xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision 4011a686)
1 /*
2  * QEMU emulation of an Intel IOMMU (VT-d)
3  *   (DMA Remapping device)
4  *
5  * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com>
6  * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12 
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17 
18  * You should have received a copy of the GNU General Public License along
19  * with this program; if not, see <http://www.gnu.org/licenses/>.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/error-report.h"
24 #include "qemu/main-loop.h"
25 #include "qapi/error.h"
26 #include "hw/sysbus.h"
27 #include "exec/address-spaces.h"
28 #include "intel_iommu_internal.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bus.h"
31 #include "hw/qdev-properties.h"
32 #include "hw/i386/pc.h"
33 #include "hw/i386/apic-msidef.h"
34 #include "hw/boards.h"
35 #include "hw/i386/x86-iommu.h"
36 #include "hw/pci-host/q35.h"
37 #include "sysemu/kvm.h"
38 #include "sysemu/dma.h"
39 #include "sysemu/sysemu.h"
40 #include "hw/i386/apic_internal.h"
41 #include "kvm/kvm_i386.h"
42 #include "migration/vmstate.h"
43 #include "trace.h"
44 
45 /* context entry operations */
46 #define VTD_CE_GET_RID2PASID(ce) \
47     ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK)
48 #define VTD_CE_GET_PASID_DIR_TABLE(ce) \
49     ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK)
50 
51 /* pe operations */
52 #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT)
53 #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW))
54 #define VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write) {\
55     if (ret_fr) {                                                             \
56         ret_fr = -ret_fr;                                                     \
57         if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) {                   \
58             trace_vtd_fault_disabled();                                       \
59         } else {                                                              \
60             vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);      \
61         }                                                                     \
62         goto error;                                                           \
63     }                                                                         \
64 }
65 
66 static void vtd_address_space_refresh_all(IntelIOMMUState *s);
67 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n);
68 
69 static void vtd_panic_require_caching_mode(void)
70 {
71     error_report("We need to set caching-mode=on for intel-iommu to enable "
72                  "device assignment with IOMMU protection.");
73     exit(1);
74 }
75 
76 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val,
77                             uint64_t wmask, uint64_t w1cmask)
78 {
79     stq_le_p(&s->csr[addr], val);
80     stq_le_p(&s->wmask[addr], wmask);
81     stq_le_p(&s->w1cmask[addr], w1cmask);
82 }
83 
84 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask)
85 {
86     stq_le_p(&s->womask[addr], mask);
87 }
88 
89 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val,
90                             uint32_t wmask, uint32_t w1cmask)
91 {
92     stl_le_p(&s->csr[addr], val);
93     stl_le_p(&s->wmask[addr], wmask);
94     stl_le_p(&s->w1cmask[addr], w1cmask);
95 }
96 
97 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask)
98 {
99     stl_le_p(&s->womask[addr], mask);
100 }
101 
102 /* "External" get/set operations */
103 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
104 {
105     uint64_t oldval = ldq_le_p(&s->csr[addr]);
106     uint64_t wmask = ldq_le_p(&s->wmask[addr]);
107     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
108     stq_le_p(&s->csr[addr],
109              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
110 }
111 
112 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
113 {
114     uint32_t oldval = ldl_le_p(&s->csr[addr]);
115     uint32_t wmask = ldl_le_p(&s->wmask[addr]);
116     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
117     stl_le_p(&s->csr[addr],
118              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
119 }
120 
121 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr)
122 {
123     uint64_t val = ldq_le_p(&s->csr[addr]);
124     uint64_t womask = ldq_le_p(&s->womask[addr]);
125     return val & ~womask;
126 }
127 
128 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr)
129 {
130     uint32_t val = ldl_le_p(&s->csr[addr]);
131     uint32_t womask = ldl_le_p(&s->womask[addr]);
132     return val & ~womask;
133 }
134 
135 /* "Internal" get/set operations */
136 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr)
137 {
138     return ldq_le_p(&s->csr[addr]);
139 }
140 
141 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr)
142 {
143     return ldl_le_p(&s->csr[addr]);
144 }
145 
146 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val)
147 {
148     stq_le_p(&s->csr[addr], val);
149 }
150 
151 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
152                                         uint32_t clear, uint32_t mask)
153 {
154     uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
155     stl_le_p(&s->csr[addr], new_val);
156     return new_val;
157 }
158 
159 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
160                                         uint64_t clear, uint64_t mask)
161 {
162     uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
163     stq_le_p(&s->csr[addr], new_val);
164     return new_val;
165 }
166 
167 static inline void vtd_iommu_lock(IntelIOMMUState *s)
168 {
169     qemu_mutex_lock(&s->iommu_lock);
170 }
171 
172 static inline void vtd_iommu_unlock(IntelIOMMUState *s)
173 {
174     qemu_mutex_unlock(&s->iommu_lock);
175 }
176 
177 static void vtd_update_scalable_state(IntelIOMMUState *s)
178 {
179     uint64_t val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
180 
181     if (s->scalable_mode) {
182         s->root_scalable = val & VTD_RTADDR_SMT;
183     }
184 }
185 
186 /* Whether the address space needs to notify new mappings */
187 static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as)
188 {
189     return as->notifier_flags & IOMMU_NOTIFIER_MAP;
190 }
191 
192 /* GHashTable functions */
193 static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2)
194 {
195     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
196 }
197 
198 static guint vtd_uint64_hash(gconstpointer v)
199 {
200     return (guint)*(const uint64_t *)v;
201 }
202 
203 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value,
204                                           gpointer user_data)
205 {
206     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
207     uint16_t domain_id = *(uint16_t *)user_data;
208     return entry->domain_id == domain_id;
209 }
210 
211 /* The shift of an addr for a certain level of paging structure */
212 static inline uint32_t vtd_slpt_level_shift(uint32_t level)
213 {
214     assert(level != 0);
215     return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
216 }
217 
218 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level)
219 {
220     return ~((1ULL << vtd_slpt_level_shift(level)) - 1);
221 }
222 
223 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value,
224                                         gpointer user_data)
225 {
226     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
227     VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data;
228     uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask;
229     uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K;
230     return (entry->domain_id == info->domain_id) &&
231             (((entry->gfn & info->mask) == gfn) ||
232              (entry->gfn == gfn_tlb));
233 }
234 
235 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
236  * IntelIOMMUState to 1.  Must be called with IOMMU lock held.
237  */
238 static void vtd_reset_context_cache_locked(IntelIOMMUState *s)
239 {
240     VTDAddressSpace *vtd_as;
241     VTDBus *vtd_bus;
242     GHashTableIter bus_it;
243     uint32_t devfn_it;
244 
245     trace_vtd_context_cache_reset();
246 
247     g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr);
248 
249     while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) {
250         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
251             vtd_as = vtd_bus->dev_as[devfn_it];
252             if (!vtd_as) {
253                 continue;
254             }
255             vtd_as->context_cache_entry.context_cache_gen = 0;
256         }
257     }
258     s->context_cache_gen = 1;
259 }
260 
261 /* Must be called with IOMMU lock held. */
262 static void vtd_reset_iotlb_locked(IntelIOMMUState *s)
263 {
264     assert(s->iotlb);
265     g_hash_table_remove_all(s->iotlb);
266 }
267 
268 static void vtd_reset_iotlb(IntelIOMMUState *s)
269 {
270     vtd_iommu_lock(s);
271     vtd_reset_iotlb_locked(s);
272     vtd_iommu_unlock(s);
273 }
274 
275 static void vtd_reset_caches(IntelIOMMUState *s)
276 {
277     vtd_iommu_lock(s);
278     vtd_reset_iotlb_locked(s);
279     vtd_reset_context_cache_locked(s);
280     vtd_iommu_unlock(s);
281 }
282 
283 static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id,
284                                   uint32_t level)
285 {
286     return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) |
287            ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT);
288 }
289 
290 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level)
291 {
292     return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K;
293 }
294 
295 /* Must be called with IOMMU lock held */
296 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id,
297                                        hwaddr addr)
298 {
299     VTDIOTLBEntry *entry;
300     uint64_t key;
301     int level;
302 
303     for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) {
304         key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level),
305                                 source_id, level);
306         entry = g_hash_table_lookup(s->iotlb, &key);
307         if (entry) {
308             goto out;
309         }
310     }
311 
312 out:
313     return entry;
314 }
315 
316 /* Must be with IOMMU lock held */
317 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id,
318                              uint16_t domain_id, hwaddr addr, uint64_t slpte,
319                              uint8_t access_flags, uint32_t level)
320 {
321     VTDIOTLBEntry *entry = g_malloc(sizeof(*entry));
322     uint64_t *key = g_malloc(sizeof(*key));
323     uint64_t gfn = vtd_get_iotlb_gfn(addr, level);
324 
325     trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id);
326     if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) {
327         trace_vtd_iotlb_reset("iotlb exceeds size limit");
328         vtd_reset_iotlb_locked(s);
329     }
330 
331     entry->gfn = gfn;
332     entry->domain_id = domain_id;
333     entry->slpte = slpte;
334     entry->access_flags = access_flags;
335     entry->mask = vtd_slpt_level_page_mask(level);
336     *key = vtd_get_iotlb_key(gfn, source_id, level);
337     g_hash_table_replace(s->iotlb, key, entry);
338 }
339 
340 /* Given the reg addr of both the message data and address, generate an
341  * interrupt via MSI.
342  */
343 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
344                                    hwaddr mesg_data_reg)
345 {
346     MSIMessage msi;
347 
348     assert(mesg_data_reg < DMAR_REG_SIZE);
349     assert(mesg_addr_reg < DMAR_REG_SIZE);
350 
351     msi.address = vtd_get_long_raw(s, mesg_addr_reg);
352     msi.data = vtd_get_long_raw(s, mesg_data_reg);
353 
354     trace_vtd_irq_generate(msi.address, msi.data);
355 
356     apic_get_class()->send_msi(&msi);
357 }
358 
359 /* Generate a fault event to software via MSI if conditions are met.
360  * Notice that the value of FSTS_REG being passed to it should be the one
361  * before any update.
362  */
363 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
364 {
365     if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
366         pre_fsts & VTD_FSTS_IQE) {
367         error_report_once("There are previous interrupt conditions "
368                           "to be serviced by software, fault event "
369                           "is not generated");
370         return;
371     }
372     vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
373     if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
374         error_report_once("Interrupt Mask set, irq is not generated");
375     } else {
376         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
377         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
378     }
379 }
380 
381 /* Check if the Fault (F) field of the Fault Recording Register referenced by
382  * @index is Set.
383  */
384 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index)
385 {
386     /* Each reg is 128-bit */
387     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
388     addr += 8; /* Access the high 64-bit half */
389 
390     assert(index < DMAR_FRCD_REG_NR);
391 
392     return vtd_get_quad_raw(s, addr) & VTD_FRCD_F;
393 }
394 
395 /* Update the PPF field of Fault Status Register.
396  * Should be called whenever change the F field of any fault recording
397  * registers.
398  */
399 static void vtd_update_fsts_ppf(IntelIOMMUState *s)
400 {
401     uint32_t i;
402     uint32_t ppf_mask = 0;
403 
404     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
405         if (vtd_is_frcd_set(s, i)) {
406             ppf_mask = VTD_FSTS_PPF;
407             break;
408         }
409     }
410     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
411     trace_vtd_fsts_ppf(!!ppf_mask);
412 }
413 
414 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index)
415 {
416     /* Each reg is 128-bit */
417     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
418     addr += 8; /* Access the high 64-bit half */
419 
420     assert(index < DMAR_FRCD_REG_NR);
421 
422     vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
423     vtd_update_fsts_ppf(s);
424 }
425 
426 /* Must not update F field now, should be done later */
427 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index,
428                             uint16_t source_id, hwaddr addr,
429                             VTDFaultReason fault, bool is_write)
430 {
431     uint64_t hi = 0, lo;
432     hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
433 
434     assert(index < DMAR_FRCD_REG_NR);
435 
436     lo = VTD_FRCD_FI(addr);
437     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
438     if (!is_write) {
439         hi |= VTD_FRCD_T;
440     }
441     vtd_set_quad_raw(s, frcd_reg_addr, lo);
442     vtd_set_quad_raw(s, frcd_reg_addr + 8, hi);
443 
444     trace_vtd_frr_new(index, hi, lo);
445 }
446 
447 /* Try to collapse multiple pending faults from the same requester */
448 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id)
449 {
450     uint32_t i;
451     uint64_t frcd_reg;
452     hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
453 
454     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
455         frcd_reg = vtd_get_quad_raw(s, addr);
456         if ((frcd_reg & VTD_FRCD_F) &&
457             ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
458             return true;
459         }
460         addr += 16; /* 128-bit for each */
461     }
462     return false;
463 }
464 
465 /* Log and report an DMAR (address translation) fault to software */
466 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
467                                   hwaddr addr, VTDFaultReason fault,
468                                   bool is_write)
469 {
470     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
471 
472     assert(fault < VTD_FR_MAX);
473 
474     if (fault == VTD_FR_RESERVED_ERR) {
475         /* This is not a normal fault reason case. Drop it. */
476         return;
477     }
478 
479     trace_vtd_dmar_fault(source_id, fault, addr, is_write);
480 
481     if (fsts_reg & VTD_FSTS_PFO) {
482         error_report_once("New fault is not recorded due to "
483                           "Primary Fault Overflow");
484         return;
485     }
486 
487     if (vtd_try_collapse_fault(s, source_id)) {
488         error_report_once("New fault is not recorded due to "
489                           "compression of faults");
490         return;
491     }
492 
493     if (vtd_is_frcd_set(s, s->next_frcd_reg)) {
494         error_report_once("Next Fault Recording Reg is used, "
495                           "new fault is not recorded, set PFO field");
496         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
497         return;
498     }
499 
500     vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write);
501 
502     if (fsts_reg & VTD_FSTS_PPF) {
503         error_report_once("There are pending faults already, "
504                           "fault event is not generated");
505         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg);
506         s->next_frcd_reg++;
507         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
508             s->next_frcd_reg = 0;
509         }
510     } else {
511         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
512                                 VTD_FSTS_FRI(s->next_frcd_reg));
513         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */
514         s->next_frcd_reg++;
515         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
516             s->next_frcd_reg = 0;
517         }
518         /* This case actually cause the PPF to be Set.
519          * So generate fault event (interrupt).
520          */
521          vtd_generate_fault_event(s, fsts_reg);
522     }
523 }
524 
525 /* Handle Invalidation Queue Errors of queued invalidation interface error
526  * conditions.
527  */
528 static void vtd_handle_inv_queue_error(IntelIOMMUState *s)
529 {
530     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
531 
532     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE);
533     vtd_generate_fault_event(s, fsts_reg);
534 }
535 
536 /* Set the IWC field and try to generate an invalidation completion interrupt */
537 static void vtd_generate_completion_event(IntelIOMMUState *s)
538 {
539     if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) {
540         trace_vtd_inv_desc_wait_irq("One pending, skip current");
541         return;
542     }
543     vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC);
544     vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP);
545     if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) {
546         trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
547                                     "new event not generated");
548         return;
549     } else {
550         /* Generate the interrupt event */
551         trace_vtd_inv_desc_wait_irq("Generating complete event");
552         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
553         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
554     }
555 }
556 
557 static inline bool vtd_root_entry_present(IntelIOMMUState *s,
558                                           VTDRootEntry *re,
559                                           uint8_t devfn)
560 {
561     if (s->root_scalable && devfn > UINT8_MAX / 2) {
562         return re->hi & VTD_ROOT_ENTRY_P;
563     }
564 
565     return re->lo & VTD_ROOT_ENTRY_P;
566 }
567 
568 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index,
569                               VTDRootEntry *re)
570 {
571     dma_addr_t addr;
572 
573     addr = s->root + index * sizeof(*re);
574     if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) {
575         re->lo = 0;
576         return -VTD_FR_ROOT_TABLE_INV;
577     }
578     re->lo = le64_to_cpu(re->lo);
579     re->hi = le64_to_cpu(re->hi);
580     return 0;
581 }
582 
583 static inline bool vtd_ce_present(VTDContextEntry *context)
584 {
585     return context->lo & VTD_CONTEXT_ENTRY_P;
586 }
587 
588 static int vtd_get_context_entry_from_root(IntelIOMMUState *s,
589                                            VTDRootEntry *re,
590                                            uint8_t index,
591                                            VTDContextEntry *ce)
592 {
593     dma_addr_t addr, ce_size;
594 
595     /* we have checked that root entry is present */
596     ce_size = s->root_scalable ? VTD_CTX_ENTRY_SCALABLE_SIZE :
597               VTD_CTX_ENTRY_LEGACY_SIZE;
598 
599     if (s->root_scalable && index > UINT8_MAX / 2) {
600         index = index & (~VTD_DEVFN_CHECK_MASK);
601         addr = re->hi & VTD_ROOT_ENTRY_CTP;
602     } else {
603         addr = re->lo & VTD_ROOT_ENTRY_CTP;
604     }
605 
606     addr = addr + index * ce_size;
607     if (dma_memory_read(&address_space_memory, addr, ce, ce_size)) {
608         return -VTD_FR_CONTEXT_TABLE_INV;
609     }
610 
611     ce->lo = le64_to_cpu(ce->lo);
612     ce->hi = le64_to_cpu(ce->hi);
613     if (ce_size == VTD_CTX_ENTRY_SCALABLE_SIZE) {
614         ce->val[2] = le64_to_cpu(ce->val[2]);
615         ce->val[3] = le64_to_cpu(ce->val[3]);
616     }
617     return 0;
618 }
619 
620 static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce)
621 {
622     return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
623 }
624 
625 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw)
626 {
627     return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw);
628 }
629 
630 /* Whether the pte indicates the address of the page frame */
631 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level)
632 {
633     return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
634 }
635 
636 /* Get the content of a spte located in @base_addr[@index] */
637 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index)
638 {
639     uint64_t slpte;
640 
641     assert(index < VTD_SL_PT_ENTRY_NR);
642 
643     if (dma_memory_read(&address_space_memory,
644                         base_addr + index * sizeof(slpte), &slpte,
645                         sizeof(slpte))) {
646         slpte = (uint64_t)-1;
647         return slpte;
648     }
649     slpte = le64_to_cpu(slpte);
650     return slpte;
651 }
652 
653 /* Given an iova and the level of paging structure, return the offset
654  * of current level.
655  */
656 static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level)
657 {
658     return (iova >> vtd_slpt_level_shift(level)) &
659             ((1ULL << VTD_SL_LEVEL_BITS) - 1);
660 }
661 
662 /* Check Capability Register to see if the @level of page-table is supported */
663 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level)
664 {
665     return VTD_CAP_SAGAW_MASK & s->cap &
666            (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
667 }
668 
669 /* Return true if check passed, otherwise false */
670 static inline bool vtd_pe_type_check(X86IOMMUState *x86_iommu,
671                                      VTDPASIDEntry *pe)
672 {
673     switch (VTD_PE_GET_TYPE(pe)) {
674     case VTD_SM_PASID_ENTRY_FLT:
675     case VTD_SM_PASID_ENTRY_SLT:
676     case VTD_SM_PASID_ENTRY_NESTED:
677         break;
678     case VTD_SM_PASID_ENTRY_PT:
679         if (!x86_iommu->pt_supported) {
680             return false;
681         }
682         break;
683     default:
684         /* Unknwon type */
685         return false;
686     }
687     return true;
688 }
689 
690 static inline bool vtd_pdire_present(VTDPASIDDirEntry *pdire)
691 {
692     return pdire->val & 1;
693 }
694 
695 /**
696  * Caller of this function should check present bit if wants
697  * to use pdir entry for futher usage except for fpd bit check.
698  */
699 static int vtd_get_pdire_from_pdir_table(dma_addr_t pasid_dir_base,
700                                          uint32_t pasid,
701                                          VTDPASIDDirEntry *pdire)
702 {
703     uint32_t index;
704     dma_addr_t addr, entry_size;
705 
706     index = VTD_PASID_DIR_INDEX(pasid);
707     entry_size = VTD_PASID_DIR_ENTRY_SIZE;
708     addr = pasid_dir_base + index * entry_size;
709     if (dma_memory_read(&address_space_memory, addr, pdire, entry_size)) {
710         return -VTD_FR_PASID_TABLE_INV;
711     }
712 
713     return 0;
714 }
715 
716 static inline bool vtd_pe_present(VTDPASIDEntry *pe)
717 {
718     return pe->val[0] & VTD_PASID_ENTRY_P;
719 }
720 
721 static int vtd_get_pe_in_pasid_leaf_table(IntelIOMMUState *s,
722                                           uint32_t pasid,
723                                           dma_addr_t addr,
724                                           VTDPASIDEntry *pe)
725 {
726     uint32_t index;
727     dma_addr_t entry_size;
728     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
729 
730     index = VTD_PASID_TABLE_INDEX(pasid);
731     entry_size = VTD_PASID_ENTRY_SIZE;
732     addr = addr + index * entry_size;
733     if (dma_memory_read(&address_space_memory, addr, pe, entry_size)) {
734         return -VTD_FR_PASID_TABLE_INV;
735     }
736 
737     /* Do translation type check */
738     if (!vtd_pe_type_check(x86_iommu, pe)) {
739         return -VTD_FR_PASID_TABLE_INV;
740     }
741 
742     if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) {
743         return -VTD_FR_PASID_TABLE_INV;
744     }
745 
746     return 0;
747 }
748 
749 /**
750  * Caller of this function should check present bit if wants
751  * to use pasid entry for futher usage except for fpd bit check.
752  */
753 static int vtd_get_pe_from_pdire(IntelIOMMUState *s,
754                                  uint32_t pasid,
755                                  VTDPASIDDirEntry *pdire,
756                                  VTDPASIDEntry *pe)
757 {
758     dma_addr_t addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK;
759 
760     return vtd_get_pe_in_pasid_leaf_table(s, pasid, addr, pe);
761 }
762 
763 /**
764  * This function gets a pasid entry from a specified pasid
765  * table (includes dir and leaf table) with a specified pasid.
766  * Sanity check should be done to ensure return a present
767  * pasid entry to caller.
768  */
769 static int vtd_get_pe_from_pasid_table(IntelIOMMUState *s,
770                                        dma_addr_t pasid_dir_base,
771                                        uint32_t pasid,
772                                        VTDPASIDEntry *pe)
773 {
774     int ret;
775     VTDPASIDDirEntry pdire;
776 
777     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base,
778                                         pasid, &pdire);
779     if (ret) {
780         return ret;
781     }
782 
783     if (!vtd_pdire_present(&pdire)) {
784         return -VTD_FR_PASID_TABLE_INV;
785     }
786 
787     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, pe);
788     if (ret) {
789         return ret;
790     }
791 
792     if (!vtd_pe_present(pe)) {
793         return -VTD_FR_PASID_TABLE_INV;
794     }
795 
796     return 0;
797 }
798 
799 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s,
800                                       VTDContextEntry *ce,
801                                       VTDPASIDEntry *pe)
802 {
803     uint32_t pasid;
804     dma_addr_t pasid_dir_base;
805     int ret = 0;
806 
807     pasid = VTD_CE_GET_RID2PASID(ce);
808     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
809     ret = vtd_get_pe_from_pasid_table(s, pasid_dir_base, pasid, pe);
810 
811     return ret;
812 }
813 
814 static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s,
815                                 VTDContextEntry *ce,
816                                 bool *pe_fpd_set)
817 {
818     int ret;
819     uint32_t pasid;
820     dma_addr_t pasid_dir_base;
821     VTDPASIDDirEntry pdire;
822     VTDPASIDEntry pe;
823 
824     pasid = VTD_CE_GET_RID2PASID(ce);
825     pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce);
826 
827     /*
828      * No present bit check since fpd is meaningful even
829      * if the present bit is clear.
830      */
831     ret = vtd_get_pdire_from_pdir_table(pasid_dir_base, pasid, &pdire);
832     if (ret) {
833         return ret;
834     }
835 
836     if (pdire.val & VTD_PASID_DIR_FPD) {
837         *pe_fpd_set = true;
838         return 0;
839     }
840 
841     if (!vtd_pdire_present(&pdire)) {
842         return -VTD_FR_PASID_TABLE_INV;
843     }
844 
845     /*
846      * No present bit check since fpd is meaningful even
847      * if the present bit is clear.
848      */
849     ret = vtd_get_pe_from_pdire(s, pasid, &pdire, &pe);
850     if (ret) {
851         return ret;
852     }
853 
854     if (pe.val[0] & VTD_PASID_ENTRY_FPD) {
855         *pe_fpd_set = true;
856     }
857 
858     return 0;
859 }
860 
861 /* Get the page-table level that hardware should use for the second-level
862  * page-table walk from the Address Width field of context-entry.
863  */
864 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
865 {
866     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
867 }
868 
869 static uint32_t vtd_get_iova_level(IntelIOMMUState *s,
870                                    VTDContextEntry *ce)
871 {
872     VTDPASIDEntry pe;
873 
874     if (s->root_scalable) {
875         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
876         return VTD_PE_GET_LEVEL(&pe);
877     }
878 
879     return vtd_ce_get_level(ce);
880 }
881 
882 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
883 {
884     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
885 }
886 
887 static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s,
888                                   VTDContextEntry *ce)
889 {
890     VTDPASIDEntry pe;
891 
892     if (s->root_scalable) {
893         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
894         return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9;
895     }
896 
897     return vtd_ce_get_agaw(ce);
898 }
899 
900 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
901 {
902     return ce->lo & VTD_CONTEXT_ENTRY_TT;
903 }
904 
905 /* Only for Legacy Mode. Return true if check passed, otherwise false */
906 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
907                                      VTDContextEntry *ce)
908 {
909     switch (vtd_ce_get_type(ce)) {
910     case VTD_CONTEXT_TT_MULTI_LEVEL:
911         /* Always supported */
912         break;
913     case VTD_CONTEXT_TT_DEV_IOTLB:
914         if (!x86_iommu->dt_supported) {
915             error_report_once("%s: DT specified but not supported", __func__);
916             return false;
917         }
918         break;
919     case VTD_CONTEXT_TT_PASS_THROUGH:
920         if (!x86_iommu->pt_supported) {
921             error_report_once("%s: PT specified but not supported", __func__);
922             return false;
923         }
924         break;
925     default:
926         /* Unknown type */
927         error_report_once("%s: unknown ce type: %"PRIu32, __func__,
928                           vtd_ce_get_type(ce));
929         return false;
930     }
931     return true;
932 }
933 
934 static inline uint64_t vtd_iova_limit(IntelIOMMUState *s,
935                                       VTDContextEntry *ce, uint8_t aw)
936 {
937     uint32_t ce_agaw = vtd_get_iova_agaw(s, ce);
938     return 1ULL << MIN(ce_agaw, aw);
939 }
940 
941 /* Return true if IOVA passes range check, otherwise false. */
942 static inline bool vtd_iova_range_check(IntelIOMMUState *s,
943                                         uint64_t iova, VTDContextEntry *ce,
944                                         uint8_t aw)
945 {
946     /*
947      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
948      * in CAP_REG and AW in context-entry.
949      */
950     return !(iova & ~(vtd_iova_limit(s, ce, aw) - 1));
951 }
952 
953 static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s,
954                                           VTDContextEntry *ce)
955 {
956     VTDPASIDEntry pe;
957 
958     if (s->root_scalable) {
959         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
960         return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR;
961     }
962 
963     return vtd_ce_get_slpt_base(ce);
964 }
965 
966 /*
967  * Rsvd field masks for spte:
968  *     vtd_spte_rsvd 4k pages
969  *     vtd_spte_rsvd_large large pages
970  */
971 static uint64_t vtd_spte_rsvd[5];
972 static uint64_t vtd_spte_rsvd_large[5];
973 
974 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
975 {
976     uint64_t rsvd_mask = vtd_spte_rsvd[level];
977 
978     if ((level == VTD_SL_PD_LEVEL || level == VTD_SL_PDP_LEVEL) &&
979         (slpte & VTD_SL_PT_PAGE_SIZE_MASK)) {
980         /* large page */
981         rsvd_mask = vtd_spte_rsvd_large[level];
982     }
983 
984     return slpte & rsvd_mask;
985 }
986 
987 /* Find the VTD address space associated with a given bus number */
988 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num)
989 {
990     VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num];
991     GHashTableIter iter;
992 
993     if (vtd_bus) {
994         return vtd_bus;
995     }
996 
997     /*
998      * Iterate over the registered buses to find the one which
999      * currently holds this bus number and update the bus_num
1000      * lookup table.
1001      */
1002     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1003     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1004         if (pci_bus_num(vtd_bus->bus) == bus_num) {
1005             s->vtd_as_by_bus_num[bus_num] = vtd_bus;
1006             return vtd_bus;
1007         }
1008     }
1009 
1010     return NULL;
1011 }
1012 
1013 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
1014  * of the translation, can be used for deciding the size of large page.
1015  */
1016 static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce,
1017                              uint64_t iova, bool is_write,
1018                              uint64_t *slptep, uint32_t *slpte_level,
1019                              bool *reads, bool *writes, uint8_t aw_bits)
1020 {
1021     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1022     uint32_t level = vtd_get_iova_level(s, ce);
1023     uint32_t offset;
1024     uint64_t slpte;
1025     uint64_t access_right_check;
1026 
1027     if (!vtd_iova_range_check(s, iova, ce, aw_bits)) {
1028         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")",
1029                           __func__, iova);
1030         return -VTD_FR_ADDR_BEYOND_MGAW;
1031     }
1032 
1033     /* FIXME: what is the Atomics request here? */
1034     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
1035 
1036     while (true) {
1037         offset = vtd_iova_level_offset(iova, level);
1038         slpte = vtd_get_slpte(addr, offset);
1039 
1040         if (slpte == (uint64_t)-1) {
1041             error_report_once("%s: detected read error on DMAR slpte "
1042                               "(iova=0x%" PRIx64 ")", __func__, iova);
1043             if (level == vtd_get_iova_level(s, ce)) {
1044                 /* Invalid programming of context-entry */
1045                 return -VTD_FR_CONTEXT_ENTRY_INV;
1046             } else {
1047                 return -VTD_FR_PAGING_ENTRY_INV;
1048             }
1049         }
1050         *reads = (*reads) && (slpte & VTD_SL_R);
1051         *writes = (*writes) && (slpte & VTD_SL_W);
1052         if (!(slpte & access_right_check)) {
1053             error_report_once("%s: detected slpte permission error "
1054                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
1055                               "slpte=0x%" PRIx64 ", write=%d)", __func__,
1056                               iova, level, slpte, is_write);
1057             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
1058         }
1059         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1060             error_report_once("%s: detected splte reserve non-zero "
1061                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
1062                               "slpte=0x%" PRIx64 ")", __func__, iova,
1063                               level, slpte);
1064             return -VTD_FR_PAGING_ENTRY_RSVD;
1065         }
1066 
1067         if (vtd_is_last_slpte(slpte, level)) {
1068             *slptep = slpte;
1069             *slpte_level = level;
1070             return 0;
1071         }
1072         addr = vtd_get_slpte_addr(slpte, aw_bits);
1073         level--;
1074     }
1075 }
1076 
1077 typedef int (*vtd_page_walk_hook)(IOMMUTLBEvent *event, void *private);
1078 
1079 /**
1080  * Constant information used during page walking
1081  *
1082  * @hook_fn: hook func to be called when detected page
1083  * @private: private data to be passed into hook func
1084  * @notify_unmap: whether we should notify invalid entries
1085  * @as: VT-d address space of the device
1086  * @aw: maximum address width
1087  * @domain: domain ID of the page walk
1088  */
1089 typedef struct {
1090     VTDAddressSpace *as;
1091     vtd_page_walk_hook hook_fn;
1092     void *private;
1093     bool notify_unmap;
1094     uint8_t aw;
1095     uint16_t domain_id;
1096 } vtd_page_walk_info;
1097 
1098 static int vtd_page_walk_one(IOMMUTLBEvent *event, vtd_page_walk_info *info)
1099 {
1100     VTDAddressSpace *as = info->as;
1101     vtd_page_walk_hook hook_fn = info->hook_fn;
1102     void *private = info->private;
1103     IOMMUTLBEntry *entry = &event->entry;
1104     DMAMap target = {
1105         .iova = entry->iova,
1106         .size = entry->addr_mask,
1107         .translated_addr = entry->translated_addr,
1108         .perm = entry->perm,
1109     };
1110     DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
1111 
1112     if (event->type == IOMMU_NOTIFIER_UNMAP && !info->notify_unmap) {
1113         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1114         return 0;
1115     }
1116 
1117     assert(hook_fn);
1118 
1119     /* Update local IOVA mapped ranges */
1120     if (event->type == IOMMU_NOTIFIER_MAP) {
1121         if (mapped) {
1122             /* If it's exactly the same translation, skip */
1123             if (!memcmp(mapped, &target, sizeof(target))) {
1124                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
1125                                                  entry->translated_addr);
1126                 return 0;
1127             } else {
1128                 /*
1129                  * Translation changed.  Normally this should not
1130                  * happen, but it can happen when with buggy guest
1131                  * OSes.  Note that there will be a small window that
1132                  * we don't have map at all.  But that's the best
1133                  * effort we can do.  The ideal way to emulate this is
1134                  * atomically modify the PTE to follow what has
1135                  * changed, but we can't.  One example is that vfio
1136                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1137                  * interface to modify a mapping (meanwhile it seems
1138                  * meaningless to even provide one).  Anyway, let's
1139                  * mark this as a TODO in case one day we'll have
1140                  * a better solution.
1141                  */
1142                 IOMMUAccessFlags cache_perm = entry->perm;
1143                 int ret;
1144 
1145                 /* Emulate an UNMAP */
1146                 event->type = IOMMU_NOTIFIER_UNMAP;
1147                 entry->perm = IOMMU_NONE;
1148                 trace_vtd_page_walk_one(info->domain_id,
1149                                         entry->iova,
1150                                         entry->translated_addr,
1151                                         entry->addr_mask,
1152                                         entry->perm);
1153                 ret = hook_fn(event, private);
1154                 if (ret) {
1155                     return ret;
1156                 }
1157                 /* Drop any existing mapping */
1158                 iova_tree_remove(as->iova_tree, &target);
1159                 /* Recover the correct type */
1160                 event->type = IOMMU_NOTIFIER_MAP;
1161                 entry->perm = cache_perm;
1162             }
1163         }
1164         iova_tree_insert(as->iova_tree, &target);
1165     } else {
1166         if (!mapped) {
1167             /* Skip since we didn't map this range at all */
1168             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
1169             return 0;
1170         }
1171         iova_tree_remove(as->iova_tree, &target);
1172     }
1173 
1174     trace_vtd_page_walk_one(info->domain_id, entry->iova,
1175                             entry->translated_addr, entry->addr_mask,
1176                             entry->perm);
1177     return hook_fn(event, private);
1178 }
1179 
1180 /**
1181  * vtd_page_walk_level - walk over specific level for IOVA range
1182  *
1183  * @addr: base GPA addr to start the walk
1184  * @start: IOVA range start address
1185  * @end: IOVA range end address (start <= addr < end)
1186  * @read: whether parent level has read permission
1187  * @write: whether parent level has write permission
1188  * @info: constant information for the page walk
1189  */
1190 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
1191                                uint64_t end, uint32_t level, bool read,
1192                                bool write, vtd_page_walk_info *info)
1193 {
1194     bool read_cur, write_cur, entry_valid;
1195     uint32_t offset;
1196     uint64_t slpte;
1197     uint64_t subpage_size, subpage_mask;
1198     IOMMUTLBEvent event;
1199     uint64_t iova = start;
1200     uint64_t iova_next;
1201     int ret = 0;
1202 
1203     trace_vtd_page_walk_level(addr, level, start, end);
1204 
1205     subpage_size = 1ULL << vtd_slpt_level_shift(level);
1206     subpage_mask = vtd_slpt_level_page_mask(level);
1207 
1208     while (iova < end) {
1209         iova_next = (iova & subpage_mask) + subpage_size;
1210 
1211         offset = vtd_iova_level_offset(iova, level);
1212         slpte = vtd_get_slpte(addr, offset);
1213 
1214         if (slpte == (uint64_t)-1) {
1215             trace_vtd_page_walk_skip_read(iova, iova_next);
1216             goto next;
1217         }
1218 
1219         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
1220             trace_vtd_page_walk_skip_reserve(iova, iova_next);
1221             goto next;
1222         }
1223 
1224         /* Permissions are stacked with parents' */
1225         read_cur = read && (slpte & VTD_SL_R);
1226         write_cur = write && (slpte & VTD_SL_W);
1227 
1228         /*
1229          * As long as we have either read/write permission, this is a
1230          * valid entry. The rule works for both page entries and page
1231          * table entries.
1232          */
1233         entry_valid = read_cur | write_cur;
1234 
1235         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
1236             /*
1237              * This is a valid PDE (or even bigger than PDE).  We need
1238              * to walk one further level.
1239              */
1240             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
1241                                       iova, MIN(iova_next, end), level - 1,
1242                                       read_cur, write_cur, info);
1243         } else {
1244             /*
1245              * This means we are either:
1246              *
1247              * (1) the real page entry (either 4K page, or huge page)
1248              * (2) the whole range is invalid
1249              *
1250              * In either case, we send an IOTLB notification down.
1251              */
1252             event.entry.target_as = &address_space_memory;
1253             event.entry.iova = iova & subpage_mask;
1254             event.entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
1255             event.entry.addr_mask = ~subpage_mask;
1256             /* NOTE: this is only meaningful if entry_valid == true */
1257             event.entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
1258             event.type = event.entry.perm ? IOMMU_NOTIFIER_MAP :
1259                                             IOMMU_NOTIFIER_UNMAP;
1260             ret = vtd_page_walk_one(&event, info);
1261         }
1262 
1263         if (ret < 0) {
1264             return ret;
1265         }
1266 
1267 next:
1268         iova = iova_next;
1269     }
1270 
1271     return 0;
1272 }
1273 
1274 /**
1275  * vtd_page_walk - walk specific IOVA range, and call the hook
1276  *
1277  * @s: intel iommu state
1278  * @ce: context entry to walk upon
1279  * @start: IOVA address to start the walk
1280  * @end: IOVA range end address (start <= addr < end)
1281  * @info: page walking information struct
1282  */
1283 static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce,
1284                          uint64_t start, uint64_t end,
1285                          vtd_page_walk_info *info)
1286 {
1287     dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce);
1288     uint32_t level = vtd_get_iova_level(s, ce);
1289 
1290     if (!vtd_iova_range_check(s, start, ce, info->aw)) {
1291         return -VTD_FR_ADDR_BEYOND_MGAW;
1292     }
1293 
1294     if (!vtd_iova_range_check(s, end, ce, info->aw)) {
1295         /* Fix end so that it reaches the maximum */
1296         end = vtd_iova_limit(s, ce, info->aw);
1297     }
1298 
1299     return vtd_page_walk_level(addr, start, end, level, true, true, info);
1300 }
1301 
1302 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s,
1303                                           VTDRootEntry *re)
1304 {
1305     /* Legacy Mode reserved bits check */
1306     if (!s->root_scalable &&
1307         (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1308         goto rsvd_err;
1309 
1310     /* Scalable Mode reserved bits check */
1311     if (s->root_scalable &&
1312         ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) ||
1313          (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits))))
1314         goto rsvd_err;
1315 
1316     return 0;
1317 
1318 rsvd_err:
1319     error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1320                       ", lo=0x%"PRIx64,
1321                       __func__, re->hi, re->lo);
1322     return -VTD_FR_ROOT_ENTRY_RSVD;
1323 }
1324 
1325 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s,
1326                                                     VTDContextEntry *ce)
1327 {
1328     if (!s->root_scalable &&
1329         (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI ||
1330          ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1331         error_report_once("%s: invalid context entry: hi=%"PRIx64
1332                           ", lo=%"PRIx64" (reserved nonzero)",
1333                           __func__, ce->hi, ce->lo);
1334         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1335     }
1336 
1337     if (s->root_scalable &&
1338         (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) ||
1339          ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 ||
1340          ce->val[2] ||
1341          ce->val[3])) {
1342         error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1343                           ", val[2]=%"PRIx64
1344                           ", val[1]=%"PRIx64
1345                           ", val[0]=%"PRIx64" (reserved nonzero)",
1346                           __func__, ce->val[3], ce->val[2],
1347                           ce->val[1], ce->val[0]);
1348         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1349     }
1350 
1351     return 0;
1352 }
1353 
1354 static int vtd_ce_rid2pasid_check(IntelIOMMUState *s,
1355                                   VTDContextEntry *ce)
1356 {
1357     VTDPASIDEntry pe;
1358 
1359     /*
1360      * Make sure in Scalable Mode, a present context entry
1361      * has valid rid2pasid setting, which includes valid
1362      * rid2pasid field and corresponding pasid entry setting
1363      */
1364     return vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1365 }
1366 
1367 /* Map a device to its corresponding domain (context-entry) */
1368 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
1369                                     uint8_t devfn, VTDContextEntry *ce)
1370 {
1371     VTDRootEntry re;
1372     int ret_fr;
1373     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
1374 
1375     ret_fr = vtd_get_root_entry(s, bus_num, &re);
1376     if (ret_fr) {
1377         return ret_fr;
1378     }
1379 
1380     if (!vtd_root_entry_present(s, &re, devfn)) {
1381         /* Not error - it's okay we don't have root entry. */
1382         trace_vtd_re_not_present(bus_num);
1383         return -VTD_FR_ROOT_ENTRY_P;
1384     }
1385 
1386     ret_fr = vtd_root_entry_rsvd_bits_check(s, &re);
1387     if (ret_fr) {
1388         return ret_fr;
1389     }
1390 
1391     ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce);
1392     if (ret_fr) {
1393         return ret_fr;
1394     }
1395 
1396     if (!vtd_ce_present(ce)) {
1397         /* Not error - it's okay we don't have context entry. */
1398         trace_vtd_ce_not_present(bus_num, devfn);
1399         return -VTD_FR_CONTEXT_ENTRY_P;
1400     }
1401 
1402     ret_fr = vtd_context_entry_rsvd_bits_check(s, ce);
1403     if (ret_fr) {
1404         return ret_fr;
1405     }
1406 
1407     /* Check if the programming of context-entry is valid */
1408     if (!s->root_scalable &&
1409         !vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1410         error_report_once("%s: invalid context entry: hi=%"PRIx64
1411                           ", lo=%"PRIx64" (level %d not supported)",
1412                           __func__, ce->hi, ce->lo,
1413                           vtd_ce_get_level(ce));
1414         return -VTD_FR_CONTEXT_ENTRY_INV;
1415     }
1416 
1417     if (!s->root_scalable) {
1418         /* Do translation type check */
1419         if (!vtd_ce_type_check(x86_iommu, ce)) {
1420             /* Errors dumped in vtd_ce_type_check() */
1421             return -VTD_FR_CONTEXT_ENTRY_INV;
1422         }
1423     } else {
1424         /*
1425          * Check if the programming of context-entry.rid2pasid
1426          * and corresponding pasid setting is valid, and thus
1427          * avoids to check pasid entry fetching result in future
1428          * helper function calling.
1429          */
1430         ret_fr = vtd_ce_rid2pasid_check(s, ce);
1431         if (ret_fr) {
1432             return ret_fr;
1433         }
1434     }
1435 
1436     return 0;
1437 }
1438 
1439 static int vtd_sync_shadow_page_hook(IOMMUTLBEvent *event,
1440                                      void *private)
1441 {
1442     memory_region_notify_iommu(private, 0, *event);
1443     return 0;
1444 }
1445 
1446 static uint16_t vtd_get_domain_id(IntelIOMMUState *s,
1447                                   VTDContextEntry *ce)
1448 {
1449     VTDPASIDEntry pe;
1450 
1451     if (s->root_scalable) {
1452         vtd_ce_get_rid2pasid_entry(s, ce, &pe);
1453         return VTD_SM_PASID_ENTRY_DID(pe.val[1]);
1454     }
1455 
1456     return VTD_CONTEXT_ENTRY_DID(ce->hi);
1457 }
1458 
1459 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1460                                             VTDContextEntry *ce,
1461                                             hwaddr addr, hwaddr size)
1462 {
1463     IntelIOMMUState *s = vtd_as->iommu_state;
1464     vtd_page_walk_info info = {
1465         .hook_fn = vtd_sync_shadow_page_hook,
1466         .private = (void *)&vtd_as->iommu,
1467         .notify_unmap = true,
1468         .aw = s->aw_bits,
1469         .as = vtd_as,
1470         .domain_id = vtd_get_domain_id(s, ce),
1471     };
1472 
1473     return vtd_page_walk(s, ce, addr, addr + size, &info);
1474 }
1475 
1476 static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
1477 {
1478     int ret;
1479     VTDContextEntry ce;
1480     IOMMUNotifier *n;
1481 
1482     if (!(vtd_as->iommu.iommu_notify_flags & IOMMU_NOTIFIER_IOTLB_EVENTS)) {
1483         return 0;
1484     }
1485 
1486     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1487                                    pci_bus_num(vtd_as->bus),
1488                                    vtd_as->devfn, &ce);
1489     if (ret) {
1490         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1491             /*
1492              * It's a valid scenario to have a context entry that is
1493              * not present.  For example, when a device is removed
1494              * from an existing domain then the context entry will be
1495              * zeroed by the guest before it was put into another
1496              * domain.  When this happens, instead of synchronizing
1497              * the shadow pages we should invalidate all existing
1498              * mappings and notify the backends.
1499              */
1500             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1501                 vtd_address_space_unmap(vtd_as, n);
1502             }
1503             ret = 0;
1504         }
1505         return ret;
1506     }
1507 
1508     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1509 }
1510 
1511 /*
1512  * Check if specific device is configed to bypass address
1513  * translation for DMA requests. In Scalable Mode, bypass
1514  * 1st-level translation or 2nd-level translation, it depends
1515  * on PGTT setting.
1516  */
1517 static bool vtd_dev_pt_enabled(VTDAddressSpace *as)
1518 {
1519     IntelIOMMUState *s;
1520     VTDContextEntry ce;
1521     VTDPASIDEntry pe;
1522     int ret;
1523 
1524     assert(as);
1525 
1526     s = as->iommu_state;
1527     ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus),
1528                                    as->devfn, &ce);
1529     if (ret) {
1530         /*
1531          * Possibly failed to parse the context entry for some reason
1532          * (e.g., during init, or any guest configuration errors on
1533          * context entries). We should assume PT not enabled for
1534          * safety.
1535          */
1536         return false;
1537     }
1538 
1539     if (s->root_scalable) {
1540         ret = vtd_ce_get_rid2pasid_entry(s, &ce, &pe);
1541         if (ret) {
1542             error_report_once("%s: vtd_ce_get_rid2pasid_entry error: %"PRId32,
1543                               __func__, ret);
1544             return false;
1545         }
1546         return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT);
1547     }
1548 
1549     return (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH);
1550 }
1551 
1552 /* Return whether the device is using IOMMU translation. */
1553 static bool vtd_switch_address_space(VTDAddressSpace *as)
1554 {
1555     bool use_iommu;
1556     /* Whether we need to take the BQL on our own */
1557     bool take_bql = !qemu_mutex_iothread_locked();
1558 
1559     assert(as);
1560 
1561     use_iommu = as->iommu_state->dmar_enabled && !vtd_dev_pt_enabled(as);
1562 
1563     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1564                                    VTD_PCI_SLOT(as->devfn),
1565                                    VTD_PCI_FUNC(as->devfn),
1566                                    use_iommu);
1567 
1568     /*
1569      * It's possible that we reach here without BQL, e.g., when called
1570      * from vtd_pt_enable_fast_path(). However the memory APIs need
1571      * it. We'd better make sure we have had it already, or, take it.
1572      */
1573     if (take_bql) {
1574         qemu_mutex_lock_iothread();
1575     }
1576 
1577     /* Turn off first then on the other */
1578     if (use_iommu) {
1579         memory_region_set_enabled(&as->nodmar, false);
1580         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1581     } else {
1582         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1583         memory_region_set_enabled(&as->nodmar, true);
1584     }
1585 
1586     if (take_bql) {
1587         qemu_mutex_unlock_iothread();
1588     }
1589 
1590     return use_iommu;
1591 }
1592 
1593 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1594 {
1595     GHashTableIter iter;
1596     VTDBus *vtd_bus;
1597     int i;
1598 
1599     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1600     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1601         for (i = 0; i < PCI_DEVFN_MAX; i++) {
1602             if (!vtd_bus->dev_as[i]) {
1603                 continue;
1604             }
1605             vtd_switch_address_space(vtd_bus->dev_as[i]);
1606         }
1607     }
1608 }
1609 
1610 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn)
1611 {
1612     return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
1613 }
1614 
1615 static const bool vtd_qualified_faults[] = {
1616     [VTD_FR_RESERVED] = false,
1617     [VTD_FR_ROOT_ENTRY_P] = false,
1618     [VTD_FR_CONTEXT_ENTRY_P] = true,
1619     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1620     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1621     [VTD_FR_WRITE] = true,
1622     [VTD_FR_READ] = true,
1623     [VTD_FR_PAGING_ENTRY_INV] = true,
1624     [VTD_FR_ROOT_TABLE_INV] = false,
1625     [VTD_FR_CONTEXT_TABLE_INV] = false,
1626     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1627     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1628     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1629     [VTD_FR_PASID_TABLE_INV] = false,
1630     [VTD_FR_RESERVED_ERR] = false,
1631     [VTD_FR_MAX] = false,
1632 };
1633 
1634 /* To see if a fault condition is "qualified", which is reported to software
1635  * only if the FPD field in the context-entry used to process the faulting
1636  * request is 0.
1637  */
1638 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1639 {
1640     return vtd_qualified_faults[fault];
1641 }
1642 
1643 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1644 {
1645     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1646 }
1647 
1648 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1649 {
1650     VTDBus *vtd_bus;
1651     VTDAddressSpace *vtd_as;
1652     bool success = false;
1653 
1654     vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id));
1655     if (!vtd_bus) {
1656         goto out;
1657     }
1658 
1659     vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)];
1660     if (!vtd_as) {
1661         goto out;
1662     }
1663 
1664     if (vtd_switch_address_space(vtd_as) == false) {
1665         /* We switched off IOMMU region successfully. */
1666         success = true;
1667     }
1668 
1669 out:
1670     trace_vtd_pt_enable_fast_path(source_id, success);
1671 }
1672 
1673 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1674  * translation.
1675  *
1676  * Called from RCU critical section.
1677  *
1678  * @bus_num: The bus number
1679  * @devfn: The devfn, which is the  combined of device and function number
1680  * @is_write: The access is a write operation
1681  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1682  *
1683  * Returns true if translation is successful, otherwise false.
1684  */
1685 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1686                                    uint8_t devfn, hwaddr addr, bool is_write,
1687                                    IOMMUTLBEntry *entry)
1688 {
1689     IntelIOMMUState *s = vtd_as->iommu_state;
1690     VTDContextEntry ce;
1691     uint8_t bus_num = pci_bus_num(bus);
1692     VTDContextCacheEntry *cc_entry;
1693     uint64_t slpte, page_mask;
1694     uint32_t level;
1695     uint16_t source_id = vtd_make_source_id(bus_num, devfn);
1696     int ret_fr;
1697     bool is_fpd_set = false;
1698     bool reads = true;
1699     bool writes = true;
1700     uint8_t access_flags;
1701     VTDIOTLBEntry *iotlb_entry;
1702 
1703     /*
1704      * We have standalone memory region for interrupt addresses, we
1705      * should never receive translation requests in this region.
1706      */
1707     assert(!vtd_is_interrupt_addr(addr));
1708 
1709     vtd_iommu_lock(s);
1710 
1711     cc_entry = &vtd_as->context_cache_entry;
1712 
1713     /* Try to fetch slpte form IOTLB */
1714     iotlb_entry = vtd_lookup_iotlb(s, source_id, addr);
1715     if (iotlb_entry) {
1716         trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1717                                  iotlb_entry->domain_id);
1718         slpte = iotlb_entry->slpte;
1719         access_flags = iotlb_entry->access_flags;
1720         page_mask = iotlb_entry->mask;
1721         goto out;
1722     }
1723 
1724     /* Try to fetch context-entry from cache first */
1725     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1726         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1727                                cc_entry->context_entry.lo,
1728                                cc_entry->context_cache_gen);
1729         ce = cc_entry->context_entry;
1730         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1731         if (!is_fpd_set && s->root_scalable) {
1732             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1733             VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1734         }
1735     } else {
1736         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1737         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1738         if (!ret_fr && !is_fpd_set && s->root_scalable) {
1739             ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set);
1740         }
1741         VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1742         /* Update context-cache */
1743         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1744                                   cc_entry->context_cache_gen,
1745                                   s->context_cache_gen);
1746         cc_entry->context_entry = ce;
1747         cc_entry->context_cache_gen = s->context_cache_gen;
1748     }
1749 
1750     /*
1751      * We don't need to translate for pass-through context entries.
1752      * Also, let's ignore IOTLB caching as well for PT devices.
1753      */
1754     if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) {
1755         entry->iova = addr & VTD_PAGE_MASK_4K;
1756         entry->translated_addr = entry->iova;
1757         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1758         entry->perm = IOMMU_RW;
1759         trace_vtd_translate_pt(source_id, entry->iova);
1760 
1761         /*
1762          * When this happens, it means firstly caching-mode is not
1763          * enabled, and this is the first passthrough translation for
1764          * the device. Let's enable the fast path for passthrough.
1765          *
1766          * When passthrough is disabled again for the device, we can
1767          * capture it via the context entry invalidation, then the
1768          * IOMMU region can be swapped back.
1769          */
1770         vtd_pt_enable_fast_path(s, source_id);
1771         vtd_iommu_unlock(s);
1772         return true;
1773     }
1774 
1775     ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level,
1776                                &reads, &writes, s->aw_bits);
1777     VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write);
1778 
1779     page_mask = vtd_slpt_level_page_mask(level);
1780     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1781     vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce), addr, slpte,
1782                      access_flags, level);
1783 out:
1784     vtd_iommu_unlock(s);
1785     entry->iova = addr & page_mask;
1786     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1787     entry->addr_mask = ~page_mask;
1788     entry->perm = access_flags;
1789     return true;
1790 
1791 error:
1792     vtd_iommu_unlock(s);
1793     entry->iova = 0;
1794     entry->translated_addr = 0;
1795     entry->addr_mask = 0;
1796     entry->perm = IOMMU_NONE;
1797     return false;
1798 }
1799 
1800 static void vtd_root_table_setup(IntelIOMMUState *s)
1801 {
1802     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1803     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
1804 
1805     vtd_update_scalable_state(s);
1806 
1807     trace_vtd_reg_dmar_root(s->root, s->root_scalable);
1808 }
1809 
1810 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
1811                                uint32_t index, uint32_t mask)
1812 {
1813     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
1814 }
1815 
1816 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
1817 {
1818     uint64_t value = 0;
1819     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
1820     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
1821     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
1822     s->intr_eime = value & VTD_IRTA_EIME;
1823 
1824     /* Notify global invalidation */
1825     vtd_iec_notify_all(s, true, 0, 0);
1826 
1827     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
1828 }
1829 
1830 static void vtd_iommu_replay_all(IntelIOMMUState *s)
1831 {
1832     VTDAddressSpace *vtd_as;
1833 
1834     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1835         vtd_sync_shadow_page_table(vtd_as);
1836     }
1837 }
1838 
1839 static void vtd_context_global_invalidate(IntelIOMMUState *s)
1840 {
1841     trace_vtd_inv_desc_cc_global();
1842     /* Protects context cache */
1843     vtd_iommu_lock(s);
1844     s->context_cache_gen++;
1845     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
1846         vtd_reset_context_cache_locked(s);
1847     }
1848     vtd_iommu_unlock(s);
1849     vtd_address_space_refresh_all(s);
1850     /*
1851      * From VT-d spec 6.5.2.1, a global context entry invalidation
1852      * should be followed by a IOTLB global invalidation, so we should
1853      * be safe even without this. Hoewever, let's replay the region as
1854      * well to be safer, and go back here when we need finer tunes for
1855      * VT-d emulation codes.
1856      */
1857     vtd_iommu_replay_all(s);
1858 }
1859 
1860 /* Do a context-cache device-selective invalidation.
1861  * @func_mask: FM field after shifting
1862  */
1863 static void vtd_context_device_invalidate(IntelIOMMUState *s,
1864                                           uint16_t source_id,
1865                                           uint16_t func_mask)
1866 {
1867     uint16_t mask;
1868     VTDBus *vtd_bus;
1869     VTDAddressSpace *vtd_as;
1870     uint8_t bus_n, devfn;
1871     uint16_t devfn_it;
1872 
1873     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
1874 
1875     switch (func_mask & 3) {
1876     case 0:
1877         mask = 0;   /* No bits in the SID field masked */
1878         break;
1879     case 1:
1880         mask = 4;   /* Mask bit 2 in the SID field */
1881         break;
1882     case 2:
1883         mask = 6;   /* Mask bit 2:1 in the SID field */
1884         break;
1885     case 3:
1886         mask = 7;   /* Mask bit 2:0 in the SID field */
1887         break;
1888     default:
1889         g_assert_not_reached();
1890     }
1891     mask = ~mask;
1892 
1893     bus_n = VTD_SID_TO_BUS(source_id);
1894     vtd_bus = vtd_find_as_from_bus_num(s, bus_n);
1895     if (vtd_bus) {
1896         devfn = VTD_SID_TO_DEVFN(source_id);
1897         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
1898             vtd_as = vtd_bus->dev_as[devfn_it];
1899             if (vtd_as && ((devfn_it & mask) == (devfn & mask))) {
1900                 trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it),
1901                                              VTD_PCI_FUNC(devfn_it));
1902                 vtd_iommu_lock(s);
1903                 vtd_as->context_cache_entry.context_cache_gen = 0;
1904                 vtd_iommu_unlock(s);
1905                 /*
1906                  * Do switch address space when needed, in case if the
1907                  * device passthrough bit is switched.
1908                  */
1909                 vtd_switch_address_space(vtd_as);
1910                 /*
1911                  * So a device is moving out of (or moving into) a
1912                  * domain, resync the shadow page table.
1913                  * This won't bring bad even if we have no such
1914                  * notifier registered - the IOMMU notification
1915                  * framework will skip MAP notifications if that
1916                  * happened.
1917                  */
1918                 vtd_sync_shadow_page_table(vtd_as);
1919             }
1920         }
1921     }
1922 }
1923 
1924 /* Context-cache invalidation
1925  * Returns the Context Actual Invalidation Granularity.
1926  * @val: the content of the CCMD_REG
1927  */
1928 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
1929 {
1930     uint64_t caig;
1931     uint64_t type = val & VTD_CCMD_CIRG_MASK;
1932 
1933     switch (type) {
1934     case VTD_CCMD_DOMAIN_INVL:
1935         /* Fall through */
1936     case VTD_CCMD_GLOBAL_INVL:
1937         caig = VTD_CCMD_GLOBAL_INVL_A;
1938         vtd_context_global_invalidate(s);
1939         break;
1940 
1941     case VTD_CCMD_DEVICE_INVL:
1942         caig = VTD_CCMD_DEVICE_INVL_A;
1943         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
1944         break;
1945 
1946     default:
1947         error_report_once("%s: invalid context: 0x%" PRIx64,
1948                           __func__, val);
1949         caig = 0;
1950     }
1951     return caig;
1952 }
1953 
1954 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
1955 {
1956     trace_vtd_inv_desc_iotlb_global();
1957     vtd_reset_iotlb(s);
1958     vtd_iommu_replay_all(s);
1959 }
1960 
1961 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
1962 {
1963     VTDContextEntry ce;
1964     VTDAddressSpace *vtd_as;
1965 
1966     trace_vtd_inv_desc_iotlb_domain(domain_id);
1967 
1968     vtd_iommu_lock(s);
1969     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
1970                                 &domain_id);
1971     vtd_iommu_unlock(s);
1972 
1973     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1974         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1975                                       vtd_as->devfn, &ce) &&
1976             domain_id == vtd_get_domain_id(s, &ce)) {
1977             vtd_sync_shadow_page_table(vtd_as);
1978         }
1979     }
1980 }
1981 
1982 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
1983                                            uint16_t domain_id, hwaddr addr,
1984                                            uint8_t am)
1985 {
1986     VTDAddressSpace *vtd_as;
1987     VTDContextEntry ce;
1988     int ret;
1989     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
1990 
1991     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
1992         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1993                                        vtd_as->devfn, &ce);
1994         if (!ret && domain_id == vtd_get_domain_id(s, &ce)) {
1995             if (vtd_as_has_map_notifier(vtd_as)) {
1996                 /*
1997                  * As long as we have MAP notifications registered in
1998                  * any of our IOMMU notifiers, we need to sync the
1999                  * shadow page table.
2000                  */
2001                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
2002             } else {
2003                 /*
2004                  * For UNMAP-only notifiers, we don't need to walk the
2005                  * page tables.  We just deliver the PSI down to
2006                  * invalidate caches.
2007                  */
2008                 IOMMUTLBEvent event = {
2009                     .type = IOMMU_NOTIFIER_UNMAP,
2010                     .entry = {
2011                         .target_as = &address_space_memory,
2012                         .iova = addr,
2013                         .translated_addr = 0,
2014                         .addr_mask = size - 1,
2015                         .perm = IOMMU_NONE,
2016                     },
2017                 };
2018                 memory_region_notify_iommu(&vtd_as->iommu, 0, event);
2019             }
2020         }
2021     }
2022 }
2023 
2024 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
2025                                       hwaddr addr, uint8_t am)
2026 {
2027     VTDIOTLBPageInvInfo info;
2028 
2029     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
2030 
2031     assert(am <= VTD_MAMV);
2032     info.domain_id = domain_id;
2033     info.addr = addr;
2034     info.mask = ~((1 << am) - 1);
2035     vtd_iommu_lock(s);
2036     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
2037     vtd_iommu_unlock(s);
2038     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am);
2039 }
2040 
2041 /* Flush IOTLB
2042  * Returns the IOTLB Actual Invalidation Granularity.
2043  * @val: the content of the IOTLB_REG
2044  */
2045 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
2046 {
2047     uint64_t iaig;
2048     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
2049     uint16_t domain_id;
2050     hwaddr addr;
2051     uint8_t am;
2052 
2053     switch (type) {
2054     case VTD_TLB_GLOBAL_FLUSH:
2055         iaig = VTD_TLB_GLOBAL_FLUSH_A;
2056         vtd_iotlb_global_invalidate(s);
2057         break;
2058 
2059     case VTD_TLB_DSI_FLUSH:
2060         domain_id = VTD_TLB_DID(val);
2061         iaig = VTD_TLB_DSI_FLUSH_A;
2062         vtd_iotlb_domain_invalidate(s, domain_id);
2063         break;
2064 
2065     case VTD_TLB_PSI_FLUSH:
2066         domain_id = VTD_TLB_DID(val);
2067         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
2068         am = VTD_IVA_AM(addr);
2069         addr = VTD_IVA_ADDR(addr);
2070         if (am > VTD_MAMV) {
2071             error_report_once("%s: address mask overflow: 0x%" PRIx64,
2072                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
2073             iaig = 0;
2074             break;
2075         }
2076         iaig = VTD_TLB_PSI_FLUSH_A;
2077         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2078         break;
2079 
2080     default:
2081         error_report_once("%s: invalid granularity: 0x%" PRIx64,
2082                           __func__, val);
2083         iaig = 0;
2084     }
2085     return iaig;
2086 }
2087 
2088 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
2089 
2090 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
2091 {
2092     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
2093            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
2094 }
2095 
2096 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
2097 {
2098     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
2099 
2100     trace_vtd_inv_qi_enable(en);
2101 
2102     if (en) {
2103         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
2104         /* 2^(x+8) entries */
2105         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0));
2106         s->qi_enabled = true;
2107         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
2108         /* Ok - report back to driver */
2109         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
2110 
2111         if (s->iq_tail != 0) {
2112             /*
2113              * This is a spec violation but Windows guests are known to set up
2114              * Queued Invalidation this way so we allow the write and process
2115              * Invalidation Descriptors right away.
2116              */
2117             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
2118             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2119                 vtd_fetch_inv_desc(s);
2120             }
2121         }
2122     } else {
2123         if (vtd_queued_inv_disable_check(s)) {
2124             /* disable Queued Invalidation */
2125             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
2126             s->iq_head = 0;
2127             s->qi_enabled = false;
2128             /* Ok - report back to driver */
2129             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
2130         } else {
2131             error_report_once("%s: detected improper state when disable QI "
2132                               "(head=0x%x, tail=0x%x, last_type=%d)",
2133                               __func__,
2134                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
2135         }
2136     }
2137 }
2138 
2139 /* Set Root Table Pointer */
2140 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
2141 {
2142     vtd_root_table_setup(s);
2143     /* Ok - report back to driver */
2144     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
2145     vtd_reset_caches(s);
2146     vtd_address_space_refresh_all(s);
2147 }
2148 
2149 /* Set Interrupt Remap Table Pointer */
2150 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
2151 {
2152     vtd_interrupt_remap_table_setup(s);
2153     /* Ok - report back to driver */
2154     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
2155 }
2156 
2157 /* Handle Translation Enable/Disable */
2158 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
2159 {
2160     if (s->dmar_enabled == en) {
2161         return;
2162     }
2163 
2164     trace_vtd_dmar_enable(en);
2165 
2166     if (en) {
2167         s->dmar_enabled = true;
2168         /* Ok - report back to driver */
2169         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
2170     } else {
2171         s->dmar_enabled = false;
2172 
2173         /* Clear the index of Fault Recording Register */
2174         s->next_frcd_reg = 0;
2175         /* Ok - report back to driver */
2176         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
2177     }
2178 
2179     vtd_reset_caches(s);
2180     vtd_address_space_refresh_all(s);
2181 }
2182 
2183 /* Handle Interrupt Remap Enable/Disable */
2184 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
2185 {
2186     trace_vtd_ir_enable(en);
2187 
2188     if (en) {
2189         s->intr_enabled = true;
2190         /* Ok - report back to driver */
2191         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
2192     } else {
2193         s->intr_enabled = false;
2194         /* Ok - report back to driver */
2195         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
2196     }
2197 }
2198 
2199 /* Handle write to Global Command Register */
2200 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
2201 {
2202     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
2203     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
2204     uint32_t changed = status ^ val;
2205 
2206     trace_vtd_reg_write_gcmd(status, val);
2207     if (changed & VTD_GCMD_TE) {
2208         /* Translation enable/disable */
2209         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
2210     }
2211     if (val & VTD_GCMD_SRTP) {
2212         /* Set/update the root-table pointer */
2213         vtd_handle_gcmd_srtp(s);
2214     }
2215     if (changed & VTD_GCMD_QIE) {
2216         /* Queued Invalidation Enable */
2217         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
2218     }
2219     if (val & VTD_GCMD_SIRTP) {
2220         /* Set/update the interrupt remapping root-table pointer */
2221         vtd_handle_gcmd_sirtp(s);
2222     }
2223     if (changed & VTD_GCMD_IRE) {
2224         /* Interrupt remap enable/disable */
2225         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
2226     }
2227 }
2228 
2229 /* Handle write to Context Command Register */
2230 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
2231 {
2232     uint64_t ret;
2233     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
2234 
2235     /* Context-cache invalidation request */
2236     if (val & VTD_CCMD_ICC) {
2237         if (s->qi_enabled) {
2238             error_report_once("Queued Invalidation enabled, "
2239                               "should not use register-based invalidation");
2240             return;
2241         }
2242         ret = vtd_context_cache_invalidate(s, val);
2243         /* Invalidation completed. Change something to show */
2244         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
2245         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
2246                                       ret);
2247     }
2248 }
2249 
2250 /* Handle write to IOTLB Invalidation Register */
2251 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
2252 {
2253     uint64_t ret;
2254     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
2255 
2256     /* IOTLB invalidation request */
2257     if (val & VTD_TLB_IVT) {
2258         if (s->qi_enabled) {
2259             error_report_once("Queued Invalidation enabled, "
2260                               "should not use register-based invalidation");
2261             return;
2262         }
2263         ret = vtd_iotlb_flush(s, val);
2264         /* Invalidation completed. Change something to show */
2265         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
2266         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
2267                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
2268     }
2269 }
2270 
2271 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2272 static bool vtd_get_inv_desc(IntelIOMMUState *s,
2273                              VTDInvDesc *inv_desc)
2274 {
2275     dma_addr_t base_addr = s->iq;
2276     uint32_t offset = s->iq_head;
2277     uint32_t dw = s->iq_dw ? 32 : 16;
2278     dma_addr_t addr = base_addr + offset * dw;
2279 
2280     if (dma_memory_read(&address_space_memory, addr, inv_desc, dw)) {
2281         error_report_once("Read INV DESC failed.");
2282         return false;
2283     }
2284     inv_desc->lo = le64_to_cpu(inv_desc->lo);
2285     inv_desc->hi = le64_to_cpu(inv_desc->hi);
2286     if (dw == 32) {
2287         inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]);
2288         inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]);
2289     }
2290     return true;
2291 }
2292 
2293 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2294 {
2295     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
2296         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
2297         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2298                           " (reserved nonzero)", __func__, inv_desc->hi,
2299                           inv_desc->lo);
2300         return false;
2301     }
2302     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
2303         /* Status Write */
2304         uint32_t status_data = (uint32_t)(inv_desc->lo >>
2305                                VTD_INV_DESC_WAIT_DATA_SHIFT);
2306 
2307         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
2308 
2309         /* FIXME: need to be masked with HAW? */
2310         dma_addr_t status_addr = inv_desc->hi;
2311         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
2312         status_data = cpu_to_le32(status_data);
2313         if (dma_memory_write(&address_space_memory, status_addr, &status_data,
2314                              sizeof(status_data))) {
2315             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
2316             return false;
2317         }
2318     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
2319         /* Interrupt flag */
2320         vtd_generate_completion_event(s);
2321     } else {
2322         error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64
2323                           " (unknown type)", __func__, inv_desc->hi,
2324                           inv_desc->lo);
2325         return false;
2326     }
2327     return true;
2328 }
2329 
2330 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
2331                                            VTDInvDesc *inv_desc)
2332 {
2333     uint16_t sid, fmask;
2334 
2335     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
2336         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2337                           " (reserved nonzero)", __func__, inv_desc->hi,
2338                           inv_desc->lo);
2339         return false;
2340     }
2341     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
2342     case VTD_INV_DESC_CC_DOMAIN:
2343         trace_vtd_inv_desc_cc_domain(
2344             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
2345         /* Fall through */
2346     case VTD_INV_DESC_CC_GLOBAL:
2347         vtd_context_global_invalidate(s);
2348         break;
2349 
2350     case VTD_INV_DESC_CC_DEVICE:
2351         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
2352         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
2353         vtd_context_device_invalidate(s, sid, fmask);
2354         break;
2355 
2356     default:
2357         error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64
2358                           " (invalid type)", __func__, inv_desc->hi,
2359                           inv_desc->lo);
2360         return false;
2361     }
2362     return true;
2363 }
2364 
2365 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
2366 {
2367     uint16_t domain_id;
2368     uint8_t am;
2369     hwaddr addr;
2370 
2371     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
2372         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
2373         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2374                           ", lo=0x%"PRIx64" (reserved bits unzero)",
2375                           __func__, inv_desc->hi, inv_desc->lo);
2376         return false;
2377     }
2378 
2379     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
2380     case VTD_INV_DESC_IOTLB_GLOBAL:
2381         vtd_iotlb_global_invalidate(s);
2382         break;
2383 
2384     case VTD_INV_DESC_IOTLB_DOMAIN:
2385         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2386         vtd_iotlb_domain_invalidate(s, domain_id);
2387         break;
2388 
2389     case VTD_INV_DESC_IOTLB_PAGE:
2390         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
2391         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
2392         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
2393         if (am > VTD_MAMV) {
2394             error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2395                               ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)",
2396                               __func__, inv_desc->hi, inv_desc->lo,
2397                               am, (unsigned)VTD_MAMV);
2398             return false;
2399         }
2400         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
2401         break;
2402 
2403     default:
2404         error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2405                           ", lo=0x%"PRIx64" (type mismatch: 0x%llx)",
2406                           __func__, inv_desc->hi, inv_desc->lo,
2407                           inv_desc->lo & VTD_INV_DESC_IOTLB_G);
2408         return false;
2409     }
2410     return true;
2411 }
2412 
2413 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
2414                                      VTDInvDesc *inv_desc)
2415 {
2416     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
2417                            inv_desc->iec.index,
2418                            inv_desc->iec.index_mask);
2419 
2420     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
2421                        inv_desc->iec.index,
2422                        inv_desc->iec.index_mask);
2423     return true;
2424 }
2425 
2426 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
2427                                           VTDInvDesc *inv_desc)
2428 {
2429     VTDAddressSpace *vtd_dev_as;
2430     IOMMUTLBEvent event;
2431     struct VTDBus *vtd_bus;
2432     hwaddr addr;
2433     uint64_t sz;
2434     uint16_t sid;
2435     uint8_t devfn;
2436     bool size;
2437     uint8_t bus_num;
2438 
2439     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2440     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2441     devfn = sid & 0xff;
2442     bus_num = sid >> 8;
2443     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2444 
2445     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2446         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2447         error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2448                           ", lo=%"PRIx64" (reserved nonzero)", __func__,
2449                           inv_desc->hi, inv_desc->lo);
2450         return false;
2451     }
2452 
2453     vtd_bus = vtd_find_as_from_bus_num(s, bus_num);
2454     if (!vtd_bus) {
2455         goto done;
2456     }
2457 
2458     vtd_dev_as = vtd_bus->dev_as[devfn];
2459     if (!vtd_dev_as) {
2460         goto done;
2461     }
2462 
2463     /* According to ATS spec table 2.4:
2464      * S = 0, bits 15:12 = xxxx     range size: 4K
2465      * S = 1, bits 15:12 = xxx0     range size: 8K
2466      * S = 1, bits 15:12 = xx01     range size: 16K
2467      * S = 1, bits 15:12 = x011     range size: 32K
2468      * S = 1, bits 15:12 = 0111     range size: 64K
2469      * ...
2470      */
2471     if (size) {
2472         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2473         addr &= ~(sz - 1);
2474     } else {
2475         sz = VTD_PAGE_SIZE;
2476     }
2477 
2478     event.type = IOMMU_NOTIFIER_DEVIOTLB_UNMAP;
2479     event.entry.target_as = &vtd_dev_as->as;
2480     event.entry.addr_mask = sz - 1;
2481     event.entry.iova = addr;
2482     event.entry.perm = IOMMU_NONE;
2483     event.entry.translated_addr = 0;
2484     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, event);
2485 
2486 done:
2487     return true;
2488 }
2489 
2490 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2491 {
2492     VTDInvDesc inv_desc;
2493     uint8_t desc_type;
2494 
2495     trace_vtd_inv_qi_head(s->iq_head);
2496     if (!vtd_get_inv_desc(s, &inv_desc)) {
2497         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2498         return false;
2499     }
2500 
2501     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2502     /* FIXME: should update at first or at last? */
2503     s->iq_last_desc_type = desc_type;
2504 
2505     switch (desc_type) {
2506     case VTD_INV_DESC_CC:
2507         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2508         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2509             return false;
2510         }
2511         break;
2512 
2513     case VTD_INV_DESC_IOTLB:
2514         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2515         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2516             return false;
2517         }
2518         break;
2519 
2520     /*
2521      * TODO: the entity of below two cases will be implemented in future series.
2522      * To make guest (which integrates scalable mode support patch set in
2523      * iommu driver) work, just return true is enough so far.
2524      */
2525     case VTD_INV_DESC_PC:
2526         break;
2527 
2528     case VTD_INV_DESC_PIOTLB:
2529         break;
2530 
2531     case VTD_INV_DESC_WAIT:
2532         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2533         if (!vtd_process_wait_desc(s, &inv_desc)) {
2534             return false;
2535         }
2536         break;
2537 
2538     case VTD_INV_DESC_IEC:
2539         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2540         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2541             return false;
2542         }
2543         break;
2544 
2545     case VTD_INV_DESC_DEVICE:
2546         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2547         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2548             return false;
2549         }
2550         break;
2551 
2552     default:
2553         error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64
2554                           " (unknown type)", __func__, inv_desc.hi,
2555                           inv_desc.lo);
2556         return false;
2557     }
2558     s->iq_head++;
2559     if (s->iq_head == s->iq_size) {
2560         s->iq_head = 0;
2561     }
2562     return true;
2563 }
2564 
2565 /* Try to fetch and process more Invalidation Descriptors */
2566 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2567 {
2568     int qi_shift;
2569 
2570     /* Refer to 10.4.23 of VT-d spec 3.0 */
2571     qi_shift = s->iq_dw ? VTD_IQH_QH_SHIFT_5 : VTD_IQH_QH_SHIFT_4;
2572 
2573     trace_vtd_inv_qi_fetch();
2574 
2575     if (s->iq_tail >= s->iq_size) {
2576         /* Detects an invalid Tail pointer */
2577         error_report_once("%s: detected invalid QI tail "
2578                           "(tail=0x%x, size=0x%x)",
2579                           __func__, s->iq_tail, s->iq_size);
2580         vtd_handle_inv_queue_error(s);
2581         return;
2582     }
2583     while (s->iq_head != s->iq_tail) {
2584         if (!vtd_process_inv_desc(s)) {
2585             /* Invalidation Queue Errors */
2586             vtd_handle_inv_queue_error(s);
2587             break;
2588         }
2589         /* Must update the IQH_REG in time */
2590         vtd_set_quad_raw(s, DMAR_IQH_REG,
2591                          (((uint64_t)(s->iq_head)) << qi_shift) &
2592                          VTD_IQH_QH_MASK);
2593     }
2594 }
2595 
2596 /* Handle write to Invalidation Queue Tail Register */
2597 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2598 {
2599     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2600 
2601     if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) {
2602         error_report_once("%s: RSV bit is set: val=0x%"PRIx64,
2603                           __func__, val);
2604         return;
2605     }
2606     s->iq_tail = VTD_IQT_QT(s->iq_dw, val);
2607     trace_vtd_inv_qi_tail(s->iq_tail);
2608 
2609     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2610         /* Process Invalidation Queue here */
2611         vtd_fetch_inv_desc(s);
2612     }
2613 }
2614 
2615 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2616 {
2617     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2618     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2619     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2620 
2621     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2622         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2623         trace_vtd_fsts_clear_ip();
2624     }
2625     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2626      * Descriptors if there are any when Queued Invalidation is enabled?
2627      */
2628 }
2629 
2630 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2631 {
2632     uint32_t fectl_reg;
2633     /* FIXME: when software clears the IM field, check the IP field. But do we
2634      * need to compare the old value and the new value to conclude that
2635      * software clears the IM field? Or just check if the IM field is zero?
2636      */
2637     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2638 
2639     trace_vtd_reg_write_fectl(fectl_reg);
2640 
2641     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2642         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2643         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2644     }
2645 }
2646 
2647 static void vtd_handle_ics_write(IntelIOMMUState *s)
2648 {
2649     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2650     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2651 
2652     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2653         trace_vtd_reg_ics_clear_ip();
2654         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2655     }
2656 }
2657 
2658 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2659 {
2660     uint32_t iectl_reg;
2661     /* FIXME: when software clears the IM field, check the IP field. But do we
2662      * need to compare the old value and the new value to conclude that
2663      * software clears the IM field? Or just check if the IM field is zero?
2664      */
2665     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2666 
2667     trace_vtd_reg_write_iectl(iectl_reg);
2668 
2669     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2670         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2671         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2672     }
2673 }
2674 
2675 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2676 {
2677     IntelIOMMUState *s = opaque;
2678     uint64_t val;
2679 
2680     trace_vtd_reg_read(addr, size);
2681 
2682     if (addr + size > DMAR_REG_SIZE) {
2683         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2684                           " size=0x%x", __func__, addr, size);
2685         return (uint64_t)-1;
2686     }
2687 
2688     switch (addr) {
2689     /* Root Table Address Register, 64-bit */
2690     case DMAR_RTADDR_REG:
2691         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
2692         if (size == 4) {
2693             val = val & ((1ULL << 32) - 1);
2694         }
2695         break;
2696 
2697     case DMAR_RTADDR_REG_HI:
2698         assert(size == 4);
2699         val = vtd_get_quad_raw(s, DMAR_RTADDR_REG) >> 32;
2700         break;
2701 
2702     /* Invalidation Queue Address Register, 64-bit */
2703     case DMAR_IQA_REG:
2704         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2705         if (size == 4) {
2706             val = val & ((1ULL << 32) - 1);
2707         }
2708         break;
2709 
2710     case DMAR_IQA_REG_HI:
2711         assert(size == 4);
2712         val = s->iq >> 32;
2713         break;
2714 
2715     default:
2716         if (size == 4) {
2717             val = vtd_get_long(s, addr);
2718         } else {
2719             val = vtd_get_quad(s, addr);
2720         }
2721     }
2722 
2723     return val;
2724 }
2725 
2726 static void vtd_mem_write(void *opaque, hwaddr addr,
2727                           uint64_t val, unsigned size)
2728 {
2729     IntelIOMMUState *s = opaque;
2730 
2731     trace_vtd_reg_write(addr, size, val);
2732 
2733     if (addr + size > DMAR_REG_SIZE) {
2734         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2735                           " size=0x%x", __func__, addr, size);
2736         return;
2737     }
2738 
2739     switch (addr) {
2740     /* Global Command Register, 32-bit */
2741     case DMAR_GCMD_REG:
2742         vtd_set_long(s, addr, val);
2743         vtd_handle_gcmd_write(s);
2744         break;
2745 
2746     /* Context Command Register, 64-bit */
2747     case DMAR_CCMD_REG:
2748         if (size == 4) {
2749             vtd_set_long(s, addr, val);
2750         } else {
2751             vtd_set_quad(s, addr, val);
2752             vtd_handle_ccmd_write(s);
2753         }
2754         break;
2755 
2756     case DMAR_CCMD_REG_HI:
2757         assert(size == 4);
2758         vtd_set_long(s, addr, val);
2759         vtd_handle_ccmd_write(s);
2760         break;
2761 
2762     /* IOTLB Invalidation Register, 64-bit */
2763     case DMAR_IOTLB_REG:
2764         if (size == 4) {
2765             vtd_set_long(s, addr, val);
2766         } else {
2767             vtd_set_quad(s, addr, val);
2768             vtd_handle_iotlb_write(s);
2769         }
2770         break;
2771 
2772     case DMAR_IOTLB_REG_HI:
2773         assert(size == 4);
2774         vtd_set_long(s, addr, val);
2775         vtd_handle_iotlb_write(s);
2776         break;
2777 
2778     /* Invalidate Address Register, 64-bit */
2779     case DMAR_IVA_REG:
2780         if (size == 4) {
2781             vtd_set_long(s, addr, val);
2782         } else {
2783             vtd_set_quad(s, addr, val);
2784         }
2785         break;
2786 
2787     case DMAR_IVA_REG_HI:
2788         assert(size == 4);
2789         vtd_set_long(s, addr, val);
2790         break;
2791 
2792     /* Fault Status Register, 32-bit */
2793     case DMAR_FSTS_REG:
2794         assert(size == 4);
2795         vtd_set_long(s, addr, val);
2796         vtd_handle_fsts_write(s);
2797         break;
2798 
2799     /* Fault Event Control Register, 32-bit */
2800     case DMAR_FECTL_REG:
2801         assert(size == 4);
2802         vtd_set_long(s, addr, val);
2803         vtd_handle_fectl_write(s);
2804         break;
2805 
2806     /* Fault Event Data Register, 32-bit */
2807     case DMAR_FEDATA_REG:
2808         assert(size == 4);
2809         vtd_set_long(s, addr, val);
2810         break;
2811 
2812     /* Fault Event Address Register, 32-bit */
2813     case DMAR_FEADDR_REG:
2814         if (size == 4) {
2815             vtd_set_long(s, addr, val);
2816         } else {
2817             /*
2818              * While the register is 32-bit only, some guests (Xen...) write to
2819              * it with 64-bit.
2820              */
2821             vtd_set_quad(s, addr, val);
2822         }
2823         break;
2824 
2825     /* Fault Event Upper Address Register, 32-bit */
2826     case DMAR_FEUADDR_REG:
2827         assert(size == 4);
2828         vtd_set_long(s, addr, val);
2829         break;
2830 
2831     /* Protected Memory Enable Register, 32-bit */
2832     case DMAR_PMEN_REG:
2833         assert(size == 4);
2834         vtd_set_long(s, addr, val);
2835         break;
2836 
2837     /* Root Table Address Register, 64-bit */
2838     case DMAR_RTADDR_REG:
2839         if (size == 4) {
2840             vtd_set_long(s, addr, val);
2841         } else {
2842             vtd_set_quad(s, addr, val);
2843         }
2844         break;
2845 
2846     case DMAR_RTADDR_REG_HI:
2847         assert(size == 4);
2848         vtd_set_long(s, addr, val);
2849         break;
2850 
2851     /* Invalidation Queue Tail Register, 64-bit */
2852     case DMAR_IQT_REG:
2853         if (size == 4) {
2854             vtd_set_long(s, addr, val);
2855         } else {
2856             vtd_set_quad(s, addr, val);
2857         }
2858         vtd_handle_iqt_write(s);
2859         break;
2860 
2861     case DMAR_IQT_REG_HI:
2862         assert(size == 4);
2863         vtd_set_long(s, addr, val);
2864         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
2865         break;
2866 
2867     /* Invalidation Queue Address Register, 64-bit */
2868     case DMAR_IQA_REG:
2869         if (size == 4) {
2870             vtd_set_long(s, addr, val);
2871         } else {
2872             vtd_set_quad(s, addr, val);
2873         }
2874         if (s->ecap & VTD_ECAP_SMTS &&
2875             val & VTD_IQA_DW_MASK) {
2876             s->iq_dw = true;
2877         } else {
2878             s->iq_dw = false;
2879         }
2880         break;
2881 
2882     case DMAR_IQA_REG_HI:
2883         assert(size == 4);
2884         vtd_set_long(s, addr, val);
2885         break;
2886 
2887     /* Invalidation Completion Status Register, 32-bit */
2888     case DMAR_ICS_REG:
2889         assert(size == 4);
2890         vtd_set_long(s, addr, val);
2891         vtd_handle_ics_write(s);
2892         break;
2893 
2894     /* Invalidation Event Control Register, 32-bit */
2895     case DMAR_IECTL_REG:
2896         assert(size == 4);
2897         vtd_set_long(s, addr, val);
2898         vtd_handle_iectl_write(s);
2899         break;
2900 
2901     /* Invalidation Event Data Register, 32-bit */
2902     case DMAR_IEDATA_REG:
2903         assert(size == 4);
2904         vtd_set_long(s, addr, val);
2905         break;
2906 
2907     /* Invalidation Event Address Register, 32-bit */
2908     case DMAR_IEADDR_REG:
2909         assert(size == 4);
2910         vtd_set_long(s, addr, val);
2911         break;
2912 
2913     /* Invalidation Event Upper Address Register, 32-bit */
2914     case DMAR_IEUADDR_REG:
2915         assert(size == 4);
2916         vtd_set_long(s, addr, val);
2917         break;
2918 
2919     /* Fault Recording Registers, 128-bit */
2920     case DMAR_FRCD_REG_0_0:
2921         if (size == 4) {
2922             vtd_set_long(s, addr, val);
2923         } else {
2924             vtd_set_quad(s, addr, val);
2925         }
2926         break;
2927 
2928     case DMAR_FRCD_REG_0_1:
2929         assert(size == 4);
2930         vtd_set_long(s, addr, val);
2931         break;
2932 
2933     case DMAR_FRCD_REG_0_2:
2934         if (size == 4) {
2935             vtd_set_long(s, addr, val);
2936         } else {
2937             vtd_set_quad(s, addr, val);
2938             /* May clear bit 127 (Fault), update PPF */
2939             vtd_update_fsts_ppf(s);
2940         }
2941         break;
2942 
2943     case DMAR_FRCD_REG_0_3:
2944         assert(size == 4);
2945         vtd_set_long(s, addr, val);
2946         /* May clear bit 127 (Fault), update PPF */
2947         vtd_update_fsts_ppf(s);
2948         break;
2949 
2950     case DMAR_IRTA_REG:
2951         if (size == 4) {
2952             vtd_set_long(s, addr, val);
2953         } else {
2954             vtd_set_quad(s, addr, val);
2955         }
2956         break;
2957 
2958     case DMAR_IRTA_REG_HI:
2959         assert(size == 4);
2960         vtd_set_long(s, addr, val);
2961         break;
2962 
2963     default:
2964         if (size == 4) {
2965             vtd_set_long(s, addr, val);
2966         } else {
2967             vtd_set_quad(s, addr, val);
2968         }
2969     }
2970 }
2971 
2972 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
2973                                          IOMMUAccessFlags flag, int iommu_idx)
2974 {
2975     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2976     IntelIOMMUState *s = vtd_as->iommu_state;
2977     IOMMUTLBEntry iotlb = {
2978         /* We'll fill in the rest later. */
2979         .target_as = &address_space_memory,
2980     };
2981     bool success;
2982 
2983     if (likely(s->dmar_enabled)) {
2984         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
2985                                          addr, flag & IOMMU_WO, &iotlb);
2986     } else {
2987         /* DMAR disabled, passthrough, use 4k-page*/
2988         iotlb.iova = addr & VTD_PAGE_MASK_4K;
2989         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
2990         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
2991         iotlb.perm = IOMMU_RW;
2992         success = true;
2993     }
2994 
2995     if (likely(success)) {
2996         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
2997                                  VTD_PCI_SLOT(vtd_as->devfn),
2998                                  VTD_PCI_FUNC(vtd_as->devfn),
2999                                  iotlb.iova, iotlb.translated_addr,
3000                                  iotlb.addr_mask);
3001     } else {
3002         error_report_once("%s: detected translation failure "
3003                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
3004                           __func__, pci_bus_num(vtd_as->bus),
3005                           VTD_PCI_SLOT(vtd_as->devfn),
3006                           VTD_PCI_FUNC(vtd_as->devfn),
3007                           addr);
3008     }
3009 
3010     return iotlb;
3011 }
3012 
3013 static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
3014                                          IOMMUNotifierFlag old,
3015                                          IOMMUNotifierFlag new,
3016                                          Error **errp)
3017 {
3018     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
3019     IntelIOMMUState *s = vtd_as->iommu_state;
3020 
3021     /* Update per-address-space notifier flags */
3022     vtd_as->notifier_flags = new;
3023 
3024     if (old == IOMMU_NOTIFIER_NONE) {
3025         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
3026     } else if (new == IOMMU_NOTIFIER_NONE) {
3027         QLIST_REMOVE(vtd_as, next);
3028     }
3029     return 0;
3030 }
3031 
3032 static int vtd_post_load(void *opaque, int version_id)
3033 {
3034     IntelIOMMUState *iommu = opaque;
3035 
3036     /*
3037      * Memory regions are dynamically turned on/off depending on
3038      * context entry configurations from the guest. After migration,
3039      * we need to make sure the memory regions are still correct.
3040      */
3041     vtd_switch_address_space_all(iommu);
3042 
3043     /*
3044      * We don't need to migrate the root_scalable because we can
3045      * simply do the calculation after the loading is complete.  We
3046      * can actually do similar things with root, dmar_enabled, etc.
3047      * however since we've had them already so we'd better keep them
3048      * for compatibility of migration.
3049      */
3050     vtd_update_scalable_state(iommu);
3051 
3052     return 0;
3053 }
3054 
3055 static const VMStateDescription vtd_vmstate = {
3056     .name = "iommu-intel",
3057     .version_id = 1,
3058     .minimum_version_id = 1,
3059     .priority = MIG_PRI_IOMMU,
3060     .post_load = vtd_post_load,
3061     .fields = (VMStateField[]) {
3062         VMSTATE_UINT64(root, IntelIOMMUState),
3063         VMSTATE_UINT64(intr_root, IntelIOMMUState),
3064         VMSTATE_UINT64(iq, IntelIOMMUState),
3065         VMSTATE_UINT32(intr_size, IntelIOMMUState),
3066         VMSTATE_UINT16(iq_head, IntelIOMMUState),
3067         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
3068         VMSTATE_UINT16(iq_size, IntelIOMMUState),
3069         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
3070         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
3071         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
3072         VMSTATE_UNUSED(1),      /* bool root_extended is obsolete by VT-d */
3073         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
3074         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
3075         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
3076         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
3077         VMSTATE_END_OF_LIST()
3078     }
3079 };
3080 
3081 static const MemoryRegionOps vtd_mem_ops = {
3082     .read = vtd_mem_read,
3083     .write = vtd_mem_write,
3084     .endianness = DEVICE_LITTLE_ENDIAN,
3085     .impl = {
3086         .min_access_size = 4,
3087         .max_access_size = 8,
3088     },
3089     .valid = {
3090         .min_access_size = 4,
3091         .max_access_size = 8,
3092     },
3093 };
3094 
3095 static Property vtd_properties[] = {
3096     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
3097     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
3098                             ON_OFF_AUTO_AUTO),
3099     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
3100     DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits,
3101                       VTD_HOST_ADDRESS_WIDTH),
3102     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
3103     DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE),
3104     DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true),
3105     DEFINE_PROP_END_OF_LIST(),
3106 };
3107 
3108 /* Read IRTE entry with specific index */
3109 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
3110                         VTD_IR_TableEntry *entry, uint16_t sid)
3111 {
3112     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
3113         {0xffff, 0xfffb, 0xfff9, 0xfff8};
3114     dma_addr_t addr = 0x00;
3115     uint16_t mask, source_id;
3116     uint8_t bus, bus_max, bus_min;
3117 
3118     if (index >= iommu->intr_size) {
3119         error_report_once("%s: index too large: ind=0x%x",
3120                           __func__, index);
3121         return -VTD_FR_IR_INDEX_OVER;
3122     }
3123 
3124     addr = iommu->intr_root + index * sizeof(*entry);
3125     if (dma_memory_read(&address_space_memory, addr, entry,
3126                         sizeof(*entry))) {
3127         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
3128                           __func__, index, addr);
3129         return -VTD_FR_IR_ROOT_INVAL;
3130     }
3131 
3132     trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]),
3133                           le64_to_cpu(entry->data[0]));
3134 
3135     if (!entry->irte.present) {
3136         error_report_once("%s: detected non-present IRTE "
3137                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3138                           __func__, index, le64_to_cpu(entry->data[1]),
3139                           le64_to_cpu(entry->data[0]));
3140         return -VTD_FR_IR_ENTRY_P;
3141     }
3142 
3143     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
3144         entry->irte.__reserved_2) {
3145         error_report_once("%s: detected non-zero reserved IRTE "
3146                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
3147                           __func__, index, le64_to_cpu(entry->data[1]),
3148                           le64_to_cpu(entry->data[0]));
3149         return -VTD_FR_IR_IRTE_RSVD;
3150     }
3151 
3152     if (sid != X86_IOMMU_SID_INVALID) {
3153         /* Validate IRTE SID */
3154         source_id = le32_to_cpu(entry->irte.source_id);
3155         switch (entry->irte.sid_vtype) {
3156         case VTD_SVT_NONE:
3157             break;
3158 
3159         case VTD_SVT_ALL:
3160             mask = vtd_svt_mask[entry->irte.sid_q];
3161             if ((source_id & mask) != (sid & mask)) {
3162                 error_report_once("%s: invalid IRTE SID "
3163                                   "(index=%u, sid=%u, source_id=%u)",
3164                                   __func__, index, sid, source_id);
3165                 return -VTD_FR_IR_SID_ERR;
3166             }
3167             break;
3168 
3169         case VTD_SVT_BUS:
3170             bus_max = source_id >> 8;
3171             bus_min = source_id & 0xff;
3172             bus = sid >> 8;
3173             if (bus > bus_max || bus < bus_min) {
3174                 error_report_once("%s: invalid SVT_BUS "
3175                                   "(index=%u, bus=%u, min=%u, max=%u)",
3176                                   __func__, index, bus, bus_min, bus_max);
3177                 return -VTD_FR_IR_SID_ERR;
3178             }
3179             break;
3180 
3181         default:
3182             error_report_once("%s: detected invalid IRTE SVT "
3183                               "(index=%u, type=%d)", __func__,
3184                               index, entry->irte.sid_vtype);
3185             /* Take this as verification failure. */
3186             return -VTD_FR_IR_SID_ERR;
3187         }
3188     }
3189 
3190     return 0;
3191 }
3192 
3193 /* Fetch IRQ information of specific IR index */
3194 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
3195                              X86IOMMUIrq *irq, uint16_t sid)
3196 {
3197     VTD_IR_TableEntry irte = {};
3198     int ret = 0;
3199 
3200     ret = vtd_irte_get(iommu, index, &irte, sid);
3201     if (ret) {
3202         return ret;
3203     }
3204 
3205     irq->trigger_mode = irte.irte.trigger_mode;
3206     irq->vector = irte.irte.vector;
3207     irq->delivery_mode = irte.irte.delivery_mode;
3208     irq->dest = le32_to_cpu(irte.irte.dest_id);
3209     if (!iommu->intr_eime) {
3210 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
3211 #define  VTD_IR_APIC_DEST_SHIFT        (8)
3212         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
3213             VTD_IR_APIC_DEST_SHIFT;
3214     }
3215     irq->dest_mode = irte.irte.dest_mode;
3216     irq->redir_hint = irte.irte.redir_hint;
3217 
3218     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
3219                        irq->delivery_mode, irq->dest, irq->dest_mode);
3220 
3221     return 0;
3222 }
3223 
3224 /* Interrupt remapping for MSI/MSI-X entry */
3225 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
3226                                    MSIMessage *origin,
3227                                    MSIMessage *translated,
3228                                    uint16_t sid)
3229 {
3230     int ret = 0;
3231     VTD_IR_MSIAddress addr;
3232     uint16_t index;
3233     X86IOMMUIrq irq = {};
3234 
3235     assert(origin && translated);
3236 
3237     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
3238 
3239     if (!iommu || !iommu->intr_enabled) {
3240         memcpy(translated, origin, sizeof(*origin));
3241         goto out;
3242     }
3243 
3244     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
3245         error_report_once("%s: MSI address high 32 bits non-zero detected: "
3246                           "address=0x%" PRIx64, __func__, origin->address);
3247         return -VTD_FR_IR_REQ_RSVD;
3248     }
3249 
3250     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
3251     if (addr.addr.__head != 0xfee) {
3252         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
3253                           __func__, addr.data);
3254         return -VTD_FR_IR_REQ_RSVD;
3255     }
3256 
3257     /* This is compatible mode. */
3258     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
3259         memcpy(translated, origin, sizeof(*origin));
3260         goto out;
3261     }
3262 
3263     index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l);
3264 
3265 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
3266 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
3267 
3268     if (addr.addr.sub_valid) {
3269         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3270         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
3271     }
3272 
3273     ret = vtd_remap_irq_get(iommu, index, &irq, sid);
3274     if (ret) {
3275         return ret;
3276     }
3277 
3278     if (addr.addr.sub_valid) {
3279         trace_vtd_ir_remap_type("MSI");
3280         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
3281             error_report_once("%s: invalid IR MSI "
3282                               "(sid=%u, address=0x%" PRIx64
3283                               ", data=0x%" PRIx32 ")",
3284                               __func__, sid, origin->address, origin->data);
3285             return -VTD_FR_IR_REQ_RSVD;
3286         }
3287     } else {
3288         uint8_t vector = origin->data & 0xff;
3289         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
3290 
3291         trace_vtd_ir_remap_type("IOAPIC");
3292         /* IOAPIC entry vector should be aligned with IRTE vector
3293          * (see vt-d spec 5.1.5.1). */
3294         if (vector != irq.vector) {
3295             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
3296         }
3297 
3298         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3299          * (see vt-d spec 5.1.5.1). */
3300         if (trigger_mode != irq.trigger_mode) {
3301             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
3302                                       irq.trigger_mode);
3303         }
3304     }
3305 
3306     /*
3307      * We'd better keep the last two bits, assuming that guest OS
3308      * might modify it. Keep it does not hurt after all.
3309      */
3310     irq.msi_addr_last_bits = addr.addr.__not_care;
3311 
3312     /* Translate X86IOMMUIrq to MSI message */
3313     x86_iommu_irq_to_msi_message(&irq, translated);
3314 
3315 out:
3316     trace_vtd_ir_remap_msi(origin->address, origin->data,
3317                            translated->address, translated->data);
3318     return 0;
3319 }
3320 
3321 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
3322                          MSIMessage *dst, uint16_t sid)
3323 {
3324     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
3325                                    src, dst, sid);
3326 }
3327 
3328 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
3329                                    uint64_t *data, unsigned size,
3330                                    MemTxAttrs attrs)
3331 {
3332     return MEMTX_OK;
3333 }
3334 
3335 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
3336                                     uint64_t value, unsigned size,
3337                                     MemTxAttrs attrs)
3338 {
3339     int ret = 0;
3340     MSIMessage from = {}, to = {};
3341     uint16_t sid = X86_IOMMU_SID_INVALID;
3342 
3343     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
3344     from.data = (uint32_t) value;
3345 
3346     if (!attrs.unspecified) {
3347         /* We have explicit Source ID */
3348         sid = attrs.requester_id;
3349     }
3350 
3351     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid);
3352     if (ret) {
3353         /* TODO: report error */
3354         /* Drop this interrupt */
3355         return MEMTX_ERROR;
3356     }
3357 
3358     apic_get_class()->send_msi(&to);
3359 
3360     return MEMTX_OK;
3361 }
3362 
3363 static const MemoryRegionOps vtd_mem_ir_ops = {
3364     .read_with_attrs = vtd_mem_ir_read,
3365     .write_with_attrs = vtd_mem_ir_write,
3366     .endianness = DEVICE_LITTLE_ENDIAN,
3367     .impl = {
3368         .min_access_size = 4,
3369         .max_access_size = 4,
3370     },
3371     .valid = {
3372         .min_access_size = 4,
3373         .max_access_size = 4,
3374     },
3375 };
3376 
3377 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
3378 {
3379     uintptr_t key = (uintptr_t)bus;
3380     VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
3381     VTDAddressSpace *vtd_dev_as;
3382     char name[128];
3383 
3384     if (!vtd_bus) {
3385         uintptr_t *new_key = g_malloc(sizeof(*new_key));
3386         *new_key = (uintptr_t)bus;
3387         /* No corresponding free() */
3388         vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
3389                             PCI_DEVFN_MAX);
3390         vtd_bus->bus = bus;
3391         g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
3392     }
3393 
3394     vtd_dev_as = vtd_bus->dev_as[devfn];
3395 
3396     if (!vtd_dev_as) {
3397         snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn),
3398                  PCI_FUNC(devfn));
3399         vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace));
3400 
3401         vtd_dev_as->bus = bus;
3402         vtd_dev_as->devfn = (uint8_t)devfn;
3403         vtd_dev_as->iommu_state = s;
3404         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
3405         vtd_dev_as->iova_tree = iova_tree_new();
3406 
3407         memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX);
3408         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root");
3409 
3410         /*
3411          * Build the DMAR-disabled container with aliases to the
3412          * shared MRs.  Note that aliasing to a shared memory region
3413          * could help the memory API to detect same FlatViews so we
3414          * can have devices to share the same FlatView when DMAR is
3415          * disabled (either by not providing "intel_iommu=on" or with
3416          * "iommu=pt").  It will greatly reduce the total number of
3417          * FlatViews of the system hence VM runs faster.
3418          */
3419         memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s),
3420                                  "vtd-nodmar", &s->mr_nodmar, 0,
3421                                  memory_region_size(&s->mr_nodmar));
3422 
3423         /*
3424          * Build the per-device DMAR-enabled container.
3425          *
3426          * TODO: currently we have per-device IOMMU memory region only
3427          * because we have per-device IOMMU notifiers for devices.  If
3428          * one day we can abstract the IOMMU notifiers out of the
3429          * memory regions then we can also share the same memory
3430          * region here just like what we've done above with the nodmar
3431          * region.
3432          */
3433         strcat(name, "-dmar");
3434         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
3435                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
3436                                  name, UINT64_MAX);
3437         memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir",
3438                                  &s->mr_ir, 0, memory_region_size(&s->mr_ir));
3439         memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu),
3440                                             VTD_INTERRUPT_ADDR_FIRST,
3441                                             &vtd_dev_as->iommu_ir, 1);
3442 
3443         /*
3444          * Hook both the containers under the root container, we
3445          * switch between DMAR & noDMAR by enable/disable
3446          * corresponding sub-containers
3447          */
3448         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3449                                             MEMORY_REGION(&vtd_dev_as->iommu),
3450                                             0);
3451         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
3452                                             &vtd_dev_as->nodmar, 0);
3453 
3454         vtd_switch_address_space(vtd_dev_as);
3455     }
3456     return vtd_dev_as;
3457 }
3458 
3459 /* Unmap the whole range in the notifier's scope. */
3460 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
3461 {
3462     hwaddr size, remain;
3463     hwaddr start = n->start;
3464     hwaddr end = n->end;
3465     IntelIOMMUState *s = as->iommu_state;
3466     DMAMap map;
3467 
3468     /*
3469      * Note: all the codes in this function has a assumption that IOVA
3470      * bits are no more than VTD_MGAW bits (which is restricted by
3471      * VT-d spec), otherwise we need to consider overflow of 64 bits.
3472      */
3473 
3474     if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) {
3475         /*
3476          * Don't need to unmap regions that is bigger than the whole
3477          * VT-d supported address space size
3478          */
3479         end = VTD_ADDRESS_SIZE(s->aw_bits) - 1;
3480     }
3481 
3482     assert(start <= end);
3483     size = remain = end - start + 1;
3484 
3485     while (remain >= VTD_PAGE_SIZE) {
3486         IOMMUTLBEvent event;
3487         uint64_t mask = dma_aligned_pow2_mask(start, end, s->aw_bits);
3488         uint64_t size = mask + 1;
3489 
3490         assert(size);
3491 
3492         event.type = IOMMU_NOTIFIER_UNMAP;
3493         event.entry.iova = start;
3494         event.entry.addr_mask = mask;
3495         event.entry.target_as = &address_space_memory;
3496         event.entry.perm = IOMMU_NONE;
3497         /* This field is meaningless for unmap */
3498         event.entry.translated_addr = 0;
3499 
3500         memory_region_notify_iommu_one(n, &event);
3501 
3502         start += size;
3503         remain -= size;
3504     }
3505 
3506     assert(!remain);
3507 
3508     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3509                              VTD_PCI_SLOT(as->devfn),
3510                              VTD_PCI_FUNC(as->devfn),
3511                              n->start, size);
3512 
3513     map.iova = n->start;
3514     map.size = size;
3515     iova_tree_remove(as->iova_tree, &map);
3516 }
3517 
3518 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3519 {
3520     VTDAddressSpace *vtd_as;
3521     IOMMUNotifier *n;
3522 
3523     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3524         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3525             vtd_address_space_unmap(vtd_as, n);
3526         }
3527     }
3528 }
3529 
3530 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3531 {
3532     vtd_address_space_unmap_all(s);
3533     vtd_switch_address_space_all(s);
3534 }
3535 
3536 static int vtd_replay_hook(IOMMUTLBEvent *event, void *private)
3537 {
3538     memory_region_notify_iommu_one(private, event);
3539     return 0;
3540 }
3541 
3542 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3543 {
3544     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3545     IntelIOMMUState *s = vtd_as->iommu_state;
3546     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3547     VTDContextEntry ce;
3548 
3549     /*
3550      * The replay can be triggered by either a invalidation or a newly
3551      * created entry. No matter what, we release existing mappings
3552      * (it means flushing caches for UNMAP-only registers).
3553      */
3554     vtd_address_space_unmap(vtd_as, n);
3555 
3556     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3557         trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" :
3558                                   "legacy mode",
3559                                   bus_n, PCI_SLOT(vtd_as->devfn),
3560                                   PCI_FUNC(vtd_as->devfn),
3561                                   vtd_get_domain_id(s, &ce),
3562                                   ce.hi, ce.lo);
3563         if (vtd_as_has_map_notifier(vtd_as)) {
3564             /* This is required only for MAP typed notifiers */
3565             vtd_page_walk_info info = {
3566                 .hook_fn = vtd_replay_hook,
3567                 .private = (void *)n,
3568                 .notify_unmap = false,
3569                 .aw = s->aw_bits,
3570                 .as = vtd_as,
3571                 .domain_id = vtd_get_domain_id(s, &ce),
3572             };
3573 
3574             vtd_page_walk(s, &ce, 0, ~0ULL, &info);
3575         }
3576     } else {
3577         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3578                                     PCI_FUNC(vtd_as->devfn));
3579     }
3580 
3581     return;
3582 }
3583 
3584 /* Do the initialization. It will also be called when reset, so pay
3585  * attention when adding new initialization stuff.
3586  */
3587 static void vtd_init(IntelIOMMUState *s)
3588 {
3589     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3590 
3591     memset(s->csr, 0, DMAR_REG_SIZE);
3592     memset(s->wmask, 0, DMAR_REG_SIZE);
3593     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3594     memset(s->womask, 0, DMAR_REG_SIZE);
3595 
3596     s->root = 0;
3597     s->root_scalable = false;
3598     s->dmar_enabled = false;
3599     s->intr_enabled = false;
3600     s->iq_head = 0;
3601     s->iq_tail = 0;
3602     s->iq = 0;
3603     s->iq_size = 0;
3604     s->qi_enabled = false;
3605     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3606     s->iq_dw = false;
3607     s->next_frcd_reg = 0;
3608     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3609              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3610              VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits);
3611     if (s->dma_drain) {
3612         s->cap |= VTD_CAP_DRAIN;
3613     }
3614     if (s->aw_bits == VTD_HOST_AW_48BIT) {
3615         s->cap |= VTD_CAP_SAGAW_48bit;
3616     }
3617     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3618 
3619     /*
3620      * Rsvd field masks for spte
3621      */
3622     vtd_spte_rsvd[0] = ~0ULL;
3623     vtd_spte_rsvd[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits,
3624                                                   x86_iommu->dt_supported);
3625     vtd_spte_rsvd[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3626     vtd_spte_rsvd[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3627     vtd_spte_rsvd[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3628 
3629     vtd_spte_rsvd_large[2] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits,
3630                                                          x86_iommu->dt_supported);
3631     vtd_spte_rsvd_large[3] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits,
3632                                                          x86_iommu->dt_supported);
3633 
3634     if (x86_iommu_ir_supported(x86_iommu)) {
3635         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3636         if (s->intr_eim == ON_OFF_AUTO_ON) {
3637             s->ecap |= VTD_ECAP_EIM;
3638         }
3639         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3640     }
3641 
3642     if (x86_iommu->dt_supported) {
3643         s->ecap |= VTD_ECAP_DT;
3644     }
3645 
3646     if (x86_iommu->pt_supported) {
3647         s->ecap |= VTD_ECAP_PT;
3648     }
3649 
3650     if (s->caching_mode) {
3651         s->cap |= VTD_CAP_CM;
3652     }
3653 
3654     /* TODO: read cap/ecap from host to decide which cap to be exposed. */
3655     if (s->scalable_mode) {
3656         s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS;
3657     }
3658 
3659     vtd_reset_caches(s);
3660 
3661     /* Define registers with default values and bit semantics */
3662     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
3663     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
3664     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
3665     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
3666     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
3667     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
3668     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0);
3669     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
3670     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
3671 
3672     /* Advanced Fault Logging not supported */
3673     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
3674     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3675     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
3676     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
3677 
3678     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3679      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3680      */
3681     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
3682 
3683     /* Treated as RO for implementations that PLMR and PHMR fields reported
3684      * as Clear in the CAP_REG.
3685      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3686      */
3687     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
3688 
3689     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
3690     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
3691     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0);
3692     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
3693     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3694     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
3695     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
3696     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3697     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
3698 
3699     /* IOTLB registers */
3700     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
3701     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
3702     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
3703 
3704     /* Fault Recording Registers, 128-bit */
3705     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
3706     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
3707 
3708     /*
3709      * Interrupt remapping registers.
3710      */
3711     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
3712 }
3713 
3714 /* Should not reset address_spaces when reset because devices will still use
3715  * the address space they got at first (won't ask the bus again).
3716  */
3717 static void vtd_reset(DeviceState *dev)
3718 {
3719     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3720 
3721     vtd_init(s);
3722     vtd_address_space_refresh_all(s);
3723 }
3724 
3725 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
3726 {
3727     IntelIOMMUState *s = opaque;
3728     VTDAddressSpace *vtd_as;
3729 
3730     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
3731 
3732     vtd_as = vtd_find_add_as(s, bus, devfn);
3733     return &vtd_as->as;
3734 }
3735 
3736 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
3737 {
3738     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3739 
3740     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) {
3741         error_setg(errp, "eim=on cannot be selected without intremap=on");
3742         return false;
3743     }
3744 
3745     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
3746         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
3747                       && x86_iommu_ir_supported(x86_iommu) ?
3748                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
3749     }
3750     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
3751         if (!kvm_irqchip_in_kernel()) {
3752             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
3753             return false;
3754         }
3755         if (!kvm_enable_x2apic()) {
3756             error_setg(errp, "eim=on requires support on the KVM side"
3757                              "(X2APIC_API, first shipped in v4.7)");
3758             return false;
3759         }
3760     }
3761 
3762     /* Currently only address widths supported are 39 and 48 bits */
3763     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
3764         (s->aw_bits != VTD_HOST_AW_48BIT)) {
3765         error_setg(errp, "Supported values for aw-bits are: %d, %d",
3766                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
3767         return false;
3768     }
3769 
3770     if (s->scalable_mode && !s->dma_drain) {
3771         error_setg(errp, "Need to set dma_drain for scalable mode");
3772         return false;
3773     }
3774 
3775     return true;
3776 }
3777 
3778 static int vtd_machine_done_notify_one(Object *child, void *unused)
3779 {
3780     IntelIOMMUState *iommu = INTEL_IOMMU_DEVICE(x86_iommu_get_default());
3781 
3782     /*
3783      * We hard-coded here because vfio-pci is the only special case
3784      * here.  Let's be more elegant in the future when we can, but so
3785      * far there seems to be no better way.
3786      */
3787     if (object_dynamic_cast(child, "vfio-pci") && !iommu->caching_mode) {
3788         vtd_panic_require_caching_mode();
3789     }
3790 
3791     return 0;
3792 }
3793 
3794 static void vtd_machine_done_hook(Notifier *notifier, void *unused)
3795 {
3796     object_child_foreach_recursive(object_get_root(),
3797                                    vtd_machine_done_notify_one, NULL);
3798 }
3799 
3800 static Notifier vtd_machine_done_notify = {
3801     .notify = vtd_machine_done_hook,
3802 };
3803 
3804 static void vtd_realize(DeviceState *dev, Error **errp)
3805 {
3806     MachineState *ms = MACHINE(qdev_get_machine());
3807     PCMachineState *pcms = PC_MACHINE(ms);
3808     X86MachineState *x86ms = X86_MACHINE(ms);
3809     PCIBus *bus = pcms->bus;
3810     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3811     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
3812 
3813     x86_iommu->type = TYPE_INTEL;
3814 
3815     if (!vtd_decide_config(s, errp)) {
3816         return;
3817     }
3818 
3819     QLIST_INIT(&s->vtd_as_with_notifiers);
3820     qemu_mutex_init(&s->iommu_lock);
3821     memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num));
3822     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
3823                           "intel_iommu", DMAR_REG_SIZE);
3824 
3825     /* Create the shared memory regions by all devices */
3826     memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar",
3827                        UINT64_MAX);
3828     memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops,
3829                           s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE);
3830     memory_region_init_alias(&s->mr_sys_alias, OBJECT(s),
3831                              "vtd-sys-alias", get_system_memory(), 0,
3832                              memory_region_size(get_system_memory()));
3833     memory_region_add_subregion_overlap(&s->mr_nodmar, 0,
3834                                         &s->mr_sys_alias, 0);
3835     memory_region_add_subregion_overlap(&s->mr_nodmar,
3836                                         VTD_INTERRUPT_ADDR_FIRST,
3837                                         &s->mr_ir, 1);
3838 
3839     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
3840     /* No corresponding destroy */
3841     s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3842                                      g_free, g_free);
3843     s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3844                                               g_free, g_free);
3845     vtd_init(s);
3846     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
3847     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
3848     /* Pseudo address space under root PCI bus. */
3849     x86ms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
3850     qemu_add_machine_init_done_notifier(&vtd_machine_done_notify);
3851 }
3852 
3853 static void vtd_class_init(ObjectClass *klass, void *data)
3854 {
3855     DeviceClass *dc = DEVICE_CLASS(klass);
3856     X86IOMMUClass *x86_class = X86_IOMMU_DEVICE_CLASS(klass);
3857 
3858     dc->reset = vtd_reset;
3859     dc->vmsd = &vtd_vmstate;
3860     device_class_set_props(dc, vtd_properties);
3861     dc->hotpluggable = false;
3862     x86_class->realize = vtd_realize;
3863     x86_class->int_remap = vtd_int_remap;
3864     /* Supported by the pc-q35-* machine types */
3865     dc->user_creatable = true;
3866     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
3867     dc->desc = "Intel IOMMU (VT-d) DMA Remapping device";
3868 }
3869 
3870 static const TypeInfo vtd_info = {
3871     .name          = TYPE_INTEL_IOMMU_DEVICE,
3872     .parent        = TYPE_X86_IOMMU_DEVICE,
3873     .instance_size = sizeof(IntelIOMMUState),
3874     .class_init    = vtd_class_init,
3875 };
3876 
3877 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
3878                                                      void *data)
3879 {
3880     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
3881 
3882     imrc->translate = vtd_iommu_translate;
3883     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
3884     imrc->replay = vtd_iommu_replay;
3885 }
3886 
3887 static const TypeInfo vtd_iommu_memory_region_info = {
3888     .parent = TYPE_IOMMU_MEMORY_REGION,
3889     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
3890     .class_init = vtd_iommu_memory_region_class_init,
3891 };
3892 
3893 static void vtd_register_types(void)
3894 {
3895     type_register_static(&vtd_info);
3896     type_register_static(&vtd_iommu_memory_region_info);
3897 }
3898 
3899 type_init(vtd_register_types)
3900