2 * QEMU emulation of an Intel IOMMU (VT-d)
3 * (DMA Remapping device)
5 * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com>
6 * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com>
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.
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.
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/>.
22 #include "qemu/osdep.h"
23 #include "qemu/error-report.h"
24 #include "qemu/main-loop.h"
25 #include "qapi/error.h"
26 #include "hw/sysbus.h"
27 #include "intel_iommu_internal.h"
28 #include "hw/pci/pci.h"
29 #include "hw/pci/pci_bus.h"
30 #include "hw/qdev-properties.h"
31 #include "hw/i386/pc.h"
32 #include "hw/i386/apic-msidef.h"
33 #include "hw/i386/x86-iommu.h"
34 #include "hw/pci-host/q35.h"
35 #include "sysemu/kvm.h"
36 #include "sysemu/dma.h"
37 #include "sysemu/sysemu.h"
38 #include "hw/i386/apic_internal.h"
39 #include "kvm/kvm_i386.h"
40 #include "migration/vmstate.h"
43 /* context entry operations */
44 #define VTD_CE_GET_RID2PASID(ce) \
45 ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK)
46 #define VTD_CE_GET_PASID_DIR_TABLE(ce) \
47 ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK)
50 #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT)
51 #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW))
54 * PCI bus number (or SID) is not reliable since the device is usaully
55 * initalized before guest can configure the PCI bridge
56 * (SECONDARY_BUS_NUMBER).
64 struct vtd_iotlb_key
{
71 static void vtd_address_space_refresh_all(IntelIOMMUState
*s
);
72 static void vtd_address_space_unmap(VTDAddressSpace
*as
, IOMMUNotifier
*n
);
74 static void vtd_panic_require_caching_mode(void)
76 error_report("We need to set caching-mode=on for intel-iommu to enable "
77 "device assignment with IOMMU protection.");
81 static void vtd_define_quad(IntelIOMMUState
*s
, hwaddr addr
, uint64_t val
,
82 uint64_t wmask
, uint64_t w1cmask
)
84 stq_le_p(&s
->csr
[addr
], val
);
85 stq_le_p(&s
->wmask
[addr
], wmask
);
86 stq_le_p(&s
->w1cmask
[addr
], w1cmask
);
89 static void vtd_define_quad_wo(IntelIOMMUState
*s
, hwaddr addr
, uint64_t mask
)
91 stq_le_p(&s
->womask
[addr
], mask
);
94 static void vtd_define_long(IntelIOMMUState
*s
, hwaddr addr
, uint32_t val
,
95 uint32_t wmask
, uint32_t w1cmask
)
97 stl_le_p(&s
->csr
[addr
], val
);
98 stl_le_p(&s
->wmask
[addr
], wmask
);
99 stl_le_p(&s
->w1cmask
[addr
], w1cmask
);
102 static void vtd_define_long_wo(IntelIOMMUState
*s
, hwaddr addr
, uint32_t mask
)
104 stl_le_p(&s
->womask
[addr
], mask
);
107 /* "External" get/set operations */
108 static void vtd_set_quad(IntelIOMMUState
*s
, hwaddr addr
, uint64_t val
)
110 uint64_t oldval
= ldq_le_p(&s
->csr
[addr
]);
111 uint64_t wmask
= ldq_le_p(&s
->wmask
[addr
]);
112 uint64_t w1cmask
= ldq_le_p(&s
->w1cmask
[addr
]);
113 stq_le_p(&s
->csr
[addr
],
114 ((oldval
& ~wmask
) | (val
& wmask
)) & ~(w1cmask
& val
));
117 static void vtd_set_long(IntelIOMMUState
*s
, hwaddr addr
, uint32_t val
)
119 uint32_t oldval
= ldl_le_p(&s
->csr
[addr
]);
120 uint32_t wmask
= ldl_le_p(&s
->wmask
[addr
]);
121 uint32_t w1cmask
= ldl_le_p(&s
->w1cmask
[addr
]);
122 stl_le_p(&s
->csr
[addr
],
123 ((oldval
& ~wmask
) | (val
& wmask
)) & ~(w1cmask
& val
));
126 static uint64_t vtd_get_quad(IntelIOMMUState
*s
, hwaddr addr
)
128 uint64_t val
= ldq_le_p(&s
->csr
[addr
]);
129 uint64_t womask
= ldq_le_p(&s
->womask
[addr
]);
130 return val
& ~womask
;
133 static uint32_t vtd_get_long(IntelIOMMUState
*s
, hwaddr addr
)
135 uint32_t val
= ldl_le_p(&s
->csr
[addr
]);
136 uint32_t womask
= ldl_le_p(&s
->womask
[addr
]);
137 return val
& ~womask
;
140 /* "Internal" get/set operations */
141 static uint64_t vtd_get_quad_raw(IntelIOMMUState
*s
, hwaddr addr
)
143 return ldq_le_p(&s
->csr
[addr
]);
146 static uint32_t vtd_get_long_raw(IntelIOMMUState
*s
, hwaddr addr
)
148 return ldl_le_p(&s
->csr
[addr
]);
151 static void vtd_set_quad_raw(IntelIOMMUState
*s
, hwaddr addr
, uint64_t val
)
153 stq_le_p(&s
->csr
[addr
], val
);
156 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState
*s
, hwaddr addr
,
157 uint32_t clear
, uint32_t mask
)
159 uint32_t new_val
= (ldl_le_p(&s
->csr
[addr
]) & ~clear
) | mask
;
160 stl_le_p(&s
->csr
[addr
], new_val
);
164 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState
*s
, hwaddr addr
,
165 uint64_t clear
, uint64_t mask
)
167 uint64_t new_val
= (ldq_le_p(&s
->csr
[addr
]) & ~clear
) | mask
;
168 stq_le_p(&s
->csr
[addr
], new_val
);
172 static inline void vtd_iommu_lock(IntelIOMMUState
*s
)
174 qemu_mutex_lock(&s
->iommu_lock
);
177 static inline void vtd_iommu_unlock(IntelIOMMUState
*s
)
179 qemu_mutex_unlock(&s
->iommu_lock
);
182 static void vtd_update_scalable_state(IntelIOMMUState
*s
)
184 uint64_t val
= vtd_get_quad_raw(s
, DMAR_RTADDR_REG
);
186 if (s
->scalable_mode
) {
187 s
->root_scalable
= val
& VTD_RTADDR_SMT
;
191 static void vtd_update_iq_dw(IntelIOMMUState
*s
)
193 uint64_t val
= vtd_get_quad_raw(s
, DMAR_IQA_REG
);
195 if (s
->ecap
& VTD_ECAP_SMTS
&&
196 val
& VTD_IQA_DW_MASK
) {
203 /* Whether the address space needs to notify new mappings */
204 static inline gboolean
vtd_as_has_map_notifier(VTDAddressSpace
*as
)
206 return as
->notifier_flags
& IOMMU_NOTIFIER_MAP
;
209 /* GHashTable functions */
210 static gboolean
vtd_iotlb_equal(gconstpointer v1
, gconstpointer v2
)
212 const struct vtd_iotlb_key
*key1
= v1
;
213 const struct vtd_iotlb_key
*key2
= v2
;
215 return key1
->sid
== key2
->sid
&&
216 key1
->pasid
== key2
->pasid
&&
217 key1
->level
== key2
->level
&&
218 key1
->gfn
== key2
->gfn
;
221 static guint
vtd_iotlb_hash(gconstpointer v
)
223 const struct vtd_iotlb_key
*key
= v
;
224 uint64_t hash64
= key
->gfn
| ((uint64_t)(key
->sid
) << VTD_IOTLB_SID_SHIFT
) |
225 (uint64_t)(key
->level
- 1) << VTD_IOTLB_LVL_SHIFT
|
226 (uint64_t)(key
->pasid
) << VTD_IOTLB_PASID_SHIFT
;
228 return (guint
)((hash64
>> 32) ^ (hash64
& 0xffffffffU
));
231 static gboolean
vtd_as_equal(gconstpointer v1
, gconstpointer v2
)
233 const struct vtd_as_key
*key1
= v1
;
234 const struct vtd_as_key
*key2
= v2
;
236 return (key1
->bus
== key2
->bus
) && (key1
->devfn
== key2
->devfn
) &&
237 (key1
->pasid
== key2
->pasid
);
241 * Note that we use pointer to PCIBus as the key, so hashing/shifting
242 * based on the pointer value is intended. Note that we deal with
243 * collisions through vtd_as_equal().
245 static guint
vtd_as_hash(gconstpointer v
)
247 const struct vtd_as_key
*key
= v
;
248 guint value
= (guint
)(uintptr_t)key
->bus
;
250 return (guint
)(value
<< 8 | key
->devfn
);
253 static gboolean
vtd_hash_remove_by_domain(gpointer key
, gpointer value
,
256 VTDIOTLBEntry
*entry
= (VTDIOTLBEntry
*)value
;
257 uint16_t domain_id
= *(uint16_t *)user_data
;
258 return entry
->domain_id
== domain_id
;
261 /* The shift of an addr for a certain level of paging structure */
262 static inline uint32_t vtd_slpt_level_shift(uint32_t level
)
265 return VTD_PAGE_SHIFT_4K
+ (level
- 1) * VTD_SL_LEVEL_BITS
;
268 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level
)
270 return ~((1ULL << vtd_slpt_level_shift(level
)) - 1);
273 static gboolean
vtd_hash_remove_by_page(gpointer key
, gpointer value
,
276 VTDIOTLBEntry
*entry
= (VTDIOTLBEntry
*)value
;
277 VTDIOTLBPageInvInfo
*info
= (VTDIOTLBPageInvInfo
*)user_data
;
278 uint64_t gfn
= (info
->addr
>> VTD_PAGE_SHIFT_4K
) & info
->mask
;
279 uint64_t gfn_tlb
= (info
->addr
& entry
->mask
) >> VTD_PAGE_SHIFT_4K
;
280 return (entry
->domain_id
== info
->domain_id
) &&
281 (((entry
->gfn
& info
->mask
) == gfn
) ||
282 (entry
->gfn
== gfn_tlb
));
285 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
286 * IntelIOMMUState to 1. Must be called with IOMMU lock held.
288 static void vtd_reset_context_cache_locked(IntelIOMMUState
*s
)
290 VTDAddressSpace
*vtd_as
;
291 GHashTableIter as_it
;
293 trace_vtd_context_cache_reset();
295 g_hash_table_iter_init(&as_it
, s
->vtd_address_spaces
);
297 while (g_hash_table_iter_next(&as_it
, NULL
, (void **)&vtd_as
)) {
298 vtd_as
->context_cache_entry
.context_cache_gen
= 0;
300 s
->context_cache_gen
= 1;
303 /* Must be called with IOMMU lock held. */
304 static void vtd_reset_iotlb_locked(IntelIOMMUState
*s
)
307 g_hash_table_remove_all(s
->iotlb
);
310 static void vtd_reset_iotlb(IntelIOMMUState
*s
)
313 vtd_reset_iotlb_locked(s
);
317 static void vtd_reset_caches(IntelIOMMUState
*s
)
320 vtd_reset_iotlb_locked(s
);
321 vtd_reset_context_cache_locked(s
);
325 static uint64_t vtd_get_iotlb_gfn(hwaddr addr
, uint32_t level
)
327 return (addr
& vtd_slpt_level_page_mask(level
)) >> VTD_PAGE_SHIFT_4K
;
330 /* Must be called with IOMMU lock held */
331 static VTDIOTLBEntry
*vtd_lookup_iotlb(IntelIOMMUState
*s
, uint16_t source_id
,
332 uint32_t pasid
, hwaddr addr
)
334 struct vtd_iotlb_key key
;
335 VTDIOTLBEntry
*entry
;
338 for (level
= VTD_SL_PT_LEVEL
; level
< VTD_SL_PML4_LEVEL
; level
++) {
339 key
.gfn
= vtd_get_iotlb_gfn(addr
, level
);
343 entry
= g_hash_table_lookup(s
->iotlb
, &key
);
353 /* Must be with IOMMU lock held */
354 static void vtd_update_iotlb(IntelIOMMUState
*s
, uint16_t source_id
,
355 uint16_t domain_id
, hwaddr addr
, uint64_t slpte
,
356 uint8_t access_flags
, uint32_t level
,
359 VTDIOTLBEntry
*entry
= g_malloc(sizeof(*entry
));
360 struct vtd_iotlb_key
*key
= g_malloc(sizeof(*key
));
361 uint64_t gfn
= vtd_get_iotlb_gfn(addr
, level
);
363 trace_vtd_iotlb_page_update(source_id
, addr
, slpte
, domain_id
);
364 if (g_hash_table_size(s
->iotlb
) >= VTD_IOTLB_MAX_SIZE
) {
365 trace_vtd_iotlb_reset("iotlb exceeds size limit");
366 vtd_reset_iotlb_locked(s
);
370 entry
->domain_id
= domain_id
;
371 entry
->slpte
= slpte
;
372 entry
->access_flags
= access_flags
;
373 entry
->mask
= vtd_slpt_level_page_mask(level
);
374 entry
->pasid
= pasid
;
377 key
->sid
= source_id
;
381 g_hash_table_replace(s
->iotlb
, key
, entry
);
384 /* Given the reg addr of both the message data and address, generate an
387 static void vtd_generate_interrupt(IntelIOMMUState
*s
, hwaddr mesg_addr_reg
,
388 hwaddr mesg_data_reg
)
392 assert(mesg_data_reg
< DMAR_REG_SIZE
);
393 assert(mesg_addr_reg
< DMAR_REG_SIZE
);
395 msi
.address
= vtd_get_long_raw(s
, mesg_addr_reg
);
396 msi
.data
= vtd_get_long_raw(s
, mesg_data_reg
);
398 trace_vtd_irq_generate(msi
.address
, msi
.data
);
400 apic_get_class(NULL
)->send_msi(&msi
);
403 /* Generate a fault event to software via MSI if conditions are met.
404 * Notice that the value of FSTS_REG being passed to it should be the one
407 static void vtd_generate_fault_event(IntelIOMMUState
*s
, uint32_t pre_fsts
)
409 if (pre_fsts
& VTD_FSTS_PPF
|| pre_fsts
& VTD_FSTS_PFO
||
410 pre_fsts
& VTD_FSTS_IQE
) {
411 error_report_once("There are previous interrupt conditions "
412 "to be serviced by software, fault event "
416 vtd_set_clear_mask_long(s
, DMAR_FECTL_REG
, 0, VTD_FECTL_IP
);
417 if (vtd_get_long_raw(s
, DMAR_FECTL_REG
) & VTD_FECTL_IM
) {
418 error_report_once("Interrupt Mask set, irq is not generated");
420 vtd_generate_interrupt(s
, DMAR_FEADDR_REG
, DMAR_FEDATA_REG
);
421 vtd_set_clear_mask_long(s
, DMAR_FECTL_REG
, VTD_FECTL_IP
, 0);
425 /* Check if the Fault (F) field of the Fault Recording Register referenced by
428 static bool vtd_is_frcd_set(IntelIOMMUState
*s
, uint16_t index
)
430 /* Each reg is 128-bit */
431 hwaddr addr
= DMAR_FRCD_REG_OFFSET
+ (((uint64_t)index
) << 4);
432 addr
+= 8; /* Access the high 64-bit half */
434 assert(index
< DMAR_FRCD_REG_NR
);
436 return vtd_get_quad_raw(s
, addr
) & VTD_FRCD_F
;
439 /* Update the PPF field of Fault Status Register.
440 * Should be called whenever change the F field of any fault recording
443 static void vtd_update_fsts_ppf(IntelIOMMUState
*s
)
446 uint32_t ppf_mask
= 0;
448 for (i
= 0; i
< DMAR_FRCD_REG_NR
; i
++) {
449 if (vtd_is_frcd_set(s
, i
)) {
450 ppf_mask
= VTD_FSTS_PPF
;
454 vtd_set_clear_mask_long(s
, DMAR_FSTS_REG
, VTD_FSTS_PPF
, ppf_mask
);
455 trace_vtd_fsts_ppf(!!ppf_mask
);
458 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState
*s
, uint16_t index
)
460 /* Each reg is 128-bit */
461 hwaddr addr
= DMAR_FRCD_REG_OFFSET
+ (((uint64_t)index
) << 4);
462 addr
+= 8; /* Access the high 64-bit half */
464 assert(index
< DMAR_FRCD_REG_NR
);
466 vtd_set_clear_mask_quad(s
, addr
, 0, VTD_FRCD_F
);
467 vtd_update_fsts_ppf(s
);
470 /* Must not update F field now, should be done later */
471 static void vtd_record_frcd(IntelIOMMUState
*s
, uint16_t index
,
472 uint16_t source_id
, hwaddr addr
,
473 VTDFaultReason fault
, bool is_write
,
474 bool is_pasid
, uint32_t pasid
)
477 hwaddr frcd_reg_addr
= DMAR_FRCD_REG_OFFSET
+ (((uint64_t)index
) << 4);
479 assert(index
< DMAR_FRCD_REG_NR
);
481 lo
= VTD_FRCD_FI(addr
);
482 hi
= VTD_FRCD_SID(source_id
) | VTD_FRCD_FR(fault
) |
483 VTD_FRCD_PV(pasid
) | VTD_FRCD_PP(is_pasid
);
487 vtd_set_quad_raw(s
, frcd_reg_addr
, lo
);
488 vtd_set_quad_raw(s
, frcd_reg_addr
+ 8, hi
);
490 trace_vtd_frr_new(index
, hi
, lo
);
493 /* Try to collapse multiple pending faults from the same requester */
494 static bool vtd_try_collapse_fault(IntelIOMMUState
*s
, uint16_t source_id
)
498 hwaddr addr
= DMAR_FRCD_REG_OFFSET
+ 8; /* The high 64-bit half */
500 for (i
= 0; i
< DMAR_FRCD_REG_NR
; i
++) {
501 frcd_reg
= vtd_get_quad_raw(s
, addr
);
502 if ((frcd_reg
& VTD_FRCD_F
) &&
503 ((frcd_reg
& VTD_FRCD_SID_MASK
) == source_id
)) {
506 addr
+= 16; /* 128-bit for each */
511 /* Log and report an DMAR (address translation) fault to software */
512 static void vtd_report_dmar_fault(IntelIOMMUState
*s
, uint16_t source_id
,
513 hwaddr addr
, VTDFaultReason fault
,
514 bool is_write
, bool is_pasid
,
517 uint32_t fsts_reg
= vtd_get_long_raw(s
, DMAR_FSTS_REG
);
519 assert(fault
< VTD_FR_MAX
);
521 trace_vtd_dmar_fault(source_id
, fault
, addr
, is_write
);
523 if (fsts_reg
& VTD_FSTS_PFO
) {
524 error_report_once("New fault is not recorded due to "
525 "Primary Fault Overflow");
529 if (vtd_try_collapse_fault(s
, source_id
)) {
530 error_report_once("New fault is not recorded due to "
531 "compression of faults");
535 if (vtd_is_frcd_set(s
, s
->next_frcd_reg
)) {
536 error_report_once("Next Fault Recording Reg is used, "
537 "new fault is not recorded, set PFO field");
538 vtd_set_clear_mask_long(s
, DMAR_FSTS_REG
, 0, VTD_FSTS_PFO
);
542 vtd_record_frcd(s
, s
->next_frcd_reg
, source_id
, addr
, fault
,
543 is_write
, is_pasid
, pasid
);
545 if (fsts_reg
& VTD_FSTS_PPF
) {
546 error_report_once("There are pending faults already, "
547 "fault event is not generated");
548 vtd_set_frcd_and_update_ppf(s
, s
->next_frcd_reg
);
550 if (s
->next_frcd_reg
== DMAR_FRCD_REG_NR
) {
551 s
->next_frcd_reg
= 0;
554 vtd_set_clear_mask_long(s
, DMAR_FSTS_REG
, VTD_FSTS_FRI_MASK
,
555 VTD_FSTS_FRI(s
->next_frcd_reg
));
556 vtd_set_frcd_and_update_ppf(s
, s
->next_frcd_reg
); /* Will set PPF */
558 if (s
->next_frcd_reg
== DMAR_FRCD_REG_NR
) {
559 s
->next_frcd_reg
= 0;
561 /* This case actually cause the PPF to be Set.
562 * So generate fault event (interrupt).
564 vtd_generate_fault_event(s
, fsts_reg
);
568 /* Handle Invalidation Queue Errors of queued invalidation interface error
571 static void vtd_handle_inv_queue_error(IntelIOMMUState
*s
)
573 uint32_t fsts_reg
= vtd_get_long_raw(s
, DMAR_FSTS_REG
);
575 vtd_set_clear_mask_long(s
, DMAR_FSTS_REG
, 0, VTD_FSTS_IQE
);
576 vtd_generate_fault_event(s
, fsts_reg
);
579 /* Set the IWC field and try to generate an invalidation completion interrupt */
580 static void vtd_generate_completion_event(IntelIOMMUState
*s
)
582 if (vtd_get_long_raw(s
, DMAR_ICS_REG
) & VTD_ICS_IWC
) {
583 trace_vtd_inv_desc_wait_irq("One pending, skip current");
586 vtd_set_clear_mask_long(s
, DMAR_ICS_REG
, 0, VTD_ICS_IWC
);
587 vtd_set_clear_mask_long(s
, DMAR_IECTL_REG
, 0, VTD_IECTL_IP
);
588 if (vtd_get_long_raw(s
, DMAR_IECTL_REG
) & VTD_IECTL_IM
) {
589 trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
590 "new event not generated");
593 /* Generate the interrupt event */
594 trace_vtd_inv_desc_wait_irq("Generating complete event");
595 vtd_generate_interrupt(s
, DMAR_IEADDR_REG
, DMAR_IEDATA_REG
);
596 vtd_set_clear_mask_long(s
, DMAR_IECTL_REG
, VTD_IECTL_IP
, 0);
600 static inline bool vtd_root_entry_present(IntelIOMMUState
*s
,
604 if (s
->root_scalable
&& devfn
> UINT8_MAX
/ 2) {
605 return re
->hi
& VTD_ROOT_ENTRY_P
;
608 return re
->lo
& VTD_ROOT_ENTRY_P
;
611 static int vtd_get_root_entry(IntelIOMMUState
*s
, uint8_t index
,
616 addr
= s
->root
+ index
* sizeof(*re
);
617 if (dma_memory_read(&address_space_memory
, addr
,
618 re
, sizeof(*re
), MEMTXATTRS_UNSPECIFIED
)) {
620 return -VTD_FR_ROOT_TABLE_INV
;
622 re
->lo
= le64_to_cpu(re
->lo
);
623 re
->hi
= le64_to_cpu(re
->hi
);
627 static inline bool vtd_ce_present(VTDContextEntry
*context
)
629 return context
->lo
& VTD_CONTEXT_ENTRY_P
;
632 static int vtd_get_context_entry_from_root(IntelIOMMUState
*s
,
637 dma_addr_t addr
, ce_size
;
639 /* we have checked that root entry is present */
640 ce_size
= s
->root_scalable
? VTD_CTX_ENTRY_SCALABLE_SIZE
:
641 VTD_CTX_ENTRY_LEGACY_SIZE
;
643 if (s
->root_scalable
&& index
> UINT8_MAX
/ 2) {
644 index
= index
& (~VTD_DEVFN_CHECK_MASK
);
645 addr
= re
->hi
& VTD_ROOT_ENTRY_CTP
;
647 addr
= re
->lo
& VTD_ROOT_ENTRY_CTP
;
650 addr
= addr
+ index
* ce_size
;
651 if (dma_memory_read(&address_space_memory
, addr
,
652 ce
, ce_size
, MEMTXATTRS_UNSPECIFIED
)) {
653 return -VTD_FR_CONTEXT_TABLE_INV
;
656 ce
->lo
= le64_to_cpu(ce
->lo
);
657 ce
->hi
= le64_to_cpu(ce
->hi
);
658 if (ce_size
== VTD_CTX_ENTRY_SCALABLE_SIZE
) {
659 ce
->val
[2] = le64_to_cpu(ce
->val
[2]);
660 ce
->val
[3] = le64_to_cpu(ce
->val
[3]);
665 static inline dma_addr_t
vtd_ce_get_slpt_base(VTDContextEntry
*ce
)
667 return ce
->lo
& VTD_CONTEXT_ENTRY_SLPTPTR
;
670 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte
, uint8_t aw
)
672 return slpte
& VTD_SL_PT_BASE_ADDR_MASK(aw
);
675 /* Whether the pte indicates the address of the page frame */
676 static inline bool vtd_is_last_slpte(uint64_t slpte
, uint32_t level
)
678 return level
== VTD_SL_PT_LEVEL
|| (slpte
& VTD_SL_PT_PAGE_SIZE_MASK
);
681 /* Get the content of a spte located in @base_addr[@index] */
682 static uint64_t vtd_get_slpte(dma_addr_t base_addr
, uint32_t index
)
686 assert(index
< VTD_SL_PT_ENTRY_NR
);
688 if (dma_memory_read(&address_space_memory
,
689 base_addr
+ index
* sizeof(slpte
),
690 &slpte
, sizeof(slpte
), MEMTXATTRS_UNSPECIFIED
)) {
691 slpte
= (uint64_t)-1;
694 slpte
= le64_to_cpu(slpte
);
698 /* Given an iova and the level of paging structure, return the offset
701 static inline uint32_t vtd_iova_level_offset(uint64_t iova
, uint32_t level
)
703 return (iova
>> vtd_slpt_level_shift(level
)) &
704 ((1ULL << VTD_SL_LEVEL_BITS
) - 1);
707 /* Check Capability Register to see if the @level of page-table is supported */
708 static inline bool vtd_is_level_supported(IntelIOMMUState
*s
, uint32_t level
)
710 return VTD_CAP_SAGAW_MASK
& s
->cap
&
711 (1ULL << (level
- 2 + VTD_CAP_SAGAW_SHIFT
));
714 /* Return true if check passed, otherwise false */
715 static inline bool vtd_pe_type_check(X86IOMMUState
*x86_iommu
,
718 switch (VTD_PE_GET_TYPE(pe
)) {
719 case VTD_SM_PASID_ENTRY_FLT
:
720 case VTD_SM_PASID_ENTRY_SLT
:
721 case VTD_SM_PASID_ENTRY_NESTED
:
723 case VTD_SM_PASID_ENTRY_PT
:
724 if (!x86_iommu
->pt_supported
) {
735 static inline bool vtd_pdire_present(VTDPASIDDirEntry
*pdire
)
737 return pdire
->val
& 1;
741 * Caller of this function should check present bit if wants
742 * to use pdir entry for further usage except for fpd bit check.
744 static int vtd_get_pdire_from_pdir_table(dma_addr_t pasid_dir_base
,
746 VTDPASIDDirEntry
*pdire
)
749 dma_addr_t addr
, entry_size
;
751 index
= VTD_PASID_DIR_INDEX(pasid
);
752 entry_size
= VTD_PASID_DIR_ENTRY_SIZE
;
753 addr
= pasid_dir_base
+ index
* entry_size
;
754 if (dma_memory_read(&address_space_memory
, addr
,
755 pdire
, entry_size
, MEMTXATTRS_UNSPECIFIED
)) {
756 return -VTD_FR_PASID_TABLE_INV
;
762 static inline bool vtd_pe_present(VTDPASIDEntry
*pe
)
764 return pe
->val
[0] & VTD_PASID_ENTRY_P
;
767 static int vtd_get_pe_in_pasid_leaf_table(IntelIOMMUState
*s
,
773 dma_addr_t entry_size
;
774 X86IOMMUState
*x86_iommu
= X86_IOMMU_DEVICE(s
);
776 index
= VTD_PASID_TABLE_INDEX(pasid
);
777 entry_size
= VTD_PASID_ENTRY_SIZE
;
778 addr
= addr
+ index
* entry_size
;
779 if (dma_memory_read(&address_space_memory
, addr
,
780 pe
, entry_size
, MEMTXATTRS_UNSPECIFIED
)) {
781 return -VTD_FR_PASID_TABLE_INV
;
784 /* Do translation type check */
785 if (!vtd_pe_type_check(x86_iommu
, pe
)) {
786 return -VTD_FR_PASID_TABLE_INV
;
789 if (!vtd_is_level_supported(s
, VTD_PE_GET_LEVEL(pe
))) {
790 return -VTD_FR_PASID_TABLE_INV
;
797 * Caller of this function should check present bit if wants
798 * to use pasid entry for further usage except for fpd bit check.
800 static int vtd_get_pe_from_pdire(IntelIOMMUState
*s
,
802 VTDPASIDDirEntry
*pdire
,
805 dma_addr_t addr
= pdire
->val
& VTD_PASID_TABLE_BASE_ADDR_MASK
;
807 return vtd_get_pe_in_pasid_leaf_table(s
, pasid
, addr
, pe
);
811 * This function gets a pasid entry from a specified pasid
812 * table (includes dir and leaf table) with a specified pasid.
813 * Sanity check should be done to ensure return a present
814 * pasid entry to caller.
816 static int vtd_get_pe_from_pasid_table(IntelIOMMUState
*s
,
817 dma_addr_t pasid_dir_base
,
822 VTDPASIDDirEntry pdire
;
824 ret
= vtd_get_pdire_from_pdir_table(pasid_dir_base
,
830 if (!vtd_pdire_present(&pdire
)) {
831 return -VTD_FR_PASID_TABLE_INV
;
834 ret
= vtd_get_pe_from_pdire(s
, pasid
, &pdire
, pe
);
839 if (!vtd_pe_present(pe
)) {
840 return -VTD_FR_PASID_TABLE_INV
;
846 static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState
*s
,
851 dma_addr_t pasid_dir_base
;
854 if (pasid
== PCI_NO_PASID
) {
855 pasid
= VTD_CE_GET_RID2PASID(ce
);
857 pasid_dir_base
= VTD_CE_GET_PASID_DIR_TABLE(ce
);
858 ret
= vtd_get_pe_from_pasid_table(s
, pasid_dir_base
, pasid
, pe
);
863 static int vtd_ce_get_pasid_fpd(IntelIOMMUState
*s
,
869 dma_addr_t pasid_dir_base
;
870 VTDPASIDDirEntry pdire
;
873 if (pasid
== PCI_NO_PASID
) {
874 pasid
= VTD_CE_GET_RID2PASID(ce
);
876 pasid_dir_base
= VTD_CE_GET_PASID_DIR_TABLE(ce
);
879 * No present bit check since fpd is meaningful even
880 * if the present bit is clear.
882 ret
= vtd_get_pdire_from_pdir_table(pasid_dir_base
, pasid
, &pdire
);
887 if (pdire
.val
& VTD_PASID_DIR_FPD
) {
892 if (!vtd_pdire_present(&pdire
)) {
893 return -VTD_FR_PASID_TABLE_INV
;
897 * No present bit check since fpd is meaningful even
898 * if the present bit is clear.
900 ret
= vtd_get_pe_from_pdire(s
, pasid
, &pdire
, &pe
);
905 if (pe
.val
[0] & VTD_PASID_ENTRY_FPD
) {
912 /* Get the page-table level that hardware should use for the second-level
913 * page-table walk from the Address Width field of context-entry.
915 static inline uint32_t vtd_ce_get_level(VTDContextEntry
*ce
)
917 return 2 + (ce
->hi
& VTD_CONTEXT_ENTRY_AW
);
920 static uint32_t vtd_get_iova_level(IntelIOMMUState
*s
,
926 if (s
->root_scalable
) {
927 vtd_ce_get_rid2pasid_entry(s
, ce
, &pe
, pasid
);
928 return VTD_PE_GET_LEVEL(&pe
);
931 return vtd_ce_get_level(ce
);
934 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry
*ce
)
936 return 30 + (ce
->hi
& VTD_CONTEXT_ENTRY_AW
) * 9;
939 static uint32_t vtd_get_iova_agaw(IntelIOMMUState
*s
,
945 if (s
->root_scalable
) {
946 vtd_ce_get_rid2pasid_entry(s
, ce
, &pe
, pasid
);
947 return 30 + ((pe
.val
[0] >> 2) & VTD_SM_PASID_ENTRY_AW
) * 9;
950 return vtd_ce_get_agaw(ce
);
953 static inline uint32_t vtd_ce_get_type(VTDContextEntry
*ce
)
955 return ce
->lo
& VTD_CONTEXT_ENTRY_TT
;
958 /* Only for Legacy Mode. Return true if check passed, otherwise false */
959 static inline bool vtd_ce_type_check(X86IOMMUState
*x86_iommu
,
962 switch (vtd_ce_get_type(ce
)) {
963 case VTD_CONTEXT_TT_MULTI_LEVEL
:
964 /* Always supported */
966 case VTD_CONTEXT_TT_DEV_IOTLB
:
967 if (!x86_iommu
->dt_supported
) {
968 error_report_once("%s: DT specified but not supported", __func__
);
972 case VTD_CONTEXT_TT_PASS_THROUGH
:
973 if (!x86_iommu
->pt_supported
) {
974 error_report_once("%s: PT specified but not supported", __func__
);
980 error_report_once("%s: unknown ce type: %"PRIu32
, __func__
,
981 vtd_ce_get_type(ce
));
987 static inline uint64_t vtd_iova_limit(IntelIOMMUState
*s
,
988 VTDContextEntry
*ce
, uint8_t aw
,
991 uint32_t ce_agaw
= vtd_get_iova_agaw(s
, ce
, pasid
);
992 return 1ULL << MIN(ce_agaw
, aw
);
995 /* Return true if IOVA passes range check, otherwise false. */
996 static inline bool vtd_iova_range_check(IntelIOMMUState
*s
,
997 uint64_t iova
, VTDContextEntry
*ce
,
998 uint8_t aw
, uint32_t pasid
)
1001 * Check if @iova is above 2^X-1, where X is the minimum of MGAW
1002 * in CAP_REG and AW in context-entry.
1004 return !(iova
& ~(vtd_iova_limit(s
, ce
, aw
, pasid
) - 1));
1007 static dma_addr_t
vtd_get_iova_pgtbl_base(IntelIOMMUState
*s
,
1008 VTDContextEntry
*ce
,
1013 if (s
->root_scalable
) {
1014 vtd_ce_get_rid2pasid_entry(s
, ce
, &pe
, pasid
);
1015 return pe
.val
[0] & VTD_SM_PASID_ENTRY_SLPTPTR
;
1018 return vtd_ce_get_slpt_base(ce
);
1022 * Rsvd field masks for spte:
1023 * vtd_spte_rsvd 4k pages
1024 * vtd_spte_rsvd_large large pages
1026 static uint64_t vtd_spte_rsvd
[5];
1027 static uint64_t vtd_spte_rsvd_large
[5];
1029 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte
, uint32_t level
)
1031 uint64_t rsvd_mask
= vtd_spte_rsvd
[level
];
1033 if ((level
== VTD_SL_PD_LEVEL
|| level
== VTD_SL_PDP_LEVEL
) &&
1034 (slpte
& VTD_SL_PT_PAGE_SIZE_MASK
)) {
1036 rsvd_mask
= vtd_spte_rsvd_large
[level
];
1039 return slpte
& rsvd_mask
;
1042 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
1043 * of the translation, can be used for deciding the size of large page.
1045 static int vtd_iova_to_slpte(IntelIOMMUState
*s
, VTDContextEntry
*ce
,
1046 uint64_t iova
, bool is_write
,
1047 uint64_t *slptep
, uint32_t *slpte_level
,
1048 bool *reads
, bool *writes
, uint8_t aw_bits
,
1051 dma_addr_t addr
= vtd_get_iova_pgtbl_base(s
, ce
, pasid
);
1052 uint32_t level
= vtd_get_iova_level(s
, ce
, pasid
);
1055 uint64_t access_right_check
;
1056 uint64_t xlat
, size
;
1058 if (!vtd_iova_range_check(s
, iova
, ce
, aw_bits
, pasid
)) {
1059 error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64
","
1060 "pasid=0x%" PRIx32
")", __func__
, iova
, pasid
);
1061 return -VTD_FR_ADDR_BEYOND_MGAW
;
1064 /* FIXME: what is the Atomics request here? */
1065 access_right_check
= is_write
? VTD_SL_W
: VTD_SL_R
;
1068 offset
= vtd_iova_level_offset(iova
, level
);
1069 slpte
= vtd_get_slpte(addr
, offset
);
1071 if (slpte
== (uint64_t)-1) {
1072 error_report_once("%s: detected read error on DMAR slpte "
1073 "(iova=0x%" PRIx64
", pasid=0x%" PRIx32
")",
1074 __func__
, iova
, pasid
);
1075 if (level
== vtd_get_iova_level(s
, ce
, pasid
)) {
1076 /* Invalid programming of context-entry */
1077 return -VTD_FR_CONTEXT_ENTRY_INV
;
1079 return -VTD_FR_PAGING_ENTRY_INV
;
1082 *reads
= (*reads
) && (slpte
& VTD_SL_R
);
1083 *writes
= (*writes
) && (slpte
& VTD_SL_W
);
1084 if (!(slpte
& access_right_check
)) {
1085 error_report_once("%s: detected slpte permission error "
1086 "(iova=0x%" PRIx64
", level=0x%" PRIx32
", "
1087 "slpte=0x%" PRIx64
", write=%d, pasid=0x%"
1088 PRIx32
")", __func__
, iova
, level
,
1089 slpte
, is_write
, pasid
);
1090 return is_write
? -VTD_FR_WRITE
: -VTD_FR_READ
;
1092 if (vtd_slpte_nonzero_rsvd(slpte
, level
)) {
1093 error_report_once("%s: detected splte reserve non-zero "
1094 "iova=0x%" PRIx64
", level=0x%" PRIx32
1095 "slpte=0x%" PRIx64
", pasid=0x%" PRIX32
")",
1096 __func__
, iova
, level
, slpte
, pasid
);
1097 return -VTD_FR_PAGING_ENTRY_RSVD
;
1100 if (vtd_is_last_slpte(slpte
, level
)) {
1102 *slpte_level
= level
;
1105 addr
= vtd_get_slpte_addr(slpte
, aw_bits
);
1109 xlat
= vtd_get_slpte_addr(*slptep
, aw_bits
);
1110 size
= ~vtd_slpt_level_page_mask(level
) + 1;
1113 * From VT-d spec 3.14: Untranslated requests and translation
1114 * requests that result in an address in the interrupt range will be
1115 * blocked with condition code LGN.4 or SGN.8.
1117 if ((xlat
> VTD_INTERRUPT_ADDR_LAST
||
1118 xlat
+ size
- 1 < VTD_INTERRUPT_ADDR_FIRST
)) {
1121 error_report_once("%s: xlat address is in interrupt range "
1122 "(iova=0x%" PRIx64
", level=0x%" PRIx32
", "
1123 "slpte=0x%" PRIx64
", write=%d, "
1124 "xlat=0x%" PRIx64
", size=0x%" PRIx64
", "
1125 "pasid=0x%" PRIx32
")",
1126 __func__
, iova
, level
, slpte
, is_write
,
1128 return s
->scalable_mode
? -VTD_FR_SM_INTERRUPT_ADDR
:
1129 -VTD_FR_INTERRUPT_ADDR
;
1133 typedef int (*vtd_page_walk_hook
)(IOMMUTLBEvent
*event
, void *private);
1136 * Constant information used during page walking
1138 * @hook_fn: hook func to be called when detected page
1139 * @private: private data to be passed into hook func
1140 * @notify_unmap: whether we should notify invalid entries
1141 * @as: VT-d address space of the device
1142 * @aw: maximum address width
1143 * @domain: domain ID of the page walk
1146 VTDAddressSpace
*as
;
1147 vtd_page_walk_hook hook_fn
;
1152 } vtd_page_walk_info
;
1154 static int vtd_page_walk_one(IOMMUTLBEvent
*event
, vtd_page_walk_info
*info
)
1156 VTDAddressSpace
*as
= info
->as
;
1157 vtd_page_walk_hook hook_fn
= info
->hook_fn
;
1158 void *private = info
->private;
1159 IOMMUTLBEntry
*entry
= &event
->entry
;
1161 .iova
= entry
->iova
,
1162 .size
= entry
->addr_mask
,
1163 .translated_addr
= entry
->translated_addr
,
1164 .perm
= entry
->perm
,
1166 const DMAMap
*mapped
= iova_tree_find(as
->iova_tree
, &target
);
1168 if (event
->type
== IOMMU_NOTIFIER_UNMAP
&& !info
->notify_unmap
) {
1169 trace_vtd_page_walk_one_skip_unmap(entry
->iova
, entry
->addr_mask
);
1175 /* Update local IOVA mapped ranges */
1176 if (event
->type
== IOMMU_NOTIFIER_MAP
) {
1178 /* If it's exactly the same translation, skip */
1179 if (!memcmp(mapped
, &target
, sizeof(target
))) {
1180 trace_vtd_page_walk_one_skip_map(entry
->iova
, entry
->addr_mask
,
1181 entry
->translated_addr
);
1185 * Translation changed. Normally this should not
1186 * happen, but it can happen when with buggy guest
1187 * OSes. Note that there will be a small window that
1188 * we don't have map at all. But that's the best
1189 * effort we can do. The ideal way to emulate this is
1190 * atomically modify the PTE to follow what has
1191 * changed, but we can't. One example is that vfio
1192 * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
1193 * interface to modify a mapping (meanwhile it seems
1194 * meaningless to even provide one). Anyway, let's
1195 * mark this as a TODO in case one day we'll have
1196 * a better solution.
1198 IOMMUAccessFlags cache_perm
= entry
->perm
;
1201 /* Emulate an UNMAP */
1202 event
->type
= IOMMU_NOTIFIER_UNMAP
;
1203 entry
->perm
= IOMMU_NONE
;
1204 trace_vtd_page_walk_one(info
->domain_id
,
1206 entry
->translated_addr
,
1209 ret
= hook_fn(event
, private);
1213 /* Drop any existing mapping */
1214 iova_tree_remove(as
->iova_tree
, target
);
1215 /* Recover the correct type */
1216 event
->type
= IOMMU_NOTIFIER_MAP
;
1217 entry
->perm
= cache_perm
;
1220 iova_tree_insert(as
->iova_tree
, &target
);
1223 /* Skip since we didn't map this range at all */
1224 trace_vtd_page_walk_one_skip_unmap(entry
->iova
, entry
->addr_mask
);
1227 iova_tree_remove(as
->iova_tree
, target
);
1230 trace_vtd_page_walk_one(info
->domain_id
, entry
->iova
,
1231 entry
->translated_addr
, entry
->addr_mask
,
1233 return hook_fn(event
, private);
1237 * vtd_page_walk_level - walk over specific level for IOVA range
1239 * @addr: base GPA addr to start the walk
1240 * @start: IOVA range start address
1241 * @end: IOVA range end address (start <= addr < end)
1242 * @read: whether parent level has read permission
1243 * @write: whether parent level has write permission
1244 * @info: constant information for the page walk
1246 static int vtd_page_walk_level(dma_addr_t addr
, uint64_t start
,
1247 uint64_t end
, uint32_t level
, bool read
,
1248 bool write
, vtd_page_walk_info
*info
)
1250 bool read_cur
, write_cur
, entry_valid
;
1253 uint64_t subpage_size
, subpage_mask
;
1254 IOMMUTLBEvent event
;
1255 uint64_t iova
= start
;
1259 trace_vtd_page_walk_level(addr
, level
, start
, end
);
1261 subpage_size
= 1ULL << vtd_slpt_level_shift(level
);
1262 subpage_mask
= vtd_slpt_level_page_mask(level
);
1264 while (iova
< end
) {
1265 iova_next
= (iova
& subpage_mask
) + subpage_size
;
1267 offset
= vtd_iova_level_offset(iova
, level
);
1268 slpte
= vtd_get_slpte(addr
, offset
);
1270 if (slpte
== (uint64_t)-1) {
1271 trace_vtd_page_walk_skip_read(iova
, iova_next
);
1275 if (vtd_slpte_nonzero_rsvd(slpte
, level
)) {
1276 trace_vtd_page_walk_skip_reserve(iova
, iova_next
);
1280 /* Permissions are stacked with parents' */
1281 read_cur
= read
&& (slpte
& VTD_SL_R
);
1282 write_cur
= write
&& (slpte
& VTD_SL_W
);
1285 * As long as we have either read/write permission, this is a
1286 * valid entry. The rule works for both page entries and page
1289 entry_valid
= read_cur
| write_cur
;
1291 if (!vtd_is_last_slpte(slpte
, level
) && entry_valid
) {
1293 * This is a valid PDE (or even bigger than PDE). We need
1294 * to walk one further level.
1296 ret
= vtd_page_walk_level(vtd_get_slpte_addr(slpte
, info
->aw
),
1297 iova
, MIN(iova_next
, end
), level
- 1,
1298 read_cur
, write_cur
, info
);
1301 * This means we are either:
1303 * (1) the real page entry (either 4K page, or huge page)
1304 * (2) the whole range is invalid
1306 * In either case, we send an IOTLB notification down.
1308 event
.entry
.target_as
= &address_space_memory
;
1309 event
.entry
.iova
= iova
& subpage_mask
;
1310 event
.entry
.perm
= IOMMU_ACCESS_FLAG(read_cur
, write_cur
);
1311 event
.entry
.addr_mask
= ~subpage_mask
;
1312 /* NOTE: this is only meaningful if entry_valid == true */
1313 event
.entry
.translated_addr
= vtd_get_slpte_addr(slpte
, info
->aw
);
1314 event
.type
= event
.entry
.perm
? IOMMU_NOTIFIER_MAP
:
1315 IOMMU_NOTIFIER_UNMAP
;
1316 ret
= vtd_page_walk_one(&event
, info
);
1331 * vtd_page_walk - walk specific IOVA range, and call the hook
1333 * @s: intel iommu state
1334 * @ce: context entry to walk upon
1335 * @start: IOVA address to start the walk
1336 * @end: IOVA range end address (start <= addr < end)
1337 * @info: page walking information struct
1339 static int vtd_page_walk(IntelIOMMUState
*s
, VTDContextEntry
*ce
,
1340 uint64_t start
, uint64_t end
,
1341 vtd_page_walk_info
*info
,
1344 dma_addr_t addr
= vtd_get_iova_pgtbl_base(s
, ce
, pasid
);
1345 uint32_t level
= vtd_get_iova_level(s
, ce
, pasid
);
1347 if (!vtd_iova_range_check(s
, start
, ce
, info
->aw
, pasid
)) {
1348 return -VTD_FR_ADDR_BEYOND_MGAW
;
1351 if (!vtd_iova_range_check(s
, end
, ce
, info
->aw
, pasid
)) {
1352 /* Fix end so that it reaches the maximum */
1353 end
= vtd_iova_limit(s
, ce
, info
->aw
, pasid
);
1356 return vtd_page_walk_level(addr
, start
, end
, level
, true, true, info
);
1359 static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState
*s
,
1362 /* Legacy Mode reserved bits check */
1363 if (!s
->root_scalable
&&
1364 (re
->hi
|| (re
->lo
& VTD_ROOT_ENTRY_RSVD(s
->aw_bits
))))
1367 /* Scalable Mode reserved bits check */
1368 if (s
->root_scalable
&&
1369 ((re
->lo
& VTD_ROOT_ENTRY_RSVD(s
->aw_bits
)) ||
1370 (re
->hi
& VTD_ROOT_ENTRY_RSVD(s
->aw_bits
))))
1376 error_report_once("%s: invalid root entry: hi=0x%"PRIx64
1378 __func__
, re
->hi
, re
->lo
);
1379 return -VTD_FR_ROOT_ENTRY_RSVD
;
1382 static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState
*s
,
1383 VTDContextEntry
*ce
)
1385 if (!s
->root_scalable
&&
1386 (ce
->hi
& VTD_CONTEXT_ENTRY_RSVD_HI
||
1387 ce
->lo
& VTD_CONTEXT_ENTRY_RSVD_LO(s
->aw_bits
))) {
1388 error_report_once("%s: invalid context entry: hi=%"PRIx64
1389 ", lo=%"PRIx64
" (reserved nonzero)",
1390 __func__
, ce
->hi
, ce
->lo
);
1391 return -VTD_FR_CONTEXT_ENTRY_RSVD
;
1394 if (s
->root_scalable
&&
1395 (ce
->val
[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s
->aw_bits
) ||
1396 ce
->val
[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1
||
1399 error_report_once("%s: invalid context entry: val[3]=%"PRIx64
1402 ", val[0]=%"PRIx64
" (reserved nonzero)",
1403 __func__
, ce
->val
[3], ce
->val
[2],
1404 ce
->val
[1], ce
->val
[0]);
1405 return -VTD_FR_CONTEXT_ENTRY_RSVD
;
1411 static int vtd_ce_rid2pasid_check(IntelIOMMUState
*s
,
1412 VTDContextEntry
*ce
)
1417 * Make sure in Scalable Mode, a present context entry
1418 * has valid rid2pasid setting, which includes valid
1419 * rid2pasid field and corresponding pasid entry setting
1421 return vtd_ce_get_rid2pasid_entry(s
, ce
, &pe
, PCI_NO_PASID
);
1424 /* Map a device to its corresponding domain (context-entry) */
1425 static int vtd_dev_to_context_entry(IntelIOMMUState
*s
, uint8_t bus_num
,
1426 uint8_t devfn
, VTDContextEntry
*ce
)
1430 X86IOMMUState
*x86_iommu
= X86_IOMMU_DEVICE(s
);
1432 ret_fr
= vtd_get_root_entry(s
, bus_num
, &re
);
1437 if (!vtd_root_entry_present(s
, &re
, devfn
)) {
1438 /* Not error - it's okay we don't have root entry. */
1439 trace_vtd_re_not_present(bus_num
);
1440 return -VTD_FR_ROOT_ENTRY_P
;
1443 ret_fr
= vtd_root_entry_rsvd_bits_check(s
, &re
);
1448 ret_fr
= vtd_get_context_entry_from_root(s
, &re
, devfn
, ce
);
1453 if (!vtd_ce_present(ce
)) {
1454 /* Not error - it's okay we don't have context entry. */
1455 trace_vtd_ce_not_present(bus_num
, devfn
);
1456 return -VTD_FR_CONTEXT_ENTRY_P
;
1459 ret_fr
= vtd_context_entry_rsvd_bits_check(s
, ce
);
1464 /* Check if the programming of context-entry is valid */
1465 if (!s
->root_scalable
&&
1466 !vtd_is_level_supported(s
, vtd_ce_get_level(ce
))) {
1467 error_report_once("%s: invalid context entry: hi=%"PRIx64
1468 ", lo=%"PRIx64
" (level %d not supported)",
1469 __func__
, ce
->hi
, ce
->lo
,
1470 vtd_ce_get_level(ce
));
1471 return -VTD_FR_CONTEXT_ENTRY_INV
;
1474 if (!s
->root_scalable
) {
1475 /* Do translation type check */
1476 if (!vtd_ce_type_check(x86_iommu
, ce
)) {
1477 /* Errors dumped in vtd_ce_type_check() */
1478 return -VTD_FR_CONTEXT_ENTRY_INV
;
1482 * Check if the programming of context-entry.rid2pasid
1483 * and corresponding pasid setting is valid, and thus
1484 * avoids to check pasid entry fetching result in future
1485 * helper function calling.
1487 ret_fr
= vtd_ce_rid2pasid_check(s
, ce
);
1496 static int vtd_sync_shadow_page_hook(IOMMUTLBEvent
*event
,
1499 memory_region_notify_iommu(private, 0, *event
);
1503 static uint16_t vtd_get_domain_id(IntelIOMMUState
*s
,
1504 VTDContextEntry
*ce
,
1509 if (s
->root_scalable
) {
1510 vtd_ce_get_rid2pasid_entry(s
, ce
, &pe
, pasid
);
1511 return VTD_SM_PASID_ENTRY_DID(pe
.val
[1]);
1514 return VTD_CONTEXT_ENTRY_DID(ce
->hi
);
1517 static int vtd_sync_shadow_page_table_range(VTDAddressSpace
*vtd_as
,
1518 VTDContextEntry
*ce
,
1519 hwaddr addr
, hwaddr size
)
1521 IntelIOMMUState
*s
= vtd_as
->iommu_state
;
1522 vtd_page_walk_info info
= {
1523 .hook_fn
= vtd_sync_shadow_page_hook
,
1524 .private = (void *)&vtd_as
->iommu
,
1525 .notify_unmap
= true,
1528 .domain_id
= vtd_get_domain_id(s
, ce
, vtd_as
->pasid
),
1531 return vtd_page_walk(s
, ce
, addr
, addr
+ size
, &info
, vtd_as
->pasid
);
1534 static int vtd_address_space_sync(VTDAddressSpace
*vtd_as
)
1540 /* If no MAP notifier registered, we simply invalidate all the cache */
1541 if (!vtd_as_has_map_notifier(vtd_as
)) {
1542 IOMMU_NOTIFIER_FOREACH(n
, &vtd_as
->iommu
) {
1543 memory_region_unmap_iommu_notifier_range(n
);
1548 ret
= vtd_dev_to_context_entry(vtd_as
->iommu_state
,
1549 pci_bus_num(vtd_as
->bus
),
1550 vtd_as
->devfn
, &ce
);
1552 if (ret
== -VTD_FR_CONTEXT_ENTRY_P
) {
1554 * It's a valid scenario to have a context entry that is
1555 * not present. For example, when a device is removed
1556 * from an existing domain then the context entry will be
1557 * zeroed by the guest before it was put into another
1558 * domain. When this happens, instead of synchronizing
1559 * the shadow pages we should invalidate all existing
1560 * mappings and notify the backends.
1562 IOMMU_NOTIFIER_FOREACH(n
, &vtd_as
->iommu
) {
1563 vtd_address_space_unmap(vtd_as
, n
);
1570 return vtd_sync_shadow_page_table_range(vtd_as
, &ce
, 0, UINT64_MAX
);
1574 * Check if specific device is configured to bypass address
1575 * translation for DMA requests. In Scalable Mode, bypass
1576 * 1st-level translation or 2nd-level translation, it depends
1579 static bool vtd_dev_pt_enabled(IntelIOMMUState
*s
, VTDContextEntry
*ce
,
1585 if (s
->root_scalable
) {
1586 ret
= vtd_ce_get_rid2pasid_entry(s
, ce
, &pe
, pasid
);
1589 * This error is guest triggerable. We should assumt PT
1590 * not enabled for safety.
1594 return (VTD_PE_GET_TYPE(&pe
) == VTD_SM_PASID_ENTRY_PT
);
1597 return (vtd_ce_get_type(ce
) == VTD_CONTEXT_TT_PASS_THROUGH
);
1601 static bool vtd_as_pt_enabled(VTDAddressSpace
*as
)
1608 s
= as
->iommu_state
;
1609 if (vtd_dev_to_context_entry(s
, pci_bus_num(as
->bus
), as
->devfn
,
1612 * Possibly failed to parse the context entry for some reason
1613 * (e.g., during init, or any guest configuration errors on
1614 * context entries). We should assume PT not enabled for
1620 return vtd_dev_pt_enabled(s
, &ce
, as
->pasid
);
1623 /* Return whether the device is using IOMMU translation. */
1624 static bool vtd_switch_address_space(VTDAddressSpace
*as
)
1627 /* Whether we need to take the BQL on our own */
1628 bool take_bql
= !qemu_mutex_iothread_locked();
1632 use_iommu
= as
->iommu_state
->dmar_enabled
&& !vtd_as_pt_enabled(as
);
1633 pt
= as
->iommu_state
->dmar_enabled
&& vtd_as_pt_enabled(as
);
1635 trace_vtd_switch_address_space(pci_bus_num(as
->bus
),
1636 VTD_PCI_SLOT(as
->devfn
),
1637 VTD_PCI_FUNC(as
->devfn
),
1641 * It's possible that we reach here without BQL, e.g., when called
1642 * from vtd_pt_enable_fast_path(). However the memory APIs need
1643 * it. We'd better make sure we have had it already, or, take it.
1646 qemu_mutex_lock_iothread();
1649 /* Turn off first then on the other */
1651 memory_region_set_enabled(&as
->nodmar
, false);
1652 memory_region_set_enabled(MEMORY_REGION(&as
->iommu
), true);
1654 * vt-d spec v3.4 3.14:
1657 * Requests-with-PASID with input address in range 0xFEEx_xxxx
1658 * are translated normally like any other request-with-PASID
1659 * through DMA-remapping hardware.
1662 * Need to disable ir for as with PASID.
1664 if (as
->pasid
!= PCI_NO_PASID
) {
1665 memory_region_set_enabled(&as
->iommu_ir
, false);
1667 memory_region_set_enabled(&as
->iommu_ir
, true);
1670 memory_region_set_enabled(MEMORY_REGION(&as
->iommu
), false);
1671 memory_region_set_enabled(&as
->nodmar
, true);
1675 * vtd-spec v3.4 3.14:
1678 * Requests-with-PASID with input address in range 0xFEEx_xxxx are
1679 * translated normally like any other request-with-PASID through
1680 * DMA-remapping hardware. However, if such a request is processed
1681 * using pass-through translation, it will be blocked as described
1682 * in the paragraph below.
1684 * Software must not program paging-structure entries to remap any
1685 * address to the interrupt address range. Untranslated requests
1686 * and translation requests that result in an address in the
1687 * interrupt range will be blocked with condition code LGN.4 or
1691 * We enable per as memory region (iommu_ir_fault) for catching
1692 * the tranlsation for interrupt range through PASID + PT.
1694 if (pt
&& as
->pasid
!= PCI_NO_PASID
) {
1695 memory_region_set_enabled(&as
->iommu_ir_fault
, true);
1697 memory_region_set_enabled(&as
->iommu_ir_fault
, false);
1701 qemu_mutex_unlock_iothread();
1707 static void vtd_switch_address_space_all(IntelIOMMUState
*s
)
1709 VTDAddressSpace
*vtd_as
;
1710 GHashTableIter iter
;
1712 g_hash_table_iter_init(&iter
, s
->vtd_address_spaces
);
1713 while (g_hash_table_iter_next(&iter
, NULL
, (void **)&vtd_as
)) {
1714 vtd_switch_address_space(vtd_as
);
1718 static const bool vtd_qualified_faults
[] = {
1719 [VTD_FR_RESERVED
] = false,
1720 [VTD_FR_ROOT_ENTRY_P
] = false,
1721 [VTD_FR_CONTEXT_ENTRY_P
] = true,
1722 [VTD_FR_CONTEXT_ENTRY_INV
] = true,
1723 [VTD_FR_ADDR_BEYOND_MGAW
] = true,
1724 [VTD_FR_WRITE
] = true,
1725 [VTD_FR_READ
] = true,
1726 [VTD_FR_PAGING_ENTRY_INV
] = true,
1727 [VTD_FR_ROOT_TABLE_INV
] = false,
1728 [VTD_FR_CONTEXT_TABLE_INV
] = false,
1729 [VTD_FR_INTERRUPT_ADDR
] = true,
1730 [VTD_FR_ROOT_ENTRY_RSVD
] = false,
1731 [VTD_FR_PAGING_ENTRY_RSVD
] = true,
1732 [VTD_FR_CONTEXT_ENTRY_TT
] = true,
1733 [VTD_FR_PASID_TABLE_INV
] = false,
1734 [VTD_FR_SM_INTERRUPT_ADDR
] = true,
1735 [VTD_FR_MAX
] = false,
1738 /* To see if a fault condition is "qualified", which is reported to software
1739 * only if the FPD field in the context-entry used to process the faulting
1742 static inline bool vtd_is_qualified_fault(VTDFaultReason fault
)
1744 return vtd_qualified_faults
[fault
];
1747 static inline bool vtd_is_interrupt_addr(hwaddr addr
)
1749 return VTD_INTERRUPT_ADDR_FIRST
<= addr
&& addr
<= VTD_INTERRUPT_ADDR_LAST
;
1752 static gboolean
vtd_find_as_by_sid(gpointer key
, gpointer value
,
1755 struct vtd_as_key
*as_key
= (struct vtd_as_key
*)key
;
1756 uint16_t target_sid
= *(uint16_t *)user_data
;
1757 uint16_t sid
= PCI_BUILD_BDF(pci_bus_num(as_key
->bus
), as_key
->devfn
);
1758 return sid
== target_sid
;
1761 static VTDAddressSpace
*vtd_get_as_by_sid(IntelIOMMUState
*s
, uint16_t sid
)
1763 uint8_t bus_num
= PCI_BUS_NUM(sid
);
1764 VTDAddressSpace
*vtd_as
= s
->vtd_as_cache
[bus_num
];
1767 (sid
== PCI_BUILD_BDF(pci_bus_num(vtd_as
->bus
), vtd_as
->devfn
))) {
1771 vtd_as
= g_hash_table_find(s
->vtd_address_spaces
, vtd_find_as_by_sid
, &sid
);
1772 s
->vtd_as_cache
[bus_num
] = vtd_as
;
1777 static void vtd_pt_enable_fast_path(IntelIOMMUState
*s
, uint16_t source_id
)
1779 VTDAddressSpace
*vtd_as
;
1780 bool success
= false;
1782 vtd_as
= vtd_get_as_by_sid(s
, source_id
);
1787 if (vtd_switch_address_space(vtd_as
) == false) {
1788 /* We switched off IOMMU region successfully. */
1793 trace_vtd_pt_enable_fast_path(source_id
, success
);
1796 static void vtd_report_fault(IntelIOMMUState
*s
,
1797 int err
, bool is_fpd_set
,
1804 if (is_fpd_set
&& vtd_is_qualified_fault(err
)) {
1805 trace_vtd_fault_disabled();
1807 vtd_report_dmar_fault(s
, source_id
, addr
, err
, is_write
,
1812 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1815 * Called from RCU critical section.
1817 * @bus_num: The bus number
1818 * @devfn: The devfn, which is the combined of device and function number
1819 * @is_write: The access is a write operation
1820 * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1822 * Returns true if translation is successful, otherwise false.
1824 static bool vtd_do_iommu_translate(VTDAddressSpace
*vtd_as
, PCIBus
*bus
,
1825 uint8_t devfn
, hwaddr addr
, bool is_write
,
1826 IOMMUTLBEntry
*entry
)
1828 IntelIOMMUState
*s
= vtd_as
->iommu_state
;
1830 uint8_t bus_num
= pci_bus_num(bus
);
1831 VTDContextCacheEntry
*cc_entry
;
1832 uint64_t slpte
, page_mask
;
1833 uint32_t level
, pasid
= vtd_as
->pasid
;
1834 uint16_t source_id
= PCI_BUILD_BDF(bus_num
, devfn
);
1836 bool is_fpd_set
= false;
1839 uint8_t access_flags
;
1840 bool rid2pasid
= (pasid
== PCI_NO_PASID
) && s
->root_scalable
;
1841 VTDIOTLBEntry
*iotlb_entry
;
1844 * We have standalone memory region for interrupt addresses, we
1845 * should never receive translation requests in this region.
1847 assert(!vtd_is_interrupt_addr(addr
));
1851 cc_entry
= &vtd_as
->context_cache_entry
;
1853 /* Try to fetch slpte form IOTLB, we don't need RID2PASID logic */
1855 iotlb_entry
= vtd_lookup_iotlb(s
, source_id
, pasid
, addr
);
1857 trace_vtd_iotlb_page_hit(source_id
, addr
, iotlb_entry
->slpte
,
1858 iotlb_entry
->domain_id
);
1859 slpte
= iotlb_entry
->slpte
;
1860 access_flags
= iotlb_entry
->access_flags
;
1861 page_mask
= iotlb_entry
->mask
;
1866 /* Try to fetch context-entry from cache first */
1867 if (cc_entry
->context_cache_gen
== s
->context_cache_gen
) {
1868 trace_vtd_iotlb_cc_hit(bus_num
, devfn
, cc_entry
->context_entry
.hi
,
1869 cc_entry
->context_entry
.lo
,
1870 cc_entry
->context_cache_gen
);
1871 ce
= cc_entry
->context_entry
;
1872 is_fpd_set
= ce
.lo
& VTD_CONTEXT_ENTRY_FPD
;
1873 if (!is_fpd_set
&& s
->root_scalable
) {
1874 ret_fr
= vtd_ce_get_pasid_fpd(s
, &ce
, &is_fpd_set
, pasid
);
1876 vtd_report_fault(s
, -ret_fr
, is_fpd_set
,
1877 source_id
, addr
, is_write
,
1883 ret_fr
= vtd_dev_to_context_entry(s
, bus_num
, devfn
, &ce
);
1884 is_fpd_set
= ce
.lo
& VTD_CONTEXT_ENTRY_FPD
;
1885 if (!ret_fr
&& !is_fpd_set
&& s
->root_scalable
) {
1886 ret_fr
= vtd_ce_get_pasid_fpd(s
, &ce
, &is_fpd_set
, pasid
);
1889 vtd_report_fault(s
, -ret_fr
, is_fpd_set
,
1890 source_id
, addr
, is_write
,
1894 /* Update context-cache */
1895 trace_vtd_iotlb_cc_update(bus_num
, devfn
, ce
.hi
, ce
.lo
,
1896 cc_entry
->context_cache_gen
,
1897 s
->context_cache_gen
);
1898 cc_entry
->context_entry
= ce
;
1899 cc_entry
->context_cache_gen
= s
->context_cache_gen
;
1903 pasid
= VTD_CE_GET_RID2PASID(&ce
);
1907 * We don't need to translate for pass-through context entries.
1908 * Also, let's ignore IOTLB caching as well for PT devices.
1910 if (vtd_dev_pt_enabled(s
, &ce
, pasid
)) {
1911 entry
->iova
= addr
& VTD_PAGE_MASK_4K
;
1912 entry
->translated_addr
= entry
->iova
;
1913 entry
->addr_mask
= ~VTD_PAGE_MASK_4K
;
1914 entry
->perm
= IOMMU_RW
;
1915 trace_vtd_translate_pt(source_id
, entry
->iova
);
1918 * When this happens, it means firstly caching-mode is not
1919 * enabled, and this is the first passthrough translation for
1920 * the device. Let's enable the fast path for passthrough.
1922 * When passthrough is disabled again for the device, we can
1923 * capture it via the context entry invalidation, then the
1924 * IOMMU region can be swapped back.
1926 vtd_pt_enable_fast_path(s
, source_id
);
1927 vtd_iommu_unlock(s
);
1931 /* Try to fetch slpte form IOTLB for RID2PASID slow path */
1933 iotlb_entry
= vtd_lookup_iotlb(s
, source_id
, pasid
, addr
);
1935 trace_vtd_iotlb_page_hit(source_id
, addr
, iotlb_entry
->slpte
,
1936 iotlb_entry
->domain_id
);
1937 slpte
= iotlb_entry
->slpte
;
1938 access_flags
= iotlb_entry
->access_flags
;
1939 page_mask
= iotlb_entry
->mask
;
1944 ret_fr
= vtd_iova_to_slpte(s
, &ce
, addr
, is_write
, &slpte
, &level
,
1945 &reads
, &writes
, s
->aw_bits
, pasid
);
1947 vtd_report_fault(s
, -ret_fr
, is_fpd_set
, source_id
,
1948 addr
, is_write
, pasid
!= PCI_NO_PASID
, pasid
);
1952 page_mask
= vtd_slpt_level_page_mask(level
);
1953 access_flags
= IOMMU_ACCESS_FLAG(reads
, writes
);
1954 vtd_update_iotlb(s
, source_id
, vtd_get_domain_id(s
, &ce
, pasid
),
1955 addr
, slpte
, access_flags
, level
, pasid
);
1957 vtd_iommu_unlock(s
);
1958 entry
->iova
= addr
& page_mask
;
1959 entry
->translated_addr
= vtd_get_slpte_addr(slpte
, s
->aw_bits
) & page_mask
;
1960 entry
->addr_mask
= ~page_mask
;
1961 entry
->perm
= access_flags
;
1965 vtd_iommu_unlock(s
);
1967 entry
->translated_addr
= 0;
1968 entry
->addr_mask
= 0;
1969 entry
->perm
= IOMMU_NONE
;
1973 static void vtd_root_table_setup(IntelIOMMUState
*s
)
1975 s
->root
= vtd_get_quad_raw(s
, DMAR_RTADDR_REG
);
1976 s
->root
&= VTD_RTADDR_ADDR_MASK(s
->aw_bits
);
1978 vtd_update_scalable_state(s
);
1980 trace_vtd_reg_dmar_root(s
->root
, s
->root_scalable
);
1983 static void vtd_iec_notify_all(IntelIOMMUState
*s
, bool global
,
1984 uint32_t index
, uint32_t mask
)
1986 x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s
), global
, index
, mask
);
1989 static void vtd_interrupt_remap_table_setup(IntelIOMMUState
*s
)
1992 value
= vtd_get_quad_raw(s
, DMAR_IRTA_REG
);
1993 s
->intr_size
= 1UL << ((value
& VTD_IRTA_SIZE_MASK
) + 1);
1994 s
->intr_root
= value
& VTD_IRTA_ADDR_MASK(s
->aw_bits
);
1995 s
->intr_eime
= value
& VTD_IRTA_EIME
;
1997 /* Notify global invalidation */
1998 vtd_iec_notify_all(s
, true, 0, 0);
2000 trace_vtd_reg_ir_root(s
->intr_root
, s
->intr_size
);
2003 static void vtd_iommu_replay_all(IntelIOMMUState
*s
)
2005 VTDAddressSpace
*vtd_as
;
2007 QLIST_FOREACH(vtd_as
, &s
->vtd_as_with_notifiers
, next
) {
2008 vtd_address_space_sync(vtd_as
);
2012 static void vtd_context_global_invalidate(IntelIOMMUState
*s
)
2014 trace_vtd_inv_desc_cc_global();
2015 /* Protects context cache */
2017 s
->context_cache_gen
++;
2018 if (s
->context_cache_gen
== VTD_CONTEXT_CACHE_GEN_MAX
) {
2019 vtd_reset_context_cache_locked(s
);
2021 vtd_iommu_unlock(s
);
2022 vtd_address_space_refresh_all(s
);
2024 * From VT-d spec 6.5.2.1, a global context entry invalidation
2025 * should be followed by a IOTLB global invalidation, so we should
2026 * be safe even without this. Hoewever, let's replay the region as
2027 * well to be safer, and go back here when we need finer tunes for
2028 * VT-d emulation codes.
2030 vtd_iommu_replay_all(s
);
2033 /* Do a context-cache device-selective invalidation.
2034 * @func_mask: FM field after shifting
2036 static void vtd_context_device_invalidate(IntelIOMMUState
*s
,
2040 GHashTableIter as_it
;
2042 VTDAddressSpace
*vtd_as
;
2043 uint8_t bus_n
, devfn
;
2045 trace_vtd_inv_desc_cc_devices(source_id
, func_mask
);
2047 switch (func_mask
& 3) {
2049 mask
= 0; /* No bits in the SID field masked */
2052 mask
= 4; /* Mask bit 2 in the SID field */
2055 mask
= 6; /* Mask bit 2:1 in the SID field */
2058 mask
= 7; /* Mask bit 2:0 in the SID field */
2061 g_assert_not_reached();
2065 bus_n
= VTD_SID_TO_BUS(source_id
);
2066 devfn
= VTD_SID_TO_DEVFN(source_id
);
2068 g_hash_table_iter_init(&as_it
, s
->vtd_address_spaces
);
2069 while (g_hash_table_iter_next(&as_it
, NULL
, (void **)&vtd_as
)) {
2070 if ((pci_bus_num(vtd_as
->bus
) == bus_n
) &&
2071 (vtd_as
->devfn
& mask
) == (devfn
& mask
)) {
2072 trace_vtd_inv_desc_cc_device(bus_n
, VTD_PCI_SLOT(vtd_as
->devfn
),
2073 VTD_PCI_FUNC(vtd_as
->devfn
));
2075 vtd_as
->context_cache_entry
.context_cache_gen
= 0;
2076 vtd_iommu_unlock(s
);
2078 * Do switch address space when needed, in case if the
2079 * device passthrough bit is switched.
2081 vtd_switch_address_space(vtd_as
);
2083 * So a device is moving out of (or moving into) a
2084 * domain, resync the shadow page table.
2085 * This won't bring bad even if we have no such
2086 * notifier registered - the IOMMU notification
2087 * framework will skip MAP notifications if that
2090 vtd_address_space_sync(vtd_as
);
2095 /* Context-cache invalidation
2096 * Returns the Context Actual Invalidation Granularity.
2097 * @val: the content of the CCMD_REG
2099 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState
*s
, uint64_t val
)
2102 uint64_t type
= val
& VTD_CCMD_CIRG_MASK
;
2105 case VTD_CCMD_DOMAIN_INVL
:
2107 case VTD_CCMD_GLOBAL_INVL
:
2108 caig
= VTD_CCMD_GLOBAL_INVL_A
;
2109 vtd_context_global_invalidate(s
);
2112 case VTD_CCMD_DEVICE_INVL
:
2113 caig
= VTD_CCMD_DEVICE_INVL_A
;
2114 vtd_context_device_invalidate(s
, VTD_CCMD_SID(val
), VTD_CCMD_FM(val
));
2118 error_report_once("%s: invalid context: 0x%" PRIx64
,
2125 static void vtd_iotlb_global_invalidate(IntelIOMMUState
*s
)
2127 trace_vtd_inv_desc_iotlb_global();
2129 vtd_iommu_replay_all(s
);
2132 static void vtd_iotlb_domain_invalidate(IntelIOMMUState
*s
, uint16_t domain_id
)
2135 VTDAddressSpace
*vtd_as
;
2137 trace_vtd_inv_desc_iotlb_domain(domain_id
);
2140 g_hash_table_foreach_remove(s
->iotlb
, vtd_hash_remove_by_domain
,
2142 vtd_iommu_unlock(s
);
2144 QLIST_FOREACH(vtd_as
, &s
->vtd_as_with_notifiers
, next
) {
2145 if (!vtd_dev_to_context_entry(s
, pci_bus_num(vtd_as
->bus
),
2146 vtd_as
->devfn
, &ce
) &&
2147 domain_id
== vtd_get_domain_id(s
, &ce
, vtd_as
->pasid
)) {
2148 vtd_address_space_sync(vtd_as
);
2153 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState
*s
,
2154 uint16_t domain_id
, hwaddr addr
,
2155 uint8_t am
, uint32_t pasid
)
2157 VTDAddressSpace
*vtd_as
;
2160 hwaddr size
= (1 << am
) * VTD_PAGE_SIZE
;
2162 QLIST_FOREACH(vtd_as
, &(s
->vtd_as_with_notifiers
), next
) {
2163 if (pasid
!= PCI_NO_PASID
&& pasid
!= vtd_as
->pasid
) {
2166 ret
= vtd_dev_to_context_entry(s
, pci_bus_num(vtd_as
->bus
),
2167 vtd_as
->devfn
, &ce
);
2168 if (!ret
&& domain_id
== vtd_get_domain_id(s
, &ce
, vtd_as
->pasid
)) {
2169 if (vtd_as_has_map_notifier(vtd_as
)) {
2171 * As long as we have MAP notifications registered in
2172 * any of our IOMMU notifiers, we need to sync the
2173 * shadow page table.
2175 vtd_sync_shadow_page_table_range(vtd_as
, &ce
, addr
, size
);
2178 * For UNMAP-only notifiers, we don't need to walk the
2179 * page tables. We just deliver the PSI down to
2180 * invalidate caches.
2182 IOMMUTLBEvent event
= {
2183 .type
= IOMMU_NOTIFIER_UNMAP
,
2185 .target_as
= &address_space_memory
,
2187 .translated_addr
= 0,
2188 .addr_mask
= size
- 1,
2192 memory_region_notify_iommu(&vtd_as
->iommu
, 0, event
);
2198 static void vtd_iotlb_page_invalidate(IntelIOMMUState
*s
, uint16_t domain_id
,
2199 hwaddr addr
, uint8_t am
)
2201 VTDIOTLBPageInvInfo info
;
2203 trace_vtd_inv_desc_iotlb_pages(domain_id
, addr
, am
);
2205 assert(am
<= VTD_MAMV
);
2206 info
.domain_id
= domain_id
;
2208 info
.mask
= ~((1 << am
) - 1);
2210 g_hash_table_foreach_remove(s
->iotlb
, vtd_hash_remove_by_page
, &info
);
2211 vtd_iommu_unlock(s
);
2212 vtd_iotlb_page_invalidate_notify(s
, domain_id
, addr
, am
, PCI_NO_PASID
);
2216 * Returns the IOTLB Actual Invalidation Granularity.
2217 * @val: the content of the IOTLB_REG
2219 static uint64_t vtd_iotlb_flush(IntelIOMMUState
*s
, uint64_t val
)
2222 uint64_t type
= val
& VTD_TLB_FLUSH_GRANU_MASK
;
2228 case VTD_TLB_GLOBAL_FLUSH
:
2229 iaig
= VTD_TLB_GLOBAL_FLUSH_A
;
2230 vtd_iotlb_global_invalidate(s
);
2233 case VTD_TLB_DSI_FLUSH
:
2234 domain_id
= VTD_TLB_DID(val
);
2235 iaig
= VTD_TLB_DSI_FLUSH_A
;
2236 vtd_iotlb_domain_invalidate(s
, domain_id
);
2239 case VTD_TLB_PSI_FLUSH
:
2240 domain_id
= VTD_TLB_DID(val
);
2241 addr
= vtd_get_quad_raw(s
, DMAR_IVA_REG
);
2242 am
= VTD_IVA_AM(addr
);
2243 addr
= VTD_IVA_ADDR(addr
);
2244 if (am
> VTD_MAMV
) {
2245 error_report_once("%s: address mask overflow: 0x%" PRIx64
,
2246 __func__
, vtd_get_quad_raw(s
, DMAR_IVA_REG
));
2250 iaig
= VTD_TLB_PSI_FLUSH_A
;
2251 vtd_iotlb_page_invalidate(s
, domain_id
, addr
, am
);
2255 error_report_once("%s: invalid granularity: 0x%" PRIx64
,
2262 static void vtd_fetch_inv_desc(IntelIOMMUState
*s
);
2264 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState
*s
)
2266 return s
->qi_enabled
&& (s
->iq_tail
== s
->iq_head
) &&
2267 (s
->iq_last_desc_type
== VTD_INV_DESC_WAIT
);
2270 static void vtd_handle_gcmd_qie(IntelIOMMUState
*s
, bool en
)
2272 uint64_t iqa_val
= vtd_get_quad_raw(s
, DMAR_IQA_REG
);
2274 trace_vtd_inv_qi_enable(en
);
2277 s
->iq
= iqa_val
& VTD_IQA_IQA_MASK(s
->aw_bits
);
2278 /* 2^(x+8) entries */
2279 s
->iq_size
= 1UL << ((iqa_val
& VTD_IQA_QS
) + 8 - (s
->iq_dw
? 1 : 0));
2280 s
->qi_enabled
= true;
2281 trace_vtd_inv_qi_setup(s
->iq
, s
->iq_size
);
2282 /* Ok - report back to driver */
2283 vtd_set_clear_mask_long(s
, DMAR_GSTS_REG
, 0, VTD_GSTS_QIES
);
2285 if (s
->iq_tail
!= 0) {
2287 * This is a spec violation but Windows guests are known to set up
2288 * Queued Invalidation this way so we allow the write and process
2289 * Invalidation Descriptors right away.
2291 trace_vtd_warn_invalid_qi_tail(s
->iq_tail
);
2292 if (!(vtd_get_long_raw(s
, DMAR_FSTS_REG
) & VTD_FSTS_IQE
)) {
2293 vtd_fetch_inv_desc(s
);
2297 if (vtd_queued_inv_disable_check(s
)) {
2298 /* disable Queued Invalidation */
2299 vtd_set_quad_raw(s
, DMAR_IQH_REG
, 0);
2301 s
->qi_enabled
= false;
2302 /* Ok - report back to driver */
2303 vtd_set_clear_mask_long(s
, DMAR_GSTS_REG
, VTD_GSTS_QIES
, 0);
2305 error_report_once("%s: detected improper state when disable QI "
2306 "(head=0x%x, tail=0x%x, last_type=%d)",
2308 s
->iq_head
, s
->iq_tail
, s
->iq_last_desc_type
);
2313 /* Set Root Table Pointer */
2314 static void vtd_handle_gcmd_srtp(IntelIOMMUState
*s
)
2316 vtd_root_table_setup(s
);
2317 /* Ok - report back to driver */
2318 vtd_set_clear_mask_long(s
, DMAR_GSTS_REG
, 0, VTD_GSTS_RTPS
);
2319 vtd_reset_caches(s
);
2320 vtd_address_space_refresh_all(s
);
2323 /* Set Interrupt Remap Table Pointer */
2324 static void vtd_handle_gcmd_sirtp(IntelIOMMUState
*s
)
2326 vtd_interrupt_remap_table_setup(s
);
2327 /* Ok - report back to driver */
2328 vtd_set_clear_mask_long(s
, DMAR_GSTS_REG
, 0, VTD_GSTS_IRTPS
);
2331 /* Handle Translation Enable/Disable */
2332 static void vtd_handle_gcmd_te(IntelIOMMUState
*s
, bool en
)
2334 if (s
->dmar_enabled
== en
) {
2338 trace_vtd_dmar_enable(en
);
2341 s
->dmar_enabled
= true;
2342 /* Ok - report back to driver */
2343 vtd_set_clear_mask_long(s
, DMAR_GSTS_REG
, 0, VTD_GSTS_TES
);
2345 s
->dmar_enabled
= false;
2347 /* Clear the index of Fault Recording Register */
2348 s
->next_frcd_reg
= 0;
2349 /* Ok - report back to driver */
2350 vtd_set_clear_mask_long(s
, DMAR_GSTS_REG
, VTD_GSTS_TES
, 0);
2353 vtd_reset_caches(s
);
2354 vtd_address_space_refresh_all(s
);
2357 /* Handle Interrupt Remap Enable/Disable */
2358 static void vtd_handle_gcmd_ire(IntelIOMMUState
*s
, bool en
)
2360 trace_vtd_ir_enable(en
);
2363 s
->intr_enabled
= true;
2364 /* Ok - report back to driver */
2365 vtd_set_clear_mask_long(s
, DMAR_GSTS_REG
, 0, VTD_GSTS_IRES
);
2367 s
->intr_enabled
= false;
2368 /* Ok - report back to driver */
2369 vtd_set_clear_mask_long(s
, DMAR_GSTS_REG
, VTD_GSTS_IRES
, 0);
2373 /* Handle write to Global Command Register */
2374 static void vtd_handle_gcmd_write(IntelIOMMUState
*s
)
2376 X86IOMMUState
*x86_iommu
= X86_IOMMU_DEVICE(s
);
2377 uint32_t status
= vtd_get_long_raw(s
, DMAR_GSTS_REG
);
2378 uint32_t val
= vtd_get_long_raw(s
, DMAR_GCMD_REG
);
2379 uint32_t changed
= status
^ val
;
2381 trace_vtd_reg_write_gcmd(status
, val
);
2382 if ((changed
& VTD_GCMD_TE
) && s
->dma_translation
) {
2383 /* Translation enable/disable */
2384 vtd_handle_gcmd_te(s
, val
& VTD_GCMD_TE
);
2386 if (val
& VTD_GCMD_SRTP
) {
2387 /* Set/update the root-table pointer */
2388 vtd_handle_gcmd_srtp(s
);
2390 if (changed
& VTD_GCMD_QIE
) {
2391 /* Queued Invalidation Enable */
2392 vtd_handle_gcmd_qie(s
, val
& VTD_GCMD_QIE
);
2394 if (val
& VTD_GCMD_SIRTP
) {
2395 /* Set/update the interrupt remapping root-table pointer */
2396 vtd_handle_gcmd_sirtp(s
);
2398 if ((changed
& VTD_GCMD_IRE
) &&
2399 x86_iommu_ir_supported(x86_iommu
)) {
2400 /* Interrupt remap enable/disable */
2401 vtd_handle_gcmd_ire(s
, val
& VTD_GCMD_IRE
);
2405 /* Handle write to Context Command Register */
2406 static void vtd_handle_ccmd_write(IntelIOMMUState
*s
)
2409 uint64_t val
= vtd_get_quad_raw(s
, DMAR_CCMD_REG
);
2411 /* Context-cache invalidation request */
2412 if (val
& VTD_CCMD_ICC
) {
2413 if (s
->qi_enabled
) {
2414 error_report_once("Queued Invalidation enabled, "
2415 "should not use register-based invalidation");
2418 ret
= vtd_context_cache_invalidate(s
, val
);
2419 /* Invalidation completed. Change something to show */
2420 vtd_set_clear_mask_quad(s
, DMAR_CCMD_REG
, VTD_CCMD_ICC
, 0ULL);
2421 ret
= vtd_set_clear_mask_quad(s
, DMAR_CCMD_REG
, VTD_CCMD_CAIG_MASK
,
2426 /* Handle write to IOTLB Invalidation Register */
2427 static void vtd_handle_iotlb_write(IntelIOMMUState
*s
)
2430 uint64_t val
= vtd_get_quad_raw(s
, DMAR_IOTLB_REG
);
2432 /* IOTLB invalidation request */
2433 if (val
& VTD_TLB_IVT
) {
2434 if (s
->qi_enabled
) {
2435 error_report_once("Queued Invalidation enabled, "
2436 "should not use register-based invalidation");
2439 ret
= vtd_iotlb_flush(s
, val
);
2440 /* Invalidation completed. Change something to show */
2441 vtd_set_clear_mask_quad(s
, DMAR_IOTLB_REG
, VTD_TLB_IVT
, 0ULL);
2442 ret
= vtd_set_clear_mask_quad(s
, DMAR_IOTLB_REG
,
2443 VTD_TLB_FLUSH_GRANU_MASK_A
, ret
);
2447 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
2448 static bool vtd_get_inv_desc(IntelIOMMUState
*s
,
2449 VTDInvDesc
*inv_desc
)
2451 dma_addr_t base_addr
= s
->iq
;
2452 uint32_t offset
= s
->iq_head
;
2453 uint32_t dw
= s
->iq_dw
? 32 : 16;
2454 dma_addr_t addr
= base_addr
+ offset
* dw
;
2456 if (dma_memory_read(&address_space_memory
, addr
,
2457 inv_desc
, dw
, MEMTXATTRS_UNSPECIFIED
)) {
2458 error_report_once("Read INV DESC failed.");
2461 inv_desc
->lo
= le64_to_cpu(inv_desc
->lo
);
2462 inv_desc
->hi
= le64_to_cpu(inv_desc
->hi
);
2464 inv_desc
->val
[2] = le64_to_cpu(inv_desc
->val
[2]);
2465 inv_desc
->val
[3] = le64_to_cpu(inv_desc
->val
[3]);
2470 static bool vtd_process_wait_desc(IntelIOMMUState
*s
, VTDInvDesc
*inv_desc
)
2472 if ((inv_desc
->hi
& VTD_INV_DESC_WAIT_RSVD_HI
) ||
2473 (inv_desc
->lo
& VTD_INV_DESC_WAIT_RSVD_LO
)) {
2474 error_report_once("%s: invalid wait desc: hi=%"PRIx64
", lo=%"PRIx64
2475 " (reserved nonzero)", __func__
, inv_desc
->hi
,
2479 if (inv_desc
->lo
& VTD_INV_DESC_WAIT_SW
) {
2481 uint32_t status_data
= (uint32_t)(inv_desc
->lo
>>
2482 VTD_INV_DESC_WAIT_DATA_SHIFT
);
2484 assert(!(inv_desc
->lo
& VTD_INV_DESC_WAIT_IF
));
2486 /* FIXME: need to be masked with HAW? */
2487 dma_addr_t status_addr
= inv_desc
->hi
;
2488 trace_vtd_inv_desc_wait_sw(status_addr
, status_data
);
2489 status_data
= cpu_to_le32(status_data
);
2490 if (dma_memory_write(&address_space_memory
, status_addr
,
2491 &status_data
, sizeof(status_data
),
2492 MEMTXATTRS_UNSPECIFIED
)) {
2493 trace_vtd_inv_desc_wait_write_fail(inv_desc
->hi
, inv_desc
->lo
);
2496 } else if (inv_desc
->lo
& VTD_INV_DESC_WAIT_IF
) {
2497 /* Interrupt flag */
2498 vtd_generate_completion_event(s
);
2500 error_report_once("%s: invalid wait desc: hi=%"PRIx64
", lo=%"PRIx64
2501 " (unknown type)", __func__
, inv_desc
->hi
,
2508 static bool vtd_process_context_cache_desc(IntelIOMMUState
*s
,
2509 VTDInvDesc
*inv_desc
)
2511 uint16_t sid
, fmask
;
2513 if ((inv_desc
->lo
& VTD_INV_DESC_CC_RSVD
) || inv_desc
->hi
) {
2514 error_report_once("%s: invalid cc inv desc: hi=%"PRIx64
", lo=%"PRIx64
2515 " (reserved nonzero)", __func__
, inv_desc
->hi
,
2519 switch (inv_desc
->lo
& VTD_INV_DESC_CC_G
) {
2520 case VTD_INV_DESC_CC_DOMAIN
:
2521 trace_vtd_inv_desc_cc_domain(
2522 (uint16_t)VTD_INV_DESC_CC_DID(inv_desc
->lo
));
2524 case VTD_INV_DESC_CC_GLOBAL
:
2525 vtd_context_global_invalidate(s
);
2528 case VTD_INV_DESC_CC_DEVICE
:
2529 sid
= VTD_INV_DESC_CC_SID(inv_desc
->lo
);
2530 fmask
= VTD_INV_DESC_CC_FM(inv_desc
->lo
);
2531 vtd_context_device_invalidate(s
, sid
, fmask
);
2535 error_report_once("%s: invalid cc inv desc: hi=%"PRIx64
", lo=%"PRIx64
2536 " (invalid type)", __func__
, inv_desc
->hi
,
2543 static bool vtd_process_iotlb_desc(IntelIOMMUState
*s
, VTDInvDesc
*inv_desc
)
2549 if ((inv_desc
->lo
& VTD_INV_DESC_IOTLB_RSVD_LO
) ||
2550 (inv_desc
->hi
& VTD_INV_DESC_IOTLB_RSVD_HI
)) {
2551 error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2552 ", lo=0x%"PRIx64
" (reserved bits unzero)",
2553 __func__
, inv_desc
->hi
, inv_desc
->lo
);
2557 switch (inv_desc
->lo
& VTD_INV_DESC_IOTLB_G
) {
2558 case VTD_INV_DESC_IOTLB_GLOBAL
:
2559 vtd_iotlb_global_invalidate(s
);
2562 case VTD_INV_DESC_IOTLB_DOMAIN
:
2563 domain_id
= VTD_INV_DESC_IOTLB_DID(inv_desc
->lo
);
2564 vtd_iotlb_domain_invalidate(s
, domain_id
);
2567 case VTD_INV_DESC_IOTLB_PAGE
:
2568 domain_id
= VTD_INV_DESC_IOTLB_DID(inv_desc
->lo
);
2569 addr
= VTD_INV_DESC_IOTLB_ADDR(inv_desc
->hi
);
2570 am
= VTD_INV_DESC_IOTLB_AM(inv_desc
->hi
);
2571 if (am
> VTD_MAMV
) {
2572 error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2573 ", lo=0x%"PRIx64
" (am=%u > VTD_MAMV=%u)",
2574 __func__
, inv_desc
->hi
, inv_desc
->lo
,
2575 am
, (unsigned)VTD_MAMV
);
2578 vtd_iotlb_page_invalidate(s
, domain_id
, addr
, am
);
2582 error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64
2583 ", lo=0x%"PRIx64
" (type mismatch: 0x%llx)",
2584 __func__
, inv_desc
->hi
, inv_desc
->lo
,
2585 inv_desc
->lo
& VTD_INV_DESC_IOTLB_G
);
2591 static bool vtd_process_inv_iec_desc(IntelIOMMUState
*s
,
2592 VTDInvDesc
*inv_desc
)
2594 trace_vtd_inv_desc_iec(inv_desc
->iec
.granularity
,
2595 inv_desc
->iec
.index
,
2596 inv_desc
->iec
.index_mask
);
2598 vtd_iec_notify_all(s
, !inv_desc
->iec
.granularity
,
2599 inv_desc
->iec
.index
,
2600 inv_desc
->iec
.index_mask
);
2604 static bool vtd_process_device_iotlb_desc(IntelIOMMUState
*s
,
2605 VTDInvDesc
*inv_desc
)
2607 VTDAddressSpace
*vtd_dev_as
;
2608 IOMMUTLBEvent event
;
2614 addr
= VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc
->hi
);
2615 sid
= VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc
->lo
);
2616 size
= VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc
->hi
);
2618 if ((inv_desc
->lo
& VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO
) ||
2619 (inv_desc
->hi
& VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI
)) {
2620 error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64
2621 ", lo=%"PRIx64
" (reserved nonzero)", __func__
,
2622 inv_desc
->hi
, inv_desc
->lo
);
2627 * Using sid is OK since the guest should have finished the
2628 * initialization of both the bus and device.
2630 vtd_dev_as
= vtd_get_as_by_sid(s
, sid
);
2635 /* According to ATS spec table 2.4:
2636 * S = 0, bits 15:12 = xxxx range size: 4K
2637 * S = 1, bits 15:12 = xxx0 range size: 8K
2638 * S = 1, bits 15:12 = xx01 range size: 16K
2639 * S = 1, bits 15:12 = x011 range size: 32K
2640 * S = 1, bits 15:12 = 0111 range size: 64K
2644 sz
= (VTD_PAGE_SIZE
* 2) << cto64(addr
>> VTD_PAGE_SHIFT
);
2650 event
.type
= IOMMU_NOTIFIER_DEVIOTLB_UNMAP
;
2651 event
.entry
.target_as
= &vtd_dev_as
->as
;
2652 event
.entry
.addr_mask
= sz
- 1;
2653 event
.entry
.iova
= addr
;
2654 event
.entry
.perm
= IOMMU_NONE
;
2655 event
.entry
.translated_addr
= 0;
2656 memory_region_notify_iommu(&vtd_dev_as
->iommu
, 0, event
);
2662 static bool vtd_process_inv_desc(IntelIOMMUState
*s
)
2664 VTDInvDesc inv_desc
;
2667 trace_vtd_inv_qi_head(s
->iq_head
);
2668 if (!vtd_get_inv_desc(s
, &inv_desc
)) {
2669 s
->iq_last_desc_type
= VTD_INV_DESC_NONE
;
2673 desc_type
= inv_desc
.lo
& VTD_INV_DESC_TYPE
;
2674 /* FIXME: should update at first or at last? */
2675 s
->iq_last_desc_type
= desc_type
;
2677 switch (desc_type
) {
2678 case VTD_INV_DESC_CC
:
2679 trace_vtd_inv_desc("context-cache", inv_desc
.hi
, inv_desc
.lo
);
2680 if (!vtd_process_context_cache_desc(s
, &inv_desc
)) {
2685 case VTD_INV_DESC_IOTLB
:
2686 trace_vtd_inv_desc("iotlb", inv_desc
.hi
, inv_desc
.lo
);
2687 if (!vtd_process_iotlb_desc(s
, &inv_desc
)) {
2693 * TODO: the entity of below two cases will be implemented in future series.
2694 * To make guest (which integrates scalable mode support patch set in
2695 * iommu driver) work, just return true is enough so far.
2697 case VTD_INV_DESC_PC
:
2700 case VTD_INV_DESC_PIOTLB
:
2703 case VTD_INV_DESC_WAIT
:
2704 trace_vtd_inv_desc("wait", inv_desc
.hi
, inv_desc
.lo
);
2705 if (!vtd_process_wait_desc(s
, &inv_desc
)) {
2710 case VTD_INV_DESC_IEC
:
2711 trace_vtd_inv_desc("iec", inv_desc
.hi
, inv_desc
.lo
);
2712 if (!vtd_process_inv_iec_desc(s
, &inv_desc
)) {
2717 case VTD_INV_DESC_DEVICE
:
2718 trace_vtd_inv_desc("device", inv_desc
.hi
, inv_desc
.lo
);
2719 if (!vtd_process_device_iotlb_desc(s
, &inv_desc
)) {
2725 error_report_once("%s: invalid inv desc: hi=%"PRIx64
", lo=%"PRIx64
2726 " (unknown type)", __func__
, inv_desc
.hi
,
2731 if (s
->iq_head
== s
->iq_size
) {
2737 /* Try to fetch and process more Invalidation Descriptors */
2738 static void vtd_fetch_inv_desc(IntelIOMMUState
*s
)
2742 /* Refer to 10.4.23 of VT-d spec 3.0 */
2743 qi_shift
= s
->iq_dw
? VTD_IQH_QH_SHIFT_5
: VTD_IQH_QH_SHIFT_4
;
2745 trace_vtd_inv_qi_fetch();
2747 if (s
->iq_tail
>= s
->iq_size
) {
2748 /* Detects an invalid Tail pointer */
2749 error_report_once("%s: detected invalid QI tail "
2750 "(tail=0x%x, size=0x%x)",
2751 __func__
, s
->iq_tail
, s
->iq_size
);
2752 vtd_handle_inv_queue_error(s
);
2755 while (s
->iq_head
!= s
->iq_tail
) {
2756 if (!vtd_process_inv_desc(s
)) {
2757 /* Invalidation Queue Errors */
2758 vtd_handle_inv_queue_error(s
);
2761 /* Must update the IQH_REG in time */
2762 vtd_set_quad_raw(s
, DMAR_IQH_REG
,
2763 (((uint64_t)(s
->iq_head
)) << qi_shift
) &
2768 /* Handle write to Invalidation Queue Tail Register */
2769 static void vtd_handle_iqt_write(IntelIOMMUState
*s
)
2771 uint64_t val
= vtd_get_quad_raw(s
, DMAR_IQT_REG
);
2773 if (s
->iq_dw
&& (val
& VTD_IQT_QT_256_RSV_BIT
)) {
2774 error_report_once("%s: RSV bit is set: val=0x%"PRIx64
,
2778 s
->iq_tail
= VTD_IQT_QT(s
->iq_dw
, val
);
2779 trace_vtd_inv_qi_tail(s
->iq_tail
);
2781 if (s
->qi_enabled
&& !(vtd_get_long_raw(s
, DMAR_FSTS_REG
) & VTD_FSTS_IQE
)) {
2782 /* Process Invalidation Queue here */
2783 vtd_fetch_inv_desc(s
);
2787 static void vtd_handle_fsts_write(IntelIOMMUState
*s
)
2789 uint32_t fsts_reg
= vtd_get_long_raw(s
, DMAR_FSTS_REG
);
2790 uint32_t fectl_reg
= vtd_get_long_raw(s
, DMAR_FECTL_REG
);
2791 uint32_t status_fields
= VTD_FSTS_PFO
| VTD_FSTS_PPF
| VTD_FSTS_IQE
;
2793 if ((fectl_reg
& VTD_FECTL_IP
) && !(fsts_reg
& status_fields
)) {
2794 vtd_set_clear_mask_long(s
, DMAR_FECTL_REG
, VTD_FECTL_IP
, 0);
2795 trace_vtd_fsts_clear_ip();
2797 /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2798 * Descriptors if there are any when Queued Invalidation is enabled?
2802 static void vtd_handle_fectl_write(IntelIOMMUState
*s
)
2805 /* FIXME: when software clears the IM field, check the IP field. But do we
2806 * need to compare the old value and the new value to conclude that
2807 * software clears the IM field? Or just check if the IM field is zero?
2809 fectl_reg
= vtd_get_long_raw(s
, DMAR_FECTL_REG
);
2811 trace_vtd_reg_write_fectl(fectl_reg
);
2813 if ((fectl_reg
& VTD_FECTL_IP
) && !(fectl_reg
& VTD_FECTL_IM
)) {
2814 vtd_generate_interrupt(s
, DMAR_FEADDR_REG
, DMAR_FEDATA_REG
);
2815 vtd_set_clear_mask_long(s
, DMAR_FECTL_REG
, VTD_FECTL_IP
, 0);
2819 static void vtd_handle_ics_write(IntelIOMMUState
*s
)
2821 uint32_t ics_reg
= vtd_get_long_raw(s
, DMAR_ICS_REG
);
2822 uint32_t iectl_reg
= vtd_get_long_raw(s
, DMAR_IECTL_REG
);
2824 if ((iectl_reg
& VTD_IECTL_IP
) && !(ics_reg
& VTD_ICS_IWC
)) {
2825 trace_vtd_reg_ics_clear_ip();
2826 vtd_set_clear_mask_long(s
, DMAR_IECTL_REG
, VTD_IECTL_IP
, 0);
2830 static void vtd_handle_iectl_write(IntelIOMMUState
*s
)
2833 /* FIXME: when software clears the IM field, check the IP field. But do we
2834 * need to compare the old value and the new value to conclude that
2835 * software clears the IM field? Or just check if the IM field is zero?
2837 iectl_reg
= vtd_get_long_raw(s
, DMAR_IECTL_REG
);
2839 trace_vtd_reg_write_iectl(iectl_reg
);
2841 if ((iectl_reg
& VTD_IECTL_IP
) && !(iectl_reg
& VTD_IECTL_IM
)) {
2842 vtd_generate_interrupt(s
, DMAR_IEADDR_REG
, DMAR_IEDATA_REG
);
2843 vtd_set_clear_mask_long(s
, DMAR_IECTL_REG
, VTD_IECTL_IP
, 0);
2847 static uint64_t vtd_mem_read(void *opaque
, hwaddr addr
, unsigned size
)
2849 IntelIOMMUState
*s
= opaque
;
2852 trace_vtd_reg_read(addr
, size
);
2854 if (addr
+ size
> DMAR_REG_SIZE
) {
2855 error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2856 " size=0x%x", __func__
, addr
, size
);
2857 return (uint64_t)-1;
2861 /* Root Table Address Register, 64-bit */
2862 case DMAR_RTADDR_REG
:
2863 val
= vtd_get_quad_raw(s
, DMAR_RTADDR_REG
);
2865 val
= val
& ((1ULL << 32) - 1);
2869 case DMAR_RTADDR_REG_HI
:
2871 val
= vtd_get_quad_raw(s
, DMAR_RTADDR_REG
) >> 32;
2874 /* Invalidation Queue Address Register, 64-bit */
2876 val
= s
->iq
| (vtd_get_quad(s
, DMAR_IQA_REG
) & VTD_IQA_QS
);
2878 val
= val
& ((1ULL << 32) - 1);
2882 case DMAR_IQA_REG_HI
:
2889 val
= vtd_get_long(s
, addr
);
2891 val
= vtd_get_quad(s
, addr
);
2898 static void vtd_mem_write(void *opaque
, hwaddr addr
,
2899 uint64_t val
, unsigned size
)
2901 IntelIOMMUState
*s
= opaque
;
2903 trace_vtd_reg_write(addr
, size
, val
);
2905 if (addr
+ size
> DMAR_REG_SIZE
) {
2906 error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2907 " size=0x%x", __func__
, addr
, size
);
2912 /* Global Command Register, 32-bit */
2914 vtd_set_long(s
, addr
, val
);
2915 vtd_handle_gcmd_write(s
);
2918 /* Context Command Register, 64-bit */
2921 vtd_set_long(s
, addr
, val
);
2923 vtd_set_quad(s
, addr
, val
);
2924 vtd_handle_ccmd_write(s
);
2928 case DMAR_CCMD_REG_HI
:
2930 vtd_set_long(s
, addr
, val
);
2931 vtd_handle_ccmd_write(s
);
2934 /* IOTLB Invalidation Register, 64-bit */
2935 case DMAR_IOTLB_REG
:
2937 vtd_set_long(s
, addr
, val
);
2939 vtd_set_quad(s
, addr
, val
);
2940 vtd_handle_iotlb_write(s
);
2944 case DMAR_IOTLB_REG_HI
:
2946 vtd_set_long(s
, addr
, val
);
2947 vtd_handle_iotlb_write(s
);
2950 /* Invalidate Address Register, 64-bit */
2953 vtd_set_long(s
, addr
, val
);
2955 vtd_set_quad(s
, addr
, val
);
2959 case DMAR_IVA_REG_HI
:
2961 vtd_set_long(s
, addr
, val
);
2964 /* Fault Status Register, 32-bit */
2967 vtd_set_long(s
, addr
, val
);
2968 vtd_handle_fsts_write(s
);
2971 /* Fault Event Control Register, 32-bit */
2972 case DMAR_FECTL_REG
:
2974 vtd_set_long(s
, addr
, val
);
2975 vtd_handle_fectl_write(s
);
2978 /* Fault Event Data Register, 32-bit */
2979 case DMAR_FEDATA_REG
:
2981 vtd_set_long(s
, addr
, val
);
2984 /* Fault Event Address Register, 32-bit */
2985 case DMAR_FEADDR_REG
:
2987 vtd_set_long(s
, addr
, val
);
2990 * While the register is 32-bit only, some guests (Xen...) write to
2993 vtd_set_quad(s
, addr
, val
);
2997 /* Fault Event Upper Address Register, 32-bit */
2998 case DMAR_FEUADDR_REG
:
3000 vtd_set_long(s
, addr
, val
);
3003 /* Protected Memory Enable Register, 32-bit */
3006 vtd_set_long(s
, addr
, val
);
3009 /* Root Table Address Register, 64-bit */
3010 case DMAR_RTADDR_REG
:
3012 vtd_set_long(s
, addr
, val
);
3014 vtd_set_quad(s
, addr
, val
);
3018 case DMAR_RTADDR_REG_HI
:
3020 vtd_set_long(s
, addr
, val
);
3023 /* Invalidation Queue Tail Register, 64-bit */
3026 vtd_set_long(s
, addr
, val
);
3028 vtd_set_quad(s
, addr
, val
);
3030 vtd_handle_iqt_write(s
);
3033 case DMAR_IQT_REG_HI
:
3035 vtd_set_long(s
, addr
, val
);
3036 /* 19:63 of IQT_REG is RsvdZ, do nothing here */
3039 /* Invalidation Queue Address Register, 64-bit */
3042 vtd_set_long(s
, addr
, val
);
3044 vtd_set_quad(s
, addr
, val
);
3046 vtd_update_iq_dw(s
);
3049 case DMAR_IQA_REG_HI
:
3051 vtd_set_long(s
, addr
, val
);
3054 /* Invalidation Completion Status Register, 32-bit */
3057 vtd_set_long(s
, addr
, val
);
3058 vtd_handle_ics_write(s
);
3061 /* Invalidation Event Control Register, 32-bit */
3062 case DMAR_IECTL_REG
:
3064 vtd_set_long(s
, addr
, val
);
3065 vtd_handle_iectl_write(s
);
3068 /* Invalidation Event Data Register, 32-bit */
3069 case DMAR_IEDATA_REG
:
3071 vtd_set_long(s
, addr
, val
);
3074 /* Invalidation Event Address Register, 32-bit */
3075 case DMAR_IEADDR_REG
:
3077 vtd_set_long(s
, addr
, val
);
3080 /* Invalidation Event Upper Address Register, 32-bit */
3081 case DMAR_IEUADDR_REG
:
3083 vtd_set_long(s
, addr
, val
);
3086 /* Fault Recording Registers, 128-bit */
3087 case DMAR_FRCD_REG_0_0
:
3089 vtd_set_long(s
, addr
, val
);
3091 vtd_set_quad(s
, addr
, val
);
3095 case DMAR_FRCD_REG_0_1
:
3097 vtd_set_long(s
, addr
, val
);
3100 case DMAR_FRCD_REG_0_2
:
3102 vtd_set_long(s
, addr
, val
);
3104 vtd_set_quad(s
, addr
, val
);
3105 /* May clear bit 127 (Fault), update PPF */
3106 vtd_update_fsts_ppf(s
);
3110 case DMAR_FRCD_REG_0_3
:
3112 vtd_set_long(s
, addr
, val
);
3113 /* May clear bit 127 (Fault), update PPF */
3114 vtd_update_fsts_ppf(s
);
3119 vtd_set_long(s
, addr
, val
);
3121 vtd_set_quad(s
, addr
, val
);
3125 case DMAR_IRTA_REG_HI
:
3127 vtd_set_long(s
, addr
, val
);
3132 vtd_set_long(s
, addr
, val
);
3134 vtd_set_quad(s
, addr
, val
);
3139 static IOMMUTLBEntry
vtd_iommu_translate(IOMMUMemoryRegion
*iommu
, hwaddr addr
,
3140 IOMMUAccessFlags flag
, int iommu_idx
)
3142 VTDAddressSpace
*vtd_as
= container_of(iommu
, VTDAddressSpace
, iommu
);
3143 IntelIOMMUState
*s
= vtd_as
->iommu_state
;
3144 IOMMUTLBEntry iotlb
= {
3145 /* We'll fill in the rest later. */
3146 .target_as
= &address_space_memory
,
3150 if (likely(s
->dmar_enabled
)) {
3151 success
= vtd_do_iommu_translate(vtd_as
, vtd_as
->bus
, vtd_as
->devfn
,
3152 addr
, flag
& IOMMU_WO
, &iotlb
);
3154 /* DMAR disabled, passthrough, use 4k-page*/
3155 iotlb
.iova
= addr
& VTD_PAGE_MASK_4K
;
3156 iotlb
.translated_addr
= addr
& VTD_PAGE_MASK_4K
;
3157 iotlb
.addr_mask
= ~VTD_PAGE_MASK_4K
;
3158 iotlb
.perm
= IOMMU_RW
;
3162 if (likely(success
)) {
3163 trace_vtd_dmar_translate(pci_bus_num(vtd_as
->bus
),
3164 VTD_PCI_SLOT(vtd_as
->devfn
),
3165 VTD_PCI_FUNC(vtd_as
->devfn
),
3166 iotlb
.iova
, iotlb
.translated_addr
,
3169 error_report_once("%s: detected translation failure "
3170 "(dev=%02x:%02x:%02x, iova=0x%" PRIx64
")",
3171 __func__
, pci_bus_num(vtd_as
->bus
),
3172 VTD_PCI_SLOT(vtd_as
->devfn
),
3173 VTD_PCI_FUNC(vtd_as
->devfn
),
3180 static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion
*iommu
,
3181 IOMMUNotifierFlag old
,
3182 IOMMUNotifierFlag
new,
3185 VTDAddressSpace
*vtd_as
= container_of(iommu
, VTDAddressSpace
, iommu
);
3186 IntelIOMMUState
*s
= vtd_as
->iommu_state
;
3187 X86IOMMUState
*x86_iommu
= X86_IOMMU_DEVICE(s
);
3189 /* TODO: add support for VFIO and vhost users */
3190 if (s
->snoop_control
) {
3191 error_setg_errno(errp
, ENOTSUP
,
3192 "Snoop Control with vhost or VFIO is not supported");
3195 if (!s
->caching_mode
&& (new & IOMMU_NOTIFIER_MAP
)) {
3196 error_setg_errno(errp
, ENOTSUP
,
3197 "device %02x.%02x.%x requires caching mode",
3198 pci_bus_num(vtd_as
->bus
), PCI_SLOT(vtd_as
->devfn
),
3199 PCI_FUNC(vtd_as
->devfn
));
3202 if (!x86_iommu
->dt_supported
&& (new & IOMMU_NOTIFIER_DEVIOTLB_UNMAP
)) {
3203 error_setg_errno(errp
, ENOTSUP
,
3204 "device %02x.%02x.%x requires device IOTLB mode",
3205 pci_bus_num(vtd_as
->bus
), PCI_SLOT(vtd_as
->devfn
),
3206 PCI_FUNC(vtd_as
->devfn
));
3210 /* Update per-address-space notifier flags */
3211 vtd_as
->notifier_flags
= new;
3213 if (old
== IOMMU_NOTIFIER_NONE
) {
3214 QLIST_INSERT_HEAD(&s
->vtd_as_with_notifiers
, vtd_as
, next
);
3215 } else if (new == IOMMU_NOTIFIER_NONE
) {
3216 QLIST_REMOVE(vtd_as
, next
);
3221 static int vtd_post_load(void *opaque
, int version_id
)
3223 IntelIOMMUState
*iommu
= opaque
;
3226 * We don't need to migrate the root_scalable because we can
3227 * simply do the calculation after the loading is complete. We
3228 * can actually do similar things with root, dmar_enabled, etc.
3229 * however since we've had them already so we'd better keep them
3230 * for compatibility of migration.
3232 vtd_update_scalable_state(iommu
);
3234 vtd_update_iq_dw(iommu
);
3237 * Memory regions are dynamically turned on/off depending on
3238 * context entry configurations from the guest. After migration,
3239 * we need to make sure the memory regions are still correct.
3241 vtd_switch_address_space_all(iommu
);
3246 static const VMStateDescription vtd_vmstate
= {
3247 .name
= "iommu-intel",
3249 .minimum_version_id
= 1,
3250 .priority
= MIG_PRI_IOMMU
,
3251 .post_load
= vtd_post_load
,
3252 .fields
= (VMStateField
[]) {
3253 VMSTATE_UINT64(root
, IntelIOMMUState
),
3254 VMSTATE_UINT64(intr_root
, IntelIOMMUState
),
3255 VMSTATE_UINT64(iq
, IntelIOMMUState
),
3256 VMSTATE_UINT32(intr_size
, IntelIOMMUState
),
3257 VMSTATE_UINT16(iq_head
, IntelIOMMUState
),
3258 VMSTATE_UINT16(iq_tail
, IntelIOMMUState
),
3259 VMSTATE_UINT16(iq_size
, IntelIOMMUState
),
3260 VMSTATE_UINT16(next_frcd_reg
, IntelIOMMUState
),
3261 VMSTATE_UINT8_ARRAY(csr
, IntelIOMMUState
, DMAR_REG_SIZE
),
3262 VMSTATE_UINT8(iq_last_desc_type
, IntelIOMMUState
),
3263 VMSTATE_UNUSED(1), /* bool root_extended is obsolete by VT-d */
3264 VMSTATE_BOOL(dmar_enabled
, IntelIOMMUState
),
3265 VMSTATE_BOOL(qi_enabled
, IntelIOMMUState
),
3266 VMSTATE_BOOL(intr_enabled
, IntelIOMMUState
),
3267 VMSTATE_BOOL(intr_eime
, IntelIOMMUState
),
3268 VMSTATE_END_OF_LIST()
3272 static const MemoryRegionOps vtd_mem_ops
= {
3273 .read
= vtd_mem_read
,
3274 .write
= vtd_mem_write
,
3275 .endianness
= DEVICE_LITTLE_ENDIAN
,
3277 .min_access_size
= 4,
3278 .max_access_size
= 8,
3281 .min_access_size
= 4,
3282 .max_access_size
= 8,
3286 static Property vtd_properties
[] = {
3287 DEFINE_PROP_UINT32("version", IntelIOMMUState
, version
, 0),
3288 DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState
, intr_eim
,
3290 DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState
, buggy_eim
, false),
3291 DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState
, aw_bits
,
3292 VTD_HOST_ADDRESS_WIDTH
),
3293 DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState
, caching_mode
, FALSE
),
3294 DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState
, scalable_mode
, FALSE
),
3295 DEFINE_PROP_BOOL("snoop-control", IntelIOMMUState
, snoop_control
, false),
3296 DEFINE_PROP_BOOL("x-pasid-mode", IntelIOMMUState
, pasid
, false),
3297 DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState
, dma_drain
, true),
3298 DEFINE_PROP_BOOL("dma-translation", IntelIOMMUState
, dma_translation
, true),
3299 DEFINE_PROP_END_OF_LIST(),
3302 /* Read IRTE entry with specific index */
3303 static int vtd_irte_get(IntelIOMMUState
*iommu
, uint16_t index
,
3304 VTD_IR_TableEntry
*entry
, uint16_t sid
)
3306 static const uint16_t vtd_svt_mask
[VTD_SQ_MAX
] = \
3307 {0xffff, 0xfffb, 0xfff9, 0xfff8};
3308 dma_addr_t addr
= 0x00;
3309 uint16_t mask
, source_id
;
3310 uint8_t bus
, bus_max
, bus_min
;
3312 if (index
>= iommu
->intr_size
) {
3313 error_report_once("%s: index too large: ind=0x%x",
3315 return -VTD_FR_IR_INDEX_OVER
;
3318 addr
= iommu
->intr_root
+ index
* sizeof(*entry
);
3319 if (dma_memory_read(&address_space_memory
, addr
,
3320 entry
, sizeof(*entry
), MEMTXATTRS_UNSPECIFIED
)) {
3321 error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64
,
3322 __func__
, index
, addr
);
3323 return -VTD_FR_IR_ROOT_INVAL
;
3326 trace_vtd_ir_irte_get(index
, le64_to_cpu(entry
->data
[1]),
3327 le64_to_cpu(entry
->data
[0]));
3329 if (!entry
->irte
.present
) {
3330 error_report_once("%s: detected non-present IRTE "
3331 "(index=%u, high=0x%" PRIx64
", low=0x%" PRIx64
")",
3332 __func__
, index
, le64_to_cpu(entry
->data
[1]),
3333 le64_to_cpu(entry
->data
[0]));
3334 return -VTD_FR_IR_ENTRY_P
;
3337 if (entry
->irte
.__reserved_0
|| entry
->irte
.__reserved_1
||
3338 entry
->irte
.__reserved_2
) {
3339 error_report_once("%s: detected non-zero reserved IRTE "
3340 "(index=%u, high=0x%" PRIx64
", low=0x%" PRIx64
")",
3341 __func__
, index
, le64_to_cpu(entry
->data
[1]),
3342 le64_to_cpu(entry
->data
[0]));
3343 return -VTD_FR_IR_IRTE_RSVD
;
3346 if (sid
!= X86_IOMMU_SID_INVALID
) {
3347 /* Validate IRTE SID */
3348 source_id
= le32_to_cpu(entry
->irte
.source_id
);
3349 switch (entry
->irte
.sid_vtype
) {
3354 mask
= vtd_svt_mask
[entry
->irte
.sid_q
];
3355 if ((source_id
& mask
) != (sid
& mask
)) {
3356 error_report_once("%s: invalid IRTE SID "
3357 "(index=%u, sid=%u, source_id=%u)",
3358 __func__
, index
, sid
, source_id
);
3359 return -VTD_FR_IR_SID_ERR
;
3364 bus_max
= source_id
>> 8;
3365 bus_min
= source_id
& 0xff;
3367 if (bus
> bus_max
|| bus
< bus_min
) {
3368 error_report_once("%s: invalid SVT_BUS "
3369 "(index=%u, bus=%u, min=%u, max=%u)",
3370 __func__
, index
, bus
, bus_min
, bus_max
);
3371 return -VTD_FR_IR_SID_ERR
;
3376 error_report_once("%s: detected invalid IRTE SVT "
3377 "(index=%u, type=%d)", __func__
,
3378 index
, entry
->irte
.sid_vtype
);
3379 /* Take this as verification failure. */
3380 return -VTD_FR_IR_SID_ERR
;
3387 /* Fetch IRQ information of specific IR index */
3388 static int vtd_remap_irq_get(IntelIOMMUState
*iommu
, uint16_t index
,
3389 X86IOMMUIrq
*irq
, uint16_t sid
)
3391 VTD_IR_TableEntry irte
= {};
3394 ret
= vtd_irte_get(iommu
, index
, &irte
, sid
);
3399 irq
->trigger_mode
= irte
.irte
.trigger_mode
;
3400 irq
->vector
= irte
.irte
.vector
;
3401 irq
->delivery_mode
= irte
.irte
.delivery_mode
;
3402 irq
->dest
= le32_to_cpu(irte
.irte
.dest_id
);
3403 if (!iommu
->intr_eime
) {
3404 #define VTD_IR_APIC_DEST_MASK (0xff00ULL)
3405 #define VTD_IR_APIC_DEST_SHIFT (8)
3406 irq
->dest
= (irq
->dest
& VTD_IR_APIC_DEST_MASK
) >>
3407 VTD_IR_APIC_DEST_SHIFT
;
3409 irq
->dest_mode
= irte
.irte
.dest_mode
;
3410 irq
->redir_hint
= irte
.irte
.redir_hint
;
3412 trace_vtd_ir_remap(index
, irq
->trigger_mode
, irq
->vector
,
3413 irq
->delivery_mode
, irq
->dest
, irq
->dest_mode
);
3418 /* Interrupt remapping for MSI/MSI-X entry */
3419 static int vtd_interrupt_remap_msi(IntelIOMMUState
*iommu
,
3421 MSIMessage
*translated
,
3425 VTD_IR_MSIAddress addr
;
3427 X86IOMMUIrq irq
= {};
3429 assert(origin
&& translated
);
3431 trace_vtd_ir_remap_msi_req(origin
->address
, origin
->data
);
3433 if (!iommu
|| !iommu
->intr_enabled
) {
3434 memcpy(translated
, origin
, sizeof(*origin
));
3438 if (origin
->address
& VTD_MSI_ADDR_HI_MASK
) {
3439 error_report_once("%s: MSI address high 32 bits non-zero detected: "
3440 "address=0x%" PRIx64
, __func__
, origin
->address
);
3441 return -VTD_FR_IR_REQ_RSVD
;
3444 addr
.data
= origin
->address
& VTD_MSI_ADDR_LO_MASK
;
3445 if (addr
.addr
.__head
!= 0xfee) {
3446 error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32
,
3447 __func__
, addr
.data
);
3448 return -VTD_FR_IR_REQ_RSVD
;
3451 /* This is compatible mode. */
3452 if (addr
.addr
.int_mode
!= VTD_IR_INT_FORMAT_REMAP
) {
3453 memcpy(translated
, origin
, sizeof(*origin
));
3457 index
= addr
.addr
.index_h
<< 15 | le16_to_cpu(addr
.addr
.index_l
);
3459 #define VTD_IR_MSI_DATA_SUBHANDLE (0x0000ffff)
3460 #define VTD_IR_MSI_DATA_RESERVED (0xffff0000)
3462 if (addr
.addr
.sub_valid
) {
3463 /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
3464 index
+= origin
->data
& VTD_IR_MSI_DATA_SUBHANDLE
;
3467 ret
= vtd_remap_irq_get(iommu
, index
, &irq
, sid
);
3472 if (addr
.addr
.sub_valid
) {
3473 trace_vtd_ir_remap_type("MSI");
3474 if (origin
->data
& VTD_IR_MSI_DATA_RESERVED
) {
3475 error_report_once("%s: invalid IR MSI "
3476 "(sid=%u, address=0x%" PRIx64
3477 ", data=0x%" PRIx32
")",
3478 __func__
, sid
, origin
->address
, origin
->data
);
3479 return -VTD_FR_IR_REQ_RSVD
;
3482 uint8_t vector
= origin
->data
& 0xff;
3483 uint8_t trigger_mode
= (origin
->data
>> MSI_DATA_TRIGGER_SHIFT
) & 0x1;
3485 trace_vtd_ir_remap_type("IOAPIC");
3486 /* IOAPIC entry vector should be aligned with IRTE vector
3487 * (see vt-d spec 5.1.5.1). */
3488 if (vector
!= irq
.vector
) {
3489 trace_vtd_warn_ir_vector(sid
, index
, vector
, irq
.vector
);
3492 /* The Trigger Mode field must match the Trigger Mode in the IRTE.
3493 * (see vt-d spec 5.1.5.1). */
3494 if (trigger_mode
!= irq
.trigger_mode
) {
3495 trace_vtd_warn_ir_trigger(sid
, index
, trigger_mode
,
3501 * We'd better keep the last two bits, assuming that guest OS
3502 * might modify it. Keep it does not hurt after all.
3504 irq
.msi_addr_last_bits
= addr
.addr
.__not_care
;
3506 /* Translate X86IOMMUIrq to MSI message */
3507 x86_iommu_irq_to_msi_message(&irq
, translated
);
3510 trace_vtd_ir_remap_msi(origin
->address
, origin
->data
,
3511 translated
->address
, translated
->data
);
3515 static int vtd_int_remap(X86IOMMUState
*iommu
, MSIMessage
*src
,
3516 MSIMessage
*dst
, uint16_t sid
)
3518 return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu
),
3522 static MemTxResult
vtd_mem_ir_read(void *opaque
, hwaddr addr
,
3523 uint64_t *data
, unsigned size
,
3529 static MemTxResult
vtd_mem_ir_write(void *opaque
, hwaddr addr
,
3530 uint64_t value
, unsigned size
,
3534 MSIMessage from
= {}, to
= {};
3535 uint16_t sid
= X86_IOMMU_SID_INVALID
;
3537 from
.address
= (uint64_t) addr
+ VTD_INTERRUPT_ADDR_FIRST
;
3538 from
.data
= (uint32_t) value
;
3540 if (!attrs
.unspecified
) {
3541 /* We have explicit Source ID */
3542 sid
= attrs
.requester_id
;
3545 ret
= vtd_interrupt_remap_msi(opaque
, &from
, &to
, sid
);
3547 /* TODO: report error */
3548 /* Drop this interrupt */
3552 apic_get_class(NULL
)->send_msi(&to
);
3557 static const MemoryRegionOps vtd_mem_ir_ops
= {
3558 .read_with_attrs
= vtd_mem_ir_read
,
3559 .write_with_attrs
= vtd_mem_ir_write
,
3560 .endianness
= DEVICE_LITTLE_ENDIAN
,
3562 .min_access_size
= 4,
3563 .max_access_size
= 4,
3566 .min_access_size
= 4,
3567 .max_access_size
= 4,
3571 static void vtd_report_ir_illegal_access(VTDAddressSpace
*vtd_as
,
3572 hwaddr addr
, bool is_write
)
3574 IntelIOMMUState
*s
= vtd_as
->iommu_state
;
3575 uint8_t bus_n
= pci_bus_num(vtd_as
->bus
);
3576 uint16_t sid
= PCI_BUILD_BDF(bus_n
, vtd_as
->devfn
);
3577 bool is_fpd_set
= false;
3580 assert(vtd_as
->pasid
!= PCI_NO_PASID
);
3582 /* Try out best to fetch FPD, we can't do anything more */
3583 if (vtd_dev_to_context_entry(s
, bus_n
, vtd_as
->devfn
, &ce
) == 0) {
3584 is_fpd_set
= ce
.lo
& VTD_CONTEXT_ENTRY_FPD
;
3585 if (!is_fpd_set
&& s
->root_scalable
) {
3586 vtd_ce_get_pasid_fpd(s
, &ce
, &is_fpd_set
, vtd_as
->pasid
);
3590 vtd_report_fault(s
, VTD_FR_SM_INTERRUPT_ADDR
,
3591 is_fpd_set
, sid
, addr
, is_write
,
3592 true, vtd_as
->pasid
);
3595 static MemTxResult
vtd_mem_ir_fault_read(void *opaque
, hwaddr addr
,
3596 uint64_t *data
, unsigned size
,
3599 vtd_report_ir_illegal_access(opaque
, addr
, false);
3604 static MemTxResult
vtd_mem_ir_fault_write(void *opaque
, hwaddr addr
,
3605 uint64_t value
, unsigned size
,
3608 vtd_report_ir_illegal_access(opaque
, addr
, true);
3613 static const MemoryRegionOps vtd_mem_ir_fault_ops
= {
3614 .read_with_attrs
= vtd_mem_ir_fault_read
,
3615 .write_with_attrs
= vtd_mem_ir_fault_write
,
3616 .endianness
= DEVICE_LITTLE_ENDIAN
,
3618 .min_access_size
= 1,
3619 .max_access_size
= 8,
3622 .min_access_size
= 1,
3623 .max_access_size
= 8,
3627 VTDAddressSpace
*vtd_find_add_as(IntelIOMMUState
*s
, PCIBus
*bus
,
3628 int devfn
, unsigned int pasid
)
3631 * We can't simply use sid here since the bus number might not be
3632 * initialized by the guest.
3634 struct vtd_as_key key
= {
3639 VTDAddressSpace
*vtd_dev_as
;
3642 vtd_dev_as
= g_hash_table_lookup(s
->vtd_address_spaces
, &key
);
3644 struct vtd_as_key
*new_key
= g_malloc(sizeof(*new_key
));
3647 new_key
->devfn
= devfn
;
3648 new_key
->pasid
= pasid
;
3650 if (pasid
== PCI_NO_PASID
) {
3651 snprintf(name
, sizeof(name
), "vtd-%02x.%x", PCI_SLOT(devfn
),
3654 snprintf(name
, sizeof(name
), "vtd-%02x.%x-pasid-%x", PCI_SLOT(devfn
),
3655 PCI_FUNC(devfn
), pasid
);
3658 vtd_dev_as
= g_new0(VTDAddressSpace
, 1);
3660 vtd_dev_as
->bus
= bus
;
3661 vtd_dev_as
->devfn
= (uint8_t)devfn
;
3662 vtd_dev_as
->pasid
= pasid
;
3663 vtd_dev_as
->iommu_state
= s
;
3664 vtd_dev_as
->context_cache_entry
.context_cache_gen
= 0;
3665 vtd_dev_as
->iova_tree
= iova_tree_new();
3667 memory_region_init(&vtd_dev_as
->root
, OBJECT(s
), name
, UINT64_MAX
);
3668 address_space_init(&vtd_dev_as
->as
, &vtd_dev_as
->root
, "vtd-root");
3671 * Build the DMAR-disabled container with aliases to the
3672 * shared MRs. Note that aliasing to a shared memory region
3673 * could help the memory API to detect same FlatViews so we
3674 * can have devices to share the same FlatView when DMAR is
3675 * disabled (either by not providing "intel_iommu=on" or with
3676 * "iommu=pt"). It will greatly reduce the total number of
3677 * FlatViews of the system hence VM runs faster.
3679 memory_region_init_alias(&vtd_dev_as
->nodmar
, OBJECT(s
),
3680 "vtd-nodmar", &s
->mr_nodmar
, 0,
3681 memory_region_size(&s
->mr_nodmar
));
3684 * Build the per-device DMAR-enabled container.
3686 * TODO: currently we have per-device IOMMU memory region only
3687 * because we have per-device IOMMU notifiers for devices. If
3688 * one day we can abstract the IOMMU notifiers out of the
3689 * memory regions then we can also share the same memory
3690 * region here just like what we've done above with the nodmar
3693 strcat(name
, "-dmar");
3694 memory_region_init_iommu(&vtd_dev_as
->iommu
, sizeof(vtd_dev_as
->iommu
),
3695 TYPE_INTEL_IOMMU_MEMORY_REGION
, OBJECT(s
),
3697 memory_region_init_alias(&vtd_dev_as
->iommu_ir
, OBJECT(s
), "vtd-ir",
3698 &s
->mr_ir
, 0, memory_region_size(&s
->mr_ir
));
3699 memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as
->iommu
),
3700 VTD_INTERRUPT_ADDR_FIRST
,
3701 &vtd_dev_as
->iommu_ir
, 1);
3704 * This region is used for catching fault to access interrupt
3705 * range via passthrough + PASID. See also
3706 * vtd_switch_address_space(). We can't use alias since we
3707 * need to know the sid which is valid for MSI who uses
3708 * bus_master_as (see msi_send_message()).
3710 memory_region_init_io(&vtd_dev_as
->iommu_ir_fault
, OBJECT(s
),
3711 &vtd_mem_ir_fault_ops
, vtd_dev_as
, "vtd-no-ir",
3712 VTD_INTERRUPT_ADDR_SIZE
);
3714 * Hook to root since when PT is enabled vtd_dev_as->iommu
3717 memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as
->root
),
3718 VTD_INTERRUPT_ADDR_FIRST
,
3719 &vtd_dev_as
->iommu_ir_fault
, 2);
3722 * Hook both the containers under the root container, we
3723 * switch between DMAR & noDMAR by enable/disable
3724 * corresponding sub-containers
3726 memory_region_add_subregion_overlap(&vtd_dev_as
->root
, 0,
3727 MEMORY_REGION(&vtd_dev_as
->iommu
),
3729 memory_region_add_subregion_overlap(&vtd_dev_as
->root
, 0,
3730 &vtd_dev_as
->nodmar
, 0);
3732 vtd_switch_address_space(vtd_dev_as
);
3734 g_hash_table_insert(s
->vtd_address_spaces
, new_key
, vtd_dev_as
);
3739 /* Unmap the whole range in the notifier's scope. */
3740 static void vtd_address_space_unmap(VTDAddressSpace
*as
, IOMMUNotifier
*n
)
3742 hwaddr size
, remain
;
3743 hwaddr start
= n
->start
;
3744 hwaddr end
= n
->end
;
3745 IntelIOMMUState
*s
= as
->iommu_state
;
3749 * Note: all the codes in this function has a assumption that IOVA
3750 * bits are no more than VTD_MGAW bits (which is restricted by
3751 * VT-d spec), otherwise we need to consider overflow of 64 bits.
3754 if (end
> VTD_ADDRESS_SIZE(s
->aw_bits
) - 1) {
3756 * Don't need to unmap regions that is bigger than the whole
3757 * VT-d supported address space size
3759 end
= VTD_ADDRESS_SIZE(s
->aw_bits
) - 1;
3762 assert(start
<= end
);
3763 size
= remain
= end
- start
+ 1;
3765 while (remain
>= VTD_PAGE_SIZE
) {
3766 IOMMUTLBEvent event
;
3767 uint64_t mask
= dma_aligned_pow2_mask(start
, end
, s
->aw_bits
);
3768 uint64_t size
= mask
+ 1;
3772 event
.type
= IOMMU_NOTIFIER_UNMAP
;
3773 event
.entry
.iova
= start
;
3774 event
.entry
.addr_mask
= mask
;
3775 event
.entry
.target_as
= &address_space_memory
;
3776 event
.entry
.perm
= IOMMU_NONE
;
3777 /* This field is meaningless for unmap */
3778 event
.entry
.translated_addr
= 0;
3780 memory_region_notify_iommu_one(n
, &event
);
3788 trace_vtd_as_unmap_whole(pci_bus_num(as
->bus
),
3789 VTD_PCI_SLOT(as
->devfn
),
3790 VTD_PCI_FUNC(as
->devfn
),
3793 map
.iova
= n
->start
;
3795 iova_tree_remove(as
->iova_tree
, map
);
3798 static void vtd_address_space_unmap_all(IntelIOMMUState
*s
)
3800 VTDAddressSpace
*vtd_as
;
3803 QLIST_FOREACH(vtd_as
, &s
->vtd_as_with_notifiers
, next
) {
3804 IOMMU_NOTIFIER_FOREACH(n
, &vtd_as
->iommu
) {
3805 vtd_address_space_unmap(vtd_as
, n
);
3810 static void vtd_address_space_refresh_all(IntelIOMMUState
*s
)
3812 vtd_address_space_unmap_all(s
);
3813 vtd_switch_address_space_all(s
);
3816 static int vtd_replay_hook(IOMMUTLBEvent
*event
, void *private)
3818 memory_region_notify_iommu_one(private, event
);
3822 static void vtd_iommu_replay(IOMMUMemoryRegion
*iommu_mr
, IOMMUNotifier
*n
)
3824 VTDAddressSpace
*vtd_as
= container_of(iommu_mr
, VTDAddressSpace
, iommu
);
3825 IntelIOMMUState
*s
= vtd_as
->iommu_state
;
3826 uint8_t bus_n
= pci_bus_num(vtd_as
->bus
);
3830 * The replay can be triggered by either a invalidation or a newly
3831 * created entry. No matter what, we release existing mappings
3832 * (it means flushing caches for UNMAP-only registers).
3834 vtd_address_space_unmap(vtd_as
, n
);
3836 if (vtd_dev_to_context_entry(s
, bus_n
, vtd_as
->devfn
, &ce
) == 0) {
3837 trace_vtd_replay_ce_valid(s
->root_scalable
? "scalable mode" :
3839 bus_n
, PCI_SLOT(vtd_as
->devfn
),
3840 PCI_FUNC(vtd_as
->devfn
),
3841 vtd_get_domain_id(s
, &ce
, vtd_as
->pasid
),
3843 if (vtd_as_has_map_notifier(vtd_as
)) {
3844 /* This is required only for MAP typed notifiers */
3845 vtd_page_walk_info info
= {
3846 .hook_fn
= vtd_replay_hook
,
3847 .private = (void *)n
,
3848 .notify_unmap
= false,
3851 .domain_id
= vtd_get_domain_id(s
, &ce
, vtd_as
->pasid
),
3854 vtd_page_walk(s
, &ce
, 0, ~0ULL, &info
, vtd_as
->pasid
);
3857 trace_vtd_replay_ce_invalid(bus_n
, PCI_SLOT(vtd_as
->devfn
),
3858 PCI_FUNC(vtd_as
->devfn
));
3864 /* Do the initialization. It will also be called when reset, so pay
3865 * attention when adding new initialization stuff.
3867 static void vtd_init(IntelIOMMUState
*s
)
3869 X86IOMMUState
*x86_iommu
= X86_IOMMU_DEVICE(s
);
3871 memset(s
->csr
, 0, DMAR_REG_SIZE
);
3872 memset(s
->wmask
, 0, DMAR_REG_SIZE
);
3873 memset(s
->w1cmask
, 0, DMAR_REG_SIZE
);
3874 memset(s
->womask
, 0, DMAR_REG_SIZE
);
3877 s
->root_scalable
= false;
3878 s
->dmar_enabled
= false;
3879 s
->intr_enabled
= false;
3884 s
->qi_enabled
= false;
3885 s
->iq_last_desc_type
= VTD_INV_DESC_NONE
;
3887 s
->next_frcd_reg
= 0;
3888 s
->cap
= VTD_CAP_FRO
| VTD_CAP_NFR
| VTD_CAP_ND
|
3889 VTD_CAP_MAMV
| VTD_CAP_PSI
| VTD_CAP_SLLPS
|
3890 VTD_CAP_MGAW(s
->aw_bits
);
3892 s
->cap
|= VTD_CAP_DRAIN
;
3894 if (s
->dma_translation
) {
3895 if (s
->aw_bits
>= VTD_HOST_AW_39BIT
) {
3896 s
->cap
|= VTD_CAP_SAGAW_39bit
;
3898 if (s
->aw_bits
>= VTD_HOST_AW_48BIT
) {
3899 s
->cap
|= VTD_CAP_SAGAW_48bit
;
3902 s
->ecap
= VTD_ECAP_QI
| VTD_ECAP_IRO
;
3905 * Rsvd field masks for spte
3907 vtd_spte_rsvd
[0] = ~0ULL;
3908 vtd_spte_rsvd
[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s
->aw_bits
,
3909 x86_iommu
->dt_supported
);
3910 vtd_spte_rsvd
[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s
->aw_bits
);
3911 vtd_spte_rsvd
[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s
->aw_bits
);
3912 vtd_spte_rsvd
[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s
->aw_bits
);
3914 vtd_spte_rsvd_large
[2] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s
->aw_bits
,
3915 x86_iommu
->dt_supported
);
3916 vtd_spte_rsvd_large
[3] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s
->aw_bits
,
3917 x86_iommu
->dt_supported
);
3919 if (s
->scalable_mode
|| s
->snoop_control
) {
3920 vtd_spte_rsvd
[1] &= ~VTD_SPTE_SNP
;
3921 vtd_spte_rsvd_large
[2] &= ~VTD_SPTE_SNP
;
3922 vtd_spte_rsvd_large
[3] &= ~VTD_SPTE_SNP
;
3925 if (x86_iommu_ir_supported(x86_iommu
)) {
3926 s
->ecap
|= VTD_ECAP_IR
| VTD_ECAP_MHMV
;
3927 if (s
->intr_eim
== ON_OFF_AUTO_ON
) {
3928 s
->ecap
|= VTD_ECAP_EIM
;
3930 assert(s
->intr_eim
!= ON_OFF_AUTO_AUTO
);
3933 if (x86_iommu
->dt_supported
) {
3934 s
->ecap
|= VTD_ECAP_DT
;
3937 if (x86_iommu
->pt_supported
) {
3938 s
->ecap
|= VTD_ECAP_PT
;
3941 if (s
->caching_mode
) {
3942 s
->cap
|= VTD_CAP_CM
;
3945 /* TODO: read cap/ecap from host to decide which cap to be exposed. */
3946 if (s
->scalable_mode
) {
3947 s
->ecap
|= VTD_ECAP_SMTS
| VTD_ECAP_SRS
| VTD_ECAP_SLTS
;
3950 if (s
->snoop_control
) {
3951 s
->ecap
|= VTD_ECAP_SC
;
3955 s
->ecap
|= VTD_ECAP_PASID
;
3958 vtd_reset_caches(s
);
3960 /* Define registers with default values and bit semantics */
3961 vtd_define_long(s
, DMAR_VER_REG
, 0x10UL
, 0, 0);
3962 vtd_define_quad(s
, DMAR_CAP_REG
, s
->cap
, 0, 0);
3963 vtd_define_quad(s
, DMAR_ECAP_REG
, s
->ecap
, 0, 0);
3964 vtd_define_long(s
, DMAR_GCMD_REG
, 0, 0xff800000UL
, 0);
3965 vtd_define_long_wo(s
, DMAR_GCMD_REG
, 0xff800000UL
);
3966 vtd_define_long(s
, DMAR_GSTS_REG
, 0, 0, 0);
3967 vtd_define_quad(s
, DMAR_RTADDR_REG
, 0, 0xfffffffffffffc00ULL
, 0);
3968 vtd_define_quad(s
, DMAR_CCMD_REG
, 0, 0xe0000003ffffffffULL
, 0);
3969 vtd_define_quad_wo(s
, DMAR_CCMD_REG
, 0x3ffff0000ULL
);
3971 /* Advanced Fault Logging not supported */
3972 vtd_define_long(s
, DMAR_FSTS_REG
, 0, 0, 0x11UL
);
3973 vtd_define_long(s
, DMAR_FECTL_REG
, 0x80000000UL
, 0x80000000UL
, 0);
3974 vtd_define_long(s
, DMAR_FEDATA_REG
, 0, 0x0000ffffUL
, 0);
3975 vtd_define_long(s
, DMAR_FEADDR_REG
, 0, 0xfffffffcUL
, 0);
3977 /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3978 * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3980 vtd_define_long(s
, DMAR_FEUADDR_REG
, 0, 0, 0);
3982 /* Treated as RO for implementations that PLMR and PHMR fields reported
3983 * as Clear in the CAP_REG.
3984 * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3986 vtd_define_long(s
, DMAR_PMEN_REG
, 0, 0, 0);
3988 vtd_define_quad(s
, DMAR_IQH_REG
, 0, 0, 0);
3989 vtd_define_quad(s
, DMAR_IQT_REG
, 0, 0x7fff0ULL
, 0);
3990 vtd_define_quad(s
, DMAR_IQA_REG
, 0, 0xfffffffffffff807ULL
, 0);
3991 vtd_define_long(s
, DMAR_ICS_REG
, 0, 0, 0x1UL
);
3992 vtd_define_long(s
, DMAR_IECTL_REG
, 0x80000000UL
, 0x80000000UL
, 0);
3993 vtd_define_long(s
, DMAR_IEDATA_REG
, 0, 0xffffffffUL
, 0);
3994 vtd_define_long(s
, DMAR_IEADDR_REG
, 0, 0xfffffffcUL
, 0);
3995 /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3996 vtd_define_long(s
, DMAR_IEUADDR_REG
, 0, 0, 0);
3998 /* IOTLB registers */
3999 vtd_define_quad(s
, DMAR_IOTLB_REG
, 0, 0Xb003ffff00000000ULL
, 0);
4000 vtd_define_quad(s
, DMAR_IVA_REG
, 0, 0xfffffffffffff07fULL
, 0);
4001 vtd_define_quad_wo(s
, DMAR_IVA_REG
, 0xfffffffffffff07fULL
);
4003 /* Fault Recording Registers, 128-bit */
4004 vtd_define_quad(s
, DMAR_FRCD_REG_0_0
, 0, 0, 0);
4005 vtd_define_quad(s
, DMAR_FRCD_REG_0_2
, 0, 0, 0x8000000000000000ULL
);
4008 * Interrupt remapping registers.
4010 vtd_define_quad(s
, DMAR_IRTA_REG
, 0, 0xfffffffffffff80fULL
, 0);
4013 /* Should not reset address_spaces when reset because devices will still use
4014 * the address space they got at first (won't ask the bus again).
4016 static void vtd_reset(DeviceState
*dev
)
4018 IntelIOMMUState
*s
= INTEL_IOMMU_DEVICE(dev
);
4021 vtd_address_space_refresh_all(s
);
4024 static AddressSpace
*vtd_host_dma_iommu(PCIBus
*bus
, void *opaque
, int devfn
)
4026 IntelIOMMUState
*s
= opaque
;
4027 VTDAddressSpace
*vtd_as
;
4029 assert(0 <= devfn
&& devfn
< PCI_DEVFN_MAX
);
4031 vtd_as
= vtd_find_add_as(s
, bus
, devfn
, PCI_NO_PASID
);
4035 static bool vtd_decide_config(IntelIOMMUState
*s
, Error
**errp
)
4037 X86IOMMUState
*x86_iommu
= X86_IOMMU_DEVICE(s
);
4039 if (s
->intr_eim
== ON_OFF_AUTO_ON
&& !x86_iommu_ir_supported(x86_iommu
)) {
4040 error_setg(errp
, "eim=on cannot be selected without intremap=on");
4044 if (s
->intr_eim
== ON_OFF_AUTO_AUTO
) {
4045 s
->intr_eim
= (kvm_irqchip_in_kernel() || s
->buggy_eim
)
4046 && x86_iommu_ir_supported(x86_iommu
) ?
4047 ON_OFF_AUTO_ON
: ON_OFF_AUTO_OFF
;
4049 if (s
->intr_eim
== ON_OFF_AUTO_ON
&& !s
->buggy_eim
) {
4050 if (!kvm_irqchip_is_split()) {
4051 error_setg(errp
, "eim=on requires accel=kvm,kernel-irqchip=split");
4054 if (!kvm_enable_x2apic()) {
4055 error_setg(errp
, "eim=on requires support on the KVM side"
4056 "(X2APIC_API, first shipped in v4.7)");
4061 /* Currently only address widths supported are 39 and 48 bits */
4062 if ((s
->aw_bits
!= VTD_HOST_AW_39BIT
) &&
4063 (s
->aw_bits
!= VTD_HOST_AW_48BIT
)) {
4064 error_setg(errp
, "Supported values for aw-bits are: %d, %d",
4065 VTD_HOST_AW_39BIT
, VTD_HOST_AW_48BIT
);
4069 if (s
->scalable_mode
&& !s
->dma_drain
) {
4070 error_setg(errp
, "Need to set dma_drain for scalable mode");
4074 if (s
->pasid
&& !s
->scalable_mode
) {
4075 error_setg(errp
, "Need to set scalable mode for PASID");
4082 static int vtd_machine_done_notify_one(Object
*child
, void *unused
)
4084 IntelIOMMUState
*iommu
= INTEL_IOMMU_DEVICE(x86_iommu_get_default());
4087 * We hard-coded here because vfio-pci is the only special case
4088 * here. Let's be more elegant in the future when we can, but so
4089 * far there seems to be no better way.
4091 if (object_dynamic_cast(child
, "vfio-pci") && !iommu
->caching_mode
) {
4092 vtd_panic_require_caching_mode();
4098 static void vtd_machine_done_hook(Notifier
*notifier
, void *unused
)
4100 object_child_foreach_recursive(object_get_root(),
4101 vtd_machine_done_notify_one
, NULL
);
4104 static Notifier vtd_machine_done_notify
= {
4105 .notify
= vtd_machine_done_hook
,
4108 static void vtd_realize(DeviceState
*dev
, Error
**errp
)
4110 MachineState
*ms
= MACHINE(qdev_get_machine());
4111 PCMachineState
*pcms
= PC_MACHINE(ms
);
4112 X86MachineState
*x86ms
= X86_MACHINE(ms
);
4113 PCIBus
*bus
= pcms
->bus
;
4114 IntelIOMMUState
*s
= INTEL_IOMMU_DEVICE(dev
);
4115 X86IOMMUState
*x86_iommu
= X86_IOMMU_DEVICE(s
);
4117 if (s
->pasid
&& x86_iommu
->dt_supported
) {
4119 * PASID-based-Device-TLB Invalidate Descriptor is not
4120 * implemented and it requires support from vhost layer which
4121 * needs to be implemented in the future.
4123 error_setg(errp
, "PASID based device IOTLB is not supported");
4127 if (!vtd_decide_config(s
, errp
)) {
4131 QLIST_INIT(&s
->vtd_as_with_notifiers
);
4132 qemu_mutex_init(&s
->iommu_lock
);
4133 memory_region_init_io(&s
->csrmem
, OBJECT(s
), &vtd_mem_ops
, s
,
4134 "intel_iommu", DMAR_REG_SIZE
);
4136 /* Create the shared memory regions by all devices */
4137 memory_region_init(&s
->mr_nodmar
, OBJECT(s
), "vtd-nodmar",
4139 memory_region_init_io(&s
->mr_ir
, OBJECT(s
), &vtd_mem_ir_ops
,
4140 s
, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE
);
4141 memory_region_init_alias(&s
->mr_sys_alias
, OBJECT(s
),
4142 "vtd-sys-alias", get_system_memory(), 0,
4143 memory_region_size(get_system_memory()));
4144 memory_region_add_subregion_overlap(&s
->mr_nodmar
, 0,
4145 &s
->mr_sys_alias
, 0);
4146 memory_region_add_subregion_overlap(&s
->mr_nodmar
,
4147 VTD_INTERRUPT_ADDR_FIRST
,
4150 sysbus_init_mmio(SYS_BUS_DEVICE(s
), &s
->csrmem
);
4151 /* No corresponding destroy */
4152 s
->iotlb
= g_hash_table_new_full(vtd_iotlb_hash
, vtd_iotlb_equal
,
4154 s
->vtd_address_spaces
= g_hash_table_new_full(vtd_as_hash
, vtd_as_equal
,
4157 sysbus_mmio_map(SYS_BUS_DEVICE(s
), 0, Q35_HOST_BRIDGE_IOMMU_ADDR
);
4158 pci_setup_iommu(bus
, vtd_host_dma_iommu
, dev
);
4159 /* Pseudo address space under root PCI bus. */
4160 x86ms
->ioapic_as
= vtd_host_dma_iommu(bus
, s
, Q35_PSEUDO_DEVFN_IOAPIC
);
4161 qemu_add_machine_init_done_notifier(&vtd_machine_done_notify
);
4164 static void vtd_class_init(ObjectClass
*klass
, void *data
)
4166 DeviceClass
*dc
= DEVICE_CLASS(klass
);
4167 X86IOMMUClass
*x86_class
= X86_IOMMU_DEVICE_CLASS(klass
);
4169 dc
->reset
= vtd_reset
;
4170 dc
->vmsd
= &vtd_vmstate
;
4171 device_class_set_props(dc
, vtd_properties
);
4172 dc
->hotpluggable
= false;
4173 x86_class
->realize
= vtd_realize
;
4174 x86_class
->int_remap
= vtd_int_remap
;
4175 /* Supported by the pc-q35-* machine types */
4176 dc
->user_creatable
= true;
4177 set_bit(DEVICE_CATEGORY_MISC
, dc
->categories
);
4178 dc
->desc
= "Intel IOMMU (VT-d) DMA Remapping device";
4181 static const TypeInfo vtd_info
= {
4182 .name
= TYPE_INTEL_IOMMU_DEVICE
,
4183 .parent
= TYPE_X86_IOMMU_DEVICE
,
4184 .instance_size
= sizeof(IntelIOMMUState
),
4185 .class_init
= vtd_class_init
,
4188 static void vtd_iommu_memory_region_class_init(ObjectClass
*klass
,
4191 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_CLASS(klass
);
4193 imrc
->translate
= vtd_iommu_translate
;
4194 imrc
->notify_flag_changed
= vtd_iommu_notify_flag_changed
;
4195 imrc
->replay
= vtd_iommu_replay
;
4198 static const TypeInfo vtd_iommu_memory_region_info
= {
4199 .parent
= TYPE_IOMMU_MEMORY_REGION
,
4200 .name
= TYPE_INTEL_IOMMU_MEMORY_REGION
,
4201 .class_init
= vtd_iommu_memory_region_class_init
,
4204 static void vtd_register_types(void)
4206 type_register_static(&vtd_info
);
4207 type_register_static(&vtd_iommu_memory_region_info
);
4210 type_init(vtd_register_types
)