]>
Commit | Line | Data |
---|---|---|
1da12ec4 LT |
1 | /* |
2 | * QEMU emulation of an Intel IOMMU (VT-d) | |
3 | * (DMA Remapping device) | |
4 | * | |
5 | * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com> | |
6 | * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com> | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or modify | |
9 | * it under the terms of the GNU General Public License as published by | |
10 | * the Free Software Foundation; either version 2 of the License, or | |
11 | * (at your option) any later version. | |
12 | ||
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | ||
18 | * You should have received a copy of the GNU General Public License along | |
19 | * with this program; if not, see <http://www.gnu.org/licenses/>. | |
20 | */ | |
21 | ||
b6a0aa05 | 22 | #include "qemu/osdep.h" |
4684a204 | 23 | #include "qemu/error-report.h" |
db725815 | 24 | #include "qemu/main-loop.h" |
6333e93c | 25 | #include "qapi/error.h" |
1da12ec4 LT |
26 | #include "hw/sysbus.h" |
27 | #include "exec/address-spaces.h" | |
28 | #include "intel_iommu_internal.h" | |
7df953bd | 29 | #include "hw/pci/pci.h" |
3cb3b154 | 30 | #include "hw/pci/pci_bus.h" |
a27bd6c7 | 31 | #include "hw/qdev-properties.h" |
621d983a | 32 | #include "hw/i386/pc.h" |
dea651a9 | 33 | #include "hw/i386/apic-msidef.h" |
04af0e18 PX |
34 | #include "hw/boards.h" |
35 | #include "hw/i386/x86-iommu.h" | |
cb135f59 | 36 | #include "hw/pci-host/q35.h" |
4684a204 | 37 | #include "sysemu/kvm.h" |
28cf553a | 38 | #include "sysemu/sysemu.h" |
32946019 | 39 | #include "hw/i386/apic_internal.h" |
fb506e70 | 40 | #include "kvm_i386.h" |
d6454270 | 41 | #include "migration/vmstate.h" |
bc535e59 | 42 | #include "trace.h" |
1da12ec4 | 43 | |
fb43cf73 LY |
44 | /* context entry operations */ |
45 | #define VTD_CE_GET_RID2PASID(ce) \ | |
46 | ((ce)->val[1] & VTD_SM_CONTEXT_ENTRY_RID2PASID_MASK) | |
47 | #define VTD_CE_GET_PASID_DIR_TABLE(ce) \ | |
48 | ((ce)->val[0] & VTD_PASID_DIR_BASE_ADDR_MASK) | |
49 | ||
50 | /* pe operations */ | |
51 | #define VTD_PE_GET_TYPE(pe) ((pe)->val[0] & VTD_SM_PASID_ENTRY_PGTT) | |
52 | #define VTD_PE_GET_LEVEL(pe) (2 + (((pe)->val[0] >> 2) & VTD_SM_PASID_ENTRY_AW)) | |
53 | #define VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write) {\ | |
54 | if (ret_fr) { \ | |
55 | ret_fr = -ret_fr; \ | |
56 | if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) { \ | |
57 | trace_vtd_fault_disabled(); \ | |
58 | } else { \ | |
59 | vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write); \ | |
60 | } \ | |
61 | goto error; \ | |
62 | } \ | |
63 | } | |
64 | ||
2cc9ddcc | 65 | static void vtd_address_space_refresh_all(IntelIOMMUState *s); |
c28b535d | 66 | static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n); |
2cc9ddcc | 67 | |
28cf553a PX |
68 | static void vtd_panic_require_caching_mode(void) |
69 | { | |
70 | error_report("We need to set caching-mode=on for intel-iommu to enable " | |
71 | "device assignment with IOMMU protection."); | |
72 | exit(1); | |
73 | } | |
74 | ||
1da12ec4 LT |
75 | static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val, |
76 | uint64_t wmask, uint64_t w1cmask) | |
77 | { | |
78 | stq_le_p(&s->csr[addr], val); | |
79 | stq_le_p(&s->wmask[addr], wmask); | |
80 | stq_le_p(&s->w1cmask[addr], w1cmask); | |
81 | } | |
82 | ||
83 | static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask) | |
84 | { | |
85 | stq_le_p(&s->womask[addr], mask); | |
86 | } | |
87 | ||
88 | static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val, | |
89 | uint32_t wmask, uint32_t w1cmask) | |
90 | { | |
91 | stl_le_p(&s->csr[addr], val); | |
92 | stl_le_p(&s->wmask[addr], wmask); | |
93 | stl_le_p(&s->w1cmask[addr], w1cmask); | |
94 | } | |
95 | ||
96 | static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask) | |
97 | { | |
98 | stl_le_p(&s->womask[addr], mask); | |
99 | } | |
100 | ||
101 | /* "External" get/set operations */ | |
102 | static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val) | |
103 | { | |
104 | uint64_t oldval = ldq_le_p(&s->csr[addr]); | |
105 | uint64_t wmask = ldq_le_p(&s->wmask[addr]); | |
106 | uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]); | |
107 | stq_le_p(&s->csr[addr], | |
108 | ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val)); | |
109 | } | |
110 | ||
111 | static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val) | |
112 | { | |
113 | uint32_t oldval = ldl_le_p(&s->csr[addr]); | |
114 | uint32_t wmask = ldl_le_p(&s->wmask[addr]); | |
115 | uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]); | |
116 | stl_le_p(&s->csr[addr], | |
117 | ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val)); | |
118 | } | |
119 | ||
120 | static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr) | |
121 | { | |
122 | uint64_t val = ldq_le_p(&s->csr[addr]); | |
123 | uint64_t womask = ldq_le_p(&s->womask[addr]); | |
124 | return val & ~womask; | |
125 | } | |
126 | ||
127 | static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr) | |
128 | { | |
129 | uint32_t val = ldl_le_p(&s->csr[addr]); | |
130 | uint32_t womask = ldl_le_p(&s->womask[addr]); | |
131 | return val & ~womask; | |
132 | } | |
133 | ||
134 | /* "Internal" get/set operations */ | |
135 | static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr) | |
136 | { | |
137 | return ldq_le_p(&s->csr[addr]); | |
138 | } | |
139 | ||
140 | static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr) | |
141 | { | |
142 | return ldl_le_p(&s->csr[addr]); | |
143 | } | |
144 | ||
145 | static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val) | |
146 | { | |
147 | stq_le_p(&s->csr[addr], val); | |
148 | } | |
149 | ||
150 | static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr, | |
151 | uint32_t clear, uint32_t mask) | |
152 | { | |
153 | uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask; | |
154 | stl_le_p(&s->csr[addr], new_val); | |
155 | return new_val; | |
156 | } | |
157 | ||
158 | static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr, | |
159 | uint64_t clear, uint64_t mask) | |
160 | { | |
161 | uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask; | |
162 | stq_le_p(&s->csr[addr], new_val); | |
163 | return new_val; | |
164 | } | |
165 | ||
1d9efa73 PX |
166 | static inline void vtd_iommu_lock(IntelIOMMUState *s) |
167 | { | |
168 | qemu_mutex_lock(&s->iommu_lock); | |
169 | } | |
170 | ||
171 | static inline void vtd_iommu_unlock(IntelIOMMUState *s) | |
172 | { | |
173 | qemu_mutex_unlock(&s->iommu_lock); | |
174 | } | |
175 | ||
2811af3b PX |
176 | static void vtd_update_scalable_state(IntelIOMMUState *s) |
177 | { | |
178 | uint64_t val = vtd_get_quad_raw(s, DMAR_RTADDR_REG); | |
179 | ||
180 | if (s->scalable_mode) { | |
181 | s->root_scalable = val & VTD_RTADDR_SMT; | |
182 | } | |
183 | } | |
184 | ||
4f8a62a9 PX |
185 | /* Whether the address space needs to notify new mappings */ |
186 | static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as) | |
187 | { | |
188 | return as->notifier_flags & IOMMU_NOTIFIER_MAP; | |
189 | } | |
190 | ||
b5a280c0 LT |
191 | /* GHashTable functions */ |
192 | static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2) | |
193 | { | |
194 | return *((const uint64_t *)v1) == *((const uint64_t *)v2); | |
195 | } | |
196 | ||
197 | static guint vtd_uint64_hash(gconstpointer v) | |
198 | { | |
199 | return (guint)*(const uint64_t *)v; | |
200 | } | |
201 | ||
202 | static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value, | |
203 | gpointer user_data) | |
204 | { | |
205 | VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value; | |
206 | uint16_t domain_id = *(uint16_t *)user_data; | |
207 | return entry->domain_id == domain_id; | |
208 | } | |
209 | ||
d66b969b JW |
210 | /* The shift of an addr for a certain level of paging structure */ |
211 | static inline uint32_t vtd_slpt_level_shift(uint32_t level) | |
212 | { | |
7e58326a | 213 | assert(level != 0); |
d66b969b JW |
214 | return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS; |
215 | } | |
216 | ||
217 | static inline uint64_t vtd_slpt_level_page_mask(uint32_t level) | |
218 | { | |
219 | return ~((1ULL << vtd_slpt_level_shift(level)) - 1); | |
220 | } | |
221 | ||
b5a280c0 LT |
222 | static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value, |
223 | gpointer user_data) | |
224 | { | |
225 | VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value; | |
226 | VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data; | |
d66b969b JW |
227 | uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask; |
228 | uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K; | |
b5a280c0 | 229 | return (entry->domain_id == info->domain_id) && |
d66b969b JW |
230 | (((entry->gfn & info->mask) == gfn) || |
231 | (entry->gfn == gfn_tlb)); | |
b5a280c0 LT |
232 | } |
233 | ||
d92fa2dc | 234 | /* Reset all the gen of VTDAddressSpace to zero and set the gen of |
1d9efa73 | 235 | * IntelIOMMUState to 1. Must be called with IOMMU lock held. |
d92fa2dc | 236 | */ |
1d9efa73 | 237 | static void vtd_reset_context_cache_locked(IntelIOMMUState *s) |
d92fa2dc | 238 | { |
d92fa2dc | 239 | VTDAddressSpace *vtd_as; |
7df953bd KO |
240 | VTDBus *vtd_bus; |
241 | GHashTableIter bus_it; | |
d92fa2dc LT |
242 | uint32_t devfn_it; |
243 | ||
7feb51b7 PX |
244 | trace_vtd_context_cache_reset(); |
245 | ||
7df953bd KO |
246 | g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr); |
247 | ||
7df953bd | 248 | while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) { |
bf33cc75 | 249 | for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) { |
7df953bd | 250 | vtd_as = vtd_bus->dev_as[devfn_it]; |
d92fa2dc LT |
251 | if (!vtd_as) { |
252 | continue; | |
253 | } | |
254 | vtd_as->context_cache_entry.context_cache_gen = 0; | |
255 | } | |
256 | } | |
257 | s->context_cache_gen = 1; | |
258 | } | |
259 | ||
1d9efa73 PX |
260 | /* Must be called with IOMMU lock held. */ |
261 | static void vtd_reset_iotlb_locked(IntelIOMMUState *s) | |
b5a280c0 LT |
262 | { |
263 | assert(s->iotlb); | |
264 | g_hash_table_remove_all(s->iotlb); | |
265 | } | |
266 | ||
1d9efa73 PX |
267 | static void vtd_reset_iotlb(IntelIOMMUState *s) |
268 | { | |
269 | vtd_iommu_lock(s); | |
270 | vtd_reset_iotlb_locked(s); | |
271 | vtd_iommu_unlock(s); | |
272 | } | |
273 | ||
06aba4ca PX |
274 | static void vtd_reset_caches(IntelIOMMUState *s) |
275 | { | |
276 | vtd_iommu_lock(s); | |
277 | vtd_reset_iotlb_locked(s); | |
278 | vtd_reset_context_cache_locked(s); | |
279 | vtd_iommu_unlock(s); | |
280 | } | |
281 | ||
bacabb0a | 282 | static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id, |
d66b969b JW |
283 | uint32_t level) |
284 | { | |
285 | return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) | | |
286 | ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT); | |
287 | } | |
288 | ||
289 | static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level) | |
290 | { | |
291 | return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K; | |
292 | } | |
293 | ||
1d9efa73 | 294 | /* Must be called with IOMMU lock held */ |
b5a280c0 LT |
295 | static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id, |
296 | hwaddr addr) | |
297 | { | |
d66b969b | 298 | VTDIOTLBEntry *entry; |
b5a280c0 | 299 | uint64_t key; |
d66b969b JW |
300 | int level; |
301 | ||
302 | for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) { | |
303 | key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level), | |
304 | source_id, level); | |
305 | entry = g_hash_table_lookup(s->iotlb, &key); | |
306 | if (entry) { | |
307 | goto out; | |
308 | } | |
309 | } | |
b5a280c0 | 310 | |
d66b969b JW |
311 | out: |
312 | return entry; | |
b5a280c0 LT |
313 | } |
314 | ||
1d9efa73 | 315 | /* Must be with IOMMU lock held */ |
b5a280c0 LT |
316 | static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id, |
317 | uint16_t domain_id, hwaddr addr, uint64_t slpte, | |
07f7b733 | 318 | uint8_t access_flags, uint32_t level) |
b5a280c0 LT |
319 | { |
320 | VTDIOTLBEntry *entry = g_malloc(sizeof(*entry)); | |
321 | uint64_t *key = g_malloc(sizeof(*key)); | |
d66b969b | 322 | uint64_t gfn = vtd_get_iotlb_gfn(addr, level); |
b5a280c0 | 323 | |
6c441e1d | 324 | trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id); |
b5a280c0 | 325 | if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) { |
6c441e1d | 326 | trace_vtd_iotlb_reset("iotlb exceeds size limit"); |
1d9efa73 | 327 | vtd_reset_iotlb_locked(s); |
b5a280c0 LT |
328 | } |
329 | ||
330 | entry->gfn = gfn; | |
331 | entry->domain_id = domain_id; | |
332 | entry->slpte = slpte; | |
07f7b733 | 333 | entry->access_flags = access_flags; |
d66b969b JW |
334 | entry->mask = vtd_slpt_level_page_mask(level); |
335 | *key = vtd_get_iotlb_key(gfn, source_id, level); | |
b5a280c0 LT |
336 | g_hash_table_replace(s->iotlb, key, entry); |
337 | } | |
338 | ||
1da12ec4 LT |
339 | /* Given the reg addr of both the message data and address, generate an |
340 | * interrupt via MSI. | |
341 | */ | |
342 | static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg, | |
343 | hwaddr mesg_data_reg) | |
344 | { | |
32946019 | 345 | MSIMessage msi; |
1da12ec4 LT |
346 | |
347 | assert(mesg_data_reg < DMAR_REG_SIZE); | |
348 | assert(mesg_addr_reg < DMAR_REG_SIZE); | |
349 | ||
32946019 RK |
350 | msi.address = vtd_get_long_raw(s, mesg_addr_reg); |
351 | msi.data = vtd_get_long_raw(s, mesg_data_reg); | |
1da12ec4 | 352 | |
7feb51b7 PX |
353 | trace_vtd_irq_generate(msi.address, msi.data); |
354 | ||
32946019 | 355 | apic_get_class()->send_msi(&msi); |
1da12ec4 LT |
356 | } |
357 | ||
358 | /* Generate a fault event to software via MSI if conditions are met. | |
359 | * Notice that the value of FSTS_REG being passed to it should be the one | |
360 | * before any update. | |
361 | */ | |
362 | static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts) | |
363 | { | |
364 | if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO || | |
365 | pre_fsts & VTD_FSTS_IQE) { | |
1376211f PX |
366 | error_report_once("There are previous interrupt conditions " |
367 | "to be serviced by software, fault event " | |
368 | "is not generated"); | |
1da12ec4 LT |
369 | return; |
370 | } | |
371 | vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP); | |
372 | if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) { | |
1376211f | 373 | error_report_once("Interrupt Mask set, irq is not generated"); |
1da12ec4 LT |
374 | } else { |
375 | vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG); | |
376 | vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); | |
377 | } | |
378 | } | |
379 | ||
380 | /* Check if the Fault (F) field of the Fault Recording Register referenced by | |
381 | * @index is Set. | |
382 | */ | |
383 | static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index) | |
384 | { | |
385 | /* Each reg is 128-bit */ | |
386 | hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); | |
387 | addr += 8; /* Access the high 64-bit half */ | |
388 | ||
389 | assert(index < DMAR_FRCD_REG_NR); | |
390 | ||
391 | return vtd_get_quad_raw(s, addr) & VTD_FRCD_F; | |
392 | } | |
393 | ||
394 | /* Update the PPF field of Fault Status Register. | |
395 | * Should be called whenever change the F field of any fault recording | |
396 | * registers. | |
397 | */ | |
398 | static void vtd_update_fsts_ppf(IntelIOMMUState *s) | |
399 | { | |
400 | uint32_t i; | |
401 | uint32_t ppf_mask = 0; | |
402 | ||
403 | for (i = 0; i < DMAR_FRCD_REG_NR; i++) { | |
404 | if (vtd_is_frcd_set(s, i)) { | |
405 | ppf_mask = VTD_FSTS_PPF; | |
406 | break; | |
407 | } | |
408 | } | |
409 | vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask); | |
7feb51b7 | 410 | trace_vtd_fsts_ppf(!!ppf_mask); |
1da12ec4 LT |
411 | } |
412 | ||
413 | static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index) | |
414 | { | |
415 | /* Each reg is 128-bit */ | |
416 | hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); | |
417 | addr += 8; /* Access the high 64-bit half */ | |
418 | ||
419 | assert(index < DMAR_FRCD_REG_NR); | |
420 | ||
421 | vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F); | |
422 | vtd_update_fsts_ppf(s); | |
423 | } | |
424 | ||
425 | /* Must not update F field now, should be done later */ | |
426 | static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index, | |
427 | uint16_t source_id, hwaddr addr, | |
428 | VTDFaultReason fault, bool is_write) | |
429 | { | |
430 | uint64_t hi = 0, lo; | |
431 | hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); | |
432 | ||
433 | assert(index < DMAR_FRCD_REG_NR); | |
434 | ||
435 | lo = VTD_FRCD_FI(addr); | |
436 | hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault); | |
437 | if (!is_write) { | |
438 | hi |= VTD_FRCD_T; | |
439 | } | |
440 | vtd_set_quad_raw(s, frcd_reg_addr, lo); | |
441 | vtd_set_quad_raw(s, frcd_reg_addr + 8, hi); | |
7feb51b7 PX |
442 | |
443 | trace_vtd_frr_new(index, hi, lo); | |
1da12ec4 LT |
444 | } |
445 | ||
446 | /* Try to collapse multiple pending faults from the same requester */ | |
447 | static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id) | |
448 | { | |
449 | uint32_t i; | |
450 | uint64_t frcd_reg; | |
451 | hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */ | |
452 | ||
453 | for (i = 0; i < DMAR_FRCD_REG_NR; i++) { | |
454 | frcd_reg = vtd_get_quad_raw(s, addr); | |
1da12ec4 LT |
455 | if ((frcd_reg & VTD_FRCD_F) && |
456 | ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) { | |
457 | return true; | |
458 | } | |
459 | addr += 16; /* 128-bit for each */ | |
460 | } | |
461 | return false; | |
462 | } | |
463 | ||
464 | /* Log and report an DMAR (address translation) fault to software */ | |
465 | static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id, | |
466 | hwaddr addr, VTDFaultReason fault, | |
467 | bool is_write) | |
468 | { | |
469 | uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); | |
470 | ||
471 | assert(fault < VTD_FR_MAX); | |
472 | ||
473 | if (fault == VTD_FR_RESERVED_ERR) { | |
474 | /* This is not a normal fault reason case. Drop it. */ | |
475 | return; | |
476 | } | |
7feb51b7 PX |
477 | |
478 | trace_vtd_dmar_fault(source_id, fault, addr, is_write); | |
479 | ||
1da12ec4 | 480 | if (fsts_reg & VTD_FSTS_PFO) { |
1376211f PX |
481 | error_report_once("New fault is not recorded due to " |
482 | "Primary Fault Overflow"); | |
1da12ec4 LT |
483 | return; |
484 | } | |
7feb51b7 | 485 | |
1da12ec4 | 486 | if (vtd_try_collapse_fault(s, source_id)) { |
1376211f PX |
487 | error_report_once("New fault is not recorded due to " |
488 | "compression of faults"); | |
1da12ec4 LT |
489 | return; |
490 | } | |
7feb51b7 | 491 | |
1da12ec4 | 492 | if (vtd_is_frcd_set(s, s->next_frcd_reg)) { |
1376211f PX |
493 | error_report_once("Next Fault Recording Reg is used, " |
494 | "new fault is not recorded, set PFO field"); | |
1da12ec4 LT |
495 | vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO); |
496 | return; | |
497 | } | |
498 | ||
499 | vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write); | |
500 | ||
501 | if (fsts_reg & VTD_FSTS_PPF) { | |
1376211f PX |
502 | error_report_once("There are pending faults already, " |
503 | "fault event is not generated"); | |
1da12ec4 LT |
504 | vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); |
505 | s->next_frcd_reg++; | |
506 | if (s->next_frcd_reg == DMAR_FRCD_REG_NR) { | |
507 | s->next_frcd_reg = 0; | |
508 | } | |
509 | } else { | |
510 | vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK, | |
511 | VTD_FSTS_FRI(s->next_frcd_reg)); | |
512 | vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */ | |
513 | s->next_frcd_reg++; | |
514 | if (s->next_frcd_reg == DMAR_FRCD_REG_NR) { | |
515 | s->next_frcd_reg = 0; | |
516 | } | |
517 | /* This case actually cause the PPF to be Set. | |
518 | * So generate fault event (interrupt). | |
519 | */ | |
520 | vtd_generate_fault_event(s, fsts_reg); | |
521 | } | |
522 | } | |
523 | ||
ed7b8fbc LT |
524 | /* Handle Invalidation Queue Errors of queued invalidation interface error |
525 | * conditions. | |
526 | */ | |
527 | static void vtd_handle_inv_queue_error(IntelIOMMUState *s) | |
528 | { | |
529 | uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); | |
530 | ||
531 | vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE); | |
532 | vtd_generate_fault_event(s, fsts_reg); | |
533 | } | |
534 | ||
535 | /* Set the IWC field and try to generate an invalidation completion interrupt */ | |
536 | static void vtd_generate_completion_event(IntelIOMMUState *s) | |
537 | { | |
ed7b8fbc | 538 | if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) { |
bc535e59 | 539 | trace_vtd_inv_desc_wait_irq("One pending, skip current"); |
ed7b8fbc LT |
540 | return; |
541 | } | |
542 | vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC); | |
543 | vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP); | |
544 | if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) { | |
bc535e59 PX |
545 | trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, " |
546 | "new event not generated"); | |
ed7b8fbc LT |
547 | return; |
548 | } else { | |
549 | /* Generate the interrupt event */ | |
bc535e59 | 550 | trace_vtd_inv_desc_wait_irq("Generating complete event"); |
ed7b8fbc LT |
551 | vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG); |
552 | vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); | |
553 | } | |
554 | } | |
555 | ||
fb43cf73 LY |
556 | static inline bool vtd_root_entry_present(IntelIOMMUState *s, |
557 | VTDRootEntry *re, | |
558 | uint8_t devfn) | |
1da12ec4 | 559 | { |
fb43cf73 LY |
560 | if (s->root_scalable && devfn > UINT8_MAX / 2) { |
561 | return re->hi & VTD_ROOT_ENTRY_P; | |
562 | } | |
563 | ||
564 | return re->lo & VTD_ROOT_ENTRY_P; | |
1da12ec4 LT |
565 | } |
566 | ||
567 | static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index, | |
568 | VTDRootEntry *re) | |
569 | { | |
570 | dma_addr_t addr; | |
571 | ||
572 | addr = s->root + index * sizeof(*re); | |
573 | if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) { | |
fb43cf73 | 574 | re->lo = 0; |
1da12ec4 LT |
575 | return -VTD_FR_ROOT_TABLE_INV; |
576 | } | |
fb43cf73 LY |
577 | re->lo = le64_to_cpu(re->lo); |
578 | re->hi = le64_to_cpu(re->hi); | |
1da12ec4 LT |
579 | return 0; |
580 | } | |
581 | ||
8f7d7161 | 582 | static inline bool vtd_ce_present(VTDContextEntry *context) |
1da12ec4 LT |
583 | { |
584 | return context->lo & VTD_CONTEXT_ENTRY_P; | |
585 | } | |
586 | ||
fb43cf73 LY |
587 | static int vtd_get_context_entry_from_root(IntelIOMMUState *s, |
588 | VTDRootEntry *re, | |
589 | uint8_t index, | |
1da12ec4 LT |
590 | VTDContextEntry *ce) |
591 | { | |
fb43cf73 | 592 | dma_addr_t addr, ce_size; |
1da12ec4 | 593 | |
6c441e1d | 594 | /* we have checked that root entry is present */ |
fb43cf73 LY |
595 | ce_size = s->root_scalable ? VTD_CTX_ENTRY_SCALABLE_SIZE : |
596 | VTD_CTX_ENTRY_LEGACY_SIZE; | |
597 | ||
598 | if (s->root_scalable && index > UINT8_MAX / 2) { | |
599 | index = index & (~VTD_DEVFN_CHECK_MASK); | |
600 | addr = re->hi & VTD_ROOT_ENTRY_CTP; | |
601 | } else { | |
602 | addr = re->lo & VTD_ROOT_ENTRY_CTP; | |
603 | } | |
604 | ||
605 | addr = addr + index * ce_size; | |
606 | if (dma_memory_read(&address_space_memory, addr, ce, ce_size)) { | |
1da12ec4 LT |
607 | return -VTD_FR_CONTEXT_TABLE_INV; |
608 | } | |
fb43cf73 | 609 | |
1da12ec4 LT |
610 | ce->lo = le64_to_cpu(ce->lo); |
611 | ce->hi = le64_to_cpu(ce->hi); | |
fb43cf73 LY |
612 | if (ce_size == VTD_CTX_ENTRY_SCALABLE_SIZE) { |
613 | ce->val[2] = le64_to_cpu(ce->val[2]); | |
614 | ce->val[3] = le64_to_cpu(ce->val[3]); | |
615 | } | |
1da12ec4 LT |
616 | return 0; |
617 | } | |
618 | ||
8f7d7161 | 619 | static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce) |
1da12ec4 LT |
620 | { |
621 | return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR; | |
622 | } | |
623 | ||
37f51384 | 624 | static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw) |
1da12ec4 | 625 | { |
37f51384 | 626 | return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw); |
1da12ec4 LT |
627 | } |
628 | ||
629 | /* Whether the pte indicates the address of the page frame */ | |
630 | static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level) | |
631 | { | |
632 | return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK); | |
633 | } | |
634 | ||
635 | /* Get the content of a spte located in @base_addr[@index] */ | |
636 | static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index) | |
637 | { | |
638 | uint64_t slpte; | |
639 | ||
640 | assert(index < VTD_SL_PT_ENTRY_NR); | |
641 | ||
642 | if (dma_memory_read(&address_space_memory, | |
643 | base_addr + index * sizeof(slpte), &slpte, | |
644 | sizeof(slpte))) { | |
645 | slpte = (uint64_t)-1; | |
646 | return slpte; | |
647 | } | |
648 | slpte = le64_to_cpu(slpte); | |
649 | return slpte; | |
650 | } | |
651 | ||
6e905564 PX |
652 | /* Given an iova and the level of paging structure, return the offset |
653 | * of current level. | |
1da12ec4 | 654 | */ |
6e905564 | 655 | static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level) |
1da12ec4 | 656 | { |
6e905564 | 657 | return (iova >> vtd_slpt_level_shift(level)) & |
1da12ec4 LT |
658 | ((1ULL << VTD_SL_LEVEL_BITS) - 1); |
659 | } | |
660 | ||
661 | /* Check Capability Register to see if the @level of page-table is supported */ | |
662 | static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level) | |
663 | { | |
664 | return VTD_CAP_SAGAW_MASK & s->cap & | |
665 | (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT)); | |
666 | } | |
667 | ||
fb43cf73 LY |
668 | /* Return true if check passed, otherwise false */ |
669 | static inline bool vtd_pe_type_check(X86IOMMUState *x86_iommu, | |
670 | VTDPASIDEntry *pe) | |
671 | { | |
672 | switch (VTD_PE_GET_TYPE(pe)) { | |
673 | case VTD_SM_PASID_ENTRY_FLT: | |
674 | case VTD_SM_PASID_ENTRY_SLT: | |
675 | case VTD_SM_PASID_ENTRY_NESTED: | |
676 | break; | |
677 | case VTD_SM_PASID_ENTRY_PT: | |
678 | if (!x86_iommu->pt_supported) { | |
679 | return false; | |
680 | } | |
681 | break; | |
682 | default: | |
683 | /* Unknwon type */ | |
684 | return false; | |
685 | } | |
686 | return true; | |
687 | } | |
688 | ||
56fc1e6a LY |
689 | static inline bool vtd_pdire_present(VTDPASIDDirEntry *pdire) |
690 | { | |
691 | return pdire->val & 1; | |
692 | } | |
693 | ||
694 | /** | |
695 | * Caller of this function should check present bit if wants | |
696 | * to use pdir entry for futher usage except for fpd bit check. | |
697 | */ | |
698 | static int vtd_get_pdire_from_pdir_table(dma_addr_t pasid_dir_base, | |
699 | uint32_t pasid, | |
700 | VTDPASIDDirEntry *pdire) | |
fb43cf73 LY |
701 | { |
702 | uint32_t index; | |
703 | dma_addr_t addr, entry_size; | |
704 | ||
705 | index = VTD_PASID_DIR_INDEX(pasid); | |
706 | entry_size = VTD_PASID_DIR_ENTRY_SIZE; | |
707 | addr = pasid_dir_base + index * entry_size; | |
708 | if (dma_memory_read(&address_space_memory, addr, pdire, entry_size)) { | |
709 | return -VTD_FR_PASID_TABLE_INV; | |
710 | } | |
711 | ||
712 | return 0; | |
713 | } | |
714 | ||
56fc1e6a LY |
715 | static inline bool vtd_pe_present(VTDPASIDEntry *pe) |
716 | { | |
717 | return pe->val[0] & VTD_PASID_ENTRY_P; | |
718 | } | |
719 | ||
720 | static int vtd_get_pe_in_pasid_leaf_table(IntelIOMMUState *s, | |
721 | uint32_t pasid, | |
722 | dma_addr_t addr, | |
723 | VTDPASIDEntry *pe) | |
fb43cf73 LY |
724 | { |
725 | uint32_t index; | |
56fc1e6a | 726 | dma_addr_t entry_size; |
fb43cf73 LY |
727 | X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s); |
728 | ||
729 | index = VTD_PASID_TABLE_INDEX(pasid); | |
730 | entry_size = VTD_PASID_ENTRY_SIZE; | |
fb43cf73 LY |
731 | addr = addr + index * entry_size; |
732 | if (dma_memory_read(&address_space_memory, addr, pe, entry_size)) { | |
733 | return -VTD_FR_PASID_TABLE_INV; | |
734 | } | |
735 | ||
736 | /* Do translation type check */ | |
737 | if (!vtd_pe_type_check(x86_iommu, pe)) { | |
738 | return -VTD_FR_PASID_TABLE_INV; | |
739 | } | |
740 | ||
741 | if (!vtd_is_level_supported(s, VTD_PE_GET_LEVEL(pe))) { | |
742 | return -VTD_FR_PASID_TABLE_INV; | |
743 | } | |
744 | ||
745 | return 0; | |
746 | } | |
747 | ||
56fc1e6a LY |
748 | /** |
749 | * Caller of this function should check present bit if wants | |
750 | * to use pasid entry for futher usage except for fpd bit check. | |
751 | */ | |
752 | static int vtd_get_pe_from_pdire(IntelIOMMUState *s, | |
753 | uint32_t pasid, | |
754 | VTDPASIDDirEntry *pdire, | |
755 | VTDPASIDEntry *pe) | |
756 | { | |
757 | dma_addr_t addr = pdire->val & VTD_PASID_TABLE_BASE_ADDR_MASK; | |
758 | ||
759 | return vtd_get_pe_in_pasid_leaf_table(s, pasid, addr, pe); | |
760 | } | |
761 | ||
762 | /** | |
763 | * This function gets a pasid entry from a specified pasid | |
764 | * table (includes dir and leaf table) with a specified pasid. | |
765 | * Sanity check should be done to ensure return a present | |
766 | * pasid entry to caller. | |
767 | */ | |
768 | static int vtd_get_pe_from_pasid_table(IntelIOMMUState *s, | |
769 | dma_addr_t pasid_dir_base, | |
770 | uint32_t pasid, | |
771 | VTDPASIDEntry *pe) | |
fb43cf73 LY |
772 | { |
773 | int ret; | |
774 | VTDPASIDDirEntry pdire; | |
775 | ||
56fc1e6a LY |
776 | ret = vtd_get_pdire_from_pdir_table(pasid_dir_base, |
777 | pasid, &pdire); | |
fb43cf73 LY |
778 | if (ret) { |
779 | return ret; | |
780 | } | |
781 | ||
56fc1e6a LY |
782 | if (!vtd_pdire_present(&pdire)) { |
783 | return -VTD_FR_PASID_TABLE_INV; | |
784 | } | |
785 | ||
786 | ret = vtd_get_pe_from_pdire(s, pasid, &pdire, pe); | |
fb43cf73 LY |
787 | if (ret) { |
788 | return ret; | |
789 | } | |
790 | ||
56fc1e6a LY |
791 | if (!vtd_pe_present(pe)) { |
792 | return -VTD_FR_PASID_TABLE_INV; | |
793 | } | |
794 | ||
795 | return 0; | |
fb43cf73 LY |
796 | } |
797 | ||
798 | static int vtd_ce_get_rid2pasid_entry(IntelIOMMUState *s, | |
799 | VTDContextEntry *ce, | |
800 | VTDPASIDEntry *pe) | |
801 | { | |
802 | uint32_t pasid; | |
803 | dma_addr_t pasid_dir_base; | |
804 | int ret = 0; | |
805 | ||
806 | pasid = VTD_CE_GET_RID2PASID(ce); | |
807 | pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce); | |
56fc1e6a | 808 | ret = vtd_get_pe_from_pasid_table(s, pasid_dir_base, pasid, pe); |
fb43cf73 LY |
809 | |
810 | return ret; | |
811 | } | |
812 | ||
813 | static int vtd_ce_get_pasid_fpd(IntelIOMMUState *s, | |
814 | VTDContextEntry *ce, | |
815 | bool *pe_fpd_set) | |
816 | { | |
817 | int ret; | |
818 | uint32_t pasid; | |
819 | dma_addr_t pasid_dir_base; | |
820 | VTDPASIDDirEntry pdire; | |
821 | VTDPASIDEntry pe; | |
822 | ||
823 | pasid = VTD_CE_GET_RID2PASID(ce); | |
824 | pasid_dir_base = VTD_CE_GET_PASID_DIR_TABLE(ce); | |
825 | ||
56fc1e6a LY |
826 | /* |
827 | * No present bit check since fpd is meaningful even | |
828 | * if the present bit is clear. | |
829 | */ | |
830 | ret = vtd_get_pdire_from_pdir_table(pasid_dir_base, pasid, &pdire); | |
fb43cf73 LY |
831 | if (ret) { |
832 | return ret; | |
833 | } | |
834 | ||
835 | if (pdire.val & VTD_PASID_DIR_FPD) { | |
836 | *pe_fpd_set = true; | |
837 | return 0; | |
838 | } | |
839 | ||
56fc1e6a LY |
840 | if (!vtd_pdire_present(&pdire)) { |
841 | return -VTD_FR_PASID_TABLE_INV; | |
842 | } | |
843 | ||
844 | /* | |
845 | * No present bit check since fpd is meaningful even | |
846 | * if the present bit is clear. | |
847 | */ | |
848 | ret = vtd_get_pe_from_pdire(s, pasid, &pdire, &pe); | |
fb43cf73 LY |
849 | if (ret) { |
850 | return ret; | |
851 | } | |
852 | ||
853 | if (pe.val[0] & VTD_PASID_ENTRY_FPD) { | |
854 | *pe_fpd_set = true; | |
855 | } | |
856 | ||
857 | return 0; | |
858 | } | |
859 | ||
1da12ec4 LT |
860 | /* Get the page-table level that hardware should use for the second-level |
861 | * page-table walk from the Address Width field of context-entry. | |
862 | */ | |
8f7d7161 | 863 | static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce) |
1da12ec4 LT |
864 | { |
865 | return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW); | |
866 | } | |
867 | ||
fb43cf73 LY |
868 | static uint32_t vtd_get_iova_level(IntelIOMMUState *s, |
869 | VTDContextEntry *ce) | |
870 | { | |
871 | VTDPASIDEntry pe; | |
872 | ||
873 | if (s->root_scalable) { | |
874 | vtd_ce_get_rid2pasid_entry(s, ce, &pe); | |
875 | return VTD_PE_GET_LEVEL(&pe); | |
876 | } | |
877 | ||
878 | return vtd_ce_get_level(ce); | |
879 | } | |
880 | ||
8f7d7161 | 881 | static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce) |
1da12ec4 LT |
882 | { |
883 | return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9; | |
884 | } | |
885 | ||
fb43cf73 LY |
886 | static uint32_t vtd_get_iova_agaw(IntelIOMMUState *s, |
887 | VTDContextEntry *ce) | |
888 | { | |
889 | VTDPASIDEntry pe; | |
890 | ||
891 | if (s->root_scalable) { | |
892 | vtd_ce_get_rid2pasid_entry(s, ce, &pe); | |
893 | return 30 + ((pe.val[0] >> 2) & VTD_SM_PASID_ENTRY_AW) * 9; | |
894 | } | |
895 | ||
896 | return vtd_ce_get_agaw(ce); | |
897 | } | |
898 | ||
127ff5c3 PX |
899 | static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce) |
900 | { | |
901 | return ce->lo & VTD_CONTEXT_ENTRY_TT; | |
902 | } | |
903 | ||
fb43cf73 | 904 | /* Only for Legacy Mode. Return true if check passed, otherwise false */ |
f80c9874 PX |
905 | static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu, |
906 | VTDContextEntry *ce) | |
907 | { | |
908 | switch (vtd_ce_get_type(ce)) { | |
909 | case VTD_CONTEXT_TT_MULTI_LEVEL: | |
910 | /* Always supported */ | |
911 | break; | |
912 | case VTD_CONTEXT_TT_DEV_IOTLB: | |
913 | if (!x86_iommu->dt_supported) { | |
095955b2 | 914 | error_report_once("%s: DT specified but not supported", __func__); |
f80c9874 PX |
915 | return false; |
916 | } | |
917 | break; | |
dbaabb25 PX |
918 | case VTD_CONTEXT_TT_PASS_THROUGH: |
919 | if (!x86_iommu->pt_supported) { | |
095955b2 | 920 | error_report_once("%s: PT specified but not supported", __func__); |
dbaabb25 PX |
921 | return false; |
922 | } | |
923 | break; | |
f80c9874 | 924 | default: |
fb43cf73 | 925 | /* Unknown type */ |
095955b2 PX |
926 | error_report_once("%s: unknown ce type: %"PRIu32, __func__, |
927 | vtd_ce_get_type(ce)); | |
f80c9874 PX |
928 | return false; |
929 | } | |
930 | return true; | |
931 | } | |
932 | ||
fb43cf73 LY |
933 | static inline uint64_t vtd_iova_limit(IntelIOMMUState *s, |
934 | VTDContextEntry *ce, uint8_t aw) | |
f06a696d | 935 | { |
fb43cf73 | 936 | uint32_t ce_agaw = vtd_get_iova_agaw(s, ce); |
37f51384 | 937 | return 1ULL << MIN(ce_agaw, aw); |
f06a696d PX |
938 | } |
939 | ||
940 | /* Return true if IOVA passes range check, otherwise false. */ | |
fb43cf73 LY |
941 | static inline bool vtd_iova_range_check(IntelIOMMUState *s, |
942 | uint64_t iova, VTDContextEntry *ce, | |
37f51384 | 943 | uint8_t aw) |
f06a696d PX |
944 | { |
945 | /* | |
946 | * Check if @iova is above 2^X-1, where X is the minimum of MGAW | |
947 | * in CAP_REG and AW in context-entry. | |
948 | */ | |
fb43cf73 LY |
949 | return !(iova & ~(vtd_iova_limit(s, ce, aw) - 1)); |
950 | } | |
951 | ||
952 | static dma_addr_t vtd_get_iova_pgtbl_base(IntelIOMMUState *s, | |
953 | VTDContextEntry *ce) | |
954 | { | |
955 | VTDPASIDEntry pe; | |
956 | ||
957 | if (s->root_scalable) { | |
958 | vtd_ce_get_rid2pasid_entry(s, ce, &pe); | |
959 | return pe.val[0] & VTD_SM_PASID_ENTRY_SLPTPTR; | |
960 | } | |
961 | ||
962 | return vtd_ce_get_slpt_base(ce); | |
f06a696d PX |
963 | } |
964 | ||
92e5d85e PS |
965 | /* |
966 | * Rsvd field masks for spte: | |
ce586f3b QY |
967 | * vtd_spte_rsvd 4k pages |
968 | * vtd_spte_rsvd_large large pages | |
92e5d85e | 969 | */ |
ce586f3b QY |
970 | static uint64_t vtd_spte_rsvd[5]; |
971 | static uint64_t vtd_spte_rsvd_large[5]; | |
1da12ec4 LT |
972 | |
973 | static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level) | |
974 | { | |
ce586f3b QY |
975 | uint64_t rsvd_mask = vtd_spte_rsvd[level]; |
976 | ||
977 | if ((level == VTD_SL_PD_LEVEL || level == VTD_SL_PDP_LEVEL) && | |
978 | (slpte & VTD_SL_PT_PAGE_SIZE_MASK)) { | |
979 | /* large page */ | |
980 | rsvd_mask = vtd_spte_rsvd_large[level]; | |
1da12ec4 | 981 | } |
ce586f3b QY |
982 | |
983 | return slpte & rsvd_mask; | |
1da12ec4 LT |
984 | } |
985 | ||
dbaabb25 PX |
986 | /* Find the VTD address space associated with a given bus number */ |
987 | static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num) | |
988 | { | |
989 | VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num]; | |
a6f65f4f | 990 | GHashTableIter iter; |
dbaabb25 | 991 | |
a6f65f4f PMD |
992 | if (vtd_bus) { |
993 | return vtd_bus; | |
994 | } | |
995 | ||
996 | /* | |
997 | * Iterate over the registered buses to find the one which | |
998 | * currently holds this bus number and update the bus_num | |
999 | * lookup table. | |
1000 | */ | |
1001 | g_hash_table_iter_init(&iter, s->vtd_as_by_busptr); | |
1002 | while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) { | |
1003 | if (pci_bus_num(vtd_bus->bus) == bus_num) { | |
1004 | s->vtd_as_by_bus_num[bus_num] = vtd_bus; | |
1005 | return vtd_bus; | |
dbaabb25 PX |
1006 | } |
1007 | } | |
a6f65f4f PMD |
1008 | |
1009 | return NULL; | |
dbaabb25 PX |
1010 | } |
1011 | ||
6e905564 | 1012 | /* Given the @iova, get relevant @slptep. @slpte_level will be the last level |
1da12ec4 LT |
1013 | * of the translation, can be used for deciding the size of large page. |
1014 | */ | |
fb43cf73 LY |
1015 | static int vtd_iova_to_slpte(IntelIOMMUState *s, VTDContextEntry *ce, |
1016 | uint64_t iova, bool is_write, | |
6e905564 | 1017 | uint64_t *slptep, uint32_t *slpte_level, |
37f51384 | 1018 | bool *reads, bool *writes, uint8_t aw_bits) |
1da12ec4 | 1019 | { |
fb43cf73 LY |
1020 | dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce); |
1021 | uint32_t level = vtd_get_iova_level(s, ce); | |
1da12ec4 LT |
1022 | uint32_t offset; |
1023 | uint64_t slpte; | |
1da12ec4 LT |
1024 | uint64_t access_right_check; |
1025 | ||
fb43cf73 | 1026 | if (!vtd_iova_range_check(s, iova, ce, aw_bits)) { |
4e4abd11 PX |
1027 | error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")", |
1028 | __func__, iova); | |
1da12ec4 LT |
1029 | return -VTD_FR_ADDR_BEYOND_MGAW; |
1030 | } | |
1031 | ||
1032 | /* FIXME: what is the Atomics request here? */ | |
1033 | access_right_check = is_write ? VTD_SL_W : VTD_SL_R; | |
1034 | ||
1035 | while (true) { | |
6e905564 | 1036 | offset = vtd_iova_level_offset(iova, level); |
1da12ec4 LT |
1037 | slpte = vtd_get_slpte(addr, offset); |
1038 | ||
1039 | if (slpte == (uint64_t)-1) { | |
4e4abd11 PX |
1040 | error_report_once("%s: detected read error on DMAR slpte " |
1041 | "(iova=0x%" PRIx64 ")", __func__, iova); | |
fb43cf73 | 1042 | if (level == vtd_get_iova_level(s, ce)) { |
1da12ec4 LT |
1043 | /* Invalid programming of context-entry */ |
1044 | return -VTD_FR_CONTEXT_ENTRY_INV; | |
1045 | } else { | |
1046 | return -VTD_FR_PAGING_ENTRY_INV; | |
1047 | } | |
1048 | } | |
1049 | *reads = (*reads) && (slpte & VTD_SL_R); | |
1050 | *writes = (*writes) && (slpte & VTD_SL_W); | |
1051 | if (!(slpte & access_right_check)) { | |
4e4abd11 PX |
1052 | error_report_once("%s: detected slpte permission error " |
1053 | "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", " | |
1054 | "slpte=0x%" PRIx64 ", write=%d)", __func__, | |
1055 | iova, level, slpte, is_write); | |
1da12ec4 LT |
1056 | return is_write ? -VTD_FR_WRITE : -VTD_FR_READ; |
1057 | } | |
1058 | if (vtd_slpte_nonzero_rsvd(slpte, level)) { | |
4e4abd11 PX |
1059 | error_report_once("%s: detected splte reserve non-zero " |
1060 | "iova=0x%" PRIx64 ", level=0x%" PRIx32 | |
1061 | "slpte=0x%" PRIx64 ")", __func__, iova, | |
1062 | level, slpte); | |
1da12ec4 LT |
1063 | return -VTD_FR_PAGING_ENTRY_RSVD; |
1064 | } | |
1065 | ||
1066 | if (vtd_is_last_slpte(slpte, level)) { | |
1067 | *slptep = slpte; | |
1068 | *slpte_level = level; | |
1069 | return 0; | |
1070 | } | |
37f51384 | 1071 | addr = vtd_get_slpte_addr(slpte, aw_bits); |
1da12ec4 LT |
1072 | level--; |
1073 | } | |
1074 | } | |
1075 | ||
f06a696d PX |
1076 | typedef int (*vtd_page_walk_hook)(IOMMUTLBEntry *entry, void *private); |
1077 | ||
fe215b0c PX |
1078 | /** |
1079 | * Constant information used during page walking | |
1080 | * | |
1081 | * @hook_fn: hook func to be called when detected page | |
1082 | * @private: private data to be passed into hook func | |
1083 | * @notify_unmap: whether we should notify invalid entries | |
2f764fa8 | 1084 | * @as: VT-d address space of the device |
fe215b0c | 1085 | * @aw: maximum address width |
d118c06e | 1086 | * @domain: domain ID of the page walk |
fe215b0c PX |
1087 | */ |
1088 | typedef struct { | |
2f764fa8 | 1089 | VTDAddressSpace *as; |
fe215b0c PX |
1090 | vtd_page_walk_hook hook_fn; |
1091 | void *private; | |
1092 | bool notify_unmap; | |
1093 | uint8_t aw; | |
d118c06e | 1094 | uint16_t domain_id; |
fe215b0c PX |
1095 | } vtd_page_walk_info; |
1096 | ||
d118c06e | 1097 | static int vtd_page_walk_one(IOMMUTLBEntry *entry, vtd_page_walk_info *info) |
36d2d52b | 1098 | { |
63b88968 | 1099 | VTDAddressSpace *as = info->as; |
fe215b0c PX |
1100 | vtd_page_walk_hook hook_fn = info->hook_fn; |
1101 | void *private = info->private; | |
63b88968 PX |
1102 | DMAMap target = { |
1103 | .iova = entry->iova, | |
1104 | .size = entry->addr_mask, | |
1105 | .translated_addr = entry->translated_addr, | |
1106 | .perm = entry->perm, | |
1107 | }; | |
1108 | DMAMap *mapped = iova_tree_find(as->iova_tree, &target); | |
1109 | ||
1110 | if (entry->perm == IOMMU_NONE && !info->notify_unmap) { | |
1111 | trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask); | |
1112 | return 0; | |
1113 | } | |
fe215b0c | 1114 | |
36d2d52b | 1115 | assert(hook_fn); |
63b88968 PX |
1116 | |
1117 | /* Update local IOVA mapped ranges */ | |
1118 | if (entry->perm) { | |
1119 | if (mapped) { | |
1120 | /* If it's exactly the same translation, skip */ | |
1121 | if (!memcmp(mapped, &target, sizeof(target))) { | |
1122 | trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask, | |
1123 | entry->translated_addr); | |
1124 | return 0; | |
1125 | } else { | |
1126 | /* | |
1127 | * Translation changed. Normally this should not | |
1128 | * happen, but it can happen when with buggy guest | |
1129 | * OSes. Note that there will be a small window that | |
1130 | * we don't have map at all. But that's the best | |
1131 | * effort we can do. The ideal way to emulate this is | |
1132 | * atomically modify the PTE to follow what has | |
1133 | * changed, but we can't. One example is that vfio | |
1134 | * driver only has VFIO_IOMMU_[UN]MAP_DMA but no | |
1135 | * interface to modify a mapping (meanwhile it seems | |
1136 | * meaningless to even provide one). Anyway, let's | |
1137 | * mark this as a TODO in case one day we'll have | |
1138 | * a better solution. | |
1139 | */ | |
1140 | IOMMUAccessFlags cache_perm = entry->perm; | |
1141 | int ret; | |
1142 | ||
1143 | /* Emulate an UNMAP */ | |
1144 | entry->perm = IOMMU_NONE; | |
1145 | trace_vtd_page_walk_one(info->domain_id, | |
1146 | entry->iova, | |
1147 | entry->translated_addr, | |
1148 | entry->addr_mask, | |
1149 | entry->perm); | |
1150 | ret = hook_fn(entry, private); | |
1151 | if (ret) { | |
1152 | return ret; | |
1153 | } | |
1154 | /* Drop any existing mapping */ | |
1155 | iova_tree_remove(as->iova_tree, &target); | |
1156 | /* Recover the correct permission */ | |
1157 | entry->perm = cache_perm; | |
1158 | } | |
1159 | } | |
1160 | iova_tree_insert(as->iova_tree, &target); | |
1161 | } else { | |
1162 | if (!mapped) { | |
1163 | /* Skip since we didn't map this range at all */ | |
1164 | trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask); | |
1165 | return 0; | |
1166 | } | |
1167 | iova_tree_remove(as->iova_tree, &target); | |
1168 | } | |
1169 | ||
d118c06e PX |
1170 | trace_vtd_page_walk_one(info->domain_id, entry->iova, |
1171 | entry->translated_addr, entry->addr_mask, | |
1172 | entry->perm); | |
36d2d52b PX |
1173 | return hook_fn(entry, private); |
1174 | } | |
1175 | ||
f06a696d PX |
1176 | /** |
1177 | * vtd_page_walk_level - walk over specific level for IOVA range | |
1178 | * | |
1179 | * @addr: base GPA addr to start the walk | |
1180 | * @start: IOVA range start address | |
1181 | * @end: IOVA range end address (start <= addr < end) | |
f06a696d PX |
1182 | * @read: whether parent level has read permission |
1183 | * @write: whether parent level has write permission | |
fe215b0c | 1184 | * @info: constant information for the page walk |
f06a696d PX |
1185 | */ |
1186 | static int vtd_page_walk_level(dma_addr_t addr, uint64_t start, | |
fe215b0c PX |
1187 | uint64_t end, uint32_t level, bool read, |
1188 | bool write, vtd_page_walk_info *info) | |
f06a696d PX |
1189 | { |
1190 | bool read_cur, write_cur, entry_valid; | |
1191 | uint32_t offset; | |
1192 | uint64_t slpte; | |
1193 | uint64_t subpage_size, subpage_mask; | |
1194 | IOMMUTLBEntry entry; | |
1195 | uint64_t iova = start; | |
1196 | uint64_t iova_next; | |
1197 | int ret = 0; | |
1198 | ||
1199 | trace_vtd_page_walk_level(addr, level, start, end); | |
1200 | ||
1201 | subpage_size = 1ULL << vtd_slpt_level_shift(level); | |
1202 | subpage_mask = vtd_slpt_level_page_mask(level); | |
1203 | ||
1204 | while (iova < end) { | |
1205 | iova_next = (iova & subpage_mask) + subpage_size; | |
1206 | ||
1207 | offset = vtd_iova_level_offset(iova, level); | |
1208 | slpte = vtd_get_slpte(addr, offset); | |
1209 | ||
1210 | if (slpte == (uint64_t)-1) { | |
1211 | trace_vtd_page_walk_skip_read(iova, iova_next); | |
1212 | goto next; | |
1213 | } | |
1214 | ||
1215 | if (vtd_slpte_nonzero_rsvd(slpte, level)) { | |
1216 | trace_vtd_page_walk_skip_reserve(iova, iova_next); | |
1217 | goto next; | |
1218 | } | |
1219 | ||
1220 | /* Permissions are stacked with parents' */ | |
1221 | read_cur = read && (slpte & VTD_SL_R); | |
1222 | write_cur = write && (slpte & VTD_SL_W); | |
1223 | ||
1224 | /* | |
1225 | * As long as we have either read/write permission, this is a | |
1226 | * valid entry. The rule works for both page entries and page | |
1227 | * table entries. | |
1228 | */ | |
1229 | entry_valid = read_cur | write_cur; | |
1230 | ||
63b88968 PX |
1231 | if (!vtd_is_last_slpte(slpte, level) && entry_valid) { |
1232 | /* | |
1233 | * This is a valid PDE (or even bigger than PDE). We need | |
1234 | * to walk one further level. | |
1235 | */ | |
fe215b0c PX |
1236 | ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw), |
1237 | iova, MIN(iova_next, end), level - 1, | |
1238 | read_cur, write_cur, info); | |
63b88968 PX |
1239 | } else { |
1240 | /* | |
1241 | * This means we are either: | |
1242 | * | |
1243 | * (1) the real page entry (either 4K page, or huge page) | |
1244 | * (2) the whole range is invalid | |
1245 | * | |
1246 | * In either case, we send an IOTLB notification down. | |
1247 | */ | |
1248 | entry.target_as = &address_space_memory; | |
1249 | entry.iova = iova & subpage_mask; | |
1250 | entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur); | |
1251 | entry.addr_mask = ~subpage_mask; | |
1252 | /* NOTE: this is only meaningful if entry_valid == true */ | |
1253 | entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw); | |
1254 | ret = vtd_page_walk_one(&entry, info); | |
1255 | } | |
1256 | ||
1257 | if (ret < 0) { | |
1258 | return ret; | |
f06a696d PX |
1259 | } |
1260 | ||
1261 | next: | |
1262 | iova = iova_next; | |
1263 | } | |
1264 | ||
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | /** | |
1269 | * vtd_page_walk - walk specific IOVA range, and call the hook | |
1270 | * | |
fb43cf73 | 1271 | * @s: intel iommu state |
f06a696d PX |
1272 | * @ce: context entry to walk upon |
1273 | * @start: IOVA address to start the walk | |
1274 | * @end: IOVA range end address (start <= addr < end) | |
fe215b0c | 1275 | * @info: page walking information struct |
f06a696d | 1276 | */ |
fb43cf73 LY |
1277 | static int vtd_page_walk(IntelIOMMUState *s, VTDContextEntry *ce, |
1278 | uint64_t start, uint64_t end, | |
fe215b0c | 1279 | vtd_page_walk_info *info) |
f06a696d | 1280 | { |
fb43cf73 LY |
1281 | dma_addr_t addr = vtd_get_iova_pgtbl_base(s, ce); |
1282 | uint32_t level = vtd_get_iova_level(s, ce); | |
f06a696d | 1283 | |
fb43cf73 | 1284 | if (!vtd_iova_range_check(s, start, ce, info->aw)) { |
f06a696d PX |
1285 | return -VTD_FR_ADDR_BEYOND_MGAW; |
1286 | } | |
1287 | ||
fb43cf73 | 1288 | if (!vtd_iova_range_check(s, end, ce, info->aw)) { |
f06a696d | 1289 | /* Fix end so that it reaches the maximum */ |
fb43cf73 | 1290 | end = vtd_iova_limit(s, ce, info->aw); |
f06a696d PX |
1291 | } |
1292 | ||
fe215b0c | 1293 | return vtd_page_walk_level(addr, start, end, level, true, true, info); |
f06a696d PX |
1294 | } |
1295 | ||
fb43cf73 LY |
1296 | static int vtd_root_entry_rsvd_bits_check(IntelIOMMUState *s, |
1297 | VTDRootEntry *re) | |
1298 | { | |
1299 | /* Legacy Mode reserved bits check */ | |
1300 | if (!s->root_scalable && | |
1301 | (re->hi || (re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)))) | |
1302 | goto rsvd_err; | |
1303 | ||
1304 | /* Scalable Mode reserved bits check */ | |
1305 | if (s->root_scalable && | |
1306 | ((re->lo & VTD_ROOT_ENTRY_RSVD(s->aw_bits)) || | |
1307 | (re->hi & VTD_ROOT_ENTRY_RSVD(s->aw_bits)))) | |
1308 | goto rsvd_err; | |
1309 | ||
1310 | return 0; | |
1311 | ||
1312 | rsvd_err: | |
1313 | error_report_once("%s: invalid root entry: hi=0x%"PRIx64 | |
1314 | ", lo=0x%"PRIx64, | |
1315 | __func__, re->hi, re->lo); | |
1316 | return -VTD_FR_ROOT_ENTRY_RSVD; | |
1317 | } | |
1318 | ||
1319 | static inline int vtd_context_entry_rsvd_bits_check(IntelIOMMUState *s, | |
1320 | VTDContextEntry *ce) | |
1321 | { | |
1322 | if (!s->root_scalable && | |
1323 | (ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI || | |
1324 | ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) { | |
1325 | error_report_once("%s: invalid context entry: hi=%"PRIx64 | |
1326 | ", lo=%"PRIx64" (reserved nonzero)", | |
1327 | __func__, ce->hi, ce->lo); | |
1328 | return -VTD_FR_CONTEXT_ENTRY_RSVD; | |
1329 | } | |
1330 | ||
1331 | if (s->root_scalable && | |
1332 | (ce->val[0] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL0(s->aw_bits) || | |
1333 | ce->val[1] & VTD_SM_CONTEXT_ENTRY_RSVD_VAL1 || | |
1334 | ce->val[2] || | |
1335 | ce->val[3])) { | |
1336 | error_report_once("%s: invalid context entry: val[3]=%"PRIx64 | |
1337 | ", val[2]=%"PRIx64 | |
1338 | ", val[1]=%"PRIx64 | |
1339 | ", val[0]=%"PRIx64" (reserved nonzero)", | |
1340 | __func__, ce->val[3], ce->val[2], | |
1341 | ce->val[1], ce->val[0]); | |
1342 | return -VTD_FR_CONTEXT_ENTRY_RSVD; | |
1343 | } | |
1344 | ||
1345 | return 0; | |
1346 | } | |
1347 | ||
1348 | static int vtd_ce_rid2pasid_check(IntelIOMMUState *s, | |
1349 | VTDContextEntry *ce) | |
1350 | { | |
1351 | VTDPASIDEntry pe; | |
1352 | ||
1353 | /* | |
1354 | * Make sure in Scalable Mode, a present context entry | |
1355 | * has valid rid2pasid setting, which includes valid | |
1356 | * rid2pasid field and corresponding pasid entry setting | |
1357 | */ | |
1358 | return vtd_ce_get_rid2pasid_entry(s, ce, &pe); | |
1359 | } | |
1360 | ||
1da12ec4 LT |
1361 | /* Map a device to its corresponding domain (context-entry) */ |
1362 | static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num, | |
1363 | uint8_t devfn, VTDContextEntry *ce) | |
1364 | { | |
1365 | VTDRootEntry re; | |
1366 | int ret_fr; | |
f80c9874 | 1367 | X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s); |
1da12ec4 LT |
1368 | |
1369 | ret_fr = vtd_get_root_entry(s, bus_num, &re); | |
1370 | if (ret_fr) { | |
1371 | return ret_fr; | |
1372 | } | |
1373 | ||
fb43cf73 | 1374 | if (!vtd_root_entry_present(s, &re, devfn)) { |
6c441e1d PX |
1375 | /* Not error - it's okay we don't have root entry. */ |
1376 | trace_vtd_re_not_present(bus_num); | |
1da12ec4 | 1377 | return -VTD_FR_ROOT_ENTRY_P; |
f80c9874 PX |
1378 | } |
1379 | ||
fb43cf73 LY |
1380 | ret_fr = vtd_root_entry_rsvd_bits_check(s, &re); |
1381 | if (ret_fr) { | |
1382 | return ret_fr; | |
1da12ec4 LT |
1383 | } |
1384 | ||
fb43cf73 | 1385 | ret_fr = vtd_get_context_entry_from_root(s, &re, devfn, ce); |
1da12ec4 LT |
1386 | if (ret_fr) { |
1387 | return ret_fr; | |
1388 | } | |
1389 | ||
8f7d7161 | 1390 | if (!vtd_ce_present(ce)) { |
6c441e1d PX |
1391 | /* Not error - it's okay we don't have context entry. */ |
1392 | trace_vtd_ce_not_present(bus_num, devfn); | |
1da12ec4 | 1393 | return -VTD_FR_CONTEXT_ENTRY_P; |
f80c9874 PX |
1394 | } |
1395 | ||
fb43cf73 LY |
1396 | ret_fr = vtd_context_entry_rsvd_bits_check(s, ce); |
1397 | if (ret_fr) { | |
1398 | return ret_fr; | |
1da12ec4 | 1399 | } |
f80c9874 | 1400 | |
1da12ec4 | 1401 | /* Check if the programming of context-entry is valid */ |
fb43cf73 LY |
1402 | if (!s->root_scalable && |
1403 | !vtd_is_level_supported(s, vtd_ce_get_level(ce))) { | |
095955b2 PX |
1404 | error_report_once("%s: invalid context entry: hi=%"PRIx64 |
1405 | ", lo=%"PRIx64" (level %d not supported)", | |
fb43cf73 LY |
1406 | __func__, ce->hi, ce->lo, |
1407 | vtd_ce_get_level(ce)); | |
1da12ec4 | 1408 | return -VTD_FR_CONTEXT_ENTRY_INV; |
1da12ec4 | 1409 | } |
f80c9874 | 1410 | |
fb43cf73 LY |
1411 | if (!s->root_scalable) { |
1412 | /* Do translation type check */ | |
1413 | if (!vtd_ce_type_check(x86_iommu, ce)) { | |
1414 | /* Errors dumped in vtd_ce_type_check() */ | |
1415 | return -VTD_FR_CONTEXT_ENTRY_INV; | |
1416 | } | |
1417 | } else { | |
1418 | /* | |
1419 | * Check if the programming of context-entry.rid2pasid | |
1420 | * and corresponding pasid setting is valid, and thus | |
1421 | * avoids to check pasid entry fetching result in future | |
1422 | * helper function calling. | |
1423 | */ | |
1424 | ret_fr = vtd_ce_rid2pasid_check(s, ce); | |
1425 | if (ret_fr) { | |
1426 | return ret_fr; | |
1427 | } | |
f80c9874 PX |
1428 | } |
1429 | ||
1da12ec4 LT |
1430 | return 0; |
1431 | } | |
1432 | ||
63b88968 PX |
1433 | static int vtd_sync_shadow_page_hook(IOMMUTLBEntry *entry, |
1434 | void *private) | |
1435 | { | |
cb1efcf4 | 1436 | memory_region_notify_iommu((IOMMUMemoryRegion *)private, 0, *entry); |
63b88968 PX |
1437 | return 0; |
1438 | } | |
1439 | ||
fb43cf73 LY |
1440 | static uint16_t vtd_get_domain_id(IntelIOMMUState *s, |
1441 | VTDContextEntry *ce) | |
1442 | { | |
1443 | VTDPASIDEntry pe; | |
1444 | ||
1445 | if (s->root_scalable) { | |
1446 | vtd_ce_get_rid2pasid_entry(s, ce, &pe); | |
1447 | return VTD_SM_PASID_ENTRY_DID(pe.val[1]); | |
1448 | } | |
1449 | ||
1450 | return VTD_CONTEXT_ENTRY_DID(ce->hi); | |
1451 | } | |
1452 | ||
63b88968 PX |
1453 | static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as, |
1454 | VTDContextEntry *ce, | |
1455 | hwaddr addr, hwaddr size) | |
1456 | { | |
1457 | IntelIOMMUState *s = vtd_as->iommu_state; | |
1458 | vtd_page_walk_info info = { | |
1459 | .hook_fn = vtd_sync_shadow_page_hook, | |
1460 | .private = (void *)&vtd_as->iommu, | |
1461 | .notify_unmap = true, | |
1462 | .aw = s->aw_bits, | |
1463 | .as = vtd_as, | |
fb43cf73 | 1464 | .domain_id = vtd_get_domain_id(s, ce), |
63b88968 | 1465 | }; |
63b88968 | 1466 | |
fb43cf73 | 1467 | return vtd_page_walk(s, ce, addr, addr + size, &info); |
63b88968 PX |
1468 | } |
1469 | ||
1470 | static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as) | |
1471 | { | |
95ecd3df PX |
1472 | int ret; |
1473 | VTDContextEntry ce; | |
c28b535d | 1474 | IOMMUNotifier *n; |
95ecd3df PX |
1475 | |
1476 | ret = vtd_dev_to_context_entry(vtd_as->iommu_state, | |
1477 | pci_bus_num(vtd_as->bus), | |
1478 | vtd_as->devfn, &ce); | |
1479 | if (ret) { | |
c28b535d PX |
1480 | if (ret == -VTD_FR_CONTEXT_ENTRY_P) { |
1481 | /* | |
1482 | * It's a valid scenario to have a context entry that is | |
1483 | * not present. For example, when a device is removed | |
1484 | * from an existing domain then the context entry will be | |
1485 | * zeroed by the guest before it was put into another | |
1486 | * domain. When this happens, instead of synchronizing | |
1487 | * the shadow pages we should invalidate all existing | |
1488 | * mappings and notify the backends. | |
1489 | */ | |
1490 | IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) { | |
1491 | vtd_address_space_unmap(vtd_as, n); | |
1492 | } | |
1493 | ret = 0; | |
1494 | } | |
95ecd3df PX |
1495 | return ret; |
1496 | } | |
1497 | ||
1498 | return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX); | |
63b88968 PX |
1499 | } |
1500 | ||
dbaabb25 | 1501 | /* |
fb43cf73 LY |
1502 | * Check if specific device is configed to bypass address |
1503 | * translation for DMA requests. In Scalable Mode, bypass | |
1504 | * 1st-level translation or 2nd-level translation, it depends | |
1505 | * on PGTT setting. | |
dbaabb25 | 1506 | */ |
fb43cf73 | 1507 | static bool vtd_dev_pt_enabled(VTDAddressSpace *as) |
dbaabb25 PX |
1508 | { |
1509 | IntelIOMMUState *s; | |
1510 | VTDContextEntry ce; | |
fb43cf73 | 1511 | VTDPASIDEntry pe; |
dbaabb25 PX |
1512 | int ret; |
1513 | ||
fb43cf73 | 1514 | assert(as); |
dbaabb25 | 1515 | |
fb43cf73 | 1516 | s = as->iommu_state; |
dbaabb25 PX |
1517 | ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus), |
1518 | as->devfn, &ce); | |
1519 | if (ret) { | |
dbaabb25 PX |
1520 | /* |
1521 | * Possibly failed to parse the context entry for some reason | |
1522 | * (e.g., during init, or any guest configuration errors on | |
1523 | * context entries). We should assume PT not enabled for | |
1524 | * safety. | |
1525 | */ | |
1526 | return false; | |
1527 | } | |
1528 | ||
fb43cf73 LY |
1529 | if (s->root_scalable) { |
1530 | ret = vtd_ce_get_rid2pasid_entry(s, &ce, &pe); | |
1531 | if (ret) { | |
1532 | error_report_once("%s: vtd_ce_get_rid2pasid_entry error: %"PRId32, | |
1533 | __func__, ret); | |
1534 | return false; | |
1535 | } | |
1536 | return (VTD_PE_GET_TYPE(&pe) == VTD_SM_PASID_ENTRY_PT); | |
1537 | } | |
1538 | ||
1539 | return (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH); | |
dbaabb25 PX |
1540 | } |
1541 | ||
1542 | /* Return whether the device is using IOMMU translation. */ | |
1543 | static bool vtd_switch_address_space(VTDAddressSpace *as) | |
1544 | { | |
1545 | bool use_iommu; | |
66a4a031 PX |
1546 | /* Whether we need to take the BQL on our own */ |
1547 | bool take_bql = !qemu_mutex_iothread_locked(); | |
dbaabb25 PX |
1548 | |
1549 | assert(as); | |
1550 | ||
2a078b10 | 1551 | use_iommu = as->iommu_state->dmar_enabled && !vtd_dev_pt_enabled(as); |
dbaabb25 PX |
1552 | |
1553 | trace_vtd_switch_address_space(pci_bus_num(as->bus), | |
1554 | VTD_PCI_SLOT(as->devfn), | |
1555 | VTD_PCI_FUNC(as->devfn), | |
1556 | use_iommu); | |
1557 | ||
66a4a031 PX |
1558 | /* |
1559 | * It's possible that we reach here without BQL, e.g., when called | |
1560 | * from vtd_pt_enable_fast_path(). However the memory APIs need | |
1561 | * it. We'd better make sure we have had it already, or, take it. | |
1562 | */ | |
1563 | if (take_bql) { | |
1564 | qemu_mutex_lock_iothread(); | |
1565 | } | |
1566 | ||
dbaabb25 PX |
1567 | /* Turn off first then on the other */ |
1568 | if (use_iommu) { | |
4b519ef1 | 1569 | memory_region_set_enabled(&as->nodmar, false); |
3df9d748 | 1570 | memory_region_set_enabled(MEMORY_REGION(&as->iommu), true); |
dbaabb25 | 1571 | } else { |
3df9d748 | 1572 | memory_region_set_enabled(MEMORY_REGION(&as->iommu), false); |
4b519ef1 | 1573 | memory_region_set_enabled(&as->nodmar, true); |
dbaabb25 PX |
1574 | } |
1575 | ||
66a4a031 PX |
1576 | if (take_bql) { |
1577 | qemu_mutex_unlock_iothread(); | |
1578 | } | |
1579 | ||
dbaabb25 PX |
1580 | return use_iommu; |
1581 | } | |
1582 | ||
1583 | static void vtd_switch_address_space_all(IntelIOMMUState *s) | |
1584 | { | |
1585 | GHashTableIter iter; | |
1586 | VTDBus *vtd_bus; | |
1587 | int i; | |
1588 | ||
1589 | g_hash_table_iter_init(&iter, s->vtd_as_by_busptr); | |
1590 | while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) { | |
bf33cc75 | 1591 | for (i = 0; i < PCI_DEVFN_MAX; i++) { |
dbaabb25 PX |
1592 | if (!vtd_bus->dev_as[i]) { |
1593 | continue; | |
1594 | } | |
1595 | vtd_switch_address_space(vtd_bus->dev_as[i]); | |
1596 | } | |
1597 | } | |
1598 | } | |
1599 | ||
1da12ec4 LT |
1600 | static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn) |
1601 | { | |
1602 | return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL); | |
1603 | } | |
1604 | ||
1605 | static const bool vtd_qualified_faults[] = { | |
1606 | [VTD_FR_RESERVED] = false, | |
1607 | [VTD_FR_ROOT_ENTRY_P] = false, | |
1608 | [VTD_FR_CONTEXT_ENTRY_P] = true, | |
1609 | [VTD_FR_CONTEXT_ENTRY_INV] = true, | |
1610 | [VTD_FR_ADDR_BEYOND_MGAW] = true, | |
1611 | [VTD_FR_WRITE] = true, | |
1612 | [VTD_FR_READ] = true, | |
1613 | [VTD_FR_PAGING_ENTRY_INV] = true, | |
1614 | [VTD_FR_ROOT_TABLE_INV] = false, | |
1615 | [VTD_FR_CONTEXT_TABLE_INV] = false, | |
1616 | [VTD_FR_ROOT_ENTRY_RSVD] = false, | |
1617 | [VTD_FR_PAGING_ENTRY_RSVD] = true, | |
1618 | [VTD_FR_CONTEXT_ENTRY_TT] = true, | |
fb43cf73 | 1619 | [VTD_FR_PASID_TABLE_INV] = false, |
1da12ec4 LT |
1620 | [VTD_FR_RESERVED_ERR] = false, |
1621 | [VTD_FR_MAX] = false, | |
1622 | }; | |
1623 | ||
1624 | /* To see if a fault condition is "qualified", which is reported to software | |
1625 | * only if the FPD field in the context-entry used to process the faulting | |
1626 | * request is 0. | |
1627 | */ | |
1628 | static inline bool vtd_is_qualified_fault(VTDFaultReason fault) | |
1629 | { | |
1630 | return vtd_qualified_faults[fault]; | |
1631 | } | |
1632 | ||
1633 | static inline bool vtd_is_interrupt_addr(hwaddr addr) | |
1634 | { | |
1635 | return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST; | |
1636 | } | |
1637 | ||
dbaabb25 PX |
1638 | static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id) |
1639 | { | |
1640 | VTDBus *vtd_bus; | |
1641 | VTDAddressSpace *vtd_as; | |
1642 | bool success = false; | |
1643 | ||
1644 | vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id)); | |
1645 | if (!vtd_bus) { | |
1646 | goto out; | |
1647 | } | |
1648 | ||
1649 | vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)]; | |
1650 | if (!vtd_as) { | |
1651 | goto out; | |
1652 | } | |
1653 | ||
1654 | if (vtd_switch_address_space(vtd_as) == false) { | |
1655 | /* We switched off IOMMU region successfully. */ | |
1656 | success = true; | |
1657 | } | |
1658 | ||
1659 | out: | |
1660 | trace_vtd_pt_enable_fast_path(source_id, success); | |
1661 | } | |
1662 | ||
1da12ec4 LT |
1663 | /* Map dev to context-entry then do a paging-structures walk to do a iommu |
1664 | * translation. | |
79e2b9ae PB |
1665 | * |
1666 | * Called from RCU critical section. | |
1667 | * | |
1da12ec4 LT |
1668 | * @bus_num: The bus number |
1669 | * @devfn: The devfn, which is the combined of device and function number | |
1670 | * @is_write: The access is a write operation | |
1671 | * @entry: IOMMUTLBEntry that contain the addr to be translated and result | |
b9313021 PX |
1672 | * |
1673 | * Returns true if translation is successful, otherwise false. | |
1da12ec4 | 1674 | */ |
b9313021 | 1675 | static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus, |
1da12ec4 LT |
1676 | uint8_t devfn, hwaddr addr, bool is_write, |
1677 | IOMMUTLBEntry *entry) | |
1678 | { | |
d92fa2dc | 1679 | IntelIOMMUState *s = vtd_as->iommu_state; |
1da12ec4 | 1680 | VTDContextEntry ce; |
7df953bd | 1681 | uint8_t bus_num = pci_bus_num(bus); |
1d9efa73 | 1682 | VTDContextCacheEntry *cc_entry; |
d66b969b | 1683 | uint64_t slpte, page_mask; |
1da12ec4 LT |
1684 | uint32_t level; |
1685 | uint16_t source_id = vtd_make_source_id(bus_num, devfn); | |
1686 | int ret_fr; | |
1687 | bool is_fpd_set = false; | |
1688 | bool reads = true; | |
1689 | bool writes = true; | |
07f7b733 | 1690 | uint8_t access_flags; |
b5a280c0 | 1691 | VTDIOTLBEntry *iotlb_entry; |
1da12ec4 | 1692 | |
046ab7e9 PX |
1693 | /* |
1694 | * We have standalone memory region for interrupt addresses, we | |
1695 | * should never receive translation requests in this region. | |
1696 | */ | |
1697 | assert(!vtd_is_interrupt_addr(addr)); | |
1698 | ||
1d9efa73 PX |
1699 | vtd_iommu_lock(s); |
1700 | ||
1701 | cc_entry = &vtd_as->context_cache_entry; | |
1702 | ||
b5a280c0 LT |
1703 | /* Try to fetch slpte form IOTLB */ |
1704 | iotlb_entry = vtd_lookup_iotlb(s, source_id, addr); | |
1705 | if (iotlb_entry) { | |
6c441e1d PX |
1706 | trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte, |
1707 | iotlb_entry->domain_id); | |
b5a280c0 | 1708 | slpte = iotlb_entry->slpte; |
07f7b733 | 1709 | access_flags = iotlb_entry->access_flags; |
d66b969b | 1710 | page_mask = iotlb_entry->mask; |
b5a280c0 LT |
1711 | goto out; |
1712 | } | |
b9313021 | 1713 | |
d92fa2dc LT |
1714 | /* Try to fetch context-entry from cache first */ |
1715 | if (cc_entry->context_cache_gen == s->context_cache_gen) { | |
6c441e1d PX |
1716 | trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi, |
1717 | cc_entry->context_entry.lo, | |
1718 | cc_entry->context_cache_gen); | |
d92fa2dc LT |
1719 | ce = cc_entry->context_entry; |
1720 | is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD; | |
fb43cf73 LY |
1721 | if (!is_fpd_set && s->root_scalable) { |
1722 | ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set); | |
1723 | VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write); | |
1724 | } | |
d92fa2dc LT |
1725 | } else { |
1726 | ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce); | |
1727 | is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD; | |
fb43cf73 LY |
1728 | if (!ret_fr && !is_fpd_set && s->root_scalable) { |
1729 | ret_fr = vtd_ce_get_pasid_fpd(s, &ce, &is_fpd_set); | |
1da12ec4 | 1730 | } |
fb43cf73 | 1731 | VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write); |
d92fa2dc | 1732 | /* Update context-cache */ |
6c441e1d PX |
1733 | trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo, |
1734 | cc_entry->context_cache_gen, | |
1735 | s->context_cache_gen); | |
d92fa2dc LT |
1736 | cc_entry->context_entry = ce; |
1737 | cc_entry->context_cache_gen = s->context_cache_gen; | |
1da12ec4 LT |
1738 | } |
1739 | ||
dbaabb25 PX |
1740 | /* |
1741 | * We don't need to translate for pass-through context entries. | |
1742 | * Also, let's ignore IOTLB caching as well for PT devices. | |
1743 | */ | |
1744 | if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) { | |
892721d9 | 1745 | entry->iova = addr & VTD_PAGE_MASK_4K; |
dbaabb25 | 1746 | entry->translated_addr = entry->iova; |
892721d9 | 1747 | entry->addr_mask = ~VTD_PAGE_MASK_4K; |
dbaabb25 PX |
1748 | entry->perm = IOMMU_RW; |
1749 | trace_vtd_translate_pt(source_id, entry->iova); | |
1750 | ||
1751 | /* | |
1752 | * When this happens, it means firstly caching-mode is not | |
1753 | * enabled, and this is the first passthrough translation for | |
1754 | * the device. Let's enable the fast path for passthrough. | |
1755 | * | |
1756 | * When passthrough is disabled again for the device, we can | |
1757 | * capture it via the context entry invalidation, then the | |
1758 | * IOMMU region can be swapped back. | |
1759 | */ | |
1760 | vtd_pt_enable_fast_path(s, source_id); | |
1d9efa73 | 1761 | vtd_iommu_unlock(s); |
b9313021 | 1762 | return true; |
dbaabb25 PX |
1763 | } |
1764 | ||
fb43cf73 | 1765 | ret_fr = vtd_iova_to_slpte(s, &ce, addr, is_write, &slpte, &level, |
37f51384 | 1766 | &reads, &writes, s->aw_bits); |
fb43cf73 | 1767 | VTD_PE_GET_FPD_ERR(ret_fr, is_fpd_set, s, source_id, addr, is_write); |
1da12ec4 | 1768 | |
d66b969b | 1769 | page_mask = vtd_slpt_level_page_mask(level); |
07f7b733 | 1770 | access_flags = IOMMU_ACCESS_FLAG(reads, writes); |
fb43cf73 | 1771 | vtd_update_iotlb(s, source_id, vtd_get_domain_id(s, &ce), addr, slpte, |
07f7b733 | 1772 | access_flags, level); |
b5a280c0 | 1773 | out: |
1d9efa73 | 1774 | vtd_iommu_unlock(s); |
d66b969b | 1775 | entry->iova = addr & page_mask; |
37f51384 | 1776 | entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask; |
d66b969b | 1777 | entry->addr_mask = ~page_mask; |
07f7b733 | 1778 | entry->perm = access_flags; |
b9313021 PX |
1779 | return true; |
1780 | ||
1781 | error: | |
1d9efa73 | 1782 | vtd_iommu_unlock(s); |
b9313021 PX |
1783 | entry->iova = 0; |
1784 | entry->translated_addr = 0; | |
1785 | entry->addr_mask = 0; | |
1786 | entry->perm = IOMMU_NONE; | |
1787 | return false; | |
1da12ec4 LT |
1788 | } |
1789 | ||
1790 | static void vtd_root_table_setup(IntelIOMMUState *s) | |
1791 | { | |
1792 | s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG); | |
37f51384 | 1793 | s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits); |
1da12ec4 | 1794 | |
2811af3b PX |
1795 | vtd_update_scalable_state(s); |
1796 | ||
81fb1e64 | 1797 | trace_vtd_reg_dmar_root(s->root, s->root_scalable); |
1da12ec4 LT |
1798 | } |
1799 | ||
02a2cbc8 PX |
1800 | static void vtd_iec_notify_all(IntelIOMMUState *s, bool global, |
1801 | uint32_t index, uint32_t mask) | |
1802 | { | |
1803 | x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask); | |
1804 | } | |
1805 | ||
a5861439 PX |
1806 | static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s) |
1807 | { | |
1808 | uint64_t value = 0; | |
1809 | value = vtd_get_quad_raw(s, DMAR_IRTA_REG); | |
1810 | s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1); | |
37f51384 | 1811 | s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits); |
28589311 | 1812 | s->intr_eime = value & VTD_IRTA_EIME; |
a5861439 | 1813 | |
02a2cbc8 PX |
1814 | /* Notify global invalidation */ |
1815 | vtd_iec_notify_all(s, true, 0, 0); | |
a5861439 | 1816 | |
7feb51b7 | 1817 | trace_vtd_reg_ir_root(s->intr_root, s->intr_size); |
a5861439 PX |
1818 | } |
1819 | ||
dd4d607e PX |
1820 | static void vtd_iommu_replay_all(IntelIOMMUState *s) |
1821 | { | |
b4a4ba0d | 1822 | VTDAddressSpace *vtd_as; |
dd4d607e | 1823 | |
b4a4ba0d | 1824 | QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) { |
63b88968 | 1825 | vtd_sync_shadow_page_table(vtd_as); |
dd4d607e PX |
1826 | } |
1827 | } | |
1828 | ||
d92fa2dc LT |
1829 | static void vtd_context_global_invalidate(IntelIOMMUState *s) |
1830 | { | |
bc535e59 | 1831 | trace_vtd_inv_desc_cc_global(); |
1d9efa73 PX |
1832 | /* Protects context cache */ |
1833 | vtd_iommu_lock(s); | |
d92fa2dc LT |
1834 | s->context_cache_gen++; |
1835 | if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) { | |
1d9efa73 | 1836 | vtd_reset_context_cache_locked(s); |
d92fa2dc | 1837 | } |
1d9efa73 | 1838 | vtd_iommu_unlock(s); |
2cc9ddcc | 1839 | vtd_address_space_refresh_all(s); |
dd4d607e PX |
1840 | /* |
1841 | * From VT-d spec 6.5.2.1, a global context entry invalidation | |
1842 | * should be followed by a IOTLB global invalidation, so we should | |
1843 | * be safe even without this. Hoewever, let's replay the region as | |
1844 | * well to be safer, and go back here when we need finer tunes for | |
1845 | * VT-d emulation codes. | |
1846 | */ | |
1847 | vtd_iommu_replay_all(s); | |
d92fa2dc LT |
1848 | } |
1849 | ||
1850 | /* Do a context-cache device-selective invalidation. | |
1851 | * @func_mask: FM field after shifting | |
1852 | */ | |
1853 | static void vtd_context_device_invalidate(IntelIOMMUState *s, | |
1854 | uint16_t source_id, | |
1855 | uint16_t func_mask) | |
1856 | { | |
1857 | uint16_t mask; | |
7df953bd | 1858 | VTDBus *vtd_bus; |
d92fa2dc | 1859 | VTDAddressSpace *vtd_as; |
bc535e59 | 1860 | uint8_t bus_n, devfn; |
d92fa2dc LT |
1861 | uint16_t devfn_it; |
1862 | ||
bc535e59 PX |
1863 | trace_vtd_inv_desc_cc_devices(source_id, func_mask); |
1864 | ||
d92fa2dc LT |
1865 | switch (func_mask & 3) { |
1866 | case 0: | |
1867 | mask = 0; /* No bits in the SID field masked */ | |
1868 | break; | |
1869 | case 1: | |
1870 | mask = 4; /* Mask bit 2 in the SID field */ | |
1871 | break; | |
1872 | case 2: | |
1873 | mask = 6; /* Mask bit 2:1 in the SID field */ | |
1874 | break; | |
1875 | case 3: | |
1876 | mask = 7; /* Mask bit 2:0 in the SID field */ | |
1877 | break; | |
1878 | } | |
6cb99acc | 1879 | mask = ~mask; |
bc535e59 PX |
1880 | |
1881 | bus_n = VTD_SID_TO_BUS(source_id); | |
1882 | vtd_bus = vtd_find_as_from_bus_num(s, bus_n); | |
7df953bd | 1883 | if (vtd_bus) { |
d92fa2dc | 1884 | devfn = VTD_SID_TO_DEVFN(source_id); |
bf33cc75 | 1885 | for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) { |
7df953bd | 1886 | vtd_as = vtd_bus->dev_as[devfn_it]; |
d92fa2dc | 1887 | if (vtd_as && ((devfn_it & mask) == (devfn & mask))) { |
bc535e59 PX |
1888 | trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it), |
1889 | VTD_PCI_FUNC(devfn_it)); | |
1d9efa73 | 1890 | vtd_iommu_lock(s); |
d92fa2dc | 1891 | vtd_as->context_cache_entry.context_cache_gen = 0; |
1d9efa73 | 1892 | vtd_iommu_unlock(s); |
dbaabb25 PX |
1893 | /* |
1894 | * Do switch address space when needed, in case if the | |
1895 | * device passthrough bit is switched. | |
1896 | */ | |
1897 | vtd_switch_address_space(vtd_as); | |
dd4d607e PX |
1898 | /* |
1899 | * So a device is moving out of (or moving into) a | |
63b88968 | 1900 | * domain, resync the shadow page table. |
dd4d607e PX |
1901 | * This won't bring bad even if we have no such |
1902 | * notifier registered - the IOMMU notification | |
1903 | * framework will skip MAP notifications if that | |
1904 | * happened. | |
1905 | */ | |
63b88968 | 1906 | vtd_sync_shadow_page_table(vtd_as); |
d92fa2dc LT |
1907 | } |
1908 | } | |
1909 | } | |
1910 | } | |
1911 | ||
1da12ec4 LT |
1912 | /* Context-cache invalidation |
1913 | * Returns the Context Actual Invalidation Granularity. | |
1914 | * @val: the content of the CCMD_REG | |
1915 | */ | |
1916 | static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val) | |
1917 | { | |
1918 | uint64_t caig; | |
1919 | uint64_t type = val & VTD_CCMD_CIRG_MASK; | |
1920 | ||
1921 | switch (type) { | |
d92fa2dc | 1922 | case VTD_CCMD_DOMAIN_INVL: |
d92fa2dc | 1923 | /* Fall through */ |
1da12ec4 | 1924 | case VTD_CCMD_GLOBAL_INVL: |
1da12ec4 | 1925 | caig = VTD_CCMD_GLOBAL_INVL_A; |
d92fa2dc | 1926 | vtd_context_global_invalidate(s); |
1da12ec4 LT |
1927 | break; |
1928 | ||
1929 | case VTD_CCMD_DEVICE_INVL: | |
1da12ec4 | 1930 | caig = VTD_CCMD_DEVICE_INVL_A; |
d92fa2dc | 1931 | vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val)); |
1da12ec4 LT |
1932 | break; |
1933 | ||
1934 | default: | |
1376211f PX |
1935 | error_report_once("%s: invalid context: 0x%" PRIx64, |
1936 | __func__, val); | |
1da12ec4 LT |
1937 | caig = 0; |
1938 | } | |
1939 | return caig; | |
1940 | } | |
1941 | ||
b5a280c0 LT |
1942 | static void vtd_iotlb_global_invalidate(IntelIOMMUState *s) |
1943 | { | |
7feb51b7 | 1944 | trace_vtd_inv_desc_iotlb_global(); |
b5a280c0 | 1945 | vtd_reset_iotlb(s); |
dd4d607e | 1946 | vtd_iommu_replay_all(s); |
b5a280c0 LT |
1947 | } |
1948 | ||
1949 | static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id) | |
1950 | { | |
dd4d607e PX |
1951 | VTDContextEntry ce; |
1952 | VTDAddressSpace *vtd_as; | |
1953 | ||
7feb51b7 PX |
1954 | trace_vtd_inv_desc_iotlb_domain(domain_id); |
1955 | ||
1d9efa73 | 1956 | vtd_iommu_lock(s); |
b5a280c0 LT |
1957 | g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain, |
1958 | &domain_id); | |
1d9efa73 | 1959 | vtd_iommu_unlock(s); |
dd4d607e | 1960 | |
b4a4ba0d | 1961 | QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) { |
dd4d607e PX |
1962 | if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus), |
1963 | vtd_as->devfn, &ce) && | |
fb43cf73 | 1964 | domain_id == vtd_get_domain_id(s, &ce)) { |
63b88968 | 1965 | vtd_sync_shadow_page_table(vtd_as); |
dd4d607e PX |
1966 | } |
1967 | } | |
1968 | } | |
1969 | ||
dd4d607e PX |
1970 | static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s, |
1971 | uint16_t domain_id, hwaddr addr, | |
1972 | uint8_t am) | |
1973 | { | |
b4a4ba0d | 1974 | VTDAddressSpace *vtd_as; |
dd4d607e PX |
1975 | VTDContextEntry ce; |
1976 | int ret; | |
4f8a62a9 | 1977 | hwaddr size = (1 << am) * VTD_PAGE_SIZE; |
dd4d607e | 1978 | |
b4a4ba0d | 1979 | QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) { |
dd4d607e PX |
1980 | ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus), |
1981 | vtd_as->devfn, &ce); | |
fb43cf73 | 1982 | if (!ret && domain_id == vtd_get_domain_id(s, &ce)) { |
4f8a62a9 PX |
1983 | if (vtd_as_has_map_notifier(vtd_as)) { |
1984 | /* | |
1985 | * As long as we have MAP notifications registered in | |
1986 | * any of our IOMMU notifiers, we need to sync the | |
1987 | * shadow page table. | |
1988 | */ | |
63b88968 | 1989 | vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size); |
4f8a62a9 PX |
1990 | } else { |
1991 | /* | |
1992 | * For UNMAP-only notifiers, we don't need to walk the | |
1993 | * page tables. We just deliver the PSI down to | |
1994 | * invalidate caches. | |
1995 | */ | |
1996 | IOMMUTLBEntry entry = { | |
1997 | .target_as = &address_space_memory, | |
1998 | .iova = addr, | |
1999 | .translated_addr = 0, | |
2000 | .addr_mask = size - 1, | |
2001 | .perm = IOMMU_NONE, | |
2002 | }; | |
cb1efcf4 | 2003 | memory_region_notify_iommu(&vtd_as->iommu, 0, entry); |
4f8a62a9 | 2004 | } |
dd4d607e PX |
2005 | } |
2006 | } | |
b5a280c0 LT |
2007 | } |
2008 | ||
2009 | static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id, | |
2010 | hwaddr addr, uint8_t am) | |
2011 | { | |
2012 | VTDIOTLBPageInvInfo info; | |
2013 | ||
7feb51b7 PX |
2014 | trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am); |
2015 | ||
b5a280c0 LT |
2016 | assert(am <= VTD_MAMV); |
2017 | info.domain_id = domain_id; | |
d66b969b | 2018 | info.addr = addr; |
b5a280c0 | 2019 | info.mask = ~((1 << am) - 1); |
1d9efa73 | 2020 | vtd_iommu_lock(s); |
b5a280c0 | 2021 | g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info); |
1d9efa73 | 2022 | vtd_iommu_unlock(s); |
dd4d607e | 2023 | vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am); |
b5a280c0 LT |
2024 | } |
2025 | ||
1da12ec4 LT |
2026 | /* Flush IOTLB |
2027 | * Returns the IOTLB Actual Invalidation Granularity. | |
2028 | * @val: the content of the IOTLB_REG | |
2029 | */ | |
2030 | static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val) | |
2031 | { | |
2032 | uint64_t iaig; | |
2033 | uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK; | |
b5a280c0 LT |
2034 | uint16_t domain_id; |
2035 | hwaddr addr; | |
2036 | uint8_t am; | |
1da12ec4 LT |
2037 | |
2038 | switch (type) { | |
2039 | case VTD_TLB_GLOBAL_FLUSH: | |
1da12ec4 | 2040 | iaig = VTD_TLB_GLOBAL_FLUSH_A; |
b5a280c0 | 2041 | vtd_iotlb_global_invalidate(s); |
1da12ec4 LT |
2042 | break; |
2043 | ||
2044 | case VTD_TLB_DSI_FLUSH: | |
b5a280c0 | 2045 | domain_id = VTD_TLB_DID(val); |
1da12ec4 | 2046 | iaig = VTD_TLB_DSI_FLUSH_A; |
b5a280c0 | 2047 | vtd_iotlb_domain_invalidate(s, domain_id); |
1da12ec4 LT |
2048 | break; |
2049 | ||
2050 | case VTD_TLB_PSI_FLUSH: | |
b5a280c0 LT |
2051 | domain_id = VTD_TLB_DID(val); |
2052 | addr = vtd_get_quad_raw(s, DMAR_IVA_REG); | |
2053 | am = VTD_IVA_AM(addr); | |
2054 | addr = VTD_IVA_ADDR(addr); | |
b5a280c0 | 2055 | if (am > VTD_MAMV) { |
1376211f PX |
2056 | error_report_once("%s: address mask overflow: 0x%" PRIx64, |
2057 | __func__, vtd_get_quad_raw(s, DMAR_IVA_REG)); | |
b5a280c0 LT |
2058 | iaig = 0; |
2059 | break; | |
2060 | } | |
1da12ec4 | 2061 | iaig = VTD_TLB_PSI_FLUSH_A; |
b5a280c0 | 2062 | vtd_iotlb_page_invalidate(s, domain_id, addr, am); |
1da12ec4 LT |
2063 | break; |
2064 | ||
2065 | default: | |
1376211f PX |
2066 | error_report_once("%s: invalid granularity: 0x%" PRIx64, |
2067 | __func__, val); | |
1da12ec4 LT |
2068 | iaig = 0; |
2069 | } | |
2070 | return iaig; | |
2071 | } | |
2072 | ||
8991c460 | 2073 | static void vtd_fetch_inv_desc(IntelIOMMUState *s); |
ed7b8fbc LT |
2074 | |
2075 | static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s) | |
2076 | { | |
2077 | return s->qi_enabled && (s->iq_tail == s->iq_head) && | |
2078 | (s->iq_last_desc_type == VTD_INV_DESC_WAIT); | |
2079 | } | |
2080 | ||
2081 | static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en) | |
2082 | { | |
2083 | uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG); | |
2084 | ||
7feb51b7 PX |
2085 | trace_vtd_inv_qi_enable(en); |
2086 | ||
ed7b8fbc | 2087 | if (en) { |
37f51384 | 2088 | s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits); |
8991c460 | 2089 | /* 2^(x+8) entries */ |
c0c1d351 | 2090 | s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8 - (s->iq_dw ? 1 : 0)); |
8991c460 LP |
2091 | s->qi_enabled = true; |
2092 | trace_vtd_inv_qi_setup(s->iq, s->iq_size); | |
2093 | /* Ok - report back to driver */ | |
2094 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES); | |
2095 | ||
2096 | if (s->iq_tail != 0) { | |
2097 | /* | |
2098 | * This is a spec violation but Windows guests are known to set up | |
2099 | * Queued Invalidation this way so we allow the write and process | |
2100 | * Invalidation Descriptors right away. | |
2101 | */ | |
2102 | trace_vtd_warn_invalid_qi_tail(s->iq_tail); | |
2103 | if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) { | |
2104 | vtd_fetch_inv_desc(s); | |
2105 | } | |
ed7b8fbc LT |
2106 | } |
2107 | } else { | |
2108 | if (vtd_queued_inv_disable_check(s)) { | |
2109 | /* disable Queued Invalidation */ | |
2110 | vtd_set_quad_raw(s, DMAR_IQH_REG, 0); | |
2111 | s->iq_head = 0; | |
2112 | s->qi_enabled = false; | |
2113 | /* Ok - report back to driver */ | |
2114 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0); | |
2115 | } else { | |
4e4abd11 PX |
2116 | error_report_once("%s: detected improper state when disable QI " |
2117 | "(head=0x%x, tail=0x%x, last_type=%d)", | |
2118 | __func__, | |
2119 | s->iq_head, s->iq_tail, s->iq_last_desc_type); | |
ed7b8fbc LT |
2120 | } |
2121 | } | |
2122 | } | |
2123 | ||
1da12ec4 LT |
2124 | /* Set Root Table Pointer */ |
2125 | static void vtd_handle_gcmd_srtp(IntelIOMMUState *s) | |
2126 | { | |
1da12ec4 LT |
2127 | vtd_root_table_setup(s); |
2128 | /* Ok - report back to driver */ | |
2129 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS); | |
2cc9ddcc PX |
2130 | vtd_reset_caches(s); |
2131 | vtd_address_space_refresh_all(s); | |
1da12ec4 LT |
2132 | } |
2133 | ||
a5861439 PX |
2134 | /* Set Interrupt Remap Table Pointer */ |
2135 | static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s) | |
2136 | { | |
a5861439 PX |
2137 | vtd_interrupt_remap_table_setup(s); |
2138 | /* Ok - report back to driver */ | |
2139 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS); | |
2140 | } | |
2141 | ||
1da12ec4 LT |
2142 | /* Handle Translation Enable/Disable */ |
2143 | static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en) | |
2144 | { | |
558e0024 PX |
2145 | if (s->dmar_enabled == en) { |
2146 | return; | |
2147 | } | |
2148 | ||
7feb51b7 | 2149 | trace_vtd_dmar_enable(en); |
1da12ec4 LT |
2150 | |
2151 | if (en) { | |
2152 | s->dmar_enabled = true; | |
2153 | /* Ok - report back to driver */ | |
2154 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES); | |
2155 | } else { | |
2156 | s->dmar_enabled = false; | |
2157 | ||
2158 | /* Clear the index of Fault Recording Register */ | |
2159 | s->next_frcd_reg = 0; | |
2160 | /* Ok - report back to driver */ | |
2161 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0); | |
2162 | } | |
558e0024 | 2163 | |
2cc9ddcc PX |
2164 | vtd_reset_caches(s); |
2165 | vtd_address_space_refresh_all(s); | |
1da12ec4 LT |
2166 | } |
2167 | ||
80de52ba PX |
2168 | /* Handle Interrupt Remap Enable/Disable */ |
2169 | static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en) | |
2170 | { | |
7feb51b7 | 2171 | trace_vtd_ir_enable(en); |
80de52ba PX |
2172 | |
2173 | if (en) { | |
2174 | s->intr_enabled = true; | |
2175 | /* Ok - report back to driver */ | |
2176 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES); | |
2177 | } else { | |
2178 | s->intr_enabled = false; | |
2179 | /* Ok - report back to driver */ | |
2180 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0); | |
2181 | } | |
2182 | } | |
2183 | ||
1da12ec4 LT |
2184 | /* Handle write to Global Command Register */ |
2185 | static void vtd_handle_gcmd_write(IntelIOMMUState *s) | |
2186 | { | |
2187 | uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG); | |
2188 | uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG); | |
2189 | uint32_t changed = status ^ val; | |
2190 | ||
7feb51b7 | 2191 | trace_vtd_reg_write_gcmd(status, val); |
1da12ec4 LT |
2192 | if (changed & VTD_GCMD_TE) { |
2193 | /* Translation enable/disable */ | |
2194 | vtd_handle_gcmd_te(s, val & VTD_GCMD_TE); | |
2195 | } | |
2196 | if (val & VTD_GCMD_SRTP) { | |
2197 | /* Set/update the root-table pointer */ | |
2198 | vtd_handle_gcmd_srtp(s); | |
2199 | } | |
ed7b8fbc LT |
2200 | if (changed & VTD_GCMD_QIE) { |
2201 | /* Queued Invalidation Enable */ | |
2202 | vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE); | |
2203 | } | |
a5861439 PX |
2204 | if (val & VTD_GCMD_SIRTP) { |
2205 | /* Set/update the interrupt remapping root-table pointer */ | |
2206 | vtd_handle_gcmd_sirtp(s); | |
2207 | } | |
80de52ba PX |
2208 | if (changed & VTD_GCMD_IRE) { |
2209 | /* Interrupt remap enable/disable */ | |
2210 | vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE); | |
2211 | } | |
1da12ec4 LT |
2212 | } |
2213 | ||
2214 | /* Handle write to Context Command Register */ | |
2215 | static void vtd_handle_ccmd_write(IntelIOMMUState *s) | |
2216 | { | |
2217 | uint64_t ret; | |
2218 | uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG); | |
2219 | ||
2220 | /* Context-cache invalidation request */ | |
2221 | if (val & VTD_CCMD_ICC) { | |
ed7b8fbc | 2222 | if (s->qi_enabled) { |
1376211f PX |
2223 | error_report_once("Queued Invalidation enabled, " |
2224 | "should not use register-based invalidation"); | |
ed7b8fbc LT |
2225 | return; |
2226 | } | |
1da12ec4 LT |
2227 | ret = vtd_context_cache_invalidate(s, val); |
2228 | /* Invalidation completed. Change something to show */ | |
2229 | vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL); | |
2230 | ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK, | |
2231 | ret); | |
1da12ec4 LT |
2232 | } |
2233 | } | |
2234 | ||
2235 | /* Handle write to IOTLB Invalidation Register */ | |
2236 | static void vtd_handle_iotlb_write(IntelIOMMUState *s) | |
2237 | { | |
2238 | uint64_t ret; | |
2239 | uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG); | |
2240 | ||
2241 | /* IOTLB invalidation request */ | |
2242 | if (val & VTD_TLB_IVT) { | |
ed7b8fbc | 2243 | if (s->qi_enabled) { |
1376211f PX |
2244 | error_report_once("Queued Invalidation enabled, " |
2245 | "should not use register-based invalidation"); | |
ed7b8fbc LT |
2246 | return; |
2247 | } | |
1da12ec4 LT |
2248 | ret = vtd_iotlb_flush(s, val); |
2249 | /* Invalidation completed. Change something to show */ | |
2250 | vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL); | |
2251 | ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, | |
2252 | VTD_TLB_FLUSH_GRANU_MASK_A, ret); | |
1da12ec4 LT |
2253 | } |
2254 | } | |
2255 | ||
ed7b8fbc | 2256 | /* Fetch an Invalidation Descriptor from the Invalidation Queue */ |
c0c1d351 | 2257 | static bool vtd_get_inv_desc(IntelIOMMUState *s, |
ed7b8fbc LT |
2258 | VTDInvDesc *inv_desc) |
2259 | { | |
c0c1d351 LY |
2260 | dma_addr_t base_addr = s->iq; |
2261 | uint32_t offset = s->iq_head; | |
2262 | uint32_t dw = s->iq_dw ? 32 : 16; | |
2263 | dma_addr_t addr = base_addr + offset * dw; | |
2264 | ||
2265 | if (dma_memory_read(&address_space_memory, addr, inv_desc, dw)) { | |
2266 | error_report_once("Read INV DESC failed."); | |
ed7b8fbc LT |
2267 | return false; |
2268 | } | |
2269 | inv_desc->lo = le64_to_cpu(inv_desc->lo); | |
2270 | inv_desc->hi = le64_to_cpu(inv_desc->hi); | |
c0c1d351 LY |
2271 | if (dw == 32) { |
2272 | inv_desc->val[2] = le64_to_cpu(inv_desc->val[2]); | |
2273 | inv_desc->val[3] = le64_to_cpu(inv_desc->val[3]); | |
2274 | } | |
ed7b8fbc LT |
2275 | return true; |
2276 | } | |
2277 | ||
2278 | static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc) | |
2279 | { | |
2280 | if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) || | |
2281 | (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) { | |
095955b2 PX |
2282 | error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64 |
2283 | " (reserved nonzero)", __func__, inv_desc->hi, | |
2284 | inv_desc->lo); | |
ed7b8fbc LT |
2285 | return false; |
2286 | } | |
2287 | if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) { | |
2288 | /* Status Write */ | |
2289 | uint32_t status_data = (uint32_t)(inv_desc->lo >> | |
2290 | VTD_INV_DESC_WAIT_DATA_SHIFT); | |
2291 | ||
2292 | assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF)); | |
2293 | ||
2294 | /* FIXME: need to be masked with HAW? */ | |
2295 | dma_addr_t status_addr = inv_desc->hi; | |
bc535e59 | 2296 | trace_vtd_inv_desc_wait_sw(status_addr, status_data); |
ed7b8fbc LT |
2297 | status_data = cpu_to_le32(status_data); |
2298 | if (dma_memory_write(&address_space_memory, status_addr, &status_data, | |
2299 | sizeof(status_data))) { | |
bc535e59 | 2300 | trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo); |
ed7b8fbc LT |
2301 | return false; |
2302 | } | |
2303 | } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) { | |
2304 | /* Interrupt flag */ | |
ed7b8fbc LT |
2305 | vtd_generate_completion_event(s); |
2306 | } else { | |
095955b2 PX |
2307 | error_report_once("%s: invalid wait desc: hi=%"PRIx64", lo=%"PRIx64 |
2308 | " (unknown type)", __func__, inv_desc->hi, | |
2309 | inv_desc->lo); | |
ed7b8fbc LT |
2310 | return false; |
2311 | } | |
2312 | return true; | |
2313 | } | |
2314 | ||
d92fa2dc LT |
2315 | static bool vtd_process_context_cache_desc(IntelIOMMUState *s, |
2316 | VTDInvDesc *inv_desc) | |
2317 | { | |
bc535e59 PX |
2318 | uint16_t sid, fmask; |
2319 | ||
d92fa2dc | 2320 | if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) { |
095955b2 PX |
2321 | error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64 |
2322 | " (reserved nonzero)", __func__, inv_desc->hi, | |
2323 | inv_desc->lo); | |
d92fa2dc LT |
2324 | return false; |
2325 | } | |
2326 | switch (inv_desc->lo & VTD_INV_DESC_CC_G) { | |
2327 | case VTD_INV_DESC_CC_DOMAIN: | |
bc535e59 PX |
2328 | trace_vtd_inv_desc_cc_domain( |
2329 | (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo)); | |
d92fa2dc LT |
2330 | /* Fall through */ |
2331 | case VTD_INV_DESC_CC_GLOBAL: | |
d92fa2dc LT |
2332 | vtd_context_global_invalidate(s); |
2333 | break; | |
2334 | ||
2335 | case VTD_INV_DESC_CC_DEVICE: | |
bc535e59 PX |
2336 | sid = VTD_INV_DESC_CC_SID(inv_desc->lo); |
2337 | fmask = VTD_INV_DESC_CC_FM(inv_desc->lo); | |
2338 | vtd_context_device_invalidate(s, sid, fmask); | |
d92fa2dc LT |
2339 | break; |
2340 | ||
2341 | default: | |
095955b2 PX |
2342 | error_report_once("%s: invalid cc inv desc: hi=%"PRIx64", lo=%"PRIx64 |
2343 | " (invalid type)", __func__, inv_desc->hi, | |
2344 | inv_desc->lo); | |
d92fa2dc LT |
2345 | return false; |
2346 | } | |
2347 | return true; | |
2348 | } | |
2349 | ||
b5a280c0 LT |
2350 | static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc) |
2351 | { | |
2352 | uint16_t domain_id; | |
2353 | uint8_t am; | |
2354 | hwaddr addr; | |
2355 | ||
2356 | if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) || | |
2357 | (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) { | |
095955b2 PX |
2358 | error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64 |
2359 | ", lo=0x%"PRIx64" (reserved bits unzero)\n", | |
2360 | __func__, inv_desc->hi, inv_desc->lo); | |
b5a280c0 LT |
2361 | return false; |
2362 | } | |
2363 | ||
2364 | switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) { | |
2365 | case VTD_INV_DESC_IOTLB_GLOBAL: | |
b5a280c0 LT |
2366 | vtd_iotlb_global_invalidate(s); |
2367 | break; | |
2368 | ||
2369 | case VTD_INV_DESC_IOTLB_DOMAIN: | |
2370 | domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo); | |
b5a280c0 LT |
2371 | vtd_iotlb_domain_invalidate(s, domain_id); |
2372 | break; | |
2373 | ||
2374 | case VTD_INV_DESC_IOTLB_PAGE: | |
2375 | domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo); | |
2376 | addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi); | |
2377 | am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi); | |
b5a280c0 | 2378 | if (am > VTD_MAMV) { |
095955b2 PX |
2379 | error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64 |
2380 | ", lo=0x%"PRIx64" (am=%u > VTD_MAMV=%u)\n", | |
2381 | __func__, inv_desc->hi, inv_desc->lo, | |
2382 | am, (unsigned)VTD_MAMV); | |
b5a280c0 LT |
2383 | return false; |
2384 | } | |
2385 | vtd_iotlb_page_invalidate(s, domain_id, addr, am); | |
2386 | break; | |
2387 | ||
2388 | default: | |
095955b2 PX |
2389 | error_report_once("%s: invalid iotlb inv desc: hi=0x%"PRIx64 |
2390 | ", lo=0x%"PRIx64" (type mismatch: 0x%llx)\n", | |
2391 | __func__, inv_desc->hi, inv_desc->lo, | |
2392 | inv_desc->lo & VTD_INV_DESC_IOTLB_G); | |
b5a280c0 LT |
2393 | return false; |
2394 | } | |
2395 | return true; | |
2396 | } | |
2397 | ||
02a2cbc8 PX |
2398 | static bool vtd_process_inv_iec_desc(IntelIOMMUState *s, |
2399 | VTDInvDesc *inv_desc) | |
2400 | { | |
7feb51b7 PX |
2401 | trace_vtd_inv_desc_iec(inv_desc->iec.granularity, |
2402 | inv_desc->iec.index, | |
2403 | inv_desc->iec.index_mask); | |
02a2cbc8 PX |
2404 | |
2405 | vtd_iec_notify_all(s, !inv_desc->iec.granularity, | |
2406 | inv_desc->iec.index, | |
2407 | inv_desc->iec.index_mask); | |
554f5e16 JW |
2408 | return true; |
2409 | } | |
2410 | ||
2411 | static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s, | |
2412 | VTDInvDesc *inv_desc) | |
2413 | { | |
2414 | VTDAddressSpace *vtd_dev_as; | |
2415 | IOMMUTLBEntry entry; | |
2416 | struct VTDBus *vtd_bus; | |
2417 | hwaddr addr; | |
2418 | uint64_t sz; | |
2419 | uint16_t sid; | |
2420 | uint8_t devfn; | |
2421 | bool size; | |
2422 | uint8_t bus_num; | |
2423 | ||
2424 | addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi); | |
2425 | sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo); | |
2426 | devfn = sid & 0xff; | |
2427 | bus_num = sid >> 8; | |
2428 | size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi); | |
2429 | ||
2430 | if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) || | |
2431 | (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) { | |
095955b2 PX |
2432 | error_report_once("%s: invalid dev-iotlb inv desc: hi=%"PRIx64 |
2433 | ", lo=%"PRIx64" (reserved nonzero)", __func__, | |
2434 | inv_desc->hi, inv_desc->lo); | |
554f5e16 JW |
2435 | return false; |
2436 | } | |
2437 | ||
2438 | vtd_bus = vtd_find_as_from_bus_num(s, bus_num); | |
2439 | if (!vtd_bus) { | |
2440 | goto done; | |
2441 | } | |
2442 | ||
2443 | vtd_dev_as = vtd_bus->dev_as[devfn]; | |
2444 | if (!vtd_dev_as) { | |
2445 | goto done; | |
2446 | } | |
2447 | ||
04eb6247 JW |
2448 | /* According to ATS spec table 2.4: |
2449 | * S = 0, bits 15:12 = xxxx range size: 4K | |
2450 | * S = 1, bits 15:12 = xxx0 range size: 8K | |
2451 | * S = 1, bits 15:12 = xx01 range size: 16K | |
2452 | * S = 1, bits 15:12 = x011 range size: 32K | |
2453 | * S = 1, bits 15:12 = 0111 range size: 64K | |
2454 | * ... | |
2455 | */ | |
554f5e16 | 2456 | if (size) { |
04eb6247 | 2457 | sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT); |
554f5e16 JW |
2458 | addr &= ~(sz - 1); |
2459 | } else { | |
2460 | sz = VTD_PAGE_SIZE; | |
2461 | } | |
02a2cbc8 | 2462 | |
554f5e16 JW |
2463 | entry.target_as = &vtd_dev_as->as; |
2464 | entry.addr_mask = sz - 1; | |
2465 | entry.iova = addr; | |
2466 | entry.perm = IOMMU_NONE; | |
2467 | entry.translated_addr = 0; | |
cb1efcf4 | 2468 | memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry); |
554f5e16 JW |
2469 | |
2470 | done: | |
02a2cbc8 PX |
2471 | return true; |
2472 | } | |
2473 | ||
ed7b8fbc LT |
2474 | static bool vtd_process_inv_desc(IntelIOMMUState *s) |
2475 | { | |
2476 | VTDInvDesc inv_desc; | |
2477 | uint8_t desc_type; | |
2478 | ||
7feb51b7 | 2479 | trace_vtd_inv_qi_head(s->iq_head); |
c0c1d351 | 2480 | if (!vtd_get_inv_desc(s, &inv_desc)) { |
ed7b8fbc LT |
2481 | s->iq_last_desc_type = VTD_INV_DESC_NONE; |
2482 | return false; | |
2483 | } | |
c0c1d351 | 2484 | |
ed7b8fbc LT |
2485 | desc_type = inv_desc.lo & VTD_INV_DESC_TYPE; |
2486 | /* FIXME: should update at first or at last? */ | |
2487 | s->iq_last_desc_type = desc_type; | |
2488 | ||
2489 | switch (desc_type) { | |
2490 | case VTD_INV_DESC_CC: | |
bc535e59 | 2491 | trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo); |
d92fa2dc LT |
2492 | if (!vtd_process_context_cache_desc(s, &inv_desc)) { |
2493 | return false; | |
2494 | } | |
ed7b8fbc LT |
2495 | break; |
2496 | ||
2497 | case VTD_INV_DESC_IOTLB: | |
bc535e59 | 2498 | trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo); |
b5a280c0 LT |
2499 | if (!vtd_process_iotlb_desc(s, &inv_desc)) { |
2500 | return false; | |
2501 | } | |
ed7b8fbc LT |
2502 | break; |
2503 | ||
4a4f219e YS |
2504 | /* |
2505 | * TODO: the entity of below two cases will be implemented in future series. | |
2506 | * To make guest (which integrates scalable mode support patch set in | |
2507 | * iommu driver) work, just return true is enough so far. | |
2508 | */ | |
2509 | case VTD_INV_DESC_PC: | |
2510 | break; | |
2511 | ||
2512 | case VTD_INV_DESC_PIOTLB: | |
2513 | break; | |
2514 | ||
ed7b8fbc | 2515 | case VTD_INV_DESC_WAIT: |
bc535e59 | 2516 | trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo); |
ed7b8fbc LT |
2517 | if (!vtd_process_wait_desc(s, &inv_desc)) { |
2518 | return false; | |
2519 | } | |
2520 | break; | |
2521 | ||
b7910472 | 2522 | case VTD_INV_DESC_IEC: |
bc535e59 | 2523 | trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo); |
02a2cbc8 PX |
2524 | if (!vtd_process_inv_iec_desc(s, &inv_desc)) { |
2525 | return false; | |
2526 | } | |
b7910472 PX |
2527 | break; |
2528 | ||
554f5e16 | 2529 | case VTD_INV_DESC_DEVICE: |
7feb51b7 | 2530 | trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo); |
554f5e16 JW |
2531 | if (!vtd_process_device_iotlb_desc(s, &inv_desc)) { |
2532 | return false; | |
2533 | } | |
2534 | break; | |
2535 | ||
ed7b8fbc | 2536 | default: |
095955b2 PX |
2537 | error_report_once("%s: invalid inv desc: hi=%"PRIx64", lo=%"PRIx64 |
2538 | " (unknown type)", __func__, inv_desc.hi, | |
2539 | inv_desc.lo); | |
ed7b8fbc LT |
2540 | return false; |
2541 | } | |
2542 | s->iq_head++; | |
2543 | if (s->iq_head == s->iq_size) { | |
2544 | s->iq_head = 0; | |
2545 | } | |
2546 | return true; | |
2547 | } | |
2548 | ||
2549 | /* Try to fetch and process more Invalidation Descriptors */ | |
2550 | static void vtd_fetch_inv_desc(IntelIOMMUState *s) | |
2551 | { | |
7feb51b7 PX |
2552 | trace_vtd_inv_qi_fetch(); |
2553 | ||
ed7b8fbc LT |
2554 | if (s->iq_tail >= s->iq_size) { |
2555 | /* Detects an invalid Tail pointer */ | |
4e4abd11 PX |
2556 | error_report_once("%s: detected invalid QI tail " |
2557 | "(tail=0x%x, size=0x%x)", | |
2558 | __func__, s->iq_tail, s->iq_size); | |
ed7b8fbc LT |
2559 | vtd_handle_inv_queue_error(s); |
2560 | return; | |
2561 | } | |
2562 | while (s->iq_head != s->iq_tail) { | |
2563 | if (!vtd_process_inv_desc(s)) { | |
2564 | /* Invalidation Queue Errors */ | |
2565 | vtd_handle_inv_queue_error(s); | |
2566 | break; | |
2567 | } | |
2568 | /* Must update the IQH_REG in time */ | |
2569 | vtd_set_quad_raw(s, DMAR_IQH_REG, | |
2570 | (((uint64_t)(s->iq_head)) << VTD_IQH_QH_SHIFT) & | |
2571 | VTD_IQH_QH_MASK); | |
2572 | } | |
2573 | } | |
2574 | ||
2575 | /* Handle write to Invalidation Queue Tail Register */ | |
2576 | static void vtd_handle_iqt_write(IntelIOMMUState *s) | |
2577 | { | |
2578 | uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG); | |
2579 | ||
c0c1d351 LY |
2580 | if (s->iq_dw && (val & VTD_IQT_QT_256_RSV_BIT)) { |
2581 | error_report_once("%s: RSV bit is set: val=0x%"PRIx64, | |
2582 | __func__, val); | |
2583 | return; | |
2584 | } | |
2585 | s->iq_tail = VTD_IQT_QT(s->iq_dw, val); | |
7feb51b7 PX |
2586 | trace_vtd_inv_qi_tail(s->iq_tail); |
2587 | ||
ed7b8fbc LT |
2588 | if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) { |
2589 | /* Process Invalidation Queue here */ | |
2590 | vtd_fetch_inv_desc(s); | |
2591 | } | |
2592 | } | |
2593 | ||
1da12ec4 LT |
2594 | static void vtd_handle_fsts_write(IntelIOMMUState *s) |
2595 | { | |
2596 | uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); | |
2597 | uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG); | |
2598 | uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE; | |
2599 | ||
2600 | if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) { | |
2601 | vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); | |
7feb51b7 | 2602 | trace_vtd_fsts_clear_ip(); |
1da12ec4 | 2603 | } |
ed7b8fbc LT |
2604 | /* FIXME: when IQE is Clear, should we try to fetch some Invalidation |
2605 | * Descriptors if there are any when Queued Invalidation is enabled? | |
2606 | */ | |
1da12ec4 LT |
2607 | } |
2608 | ||
2609 | static void vtd_handle_fectl_write(IntelIOMMUState *s) | |
2610 | { | |
2611 | uint32_t fectl_reg; | |
2612 | /* FIXME: when software clears the IM field, check the IP field. But do we | |
2613 | * need to compare the old value and the new value to conclude that | |
2614 | * software clears the IM field? Or just check if the IM field is zero? | |
2615 | */ | |
2616 | fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG); | |
7feb51b7 PX |
2617 | |
2618 | trace_vtd_reg_write_fectl(fectl_reg); | |
2619 | ||
1da12ec4 LT |
2620 | if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) { |
2621 | vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG); | |
2622 | vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); | |
1da12ec4 LT |
2623 | } |
2624 | } | |
2625 | ||
ed7b8fbc LT |
2626 | static void vtd_handle_ics_write(IntelIOMMUState *s) |
2627 | { | |
2628 | uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG); | |
2629 | uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG); | |
2630 | ||
2631 | if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) { | |
7feb51b7 | 2632 | trace_vtd_reg_ics_clear_ip(); |
ed7b8fbc | 2633 | vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); |
ed7b8fbc LT |
2634 | } |
2635 | } | |
2636 | ||
2637 | static void vtd_handle_iectl_write(IntelIOMMUState *s) | |
2638 | { | |
2639 | uint32_t iectl_reg; | |
2640 | /* FIXME: when software clears the IM field, check the IP field. But do we | |
2641 | * need to compare the old value and the new value to conclude that | |
2642 | * software clears the IM field? Or just check if the IM field is zero? | |
2643 | */ | |
2644 | iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG); | |
7feb51b7 PX |
2645 | |
2646 | trace_vtd_reg_write_iectl(iectl_reg); | |
2647 | ||
ed7b8fbc LT |
2648 | if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) { |
2649 | vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG); | |
2650 | vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); | |
ed7b8fbc LT |
2651 | } |
2652 | } | |
2653 | ||
1da12ec4 LT |
2654 | static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size) |
2655 | { | |
2656 | IntelIOMMUState *s = opaque; | |
2657 | uint64_t val; | |
2658 | ||
7feb51b7 PX |
2659 | trace_vtd_reg_read(addr, size); |
2660 | ||
1da12ec4 | 2661 | if (addr + size > DMAR_REG_SIZE) { |
1376211f PX |
2662 | error_report_once("%s: MMIO over range: addr=0x%" PRIx64 |
2663 | " size=0x%u", __func__, addr, size); | |
1da12ec4 LT |
2664 | return (uint64_t)-1; |
2665 | } | |
2666 | ||
2667 | switch (addr) { | |
2668 | /* Root Table Address Register, 64-bit */ | |
2669 | case DMAR_RTADDR_REG: | |
8fdee711 | 2670 | val = vtd_get_quad_raw(s, DMAR_RTADDR_REG); |
1da12ec4 | 2671 | if (size == 4) { |
8fdee711 | 2672 | val = val & ((1ULL << 32) - 1); |
1da12ec4 LT |
2673 | } |
2674 | break; | |
2675 | ||
2676 | case DMAR_RTADDR_REG_HI: | |
2677 | assert(size == 4); | |
8fdee711 | 2678 | val = vtd_get_quad_raw(s, DMAR_RTADDR_REG) >> 32; |
1da12ec4 LT |
2679 | break; |
2680 | ||
ed7b8fbc LT |
2681 | /* Invalidation Queue Address Register, 64-bit */ |
2682 | case DMAR_IQA_REG: | |
2683 | val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS); | |
2684 | if (size == 4) { | |
2685 | val = val & ((1ULL << 32) - 1); | |
2686 | } | |
2687 | break; | |
2688 | ||
2689 | case DMAR_IQA_REG_HI: | |
2690 | assert(size == 4); | |
2691 | val = s->iq >> 32; | |
2692 | break; | |
2693 | ||
1da12ec4 LT |
2694 | default: |
2695 | if (size == 4) { | |
2696 | val = vtd_get_long(s, addr); | |
2697 | } else { | |
2698 | val = vtd_get_quad(s, addr); | |
2699 | } | |
2700 | } | |
7feb51b7 | 2701 | |
1da12ec4 LT |
2702 | return val; |
2703 | } | |
2704 | ||
2705 | static void vtd_mem_write(void *opaque, hwaddr addr, | |
2706 | uint64_t val, unsigned size) | |
2707 | { | |
2708 | IntelIOMMUState *s = opaque; | |
2709 | ||
7feb51b7 PX |
2710 | trace_vtd_reg_write(addr, size, val); |
2711 | ||
1da12ec4 | 2712 | if (addr + size > DMAR_REG_SIZE) { |
1376211f PX |
2713 | error_report_once("%s: MMIO over range: addr=0x%" PRIx64 |
2714 | " size=0x%u", __func__, addr, size); | |
1da12ec4 LT |
2715 | return; |
2716 | } | |
2717 | ||
2718 | switch (addr) { | |
2719 | /* Global Command Register, 32-bit */ | |
2720 | case DMAR_GCMD_REG: | |
1da12ec4 LT |
2721 | vtd_set_long(s, addr, val); |
2722 | vtd_handle_gcmd_write(s); | |
2723 | break; | |
2724 | ||
2725 | /* Context Command Register, 64-bit */ | |
2726 | case DMAR_CCMD_REG: | |
1da12ec4 LT |
2727 | if (size == 4) { |
2728 | vtd_set_long(s, addr, val); | |
2729 | } else { | |
2730 | vtd_set_quad(s, addr, val); | |
2731 | vtd_handle_ccmd_write(s); | |
2732 | } | |
2733 | break; | |
2734 | ||
2735 | case DMAR_CCMD_REG_HI: | |
1da12ec4 LT |
2736 | assert(size == 4); |
2737 | vtd_set_long(s, addr, val); | |
2738 | vtd_handle_ccmd_write(s); | |
2739 | break; | |
2740 | ||
2741 | /* IOTLB Invalidation Register, 64-bit */ | |
2742 | case DMAR_IOTLB_REG: | |
1da12ec4 LT |
2743 | if (size == 4) { |
2744 | vtd_set_long(s, addr, val); | |
2745 | } else { | |
2746 | vtd_set_quad(s, addr, val); | |
2747 | vtd_handle_iotlb_write(s); | |
2748 | } | |
2749 | break; | |
2750 | ||
2751 | case DMAR_IOTLB_REG_HI: | |
1da12ec4 LT |
2752 | assert(size == 4); |
2753 | vtd_set_long(s, addr, val); | |
2754 | vtd_handle_iotlb_write(s); | |
2755 | break; | |
2756 | ||
b5a280c0 LT |
2757 | /* Invalidate Address Register, 64-bit */ |
2758 | case DMAR_IVA_REG: | |
b5a280c0 LT |
2759 | if (size == 4) { |
2760 | vtd_set_long(s, addr, val); | |
2761 | } else { | |
2762 | vtd_set_quad(s, addr, val); | |
2763 | } | |
2764 | break; | |
2765 | ||
2766 | case DMAR_IVA_REG_HI: | |
b5a280c0 LT |
2767 | assert(size == 4); |
2768 | vtd_set_long(s, addr, val); | |
2769 | break; | |
2770 | ||
1da12ec4 LT |
2771 | /* Fault Status Register, 32-bit */ |
2772 | case DMAR_FSTS_REG: | |
1da12ec4 LT |
2773 | assert(size == 4); |
2774 | vtd_set_long(s, addr, val); | |
2775 | vtd_handle_fsts_write(s); | |
2776 | break; | |
2777 | ||
2778 | /* Fault Event Control Register, 32-bit */ | |
2779 | case DMAR_FECTL_REG: | |
1da12ec4 LT |
2780 | assert(size == 4); |
2781 | vtd_set_long(s, addr, val); | |
2782 | vtd_handle_fectl_write(s); | |
2783 | break; | |
2784 | ||
2785 | /* Fault Event Data Register, 32-bit */ | |
2786 | case DMAR_FEDATA_REG: | |
1da12ec4 LT |
2787 | assert(size == 4); |
2788 | vtd_set_long(s, addr, val); | |
2789 | break; | |
2790 | ||
2791 | /* Fault Event Address Register, 32-bit */ | |
2792 | case DMAR_FEADDR_REG: | |
b7a7bb35 JK |
2793 | if (size == 4) { |
2794 | vtd_set_long(s, addr, val); | |
2795 | } else { | |
2796 | /* | |
2797 | * While the register is 32-bit only, some guests (Xen...) write to | |
2798 | * it with 64-bit. | |
2799 | */ | |
2800 | vtd_set_quad(s, addr, val); | |
2801 | } | |
1da12ec4 LT |
2802 | break; |
2803 | ||
2804 | /* Fault Event Upper Address Register, 32-bit */ | |
2805 | case DMAR_FEUADDR_REG: | |
1da12ec4 LT |
2806 | assert(size == 4); |
2807 | vtd_set_long(s, addr, val); | |
2808 | break; | |
2809 | ||
2810 | /* Protected Memory Enable Register, 32-bit */ | |
2811 | case DMAR_PMEN_REG: | |
1da12ec4 LT |
2812 | assert(size == 4); |
2813 | vtd_set_long(s, addr, val); | |
2814 | break; | |
2815 | ||
2816 | /* Root Table Address Register, 64-bit */ | |
2817 | case DMAR_RTADDR_REG: | |
1da12ec4 LT |
2818 | if (size == 4) { |
2819 | vtd_set_long(s, addr, val); | |
2820 | } else { | |
2821 | vtd_set_quad(s, addr, val); | |
2822 | } | |
2823 | break; | |
2824 | ||
2825 | case DMAR_RTADDR_REG_HI: | |
1da12ec4 LT |
2826 | assert(size == 4); |
2827 | vtd_set_long(s, addr, val); | |
2828 | break; | |
2829 | ||
ed7b8fbc LT |
2830 | /* Invalidation Queue Tail Register, 64-bit */ |
2831 | case DMAR_IQT_REG: | |
ed7b8fbc LT |
2832 | if (size == 4) { |
2833 | vtd_set_long(s, addr, val); | |
2834 | } else { | |
2835 | vtd_set_quad(s, addr, val); | |
2836 | } | |
2837 | vtd_handle_iqt_write(s); | |
2838 | break; | |
2839 | ||
2840 | case DMAR_IQT_REG_HI: | |
ed7b8fbc LT |
2841 | assert(size == 4); |
2842 | vtd_set_long(s, addr, val); | |
2843 | /* 19:63 of IQT_REG is RsvdZ, do nothing here */ | |
2844 | break; | |
2845 | ||
2846 | /* Invalidation Queue Address Register, 64-bit */ | |
2847 | case DMAR_IQA_REG: | |
ed7b8fbc LT |
2848 | if (size == 4) { |
2849 | vtd_set_long(s, addr, val); | |
2850 | } else { | |
2851 | vtd_set_quad(s, addr, val); | |
2852 | } | |
c0c1d351 LY |
2853 | if (s->ecap & VTD_ECAP_SMTS && |
2854 | val & VTD_IQA_DW_MASK) { | |
2855 | s->iq_dw = true; | |
2856 | } else { | |
2857 | s->iq_dw = false; | |
2858 | } | |
ed7b8fbc LT |
2859 | break; |
2860 | ||
2861 | case DMAR_IQA_REG_HI: | |
ed7b8fbc LT |
2862 | assert(size == 4); |
2863 | vtd_set_long(s, addr, val); | |
2864 | break; | |
2865 | ||
2866 | /* Invalidation Completion Status Register, 32-bit */ | |
2867 | case DMAR_ICS_REG: | |
ed7b8fbc LT |
2868 | assert(size == 4); |
2869 | vtd_set_long(s, addr, val); | |
2870 | vtd_handle_ics_write(s); | |
2871 | break; | |
2872 | ||
2873 | /* Invalidation Event Control Register, 32-bit */ | |
2874 | case DMAR_IECTL_REG: | |
ed7b8fbc LT |
2875 | assert(size == 4); |
2876 | vtd_set_long(s, addr, val); | |
2877 | vtd_handle_iectl_write(s); | |
2878 | break; | |
2879 | ||
2880 | /* Invalidation Event Data Register, 32-bit */ | |
2881 | case DMAR_IEDATA_REG: | |
ed7b8fbc LT |
2882 | assert(size == 4); |
2883 | vtd_set_long(s, addr, val); | |
2884 | break; | |
2885 | ||
2886 | /* Invalidation Event Address Register, 32-bit */ | |
2887 | case DMAR_IEADDR_REG: | |
ed7b8fbc LT |
2888 | assert(size == 4); |
2889 | vtd_set_long(s, addr, val); | |
2890 | break; | |
2891 | ||
2892 | /* Invalidation Event Upper Address Register, 32-bit */ | |
2893 | case DMAR_IEUADDR_REG: | |
ed7b8fbc LT |
2894 | assert(size == 4); |
2895 | vtd_set_long(s, addr, val); | |
2896 | break; | |
2897 | ||
1da12ec4 LT |
2898 | /* Fault Recording Registers, 128-bit */ |
2899 | case DMAR_FRCD_REG_0_0: | |
1da12ec4 LT |
2900 | if (size == 4) { |
2901 | vtd_set_long(s, addr, val); | |
2902 | } else { | |
2903 | vtd_set_quad(s, addr, val); | |
2904 | } | |
2905 | break; | |
2906 | ||
2907 | case DMAR_FRCD_REG_0_1: | |
1da12ec4 LT |
2908 | assert(size == 4); |
2909 | vtd_set_long(s, addr, val); | |
2910 | break; | |
2911 | ||
2912 | case DMAR_FRCD_REG_0_2: | |
1da12ec4 LT |
2913 | if (size == 4) { |
2914 | vtd_set_long(s, addr, val); | |
2915 | } else { | |
2916 | vtd_set_quad(s, addr, val); | |
2917 | /* May clear bit 127 (Fault), update PPF */ | |
2918 | vtd_update_fsts_ppf(s); | |
2919 | } | |
2920 | break; | |
2921 | ||
2922 | case DMAR_FRCD_REG_0_3: | |
1da12ec4 LT |
2923 | assert(size == 4); |
2924 | vtd_set_long(s, addr, val); | |
2925 | /* May clear bit 127 (Fault), update PPF */ | |
2926 | vtd_update_fsts_ppf(s); | |
2927 | break; | |
2928 | ||
a5861439 | 2929 | case DMAR_IRTA_REG: |
a5861439 PX |
2930 | if (size == 4) { |
2931 | vtd_set_long(s, addr, val); | |
2932 | } else { | |
2933 | vtd_set_quad(s, addr, val); | |
2934 | } | |
2935 | break; | |
2936 | ||
2937 | case DMAR_IRTA_REG_HI: | |
a5861439 PX |
2938 | assert(size == 4); |
2939 | vtd_set_long(s, addr, val); | |
2940 | break; | |
2941 | ||
1da12ec4 | 2942 | default: |
1da12ec4 LT |
2943 | if (size == 4) { |
2944 | vtd_set_long(s, addr, val); | |
2945 | } else { | |
2946 | vtd_set_quad(s, addr, val); | |
2947 | } | |
2948 | } | |
2949 | } | |
2950 | ||
3df9d748 | 2951 | static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr, |
2c91bcf2 | 2952 | IOMMUAccessFlags flag, int iommu_idx) |
1da12ec4 LT |
2953 | { |
2954 | VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu); | |
2955 | IntelIOMMUState *s = vtd_as->iommu_state; | |
b9313021 PX |
2956 | IOMMUTLBEntry iotlb = { |
2957 | /* We'll fill in the rest later. */ | |
1da12ec4 | 2958 | .target_as = &address_space_memory, |
1da12ec4 | 2959 | }; |
b9313021 | 2960 | bool success; |
1da12ec4 | 2961 | |
b9313021 PX |
2962 | if (likely(s->dmar_enabled)) { |
2963 | success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn, | |
2964 | addr, flag & IOMMU_WO, &iotlb); | |
2965 | } else { | |
1da12ec4 | 2966 | /* DMAR disabled, passthrough, use 4k-page*/ |
b9313021 PX |
2967 | iotlb.iova = addr & VTD_PAGE_MASK_4K; |
2968 | iotlb.translated_addr = addr & VTD_PAGE_MASK_4K; | |
2969 | iotlb.addr_mask = ~VTD_PAGE_MASK_4K; | |
2970 | iotlb.perm = IOMMU_RW; | |
2971 | success = true; | |
1da12ec4 LT |
2972 | } |
2973 | ||
b9313021 PX |
2974 | if (likely(success)) { |
2975 | trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus), | |
2976 | VTD_PCI_SLOT(vtd_as->devfn), | |
2977 | VTD_PCI_FUNC(vtd_as->devfn), | |
2978 | iotlb.iova, iotlb.translated_addr, | |
2979 | iotlb.addr_mask); | |
2980 | } else { | |
4e4abd11 PX |
2981 | error_report_once("%s: detected translation failure " |
2982 | "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")", | |
2983 | __func__, pci_bus_num(vtd_as->bus), | |
2984 | VTD_PCI_SLOT(vtd_as->devfn), | |
2985 | VTD_PCI_FUNC(vtd_as->devfn), | |
662b4b69 | 2986 | addr); |
b9313021 | 2987 | } |
7feb51b7 | 2988 | |
b9313021 | 2989 | return iotlb; |
1da12ec4 LT |
2990 | } |
2991 | ||
549d4005 EA |
2992 | static int vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu, |
2993 | IOMMUNotifierFlag old, | |
2994 | IOMMUNotifierFlag new, | |
2995 | Error **errp) | |
3cb3b154 AW |
2996 | { |
2997 | VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu); | |
dd4d607e | 2998 | IntelIOMMUState *s = vtd_as->iommu_state; |
3cb3b154 | 2999 | |
4f8a62a9 PX |
3000 | /* Update per-address-space notifier flags */ |
3001 | vtd_as->notifier_flags = new; | |
3002 | ||
dd4d607e | 3003 | if (old == IOMMU_NOTIFIER_NONE) { |
b4a4ba0d PX |
3004 | QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next); |
3005 | } else if (new == IOMMU_NOTIFIER_NONE) { | |
3006 | QLIST_REMOVE(vtd_as, next); | |
dd4d607e | 3007 | } |
549d4005 | 3008 | return 0; |
3cb3b154 AW |
3009 | } |
3010 | ||
552a1e01 PX |
3011 | static int vtd_post_load(void *opaque, int version_id) |
3012 | { | |
3013 | IntelIOMMUState *iommu = opaque; | |
3014 | ||
3015 | /* | |
3016 | * Memory regions are dynamically turned on/off depending on | |
3017 | * context entry configurations from the guest. After migration, | |
3018 | * we need to make sure the memory regions are still correct. | |
3019 | */ | |
3020 | vtd_switch_address_space_all(iommu); | |
3021 | ||
2811af3b PX |
3022 | /* |
3023 | * We don't need to migrate the root_scalable because we can | |
3024 | * simply do the calculation after the loading is complete. We | |
3025 | * can actually do similar things with root, dmar_enabled, etc. | |
3026 | * however since we've had them already so we'd better keep them | |
3027 | * for compatibility of migration. | |
3028 | */ | |
3029 | vtd_update_scalable_state(iommu); | |
3030 | ||
552a1e01 PX |
3031 | return 0; |
3032 | } | |
3033 | ||
1da12ec4 LT |
3034 | static const VMStateDescription vtd_vmstate = { |
3035 | .name = "iommu-intel", | |
8cdcf3c1 PX |
3036 | .version_id = 1, |
3037 | .minimum_version_id = 1, | |
3038 | .priority = MIG_PRI_IOMMU, | |
552a1e01 | 3039 | .post_load = vtd_post_load, |
8cdcf3c1 PX |
3040 | .fields = (VMStateField[]) { |
3041 | VMSTATE_UINT64(root, IntelIOMMUState), | |
3042 | VMSTATE_UINT64(intr_root, IntelIOMMUState), | |
3043 | VMSTATE_UINT64(iq, IntelIOMMUState), | |
3044 | VMSTATE_UINT32(intr_size, IntelIOMMUState), | |
3045 | VMSTATE_UINT16(iq_head, IntelIOMMUState), | |
3046 | VMSTATE_UINT16(iq_tail, IntelIOMMUState), | |
3047 | VMSTATE_UINT16(iq_size, IntelIOMMUState), | |
3048 | VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState), | |
3049 | VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE), | |
3050 | VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState), | |
81fb1e64 | 3051 | VMSTATE_UNUSED(1), /* bool root_extended is obsolete by VT-d */ |
8cdcf3c1 PX |
3052 | VMSTATE_BOOL(dmar_enabled, IntelIOMMUState), |
3053 | VMSTATE_BOOL(qi_enabled, IntelIOMMUState), | |
3054 | VMSTATE_BOOL(intr_enabled, IntelIOMMUState), | |
3055 | VMSTATE_BOOL(intr_eime, IntelIOMMUState), | |
3056 | VMSTATE_END_OF_LIST() | |
3057 | } | |
1da12ec4 LT |
3058 | }; |
3059 | ||
3060 | static const MemoryRegionOps vtd_mem_ops = { | |
3061 | .read = vtd_mem_read, | |
3062 | .write = vtd_mem_write, | |
3063 | .endianness = DEVICE_LITTLE_ENDIAN, | |
3064 | .impl = { | |
3065 | .min_access_size = 4, | |
3066 | .max_access_size = 8, | |
3067 | }, | |
3068 | .valid = { | |
3069 | .min_access_size = 4, | |
3070 | .max_access_size = 8, | |
3071 | }, | |
3072 | }; | |
3073 | ||
3074 | static Property vtd_properties[] = { | |
3075 | DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0), | |
e6b6af05 RK |
3076 | DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim, |
3077 | ON_OFF_AUTO_AUTO), | |
fb506e70 | 3078 | DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false), |
4b49b586 | 3079 | DEFINE_PROP_UINT8("aw-bits", IntelIOMMUState, aw_bits, |
37f51384 | 3080 | VTD_HOST_ADDRESS_WIDTH), |
3b40f0e5 | 3081 | DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE), |
4a4f219e | 3082 | DEFINE_PROP_BOOL("x-scalable-mode", IntelIOMMUState, scalable_mode, FALSE), |
ccc23bb0 | 3083 | DEFINE_PROP_BOOL("dma-drain", IntelIOMMUState, dma_drain, true), |
1da12ec4 LT |
3084 | DEFINE_PROP_END_OF_LIST(), |
3085 | }; | |
3086 | ||
651e4cef PX |
3087 | /* Read IRTE entry with specific index */ |
3088 | static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index, | |
bc38ee10 | 3089 | VTD_IR_TableEntry *entry, uint16_t sid) |
651e4cef | 3090 | { |
ede9c94a PX |
3091 | static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \ |
3092 | {0xffff, 0xfffb, 0xfff9, 0xfff8}; | |
651e4cef | 3093 | dma_addr_t addr = 0x00; |
ede9c94a PX |
3094 | uint16_t mask, source_id; |
3095 | uint8_t bus, bus_max, bus_min; | |
651e4cef | 3096 | |
3c507c26 JK |
3097 | if (index >= iommu->intr_size) { |
3098 | error_report_once("%s: index too large: ind=0x%x", | |
3099 | __func__, index); | |
3100 | return -VTD_FR_IR_INDEX_OVER; | |
3101 | } | |
3102 | ||
651e4cef PX |
3103 | addr = iommu->intr_root + index * sizeof(*entry); |
3104 | if (dma_memory_read(&address_space_memory, addr, entry, | |
3105 | sizeof(*entry))) { | |
1376211f PX |
3106 | error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64, |
3107 | __func__, index, addr); | |
651e4cef PX |
3108 | return -VTD_FR_IR_ROOT_INVAL; |
3109 | } | |
3110 | ||
7feb51b7 PX |
3111 | trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]), |
3112 | le64_to_cpu(entry->data[0])); | |
3113 | ||
bc38ee10 | 3114 | if (!entry->irte.present) { |
4e4abd11 PX |
3115 | error_report_once("%s: detected non-present IRTE " |
3116 | "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")", | |
3117 | __func__, index, le64_to_cpu(entry->data[1]), | |
3118 | le64_to_cpu(entry->data[0])); | |
651e4cef PX |
3119 | return -VTD_FR_IR_ENTRY_P; |
3120 | } | |
3121 | ||
bc38ee10 MT |
3122 | if (entry->irte.__reserved_0 || entry->irte.__reserved_1 || |
3123 | entry->irte.__reserved_2) { | |
4e4abd11 PX |
3124 | error_report_once("%s: detected non-zero reserved IRTE " |
3125 | "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")", | |
3126 | __func__, index, le64_to_cpu(entry->data[1]), | |
3127 | le64_to_cpu(entry->data[0])); | |
651e4cef PX |
3128 | return -VTD_FR_IR_IRTE_RSVD; |
3129 | } | |
3130 | ||
ede9c94a PX |
3131 | if (sid != X86_IOMMU_SID_INVALID) { |
3132 | /* Validate IRTE SID */ | |
bc38ee10 MT |
3133 | source_id = le32_to_cpu(entry->irte.source_id); |
3134 | switch (entry->irte.sid_vtype) { | |
ede9c94a | 3135 | case VTD_SVT_NONE: |
ede9c94a PX |
3136 | break; |
3137 | ||
3138 | case VTD_SVT_ALL: | |
bc38ee10 | 3139 | mask = vtd_svt_mask[entry->irte.sid_q]; |
ede9c94a | 3140 | if ((source_id & mask) != (sid & mask)) { |
4e4abd11 PX |
3141 | error_report_once("%s: invalid IRTE SID " |
3142 | "(index=%u, sid=%u, source_id=%u)", | |
3143 | __func__, index, sid, source_id); | |
ede9c94a PX |
3144 | return -VTD_FR_IR_SID_ERR; |
3145 | } | |
3146 | break; | |
3147 | ||
3148 | case VTD_SVT_BUS: | |
3149 | bus_max = source_id >> 8; | |
3150 | bus_min = source_id & 0xff; | |
3151 | bus = sid >> 8; | |
3152 | if (bus > bus_max || bus < bus_min) { | |
4e4abd11 PX |
3153 | error_report_once("%s: invalid SVT_BUS " |
3154 | "(index=%u, bus=%u, min=%u, max=%u)", | |
3155 | __func__, index, bus, bus_min, bus_max); | |
ede9c94a PX |
3156 | return -VTD_FR_IR_SID_ERR; |
3157 | } | |
3158 | break; | |
3159 | ||
3160 | default: | |
4e4abd11 PX |
3161 | error_report_once("%s: detected invalid IRTE SVT " |
3162 | "(index=%u, type=%d)", __func__, | |
3163 | index, entry->irte.sid_vtype); | |
ede9c94a PX |
3164 | /* Take this as verification failure. */ |
3165 | return -VTD_FR_IR_SID_ERR; | |
3166 | break; | |
3167 | } | |
3168 | } | |
651e4cef PX |
3169 | |
3170 | return 0; | |
3171 | } | |
3172 | ||
3173 | /* Fetch IRQ information of specific IR index */ | |
ede9c94a | 3174 | static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index, |
35c24501 | 3175 | X86IOMMUIrq *irq, uint16_t sid) |
651e4cef | 3176 | { |
bc38ee10 | 3177 | VTD_IR_TableEntry irte = {}; |
651e4cef PX |
3178 | int ret = 0; |
3179 | ||
ede9c94a | 3180 | ret = vtd_irte_get(iommu, index, &irte, sid); |
651e4cef PX |
3181 | if (ret) { |
3182 | return ret; | |
3183 | } | |
3184 | ||
bc38ee10 MT |
3185 | irq->trigger_mode = irte.irte.trigger_mode; |
3186 | irq->vector = irte.irte.vector; | |
3187 | irq->delivery_mode = irte.irte.delivery_mode; | |
3188 | irq->dest = le32_to_cpu(irte.irte.dest_id); | |
28589311 | 3189 | if (!iommu->intr_eime) { |
651e4cef PX |
3190 | #define VTD_IR_APIC_DEST_MASK (0xff00ULL) |
3191 | #define VTD_IR_APIC_DEST_SHIFT (8) | |
28589311 JK |
3192 | irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >> |
3193 | VTD_IR_APIC_DEST_SHIFT; | |
3194 | } | |
bc38ee10 MT |
3195 | irq->dest_mode = irte.irte.dest_mode; |
3196 | irq->redir_hint = irte.irte.redir_hint; | |
651e4cef | 3197 | |
7feb51b7 PX |
3198 | trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector, |
3199 | irq->delivery_mode, irq->dest, irq->dest_mode); | |
651e4cef PX |
3200 | |
3201 | return 0; | |
3202 | } | |
3203 | ||
651e4cef PX |
3204 | /* Interrupt remapping for MSI/MSI-X entry */ |
3205 | static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu, | |
3206 | MSIMessage *origin, | |
ede9c94a PX |
3207 | MSIMessage *translated, |
3208 | uint16_t sid) | |
651e4cef PX |
3209 | { |
3210 | int ret = 0; | |
3211 | VTD_IR_MSIAddress addr; | |
3212 | uint16_t index; | |
35c24501 | 3213 | X86IOMMUIrq irq = {}; |
651e4cef PX |
3214 | |
3215 | assert(origin && translated); | |
3216 | ||
7feb51b7 PX |
3217 | trace_vtd_ir_remap_msi_req(origin->address, origin->data); |
3218 | ||
651e4cef | 3219 | if (!iommu || !iommu->intr_enabled) { |
e7a3b91f PX |
3220 | memcpy(translated, origin, sizeof(*origin)); |
3221 | goto out; | |
651e4cef PX |
3222 | } |
3223 | ||
3224 | if (origin->address & VTD_MSI_ADDR_HI_MASK) { | |
1376211f PX |
3225 | error_report_once("%s: MSI address high 32 bits non-zero detected: " |
3226 | "address=0x%" PRIx64, __func__, origin->address); | |
651e4cef PX |
3227 | return -VTD_FR_IR_REQ_RSVD; |
3228 | } | |
3229 | ||
3230 | addr.data = origin->address & VTD_MSI_ADDR_LO_MASK; | |
1a43713b | 3231 | if (addr.addr.__head != 0xfee) { |
1376211f PX |
3232 | error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32, |
3233 | __func__, addr.data); | |
651e4cef PX |
3234 | return -VTD_FR_IR_REQ_RSVD; |
3235 | } | |
3236 | ||
3237 | /* This is compatible mode. */ | |
bc38ee10 | 3238 | if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) { |
e7a3b91f PX |
3239 | memcpy(translated, origin, sizeof(*origin)); |
3240 | goto out; | |
651e4cef PX |
3241 | } |
3242 | ||
bc38ee10 | 3243 | index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l); |
651e4cef PX |
3244 | |
3245 | #define VTD_IR_MSI_DATA_SUBHANDLE (0x0000ffff) | |
3246 | #define VTD_IR_MSI_DATA_RESERVED (0xffff0000) | |
3247 | ||
bc38ee10 | 3248 | if (addr.addr.sub_valid) { |
651e4cef PX |
3249 | /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */ |
3250 | index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE; | |
3251 | } | |
3252 | ||
ede9c94a | 3253 | ret = vtd_remap_irq_get(iommu, index, &irq, sid); |
651e4cef PX |
3254 | if (ret) { |
3255 | return ret; | |
3256 | } | |
3257 | ||
bc38ee10 | 3258 | if (addr.addr.sub_valid) { |
7feb51b7 | 3259 | trace_vtd_ir_remap_type("MSI"); |
651e4cef | 3260 | if (origin->data & VTD_IR_MSI_DATA_RESERVED) { |
4e4abd11 PX |
3261 | error_report_once("%s: invalid IR MSI " |
3262 | "(sid=%u, address=0x%" PRIx64 | |
3263 | ", data=0x%" PRIx32 ")", | |
3264 | __func__, sid, origin->address, origin->data); | |
651e4cef PX |
3265 | return -VTD_FR_IR_REQ_RSVD; |
3266 | } | |
3267 | } else { | |
3268 | uint8_t vector = origin->data & 0xff; | |
dea651a9 FW |
3269 | uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1; |
3270 | ||
7feb51b7 | 3271 | trace_vtd_ir_remap_type("IOAPIC"); |
651e4cef PX |
3272 | /* IOAPIC entry vector should be aligned with IRTE vector |
3273 | * (see vt-d spec 5.1.5.1). */ | |
3274 | if (vector != irq.vector) { | |
7feb51b7 | 3275 | trace_vtd_warn_ir_vector(sid, index, vector, irq.vector); |
651e4cef | 3276 | } |
dea651a9 FW |
3277 | |
3278 | /* The Trigger Mode field must match the Trigger Mode in the IRTE. | |
3279 | * (see vt-d spec 5.1.5.1). */ | |
3280 | if (trigger_mode != irq.trigger_mode) { | |
7feb51b7 PX |
3281 | trace_vtd_warn_ir_trigger(sid, index, trigger_mode, |
3282 | irq.trigger_mode); | |
dea651a9 | 3283 | } |
651e4cef PX |
3284 | } |
3285 | ||
3286 | /* | |
3287 | * We'd better keep the last two bits, assuming that guest OS | |
3288 | * might modify it. Keep it does not hurt after all. | |
3289 | */ | |
bc38ee10 | 3290 | irq.msi_addr_last_bits = addr.addr.__not_care; |
651e4cef | 3291 | |
35c24501 BS |
3292 | /* Translate X86IOMMUIrq to MSI message */ |
3293 | x86_iommu_irq_to_msi_message(&irq, translated); | |
651e4cef | 3294 | |
e7a3b91f | 3295 | out: |
7feb51b7 PX |
3296 | trace_vtd_ir_remap_msi(origin->address, origin->data, |
3297 | translated->address, translated->data); | |
651e4cef PX |
3298 | return 0; |
3299 | } | |
3300 | ||
8b5ed7df PX |
3301 | static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src, |
3302 | MSIMessage *dst, uint16_t sid) | |
3303 | { | |
ede9c94a PX |
3304 | return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu), |
3305 | src, dst, sid); | |
8b5ed7df PX |
3306 | } |
3307 | ||
651e4cef PX |
3308 | static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr, |
3309 | uint64_t *data, unsigned size, | |
3310 | MemTxAttrs attrs) | |
3311 | { | |
3312 | return MEMTX_OK; | |
3313 | } | |
3314 | ||
3315 | static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr, | |
3316 | uint64_t value, unsigned size, | |
3317 | MemTxAttrs attrs) | |
3318 | { | |
3319 | int ret = 0; | |
09cd058a | 3320 | MSIMessage from = {}, to = {}; |
ede9c94a | 3321 | uint16_t sid = X86_IOMMU_SID_INVALID; |
651e4cef PX |
3322 | |
3323 | from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST; | |
3324 | from.data = (uint32_t) value; | |
3325 | ||
ede9c94a PX |
3326 | if (!attrs.unspecified) { |
3327 | /* We have explicit Source ID */ | |
3328 | sid = attrs.requester_id; | |
3329 | } | |
3330 | ||
3331 | ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid); | |
651e4cef PX |
3332 | if (ret) { |
3333 | /* TODO: report error */ | |
651e4cef PX |
3334 | /* Drop this interrupt */ |
3335 | return MEMTX_ERROR; | |
3336 | } | |
3337 | ||
32946019 | 3338 | apic_get_class()->send_msi(&to); |
651e4cef PX |
3339 | |
3340 | return MEMTX_OK; | |
3341 | } | |
3342 | ||
3343 | static const MemoryRegionOps vtd_mem_ir_ops = { | |
3344 | .read_with_attrs = vtd_mem_ir_read, | |
3345 | .write_with_attrs = vtd_mem_ir_write, | |
3346 | .endianness = DEVICE_LITTLE_ENDIAN, | |
3347 | .impl = { | |
3348 | .min_access_size = 4, | |
3349 | .max_access_size = 4, | |
3350 | }, | |
3351 | .valid = { | |
3352 | .min_access_size = 4, | |
3353 | .max_access_size = 4, | |
3354 | }, | |
3355 | }; | |
7df953bd KO |
3356 | |
3357 | VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn) | |
3358 | { | |
3359 | uintptr_t key = (uintptr_t)bus; | |
3360 | VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key); | |
3361 | VTDAddressSpace *vtd_dev_as; | |
e0a3c8cc | 3362 | char name[128]; |
7df953bd KO |
3363 | |
3364 | if (!vtd_bus) { | |
2d3fc581 JW |
3365 | uintptr_t *new_key = g_malloc(sizeof(*new_key)); |
3366 | *new_key = (uintptr_t)bus; | |
7df953bd | 3367 | /* No corresponding free() */ |
04af0e18 | 3368 | vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \ |
bf33cc75 | 3369 | PCI_DEVFN_MAX); |
7df953bd | 3370 | vtd_bus->bus = bus; |
2d3fc581 | 3371 | g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus); |
7df953bd KO |
3372 | } |
3373 | ||
3374 | vtd_dev_as = vtd_bus->dev_as[devfn]; | |
3375 | ||
3376 | if (!vtd_dev_as) { | |
4b519ef1 PX |
3377 | snprintf(name, sizeof(name), "vtd-%02x.%x", PCI_SLOT(devfn), |
3378 | PCI_FUNC(devfn)); | |
7df953bd KO |
3379 | vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace)); |
3380 | ||
3381 | vtd_dev_as->bus = bus; | |
3382 | vtd_dev_as->devfn = (uint8_t)devfn; | |
3383 | vtd_dev_as->iommu_state = s; | |
3384 | vtd_dev_as->context_cache_entry.context_cache_gen = 0; | |
63b88968 | 3385 | vtd_dev_as->iova_tree = iova_tree_new(); |
558e0024 | 3386 | |
4b519ef1 PX |
3387 | memory_region_init(&vtd_dev_as->root, OBJECT(s), name, UINT64_MAX); |
3388 | address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, "vtd-root"); | |
3389 | ||
558e0024 | 3390 | /* |
4b519ef1 PX |
3391 | * Build the DMAR-disabled container with aliases to the |
3392 | * shared MRs. Note that aliasing to a shared memory region | |
3393 | * could help the memory API to detect same FlatViews so we | |
3394 | * can have devices to share the same FlatView when DMAR is | |
3395 | * disabled (either by not providing "intel_iommu=on" or with | |
3396 | * "iommu=pt"). It will greatly reduce the total number of | |
3397 | * FlatViews of the system hence VM runs faster. | |
3398 | */ | |
3399 | memory_region_init_alias(&vtd_dev_as->nodmar, OBJECT(s), | |
3400 | "vtd-nodmar", &s->mr_nodmar, 0, | |
3401 | memory_region_size(&s->mr_nodmar)); | |
3402 | ||
3403 | /* | |
3404 | * Build the per-device DMAR-enabled container. | |
558e0024 | 3405 | * |
4b519ef1 PX |
3406 | * TODO: currently we have per-device IOMMU memory region only |
3407 | * because we have per-device IOMMU notifiers for devices. If | |
3408 | * one day we can abstract the IOMMU notifiers out of the | |
3409 | * memory regions then we can also share the same memory | |
3410 | * region here just like what we've done above with the nodmar | |
3411 | * region. | |
558e0024 | 3412 | */ |
4b519ef1 | 3413 | strcat(name, "-dmar"); |
1221a474 AK |
3414 | memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu), |
3415 | TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s), | |
4b519ef1 PX |
3416 | name, UINT64_MAX); |
3417 | memory_region_init_alias(&vtd_dev_as->iommu_ir, OBJECT(s), "vtd-ir", | |
3418 | &s->mr_ir, 0, memory_region_size(&s->mr_ir)); | |
3419 | memory_region_add_subregion_overlap(MEMORY_REGION(&vtd_dev_as->iommu), | |
558e0024 | 3420 | VTD_INTERRUPT_ADDR_FIRST, |
4b519ef1 PX |
3421 | &vtd_dev_as->iommu_ir, 1); |
3422 | ||
3423 | /* | |
3424 | * Hook both the containers under the root container, we | |
3425 | * switch between DMAR & noDMAR by enable/disable | |
3426 | * corresponding sub-containers | |
3427 | */ | |
558e0024 | 3428 | memory_region_add_subregion_overlap(&vtd_dev_as->root, 0, |
3df9d748 | 3429 | MEMORY_REGION(&vtd_dev_as->iommu), |
4b519ef1 PX |
3430 | 0); |
3431 | memory_region_add_subregion_overlap(&vtd_dev_as->root, 0, | |
3432 | &vtd_dev_as->nodmar, 0); | |
3433 | ||
558e0024 | 3434 | vtd_switch_address_space(vtd_dev_as); |
7df953bd KO |
3435 | } |
3436 | return vtd_dev_as; | |
3437 | } | |
3438 | ||
9a4bb839 PX |
3439 | static uint64_t get_naturally_aligned_size(uint64_t start, |
3440 | uint64_t size, int gaw) | |
3441 | { | |
3442 | uint64_t max_mask = 1ULL << gaw; | |
3443 | uint64_t alignment = start ? start & -start : max_mask; | |
3444 | ||
3445 | alignment = MIN(alignment, max_mask); | |
3446 | size = MIN(size, max_mask); | |
3447 | ||
3448 | if (alignment <= size) { | |
3449 | /* Increase the alignment of start */ | |
3450 | return alignment; | |
3451 | } else { | |
3452 | /* Find the largest page mask from size */ | |
3453 | return 1ULL << (63 - clz64(size)); | |
3454 | } | |
3455 | } | |
3456 | ||
dd4d607e PX |
3457 | /* Unmap the whole range in the notifier's scope. */ |
3458 | static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n) | |
3459 | { | |
9a4bb839 | 3460 | hwaddr size, remain; |
dd4d607e PX |
3461 | hwaddr start = n->start; |
3462 | hwaddr end = n->end; | |
37f51384 | 3463 | IntelIOMMUState *s = as->iommu_state; |
63b88968 | 3464 | DMAMap map; |
dd4d607e PX |
3465 | |
3466 | /* | |
3467 | * Note: all the codes in this function has a assumption that IOVA | |
3468 | * bits are no more than VTD_MGAW bits (which is restricted by | |
3469 | * VT-d spec), otherwise we need to consider overflow of 64 bits. | |
3470 | */ | |
3471 | ||
d6d10793 | 3472 | if (end > VTD_ADDRESS_SIZE(s->aw_bits) - 1) { |
dd4d607e PX |
3473 | /* |
3474 | * Don't need to unmap regions that is bigger than the whole | |
3475 | * VT-d supported address space size | |
3476 | */ | |
d6d10793 | 3477 | end = VTD_ADDRESS_SIZE(s->aw_bits) - 1; |
dd4d607e PX |
3478 | } |
3479 | ||
3480 | assert(start <= end); | |
9a4bb839 | 3481 | size = remain = end - start + 1; |
dd4d607e | 3482 | |
9a4bb839 PX |
3483 | while (remain >= VTD_PAGE_SIZE) { |
3484 | IOMMUTLBEntry entry; | |
3485 | uint64_t mask = get_naturally_aligned_size(start, remain, s->aw_bits); | |
3486 | ||
3487 | assert(mask); | |
3488 | ||
3489 | entry.iova = start; | |
3490 | entry.addr_mask = mask - 1; | |
3491 | entry.target_as = &address_space_memory; | |
3492 | entry.perm = IOMMU_NONE; | |
3493 | /* This field is meaningless for unmap */ | |
3494 | entry.translated_addr = 0; | |
3495 | ||
3496 | memory_region_notify_one(n, &entry); | |
3497 | ||
3498 | start += mask; | |
3499 | remain -= mask; | |
dd4d607e PX |
3500 | } |
3501 | ||
9a4bb839 | 3502 | assert(!remain); |
dd4d607e PX |
3503 | |
3504 | trace_vtd_as_unmap_whole(pci_bus_num(as->bus), | |
3505 | VTD_PCI_SLOT(as->devfn), | |
3506 | VTD_PCI_FUNC(as->devfn), | |
9a4bb839 | 3507 | n->start, size); |
dd4d607e | 3508 | |
9a4bb839 PX |
3509 | map.iova = n->start; |
3510 | map.size = size; | |
63b88968 | 3511 | iova_tree_remove(as->iova_tree, &map); |
dd4d607e PX |
3512 | } |
3513 | ||
3514 | static void vtd_address_space_unmap_all(IntelIOMMUState *s) | |
3515 | { | |
dd4d607e PX |
3516 | VTDAddressSpace *vtd_as; |
3517 | IOMMUNotifier *n; | |
3518 | ||
b4a4ba0d | 3519 | QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) { |
dd4d607e PX |
3520 | IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) { |
3521 | vtd_address_space_unmap(vtd_as, n); | |
3522 | } | |
3523 | } | |
3524 | } | |
3525 | ||
2cc9ddcc PX |
3526 | static void vtd_address_space_refresh_all(IntelIOMMUState *s) |
3527 | { | |
3528 | vtd_address_space_unmap_all(s); | |
3529 | vtd_switch_address_space_all(s); | |
3530 | } | |
3531 | ||
f06a696d PX |
3532 | static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private) |
3533 | { | |
3534 | memory_region_notify_one((IOMMUNotifier *)private, entry); | |
3535 | return 0; | |
3536 | } | |
3537 | ||
3df9d748 | 3538 | static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n) |
f06a696d | 3539 | { |
3df9d748 | 3540 | VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu); |
f06a696d PX |
3541 | IntelIOMMUState *s = vtd_as->iommu_state; |
3542 | uint8_t bus_n = pci_bus_num(vtd_as->bus); | |
3543 | VTDContextEntry ce; | |
3544 | ||
dd4d607e PX |
3545 | /* |
3546 | * The replay can be triggered by either a invalidation or a newly | |
3547 | * created entry. No matter what, we release existing mappings | |
3548 | * (it means flushing caches for UNMAP-only registers). | |
3549 | */ | |
3550 | vtd_address_space_unmap(vtd_as, n); | |
3551 | ||
f06a696d | 3552 | if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) { |
fb43cf73 LY |
3553 | trace_vtd_replay_ce_valid(s->root_scalable ? "scalable mode" : |
3554 | "legacy mode", | |
3555 | bus_n, PCI_SLOT(vtd_as->devfn), | |
f06a696d | 3556 | PCI_FUNC(vtd_as->devfn), |
fb43cf73 | 3557 | vtd_get_domain_id(s, &ce), |
f06a696d | 3558 | ce.hi, ce.lo); |
4f8a62a9 PX |
3559 | if (vtd_as_has_map_notifier(vtd_as)) { |
3560 | /* This is required only for MAP typed notifiers */ | |
fe215b0c PX |
3561 | vtd_page_walk_info info = { |
3562 | .hook_fn = vtd_replay_hook, | |
3563 | .private = (void *)n, | |
3564 | .notify_unmap = false, | |
3565 | .aw = s->aw_bits, | |
2f764fa8 | 3566 | .as = vtd_as, |
fb43cf73 | 3567 | .domain_id = vtd_get_domain_id(s, &ce), |
fe215b0c PX |
3568 | }; |
3569 | ||
fb43cf73 | 3570 | vtd_page_walk(s, &ce, 0, ~0ULL, &info); |
4f8a62a9 | 3571 | } |
f06a696d PX |
3572 | } else { |
3573 | trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn), | |
3574 | PCI_FUNC(vtd_as->devfn)); | |
3575 | } | |
3576 | ||
3577 | return; | |
3578 | } | |
3579 | ||
1da12ec4 LT |
3580 | /* Do the initialization. It will also be called when reset, so pay |
3581 | * attention when adding new initialization stuff. | |
3582 | */ | |
3583 | static void vtd_init(IntelIOMMUState *s) | |
3584 | { | |
d54bd7f8 PX |
3585 | X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s); |
3586 | ||
1da12ec4 LT |
3587 | memset(s->csr, 0, DMAR_REG_SIZE); |
3588 | memset(s->wmask, 0, DMAR_REG_SIZE); | |
3589 | memset(s->w1cmask, 0, DMAR_REG_SIZE); | |
3590 | memset(s->womask, 0, DMAR_REG_SIZE); | |
3591 | ||
1da12ec4 | 3592 | s->root = 0; |
fb43cf73 | 3593 | s->root_scalable = false; |
1da12ec4 | 3594 | s->dmar_enabled = false; |
d7bb469a | 3595 | s->intr_enabled = false; |
1da12ec4 LT |
3596 | s->iq_head = 0; |
3597 | s->iq_tail = 0; | |
3598 | s->iq = 0; | |
3599 | s->iq_size = 0; | |
3600 | s->qi_enabled = false; | |
3601 | s->iq_last_desc_type = VTD_INV_DESC_NONE; | |
c0c1d351 | 3602 | s->iq_dw = false; |
1da12ec4 | 3603 | s->next_frcd_reg = 0; |
92e5d85e PS |
3604 | s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND | |
3605 | VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS | | |
37f51384 | 3606 | VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits); |
ccc23bb0 PX |
3607 | if (s->dma_drain) { |
3608 | s->cap |= VTD_CAP_DRAIN; | |
3609 | } | |
37f51384 PS |
3610 | if (s->aw_bits == VTD_HOST_AW_48BIT) { |
3611 | s->cap |= VTD_CAP_SAGAW_48bit; | |
3612 | } | |
ed7b8fbc | 3613 | s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO; |
1da12ec4 | 3614 | |
92e5d85e PS |
3615 | /* |
3616 | * Rsvd field masks for spte | |
3617 | */ | |
ce586f3b | 3618 | vtd_spte_rsvd[0] = ~0ULL; |
e48929c7 QY |
3619 | vtd_spte_rsvd[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits, |
3620 | x86_iommu->dt_supported); | |
ce586f3b QY |
3621 | vtd_spte_rsvd[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits); |
3622 | vtd_spte_rsvd[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits); | |
3623 | vtd_spte_rsvd[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits); | |
3624 | ||
e48929c7 QY |
3625 | vtd_spte_rsvd_large[2] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits, |
3626 | x86_iommu->dt_supported); | |
3627 | vtd_spte_rsvd_large[3] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits, | |
3628 | x86_iommu->dt_supported); | |
92e5d85e | 3629 | |
a924b3d8 | 3630 | if (x86_iommu_ir_supported(x86_iommu)) { |
e6b6af05 RK |
3631 | s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV; |
3632 | if (s->intr_eim == ON_OFF_AUTO_ON) { | |
3633 | s->ecap |= VTD_ECAP_EIM; | |
3634 | } | |
3635 | assert(s->intr_eim != ON_OFF_AUTO_AUTO); | |
d54bd7f8 PX |
3636 | } |
3637 | ||
554f5e16 JW |
3638 | if (x86_iommu->dt_supported) { |
3639 | s->ecap |= VTD_ECAP_DT; | |
3640 | } | |
3641 | ||
dbaabb25 PX |
3642 | if (x86_iommu->pt_supported) { |
3643 | s->ecap |= VTD_ECAP_PT; | |
3644 | } | |
3645 | ||
3b40f0e5 ABD |
3646 | if (s->caching_mode) { |
3647 | s->cap |= VTD_CAP_CM; | |
3648 | } | |
3649 | ||
4a4f219e YS |
3650 | /* TODO: read cap/ecap from host to decide which cap to be exposed. */ |
3651 | if (s->scalable_mode) { | |
3652 | s->ecap |= VTD_ECAP_SMTS | VTD_ECAP_SRS | VTD_ECAP_SLTS; | |
3653 | } | |
3654 | ||
06aba4ca | 3655 | vtd_reset_caches(s); |
d92fa2dc | 3656 | |
1da12ec4 LT |
3657 | /* Define registers with default values and bit semantics */ |
3658 | vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0); | |
3659 | vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0); | |
3660 | vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0); | |
3661 | vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0); | |
3662 | vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL); | |
3663 | vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0); | |
fb43cf73 | 3664 | vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffffc00ULL, 0); |
1da12ec4 LT |
3665 | vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0); |
3666 | vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL); | |
3667 | ||
3668 | /* Advanced Fault Logging not supported */ | |
3669 | vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL); | |
3670 | vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0); | |
3671 | vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0); | |
3672 | vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0); | |
3673 | ||
3674 | /* Treated as RsvdZ when EIM in ECAP_REG is not supported | |
3675 | * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0); | |
3676 | */ | |
3677 | vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0); | |
3678 | ||
3679 | /* Treated as RO for implementations that PLMR and PHMR fields reported | |
3680 | * as Clear in the CAP_REG. | |
3681 | * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0); | |
3682 | */ | |
3683 | vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0); | |
3684 | ||
ed7b8fbc LT |
3685 | vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0); |
3686 | vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0); | |
c0c1d351 | 3687 | vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff807ULL, 0); |
ed7b8fbc LT |
3688 | vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL); |
3689 | vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0); | |
3690 | vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0); | |
3691 | vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0); | |
3692 | /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */ | |
3693 | vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0); | |
3694 | ||
1da12ec4 LT |
3695 | /* IOTLB registers */ |
3696 | vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0); | |
3697 | vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0); | |
3698 | vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL); | |
3699 | ||
3700 | /* Fault Recording Registers, 128-bit */ | |
3701 | vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0); | |
3702 | vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL); | |
a5861439 PX |
3703 | |
3704 | /* | |
28589311 | 3705 | * Interrupt remapping registers. |
a5861439 | 3706 | */ |
28589311 | 3707 | vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0); |
1da12ec4 LT |
3708 | } |
3709 | ||
3710 | /* Should not reset address_spaces when reset because devices will still use | |
3711 | * the address space they got at first (won't ask the bus again). | |
3712 | */ | |
3713 | static void vtd_reset(DeviceState *dev) | |
3714 | { | |
3715 | IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev); | |
3716 | ||
1da12ec4 | 3717 | vtd_init(s); |
2cc9ddcc | 3718 | vtd_address_space_refresh_all(s); |
1da12ec4 LT |
3719 | } |
3720 | ||
621d983a MA |
3721 | static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn) |
3722 | { | |
3723 | IntelIOMMUState *s = opaque; | |
3724 | VTDAddressSpace *vtd_as; | |
3725 | ||
bf33cc75 | 3726 | assert(0 <= devfn && devfn < PCI_DEVFN_MAX); |
621d983a MA |
3727 | |
3728 | vtd_as = vtd_find_add_as(s, bus, devfn); | |
3729 | return &vtd_as->as; | |
3730 | } | |
3731 | ||
e6b6af05 | 3732 | static bool vtd_decide_config(IntelIOMMUState *s, Error **errp) |
6333e93c | 3733 | { |
e6b6af05 RK |
3734 | X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s); |
3735 | ||
a924b3d8 | 3736 | if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu_ir_supported(x86_iommu)) { |
e6b6af05 RK |
3737 | error_setg(errp, "eim=on cannot be selected without intremap=on"); |
3738 | return false; | |
3739 | } | |
3740 | ||
3741 | if (s->intr_eim == ON_OFF_AUTO_AUTO) { | |
fb506e70 | 3742 | s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim) |
a924b3d8 | 3743 | && x86_iommu_ir_supported(x86_iommu) ? |
e6b6af05 RK |
3744 | ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; |
3745 | } | |
fb506e70 RK |
3746 | if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) { |
3747 | if (!kvm_irqchip_in_kernel()) { | |
3748 | error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split"); | |
3749 | return false; | |
3750 | } | |
3751 | if (!kvm_enable_x2apic()) { | |
3752 | error_setg(errp, "eim=on requires support on the KVM side" | |
3753 | "(X2APIC_API, first shipped in v4.7)"); | |
3754 | return false; | |
3755 | } | |
3756 | } | |
e6b6af05 | 3757 | |
37f51384 PS |
3758 | /* Currently only address widths supported are 39 and 48 bits */ |
3759 | if ((s->aw_bits != VTD_HOST_AW_39BIT) && | |
3760 | (s->aw_bits != VTD_HOST_AW_48BIT)) { | |
2a345149 | 3761 | error_setg(errp, "Supported values for aw-bits are: %d, %d", |
37f51384 PS |
3762 | VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT); |
3763 | return false; | |
3764 | } | |
3765 | ||
4a4f219e YS |
3766 | if (s->scalable_mode && !s->dma_drain) { |
3767 | error_setg(errp, "Need to set dma_drain for scalable mode"); | |
3768 | return false; | |
3769 | } | |
3770 | ||
6333e93c RK |
3771 | return true; |
3772 | } | |
3773 | ||
28cf553a PX |
3774 | static int vtd_machine_done_notify_one(Object *child, void *unused) |
3775 | { | |
3776 | IntelIOMMUState *iommu = INTEL_IOMMU_DEVICE(x86_iommu_get_default()); | |
3777 | ||
3778 | /* | |
3779 | * We hard-coded here because vfio-pci is the only special case | |
3780 | * here. Let's be more elegant in the future when we can, but so | |
3781 | * far there seems to be no better way. | |
3782 | */ | |
3783 | if (object_dynamic_cast(child, "vfio-pci") && !iommu->caching_mode) { | |
3784 | vtd_panic_require_caching_mode(); | |
3785 | } | |
3786 | ||
3787 | return 0; | |
3788 | } | |
3789 | ||
3790 | static void vtd_machine_done_hook(Notifier *notifier, void *unused) | |
3791 | { | |
3792 | object_child_foreach_recursive(object_get_root(), | |
3793 | vtd_machine_done_notify_one, NULL); | |
3794 | } | |
3795 | ||
3796 | static Notifier vtd_machine_done_notify = { | |
3797 | .notify = vtd_machine_done_hook, | |
3798 | }; | |
3799 | ||
1da12ec4 LT |
3800 | static void vtd_realize(DeviceState *dev, Error **errp) |
3801 | { | |
ef0e8fc7 | 3802 | MachineState *ms = MACHINE(qdev_get_machine()); |
29396ed9 | 3803 | PCMachineState *pcms = PC_MACHINE(ms); |
f0bb276b | 3804 | X86MachineState *x86ms = X86_MACHINE(ms); |
29396ed9 | 3805 | PCIBus *bus = pcms->bus; |
1da12ec4 | 3806 | IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev); |
4684a204 | 3807 | X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev); |
1da12ec4 | 3808 | |
fb9f5926 | 3809 | x86_iommu->type = TYPE_INTEL; |
6333e93c | 3810 | |
e6b6af05 | 3811 | if (!vtd_decide_config(s, errp)) { |
6333e93c RK |
3812 | return; |
3813 | } | |
3814 | ||
b4a4ba0d | 3815 | QLIST_INIT(&s->vtd_as_with_notifiers); |
1d9efa73 | 3816 | qemu_mutex_init(&s->iommu_lock); |
7df953bd | 3817 | memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num)); |
1da12ec4 LT |
3818 | memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s, |
3819 | "intel_iommu", DMAR_REG_SIZE); | |
4b519ef1 PX |
3820 | |
3821 | /* Create the shared memory regions by all devices */ | |
3822 | memory_region_init(&s->mr_nodmar, OBJECT(s), "vtd-nodmar", | |
3823 | UINT64_MAX); | |
3824 | memory_region_init_io(&s->mr_ir, OBJECT(s), &vtd_mem_ir_ops, | |
3825 | s, "vtd-ir", VTD_INTERRUPT_ADDR_SIZE); | |
3826 | memory_region_init_alias(&s->mr_sys_alias, OBJECT(s), | |
3827 | "vtd-sys-alias", get_system_memory(), 0, | |
3828 | memory_region_size(get_system_memory())); | |
3829 | memory_region_add_subregion_overlap(&s->mr_nodmar, 0, | |
3830 | &s->mr_sys_alias, 0); | |
3831 | memory_region_add_subregion_overlap(&s->mr_nodmar, | |
3832 | VTD_INTERRUPT_ADDR_FIRST, | |
3833 | &s->mr_ir, 1); | |
3834 | ||
1da12ec4 | 3835 | sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem); |
b5a280c0 LT |
3836 | /* No corresponding destroy */ |
3837 | s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal, | |
3838 | g_free, g_free); | |
7df953bd KO |
3839 | s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal, |
3840 | g_free, g_free); | |
1da12ec4 | 3841 | vtd_init(s); |
621d983a MA |
3842 | sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR); |
3843 | pci_setup_iommu(bus, vtd_host_dma_iommu, dev); | |
cb135f59 | 3844 | /* Pseudo address space under root PCI bus. */ |
f0bb276b | 3845 | x86ms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC); |
28cf553a | 3846 | qemu_add_machine_init_done_notifier(&vtd_machine_done_notify); |
1da12ec4 LT |
3847 | } |
3848 | ||
3849 | static void vtd_class_init(ObjectClass *klass, void *data) | |
3850 | { | |
3851 | DeviceClass *dc = DEVICE_CLASS(klass); | |
1c7955c4 | 3852 | X86IOMMUClass *x86_class = X86_IOMMU_CLASS(klass); |
1da12ec4 LT |
3853 | |
3854 | dc->reset = vtd_reset; | |
1da12ec4 | 3855 | dc->vmsd = &vtd_vmstate; |
4f67d30b | 3856 | device_class_set_props(dc, vtd_properties); |
621d983a | 3857 | dc->hotpluggable = false; |
1c7955c4 | 3858 | x86_class->realize = vtd_realize; |
8b5ed7df | 3859 | x86_class->int_remap = vtd_int_remap; |
8ab5700c | 3860 | /* Supported by the pc-q35-* machine types */ |
e4f4fb1e | 3861 | dc->user_creatable = true; |
1ec202c9 EE |
3862 | set_bit(DEVICE_CATEGORY_MISC, dc->categories); |
3863 | dc->desc = "Intel IOMMU (VT-d) DMA Remapping device"; | |
1da12ec4 LT |
3864 | } |
3865 | ||
3866 | static const TypeInfo vtd_info = { | |
3867 | .name = TYPE_INTEL_IOMMU_DEVICE, | |
1c7955c4 | 3868 | .parent = TYPE_X86_IOMMU_DEVICE, |
1da12ec4 LT |
3869 | .instance_size = sizeof(IntelIOMMUState), |
3870 | .class_init = vtd_class_init, | |
3871 | }; | |
3872 | ||
1221a474 AK |
3873 | static void vtd_iommu_memory_region_class_init(ObjectClass *klass, |
3874 | void *data) | |
3875 | { | |
3876 | IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass); | |
3877 | ||
3878 | imrc->translate = vtd_iommu_translate; | |
3879 | imrc->notify_flag_changed = vtd_iommu_notify_flag_changed; | |
3880 | imrc->replay = vtd_iommu_replay; | |
3881 | } | |
3882 | ||
3883 | static const TypeInfo vtd_iommu_memory_region_info = { | |
3884 | .parent = TYPE_IOMMU_MEMORY_REGION, | |
3885 | .name = TYPE_INTEL_IOMMU_MEMORY_REGION, | |
3886 | .class_init = vtd_iommu_memory_region_class_init, | |
3887 | }; | |
3888 | ||
1da12ec4 LT |
3889 | static void vtd_register_types(void) |
3890 | { | |
1da12ec4 | 3891 | type_register_static(&vtd_info); |
1221a474 | 3892 | type_register_static(&vtd_iommu_memory_region_info); |
1da12ec4 LT |
3893 | } |
3894 | ||
3895 | type_init(vtd_register_types) |