2567 
2568     /* Refer to 10.4.23 of VT-d spec 3.0 */
2569     qi_shift = s->iq_dw ? VTD_IQH_QH_SHIFT_5 : VTD_IQH_QH_SHIFT_4;
2570 
2571     trace_vtd_inv_qi_fetch();
2572 
2573     if (s->iq_tail >= s->iq_size) {
2574         /* Detects an invalid Tail pointer */
2575         error_report_once("%s: detected invalid QI tail "
2576                           "(tail=0x%x, size=0x%x)",
2577                           __func__, s->iq_tail, s->iq_size);
2578         vtd_handle_inv_queue_error(s);
2579         return;
2580     }
2581     while (s->iq_head != s->iq_tail) {
2582         if (!vtd_process_inv_desc(s)) {
2583             /* Invalidation Queue Errors */
2584             vtd_handle_inv_queue_error(s);
2585             break;
2586         }
2587         /* Must update the IQH_REG in time */
2588         vtd_set_quad_raw(s, DMAR_IQH_REG,
2589                          (((uint64_t)(s->iq_head)) << qi_shift) &
2590                          VTD_IQH_QH_MASK);
2591     }
2592 }
2593 
2594 /* Handle write to Invalidation Queue Tail Register */
2595 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2596 {
2597     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2598 
2599     if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) {
2600         error_report_once("%s: RSV bit is set: val=0x%"PRIx64,
2601                           __func__, val);
2602         return;
2603     }
2604     s->iq_tail = VTD_IQT_QT(s->iq_dw, val);
2605     trace_vtd_inv_qi_tail(s->iq_tail);
2606 
2607     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2608         /* Process Invalidation Queue here */
2609         vtd_fetch_inv_desc(s);
2610     }
2611 }
2612 
2613 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2614 {
2615     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2616     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2617     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2618 
2619     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2620         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2621         trace_vtd_fsts_clear_ip();
2622     }
2623     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2624      * Descriptors if there are any when Queued Invalidation is enabled?
2625      */
2626 }
2627 
2628 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2629 {
2630     uint32_t fectl_reg;
2631     /* FIXME: when software clears the IM field, check the IP field. But do we
2632      * need to compare the old value and the new value to conclude that
2633      * software clears the IM field? Or just check if the IM field is zero?
2634      */
2635     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2636 
2637     trace_vtd_reg_write_fectl(fectl_reg);
2638 
2639     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2640         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2641         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2642     }
2643 }
2644 
2645 static void vtd_handle_ics_write(IntelIOMMUState *s)
2646 {
2647     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2648     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2649 
2650     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2651         trace_vtd_reg_ics_clear_ip();
2652         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2653     }
2654 }
2655 
2656 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2657 {
2658     uint32_t iectl_reg;
2659     /* FIXME: when software clears the IM field, check the IP field. But do we
2660      * need to compare the old value and the new value to conclude that
2661      * software clears the IM field? Or just check if the IM field is zero?
2662      */
2663     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2664 
2665     trace_vtd_reg_write_iectl(iectl_reg);
2666 
2667     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2668         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2669         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2670     }
2671 }
2672 
2673 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2674 {
2675     IntelIOMMUState *s = opaque;
2676     uint64_t val;
2677 
2678     trace_vtd_reg_read(addr, size);
2679 
2680     if (addr + size > DMAR_REG_SIZE) {
2681         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2682                           " size=0x%x", __func__, addr, size);
2683         return (uint64_t)-1;
2684     }
2685 
2686     switch (addr) {
2687     /* Root Table Address Register, 64-bit */
2688     case DMAR_RTADDR_REG:
2689         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
2690         if (size == 4) {
2691             val = val & ((1ULL << 32) - 1);
2692         }
2693         break;
2694 
2695     case DMAR_RTADDR_REG_HI:
2696         assert(size == 4);
2697         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG) >> 32;
2698         break;
2699 
2700     /* Invalidation Queue Address Register, 64-bit */
2701     case DMAR_IQA_REG:
2702         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2703         if (size == 4) {
2704             val = val & ((1ULL << 32) - 1);
2705         }
2706         break;
2707 
2708     case DMAR_IQA_REG_HI:
2709         assert(size == 4);
2710         val = s->iq >> 32;
2711         break;
2712 
2713     default:
2714         if (size == 4) {
2715             val = vtd_get_long(s, addr);
2716         } else {
2717             val = vtd_get_quad(s, addr);
2718         }
2719     }
2720 
2721     return val;
2722 }
2723 
2724 static void vtd_mem_write(void *opaque, hwaddr addr,
2725                           uint64_t val, unsigned size)
2726 {
2727     IntelIOMMUState *s = opaque;
2728 
2729     trace_vtd_reg_write(addr, size, val);
2730 
2731     if (addr + size > DMAR_REG_SIZE) {
2732         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2733                           " size=0x%x", __func__, addr, size);
2734         return;
2735     }
2736 
2737     switch (addr) {
2738     /* Global Command Register, 32-bit */
2739     case DMAR_GCMD_REG:
2740         vtd_set_long(s, addr, val);
2741         vtd_handle_gcmd_write(s);
2742         break;
2743 
2744     /* Context Command Register, 64-bit */
2745     case DMAR_CCMD_REG:
2746         if (size == 4) {
2747             vtd_set_long(s, addr, val);
2748         } else {
2749             vtd_set_quad(s, addr, val);
2750             vtd_handle_ccmd_write(s);
2751         }
2752         break;
2753 
2754     case DMAR_CCMD_REG_HI:
2755         assert(size == 4);
2756         vtd_set_long(s, addr, val);
2757         vtd_handle_ccmd_write(s);
2758         break;
2759 
2760     /* IOTLB Invalidation Register, 64-bit */
2761     case DMAR_IOTLB_REG:
2762         if (size == 4) {
2763             vtd_set_long(s, addr, val);
2764         } else {
2765             vtd_set_quad(s, addr, val);
2766             vtd_handle_iotlb_write(s);
2767         }
2768         break;
2769 
2770     case DMAR_IOTLB_REG_HI:
2771         assert(size == 4);
2772         vtd_set_long(s, addr, val);
2773         vtd_handle_iotlb_write(s);
2774         break;
2775 
2776     /* Invalidate Address Register, 64-bit */
2777     case DMAR_IVA_REG:
2778         if (size == 4) {
2779             vtd_set_long(s, addr, val);
2780         } else {
2781             vtd_set_quad(s, addr, val);
2782         }
2783         break;
2784 
2785     case DMAR_IVA_REG_HI:
2786         assert(size == 4);
2787         vtd_set_long(s, addr, val);
2788         break;
2789 
2790     /* Fault Status Register, 32-bit */
2791     case DMAR_FSTS_REG:
2792         assert(size == 4);
2793         vtd_set_long(s, addr, val);
2794         vtd_handle_fsts_write(s);
2795         break;
2796 
2797     /* Fault Event Control Register, 32-bit */
2798     case DMAR_FECTL_REG:
2799         assert(size == 4);
2800         vtd_set_long(s, addr, val);
2801         vtd_handle_fectl_write(s);
2802         break;
2803 
2804     /* Fault Event Data Register, 32-bit */
2805     case DMAR_FEDATA_REG:
2806         assert(size == 4);
2807         vtd_set_long(s, addr, val);
2808         break;
2809 
2810     /* Fault Event Address Register, 32-bit */
2811     case DMAR_FEADDR_REG:
2812         if (size == 4) {
2813             vtd_set_long(s, addr, val);
2814         } else {
2815             /*
2816              * While the register is 32-bit only, some guests (Xen...) write to
2817              * it with 64-bit.
2818              */
2819             vtd_set_quad(s, addr, val);
2820         }
2821         break;
2822 
2823     /* Fault Event Upper Address Register, 32-bit */
2824     case DMAR_FEUADDR_REG:
2825         assert(size == 4);
2826         vtd_set_long(s, addr, val);
2827         break;
2828 
2829     /* Protected Memory Enable Register, 32-bit */
2830     case DMAR_PMEN_REG:
2831         assert(size == 4);
2832         vtd_set_long(s, addr, val);
2833         break;
2834 
2835     /* Root Table Address Register, 64-bit */
2836     case DMAR_RTADDR_REG:
2837         if (size == 4) {
2838             vtd_set_long(s, addr, val);
2839         } else {
2840             vtd_set_quad(s, addr, val);
2841         }
2842         break;
2843 
2844     case DMAR_RTADDR_REG_HI:
2845         assert(size == 4);
2846         vtd_set_long(s, addr, val);
2847         break;
2848 
2849     /* Invalidation Queue Tail Register, 64-bit */
2850     case DMAR_IQT_REG:
2851         if (size == 4) {
2852             vtd_set_long(s, addr, val);
2853         } else {
2854             vtd_set_quad(s, addr, val);
2855         }
2856         vtd_handle_iqt_write(s);
2857         break;
2858 
2859     case DMAR_IQT_REG_HI:
2860         assert(size == 4);
2861         vtd_set_long(s, addr, val);
2862         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
2863         break;
2864 
2865     /* Invalidation Queue Address Register, 64-bit */
2866     case DMAR_IQA_REG:
2867         if (size == 4) {
2868             vtd_set_long(s, addr, val);
2869         } else {
2870             vtd_set_quad(s, addr, val);
2871         }
2872         if (s->ecap & VTD_ECAP_SMTS &&
2873             val & VTD_IQA_DW_MASK) {
2874             s->iq_dw = true;
2875         } else {
2876             s->iq_dw = false;
2877         }
2878         break;
2879 
2880     case DMAR_IQA_REG_HI:
2881         assert(size == 4);
2882         vtd_set_long(s, addr, val);
2883         break;
2884 
2885     /* Invalidation Completion Status Register, 32-bit */
2886     case DMAR_ICS_REG:
2887         assert(size == 4);
2888         vtd_set_long(s, addr, val);
2889         vtd_handle_ics_write(s);
2890         break;
2891 
2892     /* Invalidation Event Control Register, 32-bit */
2893     case DMAR_IECTL_REG:
2894         assert(size == 4);
2895         vtd_set_long(s, addr, val);
2896         vtd_handle_iectl_write(s);
2897         break;
2898 
2899     /* Invalidation Event Data Register, 32-bit */
2900     case DMAR_IEDATA_REG:
2901         assert(size == 4);
2902         vtd_set_long(s, addr, val);
2903         break;
2904 
2905     /* Invalidation Event Address Register, 32-bit */
2906     case DMAR_IEADDR_REG:
2907         assert(size == 4);
2908         vtd_set_long(s, addr, val);
2909         break;
2910 
2911     /* Invalidation Event Upper Address Register, 32-bit */
2912     case DMAR_IEUADDR_REG:
2913         assert(size == 4);
2914         vtd_set_long(s, addr, val);
2915         break;
2916 
2917     /* Fault Recording Registers, 128-bit */
2918     case DMAR_FRCD_REG_0_0:
2919         if (size == 4) {
2920             vtd_set_long(s, addr, val);
2921         } else {
2922             vtd_set_quad(s, addr, val);
2923         }
2924         break;
2925 
2926     case DMAR_FRCD_REG_0_1:
2927         assert(size == 4);
2928         vtd_set_long(s, addr, val);
2929         break;
2930 
2931     case DMAR_FRCD_REG_0_2:
2932         if (size == 4) {
2933             vtd_set_long(s, addr, val);
2934         } else {
2935             vtd_set_quad(s, addr, val);
2936             /* May clear bit 127 (Fault), update PPF */
2937             vtd_update_fsts_ppf(s);
2938         }
2939         break;
2940 
2941     case DMAR_FRCD_REG_0_3:
2942         assert(size == 4);
2943         vtd_set_long(s, addr, val);
2944         /* May clear bit 127 (Fault), update PPF */
2945         vtd_update_fsts_ppf(s);
2946         break;
2947 
2948     case DMAR_IRTA_REG:
2949         if (size == 4) {
2950             vtd_set_long(s, addr, val);
2951         } else {
2952             vtd_set_quad(s, addr, val);
2953         }
2954         break;
2955 
2956     case DMAR_IRTA_REG_HI:
2957         assert(size == 4);
2958         vtd_set_long(s, addr, val);
2959         break;
2960 
2961     default:
2962         if (size == 4) {
2963             vtd_set_long(s, addr, val);
2964         } else {
2965             vtd_set_quad(s, addr, val);
2966         }
2967     }
2968 }
2969 
2970 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
2971                                          IOMMUAccessFlags flag, int iommu_idx)
2972 {
2973     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2974     IntelIOMMUState *s = vtd_as->iommu_state;
2975     IOMMUTLBEntry iotlb = {
2976         /* We'll fill in the rest later. */
2977         .target_as = &address_space_memory,
2978     };
2979     bool success;
2980 
2981     if (likely(s->dmar_enabled)) {
2982         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
2983                                          addr, flag & IOMMU_WO, &iotlb);
2984     } else {
2985         /* DMAR disabled, passthrough, use 4k-page*/
2986         iotlb.iova = addr & VTD_PAGE_MASK_4K;
2987         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
2988         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
2989         iotlb.perm = IOMMU_RW;
2990         success = true;
2991     }
2992 
2993     if (likely(success)) {
2994         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
2995                                  VTD_PCI_SLOT(vtd_as->devfn),
2996                                  VTD_PCI_FUNC(vtd_as->devfn),
2997                                  iotlb.iova, iotlb.translated_addr,
2998                                  iotlb.addr_mask);
2999     } else {
3000         error_report_once("%s: detected translation failure "
3001                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
3002                           __func__, pci_bus_num(vtd_as->bus),
3003                           VTD_PCI_SLOT(vtd_as->devfn),
3004                           VTD_PCI_FUNC(vtd_as->devfn),
3005                           addr);
3006     }
3007 
3008     return iotlb;
3009 }
3010 
3011 static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
3012                                          IOMMUNotifierFlag old,
3013                                          IOMMUNotifierFlag new,
3014                                          Error **errp)
3015 {
3016     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
3017     IntelIOMMUState *s = vtd_as->iommu_state;
3018 
3019     /* Update per-address-space notifier flags */
3020     vtd_as->notifier_flags = new;
3021 
3022     if (old == IOMMU_NOTIFIER_NONE) {
3023         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
3024     } else if (new == IOMMU_NOTIFIER_NONE) {
3025         QLIST_REMOVE(vtd_as, next);
3026     }
3027     return 0;
3028 }
3029 
3030 static int vtd_post_load(void *opaque, int version_id)
3031 {
3032     IntelIOMMUState *iommu = opaque;
3033 
3034     /*
3035      * Memory regions are dynamically turned on/off depending on
3036      * context entry configurations from the guest. After migration,
3037      * we need to make sure the memory regions are still correct.
3038      */
3039     vtd_switch_address_space_all(iommu);
3040 
3041     /*
3042      * We don't need to migrate the root_scalable because we can
3043      * simply do the calculation after the loading is complete.  We
3044      * can actually do similar things with root, dmar_enabled, etc.
3045      * however since we've had them already so we'd better keep them
3046      * for compatibility of migration.
3047      */
3048     vtd_update_scalable_state(iommu);
3049 
3050     return 0;
3051 }
3052 
3053 static const VMStateDescription vtd_vmstate = {
3054     .name = "iommu-intel",
3055     .version_id = 1,
3056     .minimum_version_id = 1,
3057     .priority = MIG_PRI_IOMMU,
3058     .post_load = vtd_post_load,
3059     .fields = (VMStateField[]) {
3060         VMSTATE_UINT64(root, IntelIOMMUState),
3061         VMSTATE_UINT64(intr_root, IntelIOMMUState),
3062         VMSTATE_UINT64(iq, IntelIOMMUState),
3063         VMSTATE_UINT32(intr_size, IntelIOMMUState),
3064         VMSTATE_UINT16(iq_head, IntelIOMMUState),
3065         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
3066         VMSTATE_UINT16(iq_size, IntelIOMMUState),
3067         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
3068         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
3069         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
3070         VMSTATE_UNUSED(1),      /* bool root_extended is obsolete by VT-d */
3071         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
3072         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
3073         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
3074         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
3075         VMSTATE_END_OF_LIST()
3076     }
3077 };
3078 
3079 static const MemoryRegionOps vtd_mem_ops = {
3080     .read = vtd_mem_read,
3081     .write = vtd_mem_write,
3082     .endianness = DEVICE_LITTLE_ENDIAN,
3083     .impl = {
3084         .min_access_size = 4,
3085         .max_access_size = 8,
3086     },
3087     .valid = {
3088         .min_access_size = 4,
3089         .max_access_size = 8,
3090     },
3091 };
3092 
3093 static Property vtd_properties[] = {
3094     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
3095     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
3096                             ON_OFF_AUTO_AUTO),
3097     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
3098     DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits,
3099                       VTD_HOST_ADDRESS_WIDTH),
3100     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
3101     DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE),
3102     DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true),
3103     DEFINE_PROP_END_OF_LIST(),
3104 };
3105 
3106 /* Read IRTE entry with specific index */
3107 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
3108                         VTD_IR_TableEntry *entry, uint16_t sid)
3109 {
3110     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
3111         {0xffff, 0xfffb, 0xfff9, 0xfff8};
3112     dma_addr_t addr = 0x00;
3113     uint16_t mask, source_id;
3114     uint8_t bus, bus_max, bus_min;
3115 
3116     if (index >= iommu->intr_size) {
3117         error_report_once("%s: index too large: ind=0x%x",
3118                           __func__, index);
3119         return -VTD_FR_IR_INDEX_OVER;
3120     }
3121 
3122     addr = iommu->intr_root + index * sizeof(*entry);
3123     if (dma_memory_read(&address_space_memory, addr, entry,
3124                         sizeof(*entry))) {
3125         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
3126                           __func__, index, addr);
3127         return -VTD_FR_IR_ROOT_INVAL;
3128     }
3129 
3130     trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]),
3131                           le64_to_cpu(entry->data[0]));
3132 
3133     if (!entry->irte.present) {
3134         error_report_once("%s: detected non-present IRTE "
3135                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3136                           __func__, index, le64_to_cpu(entry->data[1]),
3137                           le64_to_cpu(entry->data[0]));
3138         return -VTD_FR_IR_ENTRY_P;
3139     }
3140 
3141     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
3142         entry->irte.__reserved_2) {
3143         error_report_once("%s: detected non-zero reserved IRTE "
3144                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3145                           __func__, index, le64_to_cpu(entry->data[1]),
3146                           le64_to_cpu(entry->data[0]));
3147         return -VTD_FR_IR_IRTE_RSVD;
3148     }
3149 
3150     if (sid != X86_IOMMU_SID_INVALID) {
3151         /* Validate IRTE SID */
3152         source_id = le32_to_cpu(entry->irte.source_id);
3153         switch (entry->irte.sid_vtype) {
3154         case VTD_SVT_NONE:
3155             break;
3156 
3157         case VTD_SVT_ALL:
3158             mask = vtd_svt_mask[entry->irte.sid_q];
3159             if ((source_id & mask) != (sid & mask)) {
3160                 error_report_once("%s: invalid IRTE SID "
3161                                   "(index=%u, sid=%u, source_id=%u)",
3162                                   __func__, index, sid, source_id);
3163                 return -VTD_FR_IR_SID_ERR;
3164             }
3165             break;
3166 
3167         case VTD_SVT_BUS:
3168             bus_max = source_id >> 8;
3169             bus_min = source_id & 0xff;
3170             bus = sid >> 8;
3171             if (bus > bus_max || bus < bus_min) {
3172                 error_report_once("%s: invalid SVT_BUS "
3173                                   "(index=%u, bus=%u, min=%u, max=%u)",
3174                                   __func__, index, bus, bus_min, bus_max);
3175                 return -VTD_FR_IR_SID_ERR;
3176             }
3177             break;
3178 
3179         default:
3180             error_report_once("%s: detected invalid IRTE SVT "
3181                               "(index=%u, type=%d)", __func__,
3182                               index, entry->irte.sid_vtype);
3183             /* Take this as verification failure. */
3184             return -VTD_FR_IR_SID_ERR;
3185         }
3186     }
3187 
3188     return 0;
3189 }
3190 
3191 /* Fetch IRQ information of specific IR index */
3192 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
3193                              X86IOMMUIrq *irq, uint16_t sid)
3194 {
3195     VTD_IR_TableEntry irte = {};
3196     int ret = 0;
3197 
3198     ret = vtd_irte_get(iommu, index, &irte, sid);
3199     if (ret) {
3200         return ret;
3201     }
3202 
3203     irq->trigger_mode = irte.irte.trigger_mode;
3204     irq->vector = irte.irte.vector;
3205     irq->delivery_mode = irte.irte.delivery_mode;
3206     irq->dest = le32_to_cpu(irte.irte.dest_id);
3207     if (!iommu->intr_eime) {
3208 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
3209 #define  VTD_IR_APIC_DEST_SHIFT        (8)
3210         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
3211             VTD_IR_APIC_DEST_SHIFT;
3212     }
3213     irq->dest_mode = irte.irte.dest_mode;
3214     irq->redir_hint = irte.irte.redir_hint;
3215 
3216     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
3217                        irq->delivery_mode, irq->dest, irq->dest_mode);
3218 
3219     return 0;
3220 }
3221 
3222 /* Interrupt remapping for MSI/MSI-X entry */
3223 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
3224                                    MSIMessage *origin,
3225                                    MSIMessage *translated,
3226                                    uint16_t sid)
3227 {
3228     int ret = 0;
3229     VTD_IR_MSIAddress addr;
3230     uint16_t index;
3231     X86IOMMUIrq irq = {};
3232 
3233     assert(origin && translated);
3234 
3235     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
3236 
3237     if (!iommu || !iommu->intr_enabled) {
3238         memcpy(translated, origin, sizeof(*origin));
3239         goto out;
3240     }
3241 
3242     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
3243         error_report_once("%s: MSI address high 32 bits non-zero detected: "
3244                           "address=0x%" PRIx64, __func__, origin->address);
3245         return -VTD_FR_IR_REQ_RSVD;
3246     }
3247 
3248     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
3249     if (addr.addr.__head != 0xfee) {
3250         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
3251                           __func__, addr.data);
3252         return -VTD_FR_IR_REQ_RSVD;
3253     }
3254 
3255     /* This is compatible mode. */
3256     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
3257         memcpy(translated, origin, sizeof(*origin));
3258         goto out;
3259     }
3260 
3261     index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l);
3262 
3263 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
3264 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
3265 
3266     if (addr.addr.sub_valid) {
3267         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3268         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
3269     }
3270 
3271     ret = vtd_remap_irq_get(iommu, index, &irq, sid);
3272     if (ret) {
3273         return ret;
3274     }
3275 
3276     if (addr.addr.sub_valid) {
3277         trace_vtd_ir_remap_type("MSI");
3278         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
3279             error_report_once("%s: invalid IR MSI "
3280                               "(sid=%u, address=0x%" PRIx64
3281                               ", data=0x%" PRIx32 ")",
3282                               __func__, sid, origin->address, origin->data);
3283             return -VTD_FR_IR_REQ_RSVD;
3284         }
3285     } else {
3286         uint8_t vector = origin->data & 0xff;
3287         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
3288 
3289         trace_vtd_ir_remap_type("IOAPIC");
3290         /* IOAPIC entry vector should be aligned with IRTE vector
3291          * (see vt-d spec 5.1.5.1). */
3292         if (vector != irq.vector) {
3293             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
3294         }
3295 
3296         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3297          * (see vt-d spec 5.1.5.1). */
3298         if (trigger_mode != irq.trigger_mode) {
3299             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
3300                                       irq.trigger_mode);
3301         }
3302     }
3303 
3304     /*
3305      * We'd better keep the last two bits, assuming that guest OS
3306      * might modify it. Keep it does not hurt after all.
3307      */
3308     irq.msi_addr_last_bits = addr.addr.__not_care;
3309 
3310     /* Translate X86IOMMUIrq to MSI message */
3311     x86_iommu_irq_to_msi_message(&irq, translated);
3312 
3313 out:
3314     trace_vtd_ir_remap_msi(origin->address, origin->data,
3315                            translated->address, translated->data);
3316     return 0;
3317 }
3318 
3319 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
3320                          MSIMessage *dst, uint16_t sid)
3321 {
3322     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
3323                                    src, dst, sid);
3324 }
3325 
3326 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
3327                                    uint64_t *data, unsigned size,
3328                                    MemTxAttrs attrs)
3329 {
3330     return MEMTX_OK;
3331 }
3332 
3333 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
3334                                     uint64_t value, unsigned size,
3335                                     MemTxAttrs attrs)
3336 {
3337     int ret = 0;
3338     MSIMessage from = {}, to = {};
3339     uint16_t sid = X86_IOMMU_SID_INVALID;
3340 
3341     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
3342     from.data = (uint32_t) value;
3343 
3344     if (!attrs.unspecified) {
3345         /* We have explicit Source ID */
3346         sid = attrs.requester_id;
3347     }
3348 
3349     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid);
3350     if (ret) {
3351         /* TODO: report error */
3352         /* Drop this interrupt */
3353         return MEMTX_ERROR;
3354     }
3355 
3356     apic_get_class()->send_msi(&to);
3357 
3358     return MEMTX_OK;
3359 }
3360 
3361 static const MemoryRegionOps vtd_mem_ir_ops = {
3362     .read_with_attrs = vtd_mem_ir_read,
3363     .write_with_attrs = vtd_mem_ir_write,
3364     .endianness = DEVICE_LITTLE_ENDIAN,
3365     .impl = {
3366         .min_access_size = 4,
3367         .max_access_size = 4,
3368     },
3369     .valid = {
3370         .min_access_size = 4,
3371         .max_access_size = 4,
3372     },
3373 };
3374 
3375 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
3376 {
3377     uintptr_t key = (uintptr_t)bus;
3378     VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
3379     VTDAddressSpace *vtd_dev_as;
3380     char name[128];
3381 
3382     if (!vtd_bus) {
3383         uintptr_t *new_key = g_malloc(sizeof(*new_key));
3384         *new_key = (uintptr_t)bus;
3385         /* No corresponding free() */
3386         vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
3387                             PCI_DEVFN_MAX);
3388         vtd_bus->bus = bus;
3389         g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
3390     }
3391 
3392     vtd_dev_as = vtd_bus->dev_as[devfn];
3393 
3394     if (!vtd_dev_as) {
3395         snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn),
3396                  PCI_FUNC(devfn));
3397         vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace));
3398 
3399         vtd_dev_as->bus = bus;
3400         vtd_dev_as->devfn = (uint8_t)devfn;
3401         vtd_dev_as->iommu_state = s;
3402         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
3403         vtd_dev_as->iova_tree = iova_tree_new();
3404 
3405         memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX);
3406         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root");
3407 
3408         /*
3409          * Build the DMAR-disabled container with aliases to the
3410          * shared MRs.  Note that aliasing to a shared memory region
3411          * could help the memory API to detect same FlatViews so we
3412          * can have devices to share the same FlatView when DMAR is
3413          * disabled (either by not providing "intel_iommu=on" or with
3414          * "iommu=pt").  It will greatly reduce the total number of
3415          * FlatViews of the system hence VM runs faster.
3416          */
3417         memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s),
3418                                  "vtd-nodmar", &s->mr_nodmar, 0,
3419                                  memory_region_size(&s->mr_nodmar));
3420 
3421         /*
3422          * Build the per-device DMAR-enabled container.
3423          *
3424          * TODO: currently we have per-device IOMMU memory region only
3425          * because we have per-device IOMMU notifiers for devices.  If
3426          * one day we can abstract the IOMMU notifiers out of the
3427          * memory regions then we can also share the same memory
3428          * region here just like what we've done above with the nodmar
3429          * region.
3430          */
3431         strcat(name, "-dmar");
3432         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
3433                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
3434                                  name, UINT64_MAX);
3435         memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir",
3436                                  &s->mr_ir, 0, memory_region_size(&s->mr_ir));
3437         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
3438                                             VTD_INTERRUPT_ADDR_FIRST,
3439                                             &vtd_dev_as->iommu_ir, 1);
3440 
3441         /*
3442          * Hook both the containers under the root container, we
3443          * switch between DMAR & noDMAR by enable/disable
3444          * corresponding sub-containers
3445          */
3446         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3447                                             MEMORY_REGION(&vtd_dev_as->iommu),
3448                                             0);
3449         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3450                                             &vtd_dev_as->nodmar, 0);
3451 
3452         vtd_switch_address_space(vtd_dev_as);
3453     }
3454     return vtd_dev_as;
3455 }
3456 
3457 /* Unmap the whole range in the notifier's scope. */
3458 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
3459 {
3460     hwaddr size, remain;
3461     hwaddr start = n->start;
3462     hwaddr end = n->end;
3463     IntelIOMMUState *s = as->iommu_state;
3464     DMAMap map;
3465 
3466     /*
3467      * Note: all the codes in this function has a assumption that IOVA
3468      * bits are no more than VTD_MGAW bits (which is restricted by
3469      * VT-d spec), otherwise we need to consider overflow of 64 bits.
3470      */
3471 
3472     if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) {
3473         /*
3474          * Don't need to unmap regions that is bigger than the whole
3475          * VT-d supported address space size
3476          */
3477         end = VTD_ADDRESS_SIZE(s->aw_bits) - 1;
3478     }
3479 
3480     assert(start <= end);
3481     size = remain = end - start + 1;
3482 
3483     while (remain >= VTD_PAGE_SIZE) {
3484         IOMMUTLBEvent event;
3485         uint64_t mask = dma_aligned_pow2_mask(start, end, s->aw_bits);
3486         uint64_t size = mask + 1;
3487 
3488         assert(size);
3489 
3490         event.type = IOMMU_NOTIFIER_UNMAP;
3491         event.entry.iova = start;
3492         event.entry.addr_mask = mask;
3493         event.entry.target_as = &address_space_memory;
3494         event.entry.perm = IOMMU_NONE;
3495         /* This field is meaningless for unmap */
3496         event.entry.translated_addr = 0;
3497 
3498         memory_region_notify_iommu_one(n, &event);
3499 
3500         start += size;
3501         remain -= size;
3502     }
3503 
3504     assert(!remain);
3505 
3506     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3507                              VTD_PCI_SLOT(as->devfn),
3508                              VTD_PCI_FUNC(as->devfn),
3509                              n->start, size);
3510 
3511     map.iova = n->start;
3512     map.size = size;
3513     iova_tree_remove(as->iova_tree, &map);
3514 }
3515 
3516 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3517 {
3518     VTDAddressSpace *vtd_as;
3519     IOMMUNotifier *n;
3520 
3521     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3522         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3523             vtd_address_space_unmap(vtd_as, n);
3524         }
3525     }
3526 }
3527 
3528 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3529 {
3530     vtd_address_space_unmap_all(s);
3531     vtd_switch_address_space_all(s);
3532 }
3533 
3534 static int vtd_replay_hook(IOMMUTLBEvent *event, void *private)
3535 {
3536     memory_region_notify_iommu_one(private, event);
3537     return 0;
3538 }
3539 
3540 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3541 {
3542     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3543     IntelIOMMUState *s = vtd_as->iommu_state;
3544     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3545     VTDContextEntry ce;
3546 
3547     /*
3548      * The replay can be triggered by either a invalidation or a newly
3549      * created entry. No matter what, we release existing mappings
3550      * (it means flushing caches for UNMAP-only registers).
3551      */
3552     vtd_address_space_unmap(vtd_as, n);
3553 
3554     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3555         trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" :
3556                                   "legacy mode",
3557                                   bus_n, PCI_SLOT(vtd_as->devfn),
3558                                   PCI_FUNC(vtd_as->devfn),
3559                                   vtd_get_domain_id(s, &ce),
3560                                   ce.hi, ce.lo);
3561         if (vtd_as_has_map_notifier(vtd_as)) {
3562             /* This is required only for MAP typed notifiers */
3563             vtd_page_walk_info info = {
3564                 .hook_fn = vtd_replay_hook,
3565                 .private = (void *)n,
3566                 .notify_unmap = false,
3567                 .aw = s->aw_bits,
3568                 .as = vtd_as,
3569                 .domain_id = vtd_get_domain_id(s, &ce),
3570             };
3571 
3572             vtd_page_walk(s, &ce, 0, ~0ULL, &info);
3573         }
3574     } else {
3575         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3576                                     PCI_FUNC(vtd_as->devfn));
3577     }
3578 
3579     return;
3580 }
3581 
3582 /* Do the initialization. It will also be called when reset, so pay
3583  * attention when adding new initialization stuff.
3584  */
3585 static void vtd_init(IntelIOMMUState *s)
3586 {
3587     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3588 
3589     memset(s->csr, 0, DMAR_REG_SIZE);
3590     memset(s->wmask, 0, DMAR_REG_SIZE);
3591     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3592     memset(s->womask, 0, DMAR_REG_SIZE);
3593 
3594     s->root = 0;
3595     s->root_scalable = false;
3596     s->dmar_enabled = false;
3597     s->intr_enabled = false;
3598     s->iq_head = 0;
3599     s->iq_tail = 0;
3600     s->iq = 0;
3601     s->iq_size = 0;
3602     s->qi_enabled = false;
3603     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3604     s->iq_dw = false;
3605     s->next_frcd_reg = 0;
3606     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3607              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3608              VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits);
3609     if (s->dma_drain) {
3610         s->cap |= VTD_CAP_DRAIN;
3611     }
3612     if (s->aw_bits == VTD_HOST_AW_48BIT) {
3613         s->cap |= VTD_CAP_SAGAW_48bit;
3614     }
3615     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3616 
3617     /*
3618      * Rsvd field masks for spte
3619      */
3620     vtd_spte_rsvd[0] = ~0ULL;
3621     vtd_spte_rsvd[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits,
3622                                                   x86_iommu->dt_supported);
3623     vtd_spte_rsvd[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3624     vtd_spte_rsvd[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3625     vtd_spte_rsvd[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3626 
3627     vtd_spte_rsvd_large[2] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits,
3628                                                          x86_iommu->dt_supported);
3629     vtd_spte_rsvd_large[3] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits,
3630                                                          x86_iommu->dt_supported);
3631 
3632     if (x86_iommu_ir_supported(x86_iommu)) {
3633         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3634         if (s->intr_eim == ON_OFF_AUTO_ON) {
3635             s->ecap |= VTD_ECAP_EIM;
3636         }
3637         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3638     }
3639 
3640     if (x86_iommu->dt_supported) {
3641         s->ecap |= VTD_ECAP_DT;
3642     }
3643 
3644     if (x86_iommu->pt_supported) {
3645         s->ecap |= VTD_ECAP_PT;
3646     }
3647 
3648     if (s->caching_mode) {
3649         s->cap |= VTD_CAP_CM;
3650     }
3651 
3652     /* TODO: read cap/ecap from host to decide which cap to be exposed. */
3653     if (s->scalable_mode) {
3654         s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS;
3655     }
3656 
3657     vtd_reset_caches(s);
3658 
3659     /* Define registers with default values and bit semantics */
3660     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
3661     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
3662     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
3663     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
3664     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
3665     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
3666     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0);
3667     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
3668     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
3669 
3670     /* Advanced Fault Logging not supported */
3671     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
3672     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3673     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
3674     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
3675 
3676     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3677      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3678      */
3679     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
3680 
3681     /* Treated as RO for implementations that PLMR and PHMR fields reported
3682      * as Clear in the CAP_REG.
3683      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3684      */
3685     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
3686 
3687     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
3688     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
3689     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0);
3690     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
3691     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3692     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
3693     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
3694     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3695     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
3696 
3697     /* IOTLB registers */
3698     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
3699     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
3700     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
3701 
3702     /* Fault Recording Registers, 128-bit */
3703     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
3704     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
3705 
3706     /*
3707      * Interrupt remapping registers.
3708      */
3709     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
3710 }
3711 
3712 /* Should not reset address_spaces when reset because devices will still use
3713  * the address space they got at first (won't ask the bus again).
3714  */
3715 static void vtd_reset(DeviceState *dev)
3716 {
3717     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3718 
3719     vtd_init(s);
3720     vtd_address_space_refresh_all(s);
3721 }
3722 
3723 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
3724 {
3725     IntelIOMMUState *s = opaque;
3726     VTDAddressSpace *vtd_as;
3727 
3728     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
3729 
3730     vtd_as = vtd_find_add_as(s, bus, devfn);
3731     return &vtd_as->as;
3732 }
3733 
3734 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
3735 {
3736     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3737 
3738     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) {
3739         error_setg(errp, "eim=on cannot be selected without intremap=on");
3740         return false;
3741     }
3742 
3743     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
3744         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
3745                       && x86_iommu_ir_supported(x86_iommu) ?
3746                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
3747     }
3748     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
3749         if (!kvm_irqchip_in_kernel()) {
3750             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
3751             return false;
3752         }
3753         if (!kvm_enable_x2apic()) {
3754             error_setg(errp, "eim=on requires support on the KVM side"
3755                              "(X2APIC_API, first shipped in v4.7)");
3756             return false;
3757         }
3758     }
3759 
3760     /* Currently only address widths supported are 39 and 48 bits */
3761     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
3762         (s->aw_bits != VTD_HOST_AW_48BIT)) {
3763         error_setg(errp, "Supported values for aw-bits are: %d, %d",
3764                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
3765         return false;
3766     }
3767 
3768     if (s->scalable_mode && !s->dma_drain) {
3769         error_setg(errp, "Need to set dma_drain for scalable mode");
3770         return false;
3771     }
3772 
3773     return true;
3774 }
3775 
3776 static int vtd_machine_done_notify_one(Object *child, void *unused)
3777 {
3778     IntelIOMMUState *iommu = INTEL_IOMMU_DEVICE(x86_iommu_get_default());
3779 
3780     /*
3781      * We hard-coded here because vfio-pci is the only special case
3782      * here.  Let's be more elegant in the future when we can, but so
3783      * far there seems to be no better way.
3784      */
3785     if (object_dynamic_cast(child, "vfio-pci") && !iommu->caching_mode) {
3786         vtd_panic_require_caching_mode();
3787     }
3788 
3789     return 0;
3790 }
3791 
3792 static void vtd_machine_done_hook(Notifier *notifier, void *unused)
3793 {
3794     object_child_foreach_recursive(object_get_root(),
3795                                    vtd_machine_done_notify_one, NULL);
3796 }
3797 
3798 static Notifier vtd_machine_done_notify = {
3799     .notify = vtd_machine_done_hook,
3800 };
3801 
3802 static void vtd_realize(DeviceState *dev, Error **errp)
3803 {
3804     MachineState *ms = MACHINE(qdev_get_machine());
3805     PCMachineState *pcms = PC_MACHINE(ms);
3806     X86MachineState *x86ms = X86_MACHINE(ms);
3807     PCIBus *bus = pcms->bus;
3808     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3809     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
3810 
3811     x86_iommu->type = TYPE_INTEL;
3812 
3813     if (!vtd_decide_config(s, errp)) {
3814         return;
3815     }
3816 
3817     QLIST_INIT(&s->vtd_as_with_notifiers);
3818     qemu_mutex_init(&s->iommu_lock);
3819     memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num));
3820     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
3821                           "intel_iommu", DMAR_REG_SIZE);
3822 
3823     /* Create the shared memory regions by all devices */
3824     memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar",
3825                        UINT64_MAX);
3826     memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops,
3827                           s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE);
3828     memory_region_init_alias(&s->mr_sys_alias, OBJECT(s),
3829                              "vtd-sys-alias", get_system_memory(), 0,
3830                              memory_region_size(get_system_memory()));
3831     memory_region_add_subregion_overlap(&s->mr_nodmar, 0,
3832                                         &s->mr_sys_alias, 0);
3833     memory_region_add_subregion_overlap(&s->mr_nodmar,
3834                                         VTD_INTERRUPT_ADDR_FIRST,
3835                                         &s->mr_ir, 1);
3836 
3837     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
3838     /* No corresponding destroy */
3839     s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3840                                      g_free, g_free);
3841     s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3842                                               g_free, g_free);
3843     vtd_init(s);
3844     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
3845     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
3846     /* Pseudo address space under root PCI bus. */
3847     x86ms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
3848     qemu_add_machine_init_done_notifier(&vtd_machine_done_notify);
3849 }
3850 
3851 static void vtd_class_init(ObjectClass *klass, void *data)
3852 {
3853     DeviceClass *dc = DEVICE_CLASS(klass);
3854     X86IOMMUClass *x86_class = X86_IOMMU_DEVICE_CLASS(klass);
3855 
3856     dc->reset = vtd_reset;
3857     dc->vmsd = &vtd_vmstate;
3858     device_class_set_props(dc, vtd_properties);
3859     dc->hotpluggable = false;
3860     x86_class->realize = vtd_realize;
3861     x86_class->int_remap = vtd_int_remap;
3862     /* Supported by the pc-q35-* machine types */
3863     dc->user_creatable = true;
3864     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
3865     dc->desc = "Intel IOMMU (VT-d) DMA Remapping device";
3866 }
3867 
3868 static const TypeInfo vtd_info = {
3869     .name          = TYPE_INTEL_IOMMU_DEVICE,
3870     .parent        = TYPE_X86_IOMMU_DEVICE,
3871     .instance_size = sizeof(IntelIOMMUState),
3872     .class_init    = vtd_class_init,
3873 };
3874 
3875 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
3876                                                      void *data)
3877 {
3878     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
3879 
3880     imrc->translate = vtd_iommu_translate;
3881     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
3882     imrc->replay = vtd_iommu_replay;
3883 }
3884 
3885 static const TypeInfo vtd_iommu_memory_region_info = {
3886     .parent = TYPE_IOMMU_MEMORY_REGION,
3887     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
3888     .class_init = vtd_iommu_memory_region_class_init,
3889 };
3890 
3891 static void vtd_register_types(void)
3892 {
3893     type_register_static(&vtd_info);
3894     type_register_static(&vtd_iommu_memory_region_info);
3895 }
3896 
3897 type_init(vtd_register_types)
3898