]>
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 <[email protected]> | |
6 | * Copyright (C) 2014 Le Tan, <[email protected]> | |
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" |
6333e93c | 24 | #include "qapi/error.h" |
1da12ec4 LT |
25 | #include "hw/sysbus.h" |
26 | #include "exec/address-spaces.h" | |
27 | #include "intel_iommu_internal.h" | |
7df953bd | 28 | #include "hw/pci/pci.h" |
3cb3b154 | 29 | #include "hw/pci/pci_bus.h" |
621d983a | 30 | #include "hw/i386/pc.h" |
dea651a9 | 31 | #include "hw/i386/apic-msidef.h" |
04af0e18 PX |
32 | #include "hw/boards.h" |
33 | #include "hw/i386/x86-iommu.h" | |
cb135f59 | 34 | #include "hw/pci-host/q35.h" |
4684a204 | 35 | #include "sysemu/kvm.h" |
32946019 | 36 | #include "hw/i386/apic_internal.h" |
fb506e70 | 37 | #include "kvm_i386.h" |
1da12ec4 LT |
38 | |
39 | /*#define DEBUG_INTEL_IOMMU*/ | |
40 | #ifdef DEBUG_INTEL_IOMMU | |
41 | enum { | |
42 | DEBUG_GENERAL, DEBUG_CSR, DEBUG_INV, DEBUG_MMU, DEBUG_FLOG, | |
a5861439 | 43 | DEBUG_CACHE, DEBUG_IR, |
1da12ec4 LT |
44 | }; |
45 | #define VTD_DBGBIT(x) (1 << DEBUG_##x) | |
46 | static int vtd_dbgflags = VTD_DBGBIT(GENERAL) | VTD_DBGBIT(CSR); | |
47 | ||
48 | #define VTD_DPRINTF(what, fmt, ...) do { \ | |
49 | if (vtd_dbgflags & VTD_DBGBIT(what)) { \ | |
50 | fprintf(stderr, "(vtd)%s: " fmt "\n", __func__, \ | |
51 | ## __VA_ARGS__); } \ | |
52 | } while (0) | |
53 | #else | |
54 | #define VTD_DPRINTF(what, fmt, ...) do {} while (0) | |
55 | #endif | |
56 | ||
57 | static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val, | |
58 | uint64_t wmask, uint64_t w1cmask) | |
59 | { | |
60 | stq_le_p(&s->csr[addr], val); | |
61 | stq_le_p(&s->wmask[addr], wmask); | |
62 | stq_le_p(&s->w1cmask[addr], w1cmask); | |
63 | } | |
64 | ||
65 | static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask) | |
66 | { | |
67 | stq_le_p(&s->womask[addr], mask); | |
68 | } | |
69 | ||
70 | static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val, | |
71 | uint32_t wmask, uint32_t w1cmask) | |
72 | { | |
73 | stl_le_p(&s->csr[addr], val); | |
74 | stl_le_p(&s->wmask[addr], wmask); | |
75 | stl_le_p(&s->w1cmask[addr], w1cmask); | |
76 | } | |
77 | ||
78 | static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask) | |
79 | { | |
80 | stl_le_p(&s->womask[addr], mask); | |
81 | } | |
82 | ||
83 | /* "External" get/set operations */ | |
84 | static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val) | |
85 | { | |
86 | uint64_t oldval = ldq_le_p(&s->csr[addr]); | |
87 | uint64_t wmask = ldq_le_p(&s->wmask[addr]); | |
88 | uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]); | |
89 | stq_le_p(&s->csr[addr], | |
90 | ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val)); | |
91 | } | |
92 | ||
93 | static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val) | |
94 | { | |
95 | uint32_t oldval = ldl_le_p(&s->csr[addr]); | |
96 | uint32_t wmask = ldl_le_p(&s->wmask[addr]); | |
97 | uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]); | |
98 | stl_le_p(&s->csr[addr], | |
99 | ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val)); | |
100 | } | |
101 | ||
102 | static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr) | |
103 | { | |
104 | uint64_t val = ldq_le_p(&s->csr[addr]); | |
105 | uint64_t womask = ldq_le_p(&s->womask[addr]); | |
106 | return val & ~womask; | |
107 | } | |
108 | ||
109 | static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr) | |
110 | { | |
111 | uint32_t val = ldl_le_p(&s->csr[addr]); | |
112 | uint32_t womask = ldl_le_p(&s->womask[addr]); | |
113 | return val & ~womask; | |
114 | } | |
115 | ||
116 | /* "Internal" get/set operations */ | |
117 | static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr) | |
118 | { | |
119 | return ldq_le_p(&s->csr[addr]); | |
120 | } | |
121 | ||
122 | static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr) | |
123 | { | |
124 | return ldl_le_p(&s->csr[addr]); | |
125 | } | |
126 | ||
127 | static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val) | |
128 | { | |
129 | stq_le_p(&s->csr[addr], val); | |
130 | } | |
131 | ||
132 | static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr, | |
133 | uint32_t clear, uint32_t mask) | |
134 | { | |
135 | uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask; | |
136 | stl_le_p(&s->csr[addr], new_val); | |
137 | return new_val; | |
138 | } | |
139 | ||
140 | static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr, | |
141 | uint64_t clear, uint64_t mask) | |
142 | { | |
143 | uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask; | |
144 | stq_le_p(&s->csr[addr], new_val); | |
145 | return new_val; | |
146 | } | |
147 | ||
b5a280c0 LT |
148 | /* GHashTable functions */ |
149 | static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2) | |
150 | { | |
151 | return *((const uint64_t *)v1) == *((const uint64_t *)v2); | |
152 | } | |
153 | ||
154 | static guint vtd_uint64_hash(gconstpointer v) | |
155 | { | |
156 | return (guint)*(const uint64_t *)v; | |
157 | } | |
158 | ||
159 | static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value, | |
160 | gpointer user_data) | |
161 | { | |
162 | VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value; | |
163 | uint16_t domain_id = *(uint16_t *)user_data; | |
164 | return entry->domain_id == domain_id; | |
165 | } | |
166 | ||
d66b969b JW |
167 | /* The shift of an addr for a certain level of paging structure */ |
168 | static inline uint32_t vtd_slpt_level_shift(uint32_t level) | |
169 | { | |
170 | return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS; | |
171 | } | |
172 | ||
173 | static inline uint64_t vtd_slpt_level_page_mask(uint32_t level) | |
174 | { | |
175 | return ~((1ULL << vtd_slpt_level_shift(level)) - 1); | |
176 | } | |
177 | ||
b5a280c0 LT |
178 | static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value, |
179 | gpointer user_data) | |
180 | { | |
181 | VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value; | |
182 | VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data; | |
d66b969b JW |
183 | uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask; |
184 | uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K; | |
b5a280c0 | 185 | return (entry->domain_id == info->domain_id) && |
d66b969b JW |
186 | (((entry->gfn & info->mask) == gfn) || |
187 | (entry->gfn == gfn_tlb)); | |
b5a280c0 LT |
188 | } |
189 | ||
d92fa2dc LT |
190 | /* Reset all the gen of VTDAddressSpace to zero and set the gen of |
191 | * IntelIOMMUState to 1. | |
192 | */ | |
193 | static void vtd_reset_context_cache(IntelIOMMUState *s) | |
194 | { | |
d92fa2dc | 195 | VTDAddressSpace *vtd_as; |
7df953bd KO |
196 | VTDBus *vtd_bus; |
197 | GHashTableIter bus_it; | |
d92fa2dc LT |
198 | uint32_t devfn_it; |
199 | ||
7df953bd KO |
200 | g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr); |
201 | ||
d92fa2dc | 202 | VTD_DPRINTF(CACHE, "global context_cache_gen=1"); |
7df953bd | 203 | while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) { |
04af0e18 | 204 | for (devfn_it = 0; devfn_it < X86_IOMMU_PCI_DEVFN_MAX; ++devfn_it) { |
7df953bd | 205 | vtd_as = vtd_bus->dev_as[devfn_it]; |
d92fa2dc LT |
206 | if (!vtd_as) { |
207 | continue; | |
208 | } | |
209 | vtd_as->context_cache_entry.context_cache_gen = 0; | |
210 | } | |
211 | } | |
212 | s->context_cache_gen = 1; | |
213 | } | |
214 | ||
b5a280c0 LT |
215 | static void vtd_reset_iotlb(IntelIOMMUState *s) |
216 | { | |
217 | assert(s->iotlb); | |
218 | g_hash_table_remove_all(s->iotlb); | |
219 | } | |
220 | ||
d66b969b JW |
221 | static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint8_t source_id, |
222 | uint32_t level) | |
223 | { | |
224 | return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) | | |
225 | ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT); | |
226 | } | |
227 | ||
228 | static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level) | |
229 | { | |
230 | return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K; | |
231 | } | |
232 | ||
b5a280c0 LT |
233 | static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id, |
234 | hwaddr addr) | |
235 | { | |
d66b969b | 236 | VTDIOTLBEntry *entry; |
b5a280c0 | 237 | uint64_t key; |
d66b969b JW |
238 | int level; |
239 | ||
240 | for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) { | |
241 | key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level), | |
242 | source_id, level); | |
243 | entry = g_hash_table_lookup(s->iotlb, &key); | |
244 | if (entry) { | |
245 | goto out; | |
246 | } | |
247 | } | |
b5a280c0 | 248 | |
d66b969b JW |
249 | out: |
250 | return entry; | |
b5a280c0 LT |
251 | } |
252 | ||
253 | static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id, | |
254 | uint16_t domain_id, hwaddr addr, uint64_t slpte, | |
d66b969b JW |
255 | bool read_flags, bool write_flags, |
256 | uint32_t level) | |
b5a280c0 LT |
257 | { |
258 | VTDIOTLBEntry *entry = g_malloc(sizeof(*entry)); | |
259 | uint64_t *key = g_malloc(sizeof(*key)); | |
d66b969b | 260 | uint64_t gfn = vtd_get_iotlb_gfn(addr, level); |
b5a280c0 LT |
261 | |
262 | VTD_DPRINTF(CACHE, "update iotlb sid 0x%"PRIx16 " gpa 0x%"PRIx64 | |
263 | " slpte 0x%"PRIx64 " did 0x%"PRIx16, source_id, addr, slpte, | |
264 | domain_id); | |
265 | if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) { | |
266 | VTD_DPRINTF(CACHE, "iotlb exceeds size limit, forced to reset"); | |
267 | vtd_reset_iotlb(s); | |
268 | } | |
269 | ||
270 | entry->gfn = gfn; | |
271 | entry->domain_id = domain_id; | |
272 | entry->slpte = slpte; | |
273 | entry->read_flags = read_flags; | |
274 | entry->write_flags = write_flags; | |
d66b969b JW |
275 | entry->mask = vtd_slpt_level_page_mask(level); |
276 | *key = vtd_get_iotlb_key(gfn, source_id, level); | |
b5a280c0 LT |
277 | g_hash_table_replace(s->iotlb, key, entry); |
278 | } | |
279 | ||
1da12ec4 LT |
280 | /* Given the reg addr of both the message data and address, generate an |
281 | * interrupt via MSI. | |
282 | */ | |
283 | static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg, | |
284 | hwaddr mesg_data_reg) | |
285 | { | |
32946019 | 286 | MSIMessage msi; |
1da12ec4 LT |
287 | |
288 | assert(mesg_data_reg < DMAR_REG_SIZE); | |
289 | assert(mesg_addr_reg < DMAR_REG_SIZE); | |
290 | ||
32946019 RK |
291 | msi.address = vtd_get_long_raw(s, mesg_addr_reg); |
292 | msi.data = vtd_get_long_raw(s, mesg_data_reg); | |
1da12ec4 | 293 | |
32946019 RK |
294 | VTD_DPRINTF(FLOG, "msi: addr 0x%"PRIx64 " data 0x%"PRIx32, |
295 | msi.address, msi.data); | |
296 | apic_get_class()->send_msi(&msi); | |
1da12ec4 LT |
297 | } |
298 | ||
299 | /* Generate a fault event to software via MSI if conditions are met. | |
300 | * Notice that the value of FSTS_REG being passed to it should be the one | |
301 | * before any update. | |
302 | */ | |
303 | static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts) | |
304 | { | |
305 | if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO || | |
306 | pre_fsts & VTD_FSTS_IQE) { | |
307 | VTD_DPRINTF(FLOG, "there are previous interrupt conditions " | |
308 | "to be serviced by software, fault event is not generated " | |
309 | "(FSTS_REG 0x%"PRIx32 ")", pre_fsts); | |
310 | return; | |
311 | } | |
312 | vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP); | |
313 | if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) { | |
314 | VTD_DPRINTF(FLOG, "Interrupt Mask set, fault event is not generated"); | |
315 | } else { | |
316 | vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG); | |
317 | vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); | |
318 | } | |
319 | } | |
320 | ||
321 | /* Check if the Fault (F) field of the Fault Recording Register referenced by | |
322 | * @index is Set. | |
323 | */ | |
324 | static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index) | |
325 | { | |
326 | /* Each reg is 128-bit */ | |
327 | hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); | |
328 | addr += 8; /* Access the high 64-bit half */ | |
329 | ||
330 | assert(index < DMAR_FRCD_REG_NR); | |
331 | ||
332 | return vtd_get_quad_raw(s, addr) & VTD_FRCD_F; | |
333 | } | |
334 | ||
335 | /* Update the PPF field of Fault Status Register. | |
336 | * Should be called whenever change the F field of any fault recording | |
337 | * registers. | |
338 | */ | |
339 | static void vtd_update_fsts_ppf(IntelIOMMUState *s) | |
340 | { | |
341 | uint32_t i; | |
342 | uint32_t ppf_mask = 0; | |
343 | ||
344 | for (i = 0; i < DMAR_FRCD_REG_NR; i++) { | |
345 | if (vtd_is_frcd_set(s, i)) { | |
346 | ppf_mask = VTD_FSTS_PPF; | |
347 | break; | |
348 | } | |
349 | } | |
350 | vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask); | |
351 | VTD_DPRINTF(FLOG, "set PPF of FSTS_REG to %d", ppf_mask ? 1 : 0); | |
352 | } | |
353 | ||
354 | static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index) | |
355 | { | |
356 | /* Each reg is 128-bit */ | |
357 | hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); | |
358 | addr += 8; /* Access the high 64-bit half */ | |
359 | ||
360 | assert(index < DMAR_FRCD_REG_NR); | |
361 | ||
362 | vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F); | |
363 | vtd_update_fsts_ppf(s); | |
364 | } | |
365 | ||
366 | /* Must not update F field now, should be done later */ | |
367 | static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index, | |
368 | uint16_t source_id, hwaddr addr, | |
369 | VTDFaultReason fault, bool is_write) | |
370 | { | |
371 | uint64_t hi = 0, lo; | |
372 | hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4); | |
373 | ||
374 | assert(index < DMAR_FRCD_REG_NR); | |
375 | ||
376 | lo = VTD_FRCD_FI(addr); | |
377 | hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault); | |
378 | if (!is_write) { | |
379 | hi |= VTD_FRCD_T; | |
380 | } | |
381 | vtd_set_quad_raw(s, frcd_reg_addr, lo); | |
382 | vtd_set_quad_raw(s, frcd_reg_addr + 8, hi); | |
383 | VTD_DPRINTF(FLOG, "record to FRCD_REG #%"PRIu16 ": hi 0x%"PRIx64 | |
384 | ", lo 0x%"PRIx64, index, hi, lo); | |
385 | } | |
386 | ||
387 | /* Try to collapse multiple pending faults from the same requester */ | |
388 | static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id) | |
389 | { | |
390 | uint32_t i; | |
391 | uint64_t frcd_reg; | |
392 | hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */ | |
393 | ||
394 | for (i = 0; i < DMAR_FRCD_REG_NR; i++) { | |
395 | frcd_reg = vtd_get_quad_raw(s, addr); | |
396 | VTD_DPRINTF(FLOG, "frcd_reg #%d 0x%"PRIx64, i, frcd_reg); | |
397 | if ((frcd_reg & VTD_FRCD_F) && | |
398 | ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) { | |
399 | return true; | |
400 | } | |
401 | addr += 16; /* 128-bit for each */ | |
402 | } | |
403 | return false; | |
404 | } | |
405 | ||
406 | /* Log and report an DMAR (address translation) fault to software */ | |
407 | static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id, | |
408 | hwaddr addr, VTDFaultReason fault, | |
409 | bool is_write) | |
410 | { | |
411 | uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); | |
412 | ||
413 | assert(fault < VTD_FR_MAX); | |
414 | ||
415 | if (fault == VTD_FR_RESERVED_ERR) { | |
416 | /* This is not a normal fault reason case. Drop it. */ | |
417 | return; | |
418 | } | |
419 | VTD_DPRINTF(FLOG, "sid 0x%"PRIx16 ", fault %d, addr 0x%"PRIx64 | |
420 | ", is_write %d", source_id, fault, addr, is_write); | |
421 | if (fsts_reg & VTD_FSTS_PFO) { | |
422 | VTD_DPRINTF(FLOG, "new fault is not recorded due to " | |
423 | "Primary Fault Overflow"); | |
424 | return; | |
425 | } | |
426 | if (vtd_try_collapse_fault(s, source_id)) { | |
427 | VTD_DPRINTF(FLOG, "new fault is not recorded due to " | |
428 | "compression of faults"); | |
429 | return; | |
430 | } | |
431 | if (vtd_is_frcd_set(s, s->next_frcd_reg)) { | |
432 | VTD_DPRINTF(FLOG, "Primary Fault Overflow and " | |
433 | "new fault is not recorded, set PFO field"); | |
434 | vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO); | |
435 | return; | |
436 | } | |
437 | ||
438 | vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write); | |
439 | ||
440 | if (fsts_reg & VTD_FSTS_PPF) { | |
441 | VTD_DPRINTF(FLOG, "there are pending faults already, " | |
442 | "fault event is not generated"); | |
443 | vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); | |
444 | s->next_frcd_reg++; | |
445 | if (s->next_frcd_reg == DMAR_FRCD_REG_NR) { | |
446 | s->next_frcd_reg = 0; | |
447 | } | |
448 | } else { | |
449 | vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK, | |
450 | VTD_FSTS_FRI(s->next_frcd_reg)); | |
451 | vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */ | |
452 | s->next_frcd_reg++; | |
453 | if (s->next_frcd_reg == DMAR_FRCD_REG_NR) { | |
454 | s->next_frcd_reg = 0; | |
455 | } | |
456 | /* This case actually cause the PPF to be Set. | |
457 | * So generate fault event (interrupt). | |
458 | */ | |
459 | vtd_generate_fault_event(s, fsts_reg); | |
460 | } | |
461 | } | |
462 | ||
ed7b8fbc LT |
463 | /* Handle Invalidation Queue Errors of queued invalidation interface error |
464 | * conditions. | |
465 | */ | |
466 | static void vtd_handle_inv_queue_error(IntelIOMMUState *s) | |
467 | { | |
468 | uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); | |
469 | ||
470 | vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE); | |
471 | vtd_generate_fault_event(s, fsts_reg); | |
472 | } | |
473 | ||
474 | /* Set the IWC field and try to generate an invalidation completion interrupt */ | |
475 | static void vtd_generate_completion_event(IntelIOMMUState *s) | |
476 | { | |
477 | VTD_DPRINTF(INV, "completes an invalidation wait command with " | |
478 | "Interrupt Flag"); | |
479 | if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) { | |
480 | VTD_DPRINTF(INV, "there is a previous interrupt condition to be " | |
481 | "serviced by software, " | |
482 | "new invalidation event is not generated"); | |
483 | return; | |
484 | } | |
485 | vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC); | |
486 | vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP); | |
487 | if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) { | |
488 | VTD_DPRINTF(INV, "IM filed in IECTL_REG is set, new invalidation " | |
489 | "event is not generated"); | |
490 | return; | |
491 | } else { | |
492 | /* Generate the interrupt event */ | |
493 | vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG); | |
494 | vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); | |
495 | } | |
496 | } | |
497 | ||
1da12ec4 LT |
498 | static inline bool vtd_root_entry_present(VTDRootEntry *root) |
499 | { | |
500 | return root->val & VTD_ROOT_ENTRY_P; | |
501 | } | |
502 | ||
503 | static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index, | |
504 | VTDRootEntry *re) | |
505 | { | |
506 | dma_addr_t addr; | |
507 | ||
508 | addr = s->root + index * sizeof(*re); | |
509 | if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) { | |
510 | VTD_DPRINTF(GENERAL, "error: fail to access root-entry at 0x%"PRIx64 | |
511 | " + %"PRIu8, s->root, index); | |
512 | re->val = 0; | |
513 | return -VTD_FR_ROOT_TABLE_INV; | |
514 | } | |
515 | re->val = le64_to_cpu(re->val); | |
516 | return 0; | |
517 | } | |
518 | ||
519 | static inline bool vtd_context_entry_present(VTDContextEntry *context) | |
520 | { | |
521 | return context->lo & VTD_CONTEXT_ENTRY_P; | |
522 | } | |
523 | ||
524 | static int vtd_get_context_entry_from_root(VTDRootEntry *root, uint8_t index, | |
525 | VTDContextEntry *ce) | |
526 | { | |
527 | dma_addr_t addr; | |
528 | ||
529 | if (!vtd_root_entry_present(root)) { | |
530 | VTD_DPRINTF(GENERAL, "error: root-entry is not present"); | |
531 | return -VTD_FR_ROOT_ENTRY_P; | |
532 | } | |
533 | addr = (root->val & VTD_ROOT_ENTRY_CTP) + index * sizeof(*ce); | |
534 | if (dma_memory_read(&address_space_memory, addr, ce, sizeof(*ce))) { | |
535 | VTD_DPRINTF(GENERAL, "error: fail to access context-entry at 0x%"PRIx64 | |
536 | " + %"PRIu8, | |
537 | (uint64_t)(root->val & VTD_ROOT_ENTRY_CTP), index); | |
538 | return -VTD_FR_CONTEXT_TABLE_INV; | |
539 | } | |
540 | ce->lo = le64_to_cpu(ce->lo); | |
541 | ce->hi = le64_to_cpu(ce->hi); | |
542 | return 0; | |
543 | } | |
544 | ||
545 | static inline dma_addr_t vtd_get_slpt_base_from_context(VTDContextEntry *ce) | |
546 | { | |
547 | return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR; | |
548 | } | |
549 | ||
1da12ec4 LT |
550 | static inline uint64_t vtd_get_slpte_addr(uint64_t slpte) |
551 | { | |
552 | return slpte & VTD_SL_PT_BASE_ADDR_MASK; | |
553 | } | |
554 | ||
555 | /* Whether the pte indicates the address of the page frame */ | |
556 | static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level) | |
557 | { | |
558 | return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK); | |
559 | } | |
560 | ||
561 | /* Get the content of a spte located in @base_addr[@index] */ | |
562 | static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index) | |
563 | { | |
564 | uint64_t slpte; | |
565 | ||
566 | assert(index < VTD_SL_PT_ENTRY_NR); | |
567 | ||
568 | if (dma_memory_read(&address_space_memory, | |
569 | base_addr + index * sizeof(slpte), &slpte, | |
570 | sizeof(slpte))) { | |
571 | slpte = (uint64_t)-1; | |
572 | return slpte; | |
573 | } | |
574 | slpte = le64_to_cpu(slpte); | |
575 | return slpte; | |
576 | } | |
577 | ||
578 | /* Given a gpa and the level of paging structure, return the offset of current | |
579 | * level. | |
580 | */ | |
581 | static inline uint32_t vtd_gpa_level_offset(uint64_t gpa, uint32_t level) | |
582 | { | |
583 | return (gpa >> vtd_slpt_level_shift(level)) & | |
584 | ((1ULL << VTD_SL_LEVEL_BITS) - 1); | |
585 | } | |
586 | ||
587 | /* Check Capability Register to see if the @level of page-table is supported */ | |
588 | static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level) | |
589 | { | |
590 | return VTD_CAP_SAGAW_MASK & s->cap & | |
591 | (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT)); | |
592 | } | |
593 | ||
594 | /* Get the page-table level that hardware should use for the second-level | |
595 | * page-table walk from the Address Width field of context-entry. | |
596 | */ | |
597 | static inline uint32_t vtd_get_level_from_context_entry(VTDContextEntry *ce) | |
598 | { | |
599 | return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW); | |
600 | } | |
601 | ||
602 | static inline uint32_t vtd_get_agaw_from_context_entry(VTDContextEntry *ce) | |
603 | { | |
604 | return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9; | |
605 | } | |
606 | ||
607 | static const uint64_t vtd_paging_entry_rsvd_field[] = { | |
608 | [0] = ~0ULL, | |
609 | /* For not large page */ | |
610 | [1] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), | |
611 | [2] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), | |
612 | [3] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), | |
613 | [4] = 0x880ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), | |
614 | /* For large page */ | |
615 | [5] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), | |
616 | [6] = 0x1ff800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), | |
617 | [7] = 0x3ffff800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), | |
618 | [8] = 0x880ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM), | |
619 | }; | |
620 | ||
621 | static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level) | |
622 | { | |
623 | if (slpte & VTD_SL_PT_PAGE_SIZE_MASK) { | |
624 | /* Maybe large page */ | |
625 | return slpte & vtd_paging_entry_rsvd_field[level + 4]; | |
626 | } else { | |
627 | return slpte & vtd_paging_entry_rsvd_field[level]; | |
628 | } | |
629 | } | |
630 | ||
631 | /* Given the @gpa, get relevant @slptep. @slpte_level will be the last level | |
632 | * of the translation, can be used for deciding the size of large page. | |
633 | */ | |
634 | static int vtd_gpa_to_slpte(VTDContextEntry *ce, uint64_t gpa, bool is_write, | |
635 | uint64_t *slptep, uint32_t *slpte_level, | |
636 | bool *reads, bool *writes) | |
637 | { | |
638 | dma_addr_t addr = vtd_get_slpt_base_from_context(ce); | |
639 | uint32_t level = vtd_get_level_from_context_entry(ce); | |
640 | uint32_t offset; | |
641 | uint64_t slpte; | |
642 | uint32_t ce_agaw = vtd_get_agaw_from_context_entry(ce); | |
643 | uint64_t access_right_check; | |
644 | ||
645 | /* Check if @gpa is above 2^X-1, where X is the minimum of MGAW in CAP_REG | |
646 | * and AW in context-entry. | |
647 | */ | |
648 | if (gpa & ~((1ULL << MIN(ce_agaw, VTD_MGAW)) - 1)) { | |
649 | VTD_DPRINTF(GENERAL, "error: gpa 0x%"PRIx64 " exceeds limits", gpa); | |
650 | return -VTD_FR_ADDR_BEYOND_MGAW; | |
651 | } | |
652 | ||
653 | /* FIXME: what is the Atomics request here? */ | |
654 | access_right_check = is_write ? VTD_SL_W : VTD_SL_R; | |
655 | ||
656 | while (true) { | |
657 | offset = vtd_gpa_level_offset(gpa, level); | |
658 | slpte = vtd_get_slpte(addr, offset); | |
659 | ||
660 | if (slpte == (uint64_t)-1) { | |
661 | VTD_DPRINTF(GENERAL, "error: fail to access second-level paging " | |
662 | "entry at level %"PRIu32 " for gpa 0x%"PRIx64, | |
663 | level, gpa); | |
664 | if (level == vtd_get_level_from_context_entry(ce)) { | |
665 | /* Invalid programming of context-entry */ | |
666 | return -VTD_FR_CONTEXT_ENTRY_INV; | |
667 | } else { | |
668 | return -VTD_FR_PAGING_ENTRY_INV; | |
669 | } | |
670 | } | |
671 | *reads = (*reads) && (slpte & VTD_SL_R); | |
672 | *writes = (*writes) && (slpte & VTD_SL_W); | |
673 | if (!(slpte & access_right_check)) { | |
674 | VTD_DPRINTF(GENERAL, "error: lack of %s permission for " | |
675 | "gpa 0x%"PRIx64 " slpte 0x%"PRIx64, | |
676 | (is_write ? "write" : "read"), gpa, slpte); | |
677 | return is_write ? -VTD_FR_WRITE : -VTD_FR_READ; | |
678 | } | |
679 | if (vtd_slpte_nonzero_rsvd(slpte, level)) { | |
680 | VTD_DPRINTF(GENERAL, "error: non-zero reserved field in second " | |
681 | "level paging entry level %"PRIu32 " slpte 0x%"PRIx64, | |
682 | level, slpte); | |
683 | return -VTD_FR_PAGING_ENTRY_RSVD; | |
684 | } | |
685 | ||
686 | if (vtd_is_last_slpte(slpte, level)) { | |
687 | *slptep = slpte; | |
688 | *slpte_level = level; | |
689 | return 0; | |
690 | } | |
691 | addr = vtd_get_slpte_addr(slpte); | |
692 | level--; | |
693 | } | |
694 | } | |
695 | ||
696 | /* Map a device to its corresponding domain (context-entry) */ | |
697 | static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num, | |
698 | uint8_t devfn, VTDContextEntry *ce) | |
699 | { | |
700 | VTDRootEntry re; | |
701 | int ret_fr; | |
702 | ||
703 | ret_fr = vtd_get_root_entry(s, bus_num, &re); | |
704 | if (ret_fr) { | |
705 | return ret_fr; | |
706 | } | |
707 | ||
708 | if (!vtd_root_entry_present(&re)) { | |
709 | VTD_DPRINTF(GENERAL, "error: root-entry #%"PRIu8 " is not present", | |
710 | bus_num); | |
711 | return -VTD_FR_ROOT_ENTRY_P; | |
712 | } else if (re.rsvd || (re.val & VTD_ROOT_ENTRY_RSVD)) { | |
713 | VTD_DPRINTF(GENERAL, "error: non-zero reserved field in root-entry " | |
714 | "hi 0x%"PRIx64 " lo 0x%"PRIx64, re.rsvd, re.val); | |
715 | return -VTD_FR_ROOT_ENTRY_RSVD; | |
716 | } | |
717 | ||
718 | ret_fr = vtd_get_context_entry_from_root(&re, devfn, ce); | |
719 | if (ret_fr) { | |
720 | return ret_fr; | |
721 | } | |
722 | ||
723 | if (!vtd_context_entry_present(ce)) { | |
724 | VTD_DPRINTF(GENERAL, | |
725 | "error: context-entry #%"PRIu8 "(bus #%"PRIu8 ") " | |
726 | "is not present", devfn, bus_num); | |
727 | return -VTD_FR_CONTEXT_ENTRY_P; | |
728 | } else if ((ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI) || | |
729 | (ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO)) { | |
730 | VTD_DPRINTF(GENERAL, | |
731 | "error: non-zero reserved field in context-entry " | |
732 | "hi 0x%"PRIx64 " lo 0x%"PRIx64, ce->hi, ce->lo); | |
733 | return -VTD_FR_CONTEXT_ENTRY_RSVD; | |
734 | } | |
735 | /* Check if the programming of context-entry is valid */ | |
736 | if (!vtd_is_level_supported(s, vtd_get_level_from_context_entry(ce))) { | |
737 | VTD_DPRINTF(GENERAL, "error: unsupported Address Width value in " | |
738 | "context-entry hi 0x%"PRIx64 " lo 0x%"PRIx64, | |
739 | ce->hi, ce->lo); | |
740 | return -VTD_FR_CONTEXT_ENTRY_INV; | |
741 | } else if (ce->lo & VTD_CONTEXT_ENTRY_TT) { | |
742 | VTD_DPRINTF(GENERAL, "error: unsupported Translation Type in " | |
743 | "context-entry hi 0x%"PRIx64 " lo 0x%"PRIx64, | |
744 | ce->hi, ce->lo); | |
745 | return -VTD_FR_CONTEXT_ENTRY_INV; | |
746 | } | |
747 | return 0; | |
748 | } | |
749 | ||
750 | static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn) | |
751 | { | |
752 | return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL); | |
753 | } | |
754 | ||
755 | static const bool vtd_qualified_faults[] = { | |
756 | [VTD_FR_RESERVED] = false, | |
757 | [VTD_FR_ROOT_ENTRY_P] = false, | |
758 | [VTD_FR_CONTEXT_ENTRY_P] = true, | |
759 | [VTD_FR_CONTEXT_ENTRY_INV] = true, | |
760 | [VTD_FR_ADDR_BEYOND_MGAW] = true, | |
761 | [VTD_FR_WRITE] = true, | |
762 | [VTD_FR_READ] = true, | |
763 | [VTD_FR_PAGING_ENTRY_INV] = true, | |
764 | [VTD_FR_ROOT_TABLE_INV] = false, | |
765 | [VTD_FR_CONTEXT_TABLE_INV] = false, | |
766 | [VTD_FR_ROOT_ENTRY_RSVD] = false, | |
767 | [VTD_FR_PAGING_ENTRY_RSVD] = true, | |
768 | [VTD_FR_CONTEXT_ENTRY_TT] = true, | |
769 | [VTD_FR_RESERVED_ERR] = false, | |
770 | [VTD_FR_MAX] = false, | |
771 | }; | |
772 | ||
773 | /* To see if a fault condition is "qualified", which is reported to software | |
774 | * only if the FPD field in the context-entry used to process the faulting | |
775 | * request is 0. | |
776 | */ | |
777 | static inline bool vtd_is_qualified_fault(VTDFaultReason fault) | |
778 | { | |
779 | return vtd_qualified_faults[fault]; | |
780 | } | |
781 | ||
782 | static inline bool vtd_is_interrupt_addr(hwaddr addr) | |
783 | { | |
784 | return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST; | |
785 | } | |
786 | ||
787 | /* Map dev to context-entry then do a paging-structures walk to do a iommu | |
788 | * translation. | |
79e2b9ae PB |
789 | * |
790 | * Called from RCU critical section. | |
791 | * | |
1da12ec4 LT |
792 | * @bus_num: The bus number |
793 | * @devfn: The devfn, which is the combined of device and function number | |
794 | * @is_write: The access is a write operation | |
795 | * @entry: IOMMUTLBEntry that contain the addr to be translated and result | |
796 | */ | |
7df953bd | 797 | static void vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus, |
1da12ec4 LT |
798 | uint8_t devfn, hwaddr addr, bool is_write, |
799 | IOMMUTLBEntry *entry) | |
800 | { | |
d92fa2dc | 801 | IntelIOMMUState *s = vtd_as->iommu_state; |
1da12ec4 | 802 | VTDContextEntry ce; |
7df953bd | 803 | uint8_t bus_num = pci_bus_num(bus); |
d92fa2dc | 804 | VTDContextCacheEntry *cc_entry = &vtd_as->context_cache_entry; |
d66b969b | 805 | uint64_t slpte, page_mask; |
1da12ec4 LT |
806 | uint32_t level; |
807 | uint16_t source_id = vtd_make_source_id(bus_num, devfn); | |
808 | int ret_fr; | |
809 | bool is_fpd_set = false; | |
810 | bool reads = true; | |
811 | bool writes = true; | |
b5a280c0 | 812 | VTDIOTLBEntry *iotlb_entry; |
1da12ec4 LT |
813 | |
814 | /* Check if the request is in interrupt address range */ | |
815 | if (vtd_is_interrupt_addr(addr)) { | |
816 | if (is_write) { | |
817 | /* FIXME: since we don't know the length of the access here, we | |
818 | * treat Non-DWORD length write requests without PASID as | |
819 | * interrupt requests, too. Withoud interrupt remapping support, | |
820 | * we just use 1:1 mapping. | |
821 | */ | |
822 | VTD_DPRINTF(MMU, "write request to interrupt address " | |
823 | "gpa 0x%"PRIx64, addr); | |
824 | entry->iova = addr & VTD_PAGE_MASK_4K; | |
825 | entry->translated_addr = addr & VTD_PAGE_MASK_4K; | |
826 | entry->addr_mask = ~VTD_PAGE_MASK_4K; | |
827 | entry->perm = IOMMU_WO; | |
828 | return; | |
829 | } else { | |
830 | VTD_DPRINTF(GENERAL, "error: read request from interrupt address " | |
831 | "gpa 0x%"PRIx64, addr); | |
832 | vtd_report_dmar_fault(s, source_id, addr, VTD_FR_READ, is_write); | |
833 | return; | |
834 | } | |
835 | } | |
b5a280c0 LT |
836 | /* Try to fetch slpte form IOTLB */ |
837 | iotlb_entry = vtd_lookup_iotlb(s, source_id, addr); | |
838 | if (iotlb_entry) { | |
839 | VTD_DPRINTF(CACHE, "hit iotlb sid 0x%"PRIx16 " gpa 0x%"PRIx64 | |
840 | " slpte 0x%"PRIx64 " did 0x%"PRIx16, source_id, addr, | |
841 | iotlb_entry->slpte, iotlb_entry->domain_id); | |
842 | slpte = iotlb_entry->slpte; | |
843 | reads = iotlb_entry->read_flags; | |
844 | writes = iotlb_entry->write_flags; | |
d66b969b | 845 | page_mask = iotlb_entry->mask; |
b5a280c0 LT |
846 | goto out; |
847 | } | |
d92fa2dc LT |
848 | /* Try to fetch context-entry from cache first */ |
849 | if (cc_entry->context_cache_gen == s->context_cache_gen) { | |
850 | VTD_DPRINTF(CACHE, "hit context-cache bus %d devfn %d " | |
851 | "(hi %"PRIx64 " lo %"PRIx64 " gen %"PRIu32 ")", | |
852 | bus_num, devfn, cc_entry->context_entry.hi, | |
853 | cc_entry->context_entry.lo, cc_entry->context_cache_gen); | |
854 | ce = cc_entry->context_entry; | |
855 | is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD; | |
856 | } else { | |
857 | ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce); | |
858 | is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD; | |
859 | if (ret_fr) { | |
860 | ret_fr = -ret_fr; | |
861 | if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) { | |
862 | VTD_DPRINTF(FLOG, "fault processing is disabled for DMA " | |
863 | "requests through this context-entry " | |
864 | "(with FPD Set)"); | |
865 | } else { | |
866 | vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write); | |
867 | } | |
868 | return; | |
1da12ec4 | 869 | } |
d92fa2dc LT |
870 | /* Update context-cache */ |
871 | VTD_DPRINTF(CACHE, "update context-cache bus %d devfn %d " | |
872 | "(hi %"PRIx64 " lo %"PRIx64 " gen %"PRIu32 "->%"PRIu32 ")", | |
873 | bus_num, devfn, ce.hi, ce.lo, | |
874 | cc_entry->context_cache_gen, s->context_cache_gen); | |
875 | cc_entry->context_entry = ce; | |
876 | cc_entry->context_cache_gen = s->context_cache_gen; | |
1da12ec4 LT |
877 | } |
878 | ||
879 | ret_fr = vtd_gpa_to_slpte(&ce, addr, is_write, &slpte, &level, | |
880 | &reads, &writes); | |
881 | if (ret_fr) { | |
882 | ret_fr = -ret_fr; | |
883 | if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) { | |
884 | VTD_DPRINTF(FLOG, "fault processing is disabled for DMA requests " | |
885 | "through this context-entry (with FPD Set)"); | |
886 | } else { | |
887 | vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write); | |
888 | } | |
889 | return; | |
890 | } | |
891 | ||
d66b969b | 892 | page_mask = vtd_slpt_level_page_mask(level); |
b5a280c0 | 893 | vtd_update_iotlb(s, source_id, VTD_CONTEXT_ENTRY_DID(ce.hi), addr, slpte, |
d66b969b | 894 | reads, writes, level); |
b5a280c0 | 895 | out: |
d66b969b JW |
896 | entry->iova = addr & page_mask; |
897 | entry->translated_addr = vtd_get_slpte_addr(slpte) & page_mask; | |
898 | entry->addr_mask = ~page_mask; | |
1da12ec4 LT |
899 | entry->perm = (writes ? 2 : 0) + (reads ? 1 : 0); |
900 | } | |
901 | ||
902 | static void vtd_root_table_setup(IntelIOMMUState *s) | |
903 | { | |
904 | s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG); | |
905 | s->root_extended = s->root & VTD_RTADDR_RTT; | |
906 | s->root &= VTD_RTADDR_ADDR_MASK; | |
907 | ||
908 | VTD_DPRINTF(CSR, "root_table addr 0x%"PRIx64 " %s", s->root, | |
909 | (s->root_extended ? "(extended)" : "")); | |
910 | } | |
911 | ||
02a2cbc8 PX |
912 | static void vtd_iec_notify_all(IntelIOMMUState *s, bool global, |
913 | uint32_t index, uint32_t mask) | |
914 | { | |
915 | x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask); | |
916 | } | |
917 | ||
a5861439 PX |
918 | static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s) |
919 | { | |
920 | uint64_t value = 0; | |
921 | value = vtd_get_quad_raw(s, DMAR_IRTA_REG); | |
922 | s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1); | |
923 | s->intr_root = value & VTD_IRTA_ADDR_MASK; | |
28589311 | 924 | s->intr_eime = value & VTD_IRTA_EIME; |
a5861439 | 925 | |
02a2cbc8 PX |
926 | /* Notify global invalidation */ |
927 | vtd_iec_notify_all(s, true, 0, 0); | |
a5861439 PX |
928 | |
929 | VTD_DPRINTF(CSR, "int remap table addr 0x%"PRIx64 " size %"PRIu32, | |
930 | s->intr_root, s->intr_size); | |
931 | } | |
932 | ||
d92fa2dc LT |
933 | static void vtd_context_global_invalidate(IntelIOMMUState *s) |
934 | { | |
935 | s->context_cache_gen++; | |
936 | if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) { | |
937 | vtd_reset_context_cache(s); | |
938 | } | |
939 | } | |
940 | ||
7df953bd KO |
941 | |
942 | /* Find the VTD address space currently associated with a given bus number, | |
943 | */ | |
944 | static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num) | |
945 | { | |
946 | VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num]; | |
947 | if (!vtd_bus) { | |
948 | /* Iterate over the registered buses to find the one | |
949 | * which currently hold this bus number, and update the bus_num lookup table: | |
950 | */ | |
951 | GHashTableIter iter; | |
952 | ||
953 | g_hash_table_iter_init(&iter, s->vtd_as_by_busptr); | |
954 | while (g_hash_table_iter_next (&iter, NULL, (void**)&vtd_bus)) { | |
955 | if (pci_bus_num(vtd_bus->bus) == bus_num) { | |
956 | s->vtd_as_by_bus_num[bus_num] = vtd_bus; | |
957 | return vtd_bus; | |
958 | } | |
959 | } | |
960 | } | |
961 | return vtd_bus; | |
962 | } | |
963 | ||
d92fa2dc LT |
964 | /* Do a context-cache device-selective invalidation. |
965 | * @func_mask: FM field after shifting | |
966 | */ | |
967 | static void vtd_context_device_invalidate(IntelIOMMUState *s, | |
968 | uint16_t source_id, | |
969 | uint16_t func_mask) | |
970 | { | |
971 | uint16_t mask; | |
7df953bd | 972 | VTDBus *vtd_bus; |
d92fa2dc LT |
973 | VTDAddressSpace *vtd_as; |
974 | uint16_t devfn; | |
975 | uint16_t devfn_it; | |
976 | ||
977 | switch (func_mask & 3) { | |
978 | case 0: | |
979 | mask = 0; /* No bits in the SID field masked */ | |
980 | break; | |
981 | case 1: | |
982 | mask = 4; /* Mask bit 2 in the SID field */ | |
983 | break; | |
984 | case 2: | |
985 | mask = 6; /* Mask bit 2:1 in the SID field */ | |
986 | break; | |
987 | case 3: | |
988 | mask = 7; /* Mask bit 2:0 in the SID field */ | |
989 | break; | |
990 | } | |
991 | VTD_DPRINTF(INV, "device-selective invalidation source 0x%"PRIx16 | |
992 | " mask %"PRIu16, source_id, mask); | |
7df953bd KO |
993 | vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id)); |
994 | if (vtd_bus) { | |
d92fa2dc | 995 | devfn = VTD_SID_TO_DEVFN(source_id); |
04af0e18 | 996 | for (devfn_it = 0; devfn_it < X86_IOMMU_PCI_DEVFN_MAX; ++devfn_it) { |
7df953bd | 997 | vtd_as = vtd_bus->dev_as[devfn_it]; |
d92fa2dc LT |
998 | if (vtd_as && ((devfn_it & mask) == (devfn & mask))) { |
999 | VTD_DPRINTF(INV, "invalidate context-cahce of devfn 0x%"PRIx16, | |
1000 | devfn_it); | |
1001 | vtd_as->context_cache_entry.context_cache_gen = 0; | |
1002 | } | |
1003 | } | |
1004 | } | |
1005 | } | |
1006 | ||
1da12ec4 LT |
1007 | /* Context-cache invalidation |
1008 | * Returns the Context Actual Invalidation Granularity. | |
1009 | * @val: the content of the CCMD_REG | |
1010 | */ | |
1011 | static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val) | |
1012 | { | |
1013 | uint64_t caig; | |
1014 | uint64_t type = val & VTD_CCMD_CIRG_MASK; | |
1015 | ||
1016 | switch (type) { | |
d92fa2dc LT |
1017 | case VTD_CCMD_DOMAIN_INVL: |
1018 | VTD_DPRINTF(INV, "domain-selective invalidation domain 0x%"PRIx16, | |
1019 | (uint16_t)VTD_CCMD_DID(val)); | |
1020 | /* Fall through */ | |
1da12ec4 | 1021 | case VTD_CCMD_GLOBAL_INVL: |
d92fa2dc | 1022 | VTD_DPRINTF(INV, "global invalidation"); |
1da12ec4 | 1023 | caig = VTD_CCMD_GLOBAL_INVL_A; |
d92fa2dc | 1024 | vtd_context_global_invalidate(s); |
1da12ec4 LT |
1025 | break; |
1026 | ||
1027 | case VTD_CCMD_DEVICE_INVL: | |
1da12ec4 | 1028 | caig = VTD_CCMD_DEVICE_INVL_A; |
d92fa2dc | 1029 | vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val)); |
1da12ec4 LT |
1030 | break; |
1031 | ||
1032 | default: | |
d92fa2dc | 1033 | VTD_DPRINTF(GENERAL, "error: invalid granularity"); |
1da12ec4 LT |
1034 | caig = 0; |
1035 | } | |
1036 | return caig; | |
1037 | } | |
1038 | ||
b5a280c0 LT |
1039 | static void vtd_iotlb_global_invalidate(IntelIOMMUState *s) |
1040 | { | |
1041 | vtd_reset_iotlb(s); | |
1042 | } | |
1043 | ||
1044 | static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id) | |
1045 | { | |
1046 | g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain, | |
1047 | &domain_id); | |
1048 | } | |
1049 | ||
1050 | static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id, | |
1051 | hwaddr addr, uint8_t am) | |
1052 | { | |
1053 | VTDIOTLBPageInvInfo info; | |
1054 | ||
1055 | assert(am <= VTD_MAMV); | |
1056 | info.domain_id = domain_id; | |
d66b969b | 1057 | info.addr = addr; |
b5a280c0 LT |
1058 | info.mask = ~((1 << am) - 1); |
1059 | g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info); | |
1060 | } | |
1061 | ||
1da12ec4 LT |
1062 | /* Flush IOTLB |
1063 | * Returns the IOTLB Actual Invalidation Granularity. | |
1064 | * @val: the content of the IOTLB_REG | |
1065 | */ | |
1066 | static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val) | |
1067 | { | |
1068 | uint64_t iaig; | |
1069 | uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK; | |
b5a280c0 LT |
1070 | uint16_t domain_id; |
1071 | hwaddr addr; | |
1072 | uint8_t am; | |
1da12ec4 LT |
1073 | |
1074 | switch (type) { | |
1075 | case VTD_TLB_GLOBAL_FLUSH: | |
b5a280c0 | 1076 | VTD_DPRINTF(INV, "global invalidation"); |
1da12ec4 | 1077 | iaig = VTD_TLB_GLOBAL_FLUSH_A; |
b5a280c0 | 1078 | vtd_iotlb_global_invalidate(s); |
1da12ec4 LT |
1079 | break; |
1080 | ||
1081 | case VTD_TLB_DSI_FLUSH: | |
b5a280c0 LT |
1082 | domain_id = VTD_TLB_DID(val); |
1083 | VTD_DPRINTF(INV, "domain-selective invalidation domain 0x%"PRIx16, | |
1084 | domain_id); | |
1da12ec4 | 1085 | iaig = VTD_TLB_DSI_FLUSH_A; |
b5a280c0 | 1086 | vtd_iotlb_domain_invalidate(s, domain_id); |
1da12ec4 LT |
1087 | break; |
1088 | ||
1089 | case VTD_TLB_PSI_FLUSH: | |
b5a280c0 LT |
1090 | domain_id = VTD_TLB_DID(val); |
1091 | addr = vtd_get_quad_raw(s, DMAR_IVA_REG); | |
1092 | am = VTD_IVA_AM(addr); | |
1093 | addr = VTD_IVA_ADDR(addr); | |
1094 | VTD_DPRINTF(INV, "page-selective invalidation domain 0x%"PRIx16 | |
1095 | " addr 0x%"PRIx64 " mask %"PRIu8, domain_id, addr, am); | |
1096 | if (am > VTD_MAMV) { | |
1097 | VTD_DPRINTF(GENERAL, "error: supported max address mask value is " | |
1098 | "%"PRIu8, (uint8_t)VTD_MAMV); | |
1099 | iaig = 0; | |
1100 | break; | |
1101 | } | |
1da12ec4 | 1102 | iaig = VTD_TLB_PSI_FLUSH_A; |
b5a280c0 | 1103 | vtd_iotlb_page_invalidate(s, domain_id, addr, am); |
1da12ec4 LT |
1104 | break; |
1105 | ||
1106 | default: | |
b5a280c0 | 1107 | VTD_DPRINTF(GENERAL, "error: invalid granularity"); |
1da12ec4 LT |
1108 | iaig = 0; |
1109 | } | |
1110 | return iaig; | |
1111 | } | |
1112 | ||
ed7b8fbc LT |
1113 | static inline bool vtd_queued_inv_enable_check(IntelIOMMUState *s) |
1114 | { | |
1115 | return s->iq_tail == 0; | |
1116 | } | |
1117 | ||
1118 | static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s) | |
1119 | { | |
1120 | return s->qi_enabled && (s->iq_tail == s->iq_head) && | |
1121 | (s->iq_last_desc_type == VTD_INV_DESC_WAIT); | |
1122 | } | |
1123 | ||
1124 | static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en) | |
1125 | { | |
1126 | uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG); | |
1127 | ||
1128 | VTD_DPRINTF(INV, "Queued Invalidation Enable %s", (en ? "on" : "off")); | |
1129 | if (en) { | |
1130 | if (vtd_queued_inv_enable_check(s)) { | |
1131 | s->iq = iqa_val & VTD_IQA_IQA_MASK; | |
1132 | /* 2^(x+8) entries */ | |
1133 | s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8); | |
1134 | s->qi_enabled = true; | |
1135 | VTD_DPRINTF(INV, "DMAR_IQA_REG 0x%"PRIx64, iqa_val); | |
1136 | VTD_DPRINTF(INV, "Invalidation Queue addr 0x%"PRIx64 " size %d", | |
1137 | s->iq, s->iq_size); | |
1138 | /* Ok - report back to driver */ | |
1139 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES); | |
1140 | } else { | |
1141 | VTD_DPRINTF(GENERAL, "error: can't enable Queued Invalidation: " | |
1142 | "tail %"PRIu16, s->iq_tail); | |
1143 | } | |
1144 | } else { | |
1145 | if (vtd_queued_inv_disable_check(s)) { | |
1146 | /* disable Queued Invalidation */ | |
1147 | vtd_set_quad_raw(s, DMAR_IQH_REG, 0); | |
1148 | s->iq_head = 0; | |
1149 | s->qi_enabled = false; | |
1150 | /* Ok - report back to driver */ | |
1151 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0); | |
1152 | } else { | |
1153 | VTD_DPRINTF(GENERAL, "error: can't disable Queued Invalidation: " | |
1154 | "head %"PRIu16 ", tail %"PRIu16 | |
1155 | ", last_descriptor %"PRIu8, | |
1156 | s->iq_head, s->iq_tail, s->iq_last_desc_type); | |
1157 | } | |
1158 | } | |
1159 | } | |
1160 | ||
1da12ec4 LT |
1161 | /* Set Root Table Pointer */ |
1162 | static void vtd_handle_gcmd_srtp(IntelIOMMUState *s) | |
1163 | { | |
1164 | VTD_DPRINTF(CSR, "set Root Table Pointer"); | |
1165 | ||
1166 | vtd_root_table_setup(s); | |
1167 | /* Ok - report back to driver */ | |
1168 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS); | |
1169 | } | |
1170 | ||
a5861439 PX |
1171 | /* Set Interrupt Remap Table Pointer */ |
1172 | static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s) | |
1173 | { | |
1174 | VTD_DPRINTF(CSR, "set Interrupt Remap Table Pointer"); | |
1175 | ||
1176 | vtd_interrupt_remap_table_setup(s); | |
1177 | /* Ok - report back to driver */ | |
1178 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS); | |
1179 | } | |
1180 | ||
1da12ec4 LT |
1181 | /* Handle Translation Enable/Disable */ |
1182 | static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en) | |
1183 | { | |
1184 | VTD_DPRINTF(CSR, "Translation Enable %s", (en ? "on" : "off")); | |
1185 | ||
1186 | if (en) { | |
1187 | s->dmar_enabled = true; | |
1188 | /* Ok - report back to driver */ | |
1189 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES); | |
1190 | } else { | |
1191 | s->dmar_enabled = false; | |
1192 | ||
1193 | /* Clear the index of Fault Recording Register */ | |
1194 | s->next_frcd_reg = 0; | |
1195 | /* Ok - report back to driver */ | |
1196 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0); | |
1197 | } | |
1198 | } | |
1199 | ||
80de52ba PX |
1200 | /* Handle Interrupt Remap Enable/Disable */ |
1201 | static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en) | |
1202 | { | |
1203 | VTD_DPRINTF(CSR, "Interrupt Remap Enable %s", (en ? "on" : "off")); | |
1204 | ||
1205 | if (en) { | |
1206 | s->intr_enabled = true; | |
1207 | /* Ok - report back to driver */ | |
1208 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES); | |
1209 | } else { | |
1210 | s->intr_enabled = false; | |
1211 | /* Ok - report back to driver */ | |
1212 | vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0); | |
1213 | } | |
1214 | } | |
1215 | ||
1da12ec4 LT |
1216 | /* Handle write to Global Command Register */ |
1217 | static void vtd_handle_gcmd_write(IntelIOMMUState *s) | |
1218 | { | |
1219 | uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG); | |
1220 | uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG); | |
1221 | uint32_t changed = status ^ val; | |
1222 | ||
1223 | VTD_DPRINTF(CSR, "value 0x%"PRIx32 " status 0x%"PRIx32, val, status); | |
1224 | if (changed & VTD_GCMD_TE) { | |
1225 | /* Translation enable/disable */ | |
1226 | vtd_handle_gcmd_te(s, val & VTD_GCMD_TE); | |
1227 | } | |
1228 | if (val & VTD_GCMD_SRTP) { | |
1229 | /* Set/update the root-table pointer */ | |
1230 | vtd_handle_gcmd_srtp(s); | |
1231 | } | |
ed7b8fbc LT |
1232 | if (changed & VTD_GCMD_QIE) { |
1233 | /* Queued Invalidation Enable */ | |
1234 | vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE); | |
1235 | } | |
a5861439 PX |
1236 | if (val & VTD_GCMD_SIRTP) { |
1237 | /* Set/update the interrupt remapping root-table pointer */ | |
1238 | vtd_handle_gcmd_sirtp(s); | |
1239 | } | |
80de52ba PX |
1240 | if (changed & VTD_GCMD_IRE) { |
1241 | /* Interrupt remap enable/disable */ | |
1242 | vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE); | |
1243 | } | |
1da12ec4 LT |
1244 | } |
1245 | ||
1246 | /* Handle write to Context Command Register */ | |
1247 | static void vtd_handle_ccmd_write(IntelIOMMUState *s) | |
1248 | { | |
1249 | uint64_t ret; | |
1250 | uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG); | |
1251 | ||
1252 | /* Context-cache invalidation request */ | |
1253 | if (val & VTD_CCMD_ICC) { | |
ed7b8fbc LT |
1254 | if (s->qi_enabled) { |
1255 | VTD_DPRINTF(GENERAL, "error: Queued Invalidation enabled, " | |
1256 | "should not use register-based invalidation"); | |
1257 | return; | |
1258 | } | |
1da12ec4 LT |
1259 | ret = vtd_context_cache_invalidate(s, val); |
1260 | /* Invalidation completed. Change something to show */ | |
1261 | vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL); | |
1262 | ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK, | |
1263 | ret); | |
1264 | VTD_DPRINTF(INV, "CCMD_REG write-back val: 0x%"PRIx64, ret); | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | /* Handle write to IOTLB Invalidation Register */ | |
1269 | static void vtd_handle_iotlb_write(IntelIOMMUState *s) | |
1270 | { | |
1271 | uint64_t ret; | |
1272 | uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG); | |
1273 | ||
1274 | /* IOTLB invalidation request */ | |
1275 | if (val & VTD_TLB_IVT) { | |
ed7b8fbc LT |
1276 | if (s->qi_enabled) { |
1277 | VTD_DPRINTF(GENERAL, "error: Queued Invalidation enabled, " | |
1278 | "should not use register-based invalidation"); | |
1279 | return; | |
1280 | } | |
1da12ec4 LT |
1281 | ret = vtd_iotlb_flush(s, val); |
1282 | /* Invalidation completed. Change something to show */ | |
1283 | vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL); | |
1284 | ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, | |
1285 | VTD_TLB_FLUSH_GRANU_MASK_A, ret); | |
1286 | VTD_DPRINTF(INV, "IOTLB_REG write-back val: 0x%"PRIx64, ret); | |
1287 | } | |
1288 | } | |
1289 | ||
ed7b8fbc LT |
1290 | /* Fetch an Invalidation Descriptor from the Invalidation Queue */ |
1291 | static bool vtd_get_inv_desc(dma_addr_t base_addr, uint32_t offset, | |
1292 | VTDInvDesc *inv_desc) | |
1293 | { | |
1294 | dma_addr_t addr = base_addr + offset * sizeof(*inv_desc); | |
1295 | if (dma_memory_read(&address_space_memory, addr, inv_desc, | |
1296 | sizeof(*inv_desc))) { | |
1297 | VTD_DPRINTF(GENERAL, "error: fail to fetch Invalidation Descriptor " | |
1298 | "base_addr 0x%"PRIx64 " offset %"PRIu32, base_addr, offset); | |
1299 | inv_desc->lo = 0; | |
1300 | inv_desc->hi = 0; | |
1301 | ||
1302 | return false; | |
1303 | } | |
1304 | inv_desc->lo = le64_to_cpu(inv_desc->lo); | |
1305 | inv_desc->hi = le64_to_cpu(inv_desc->hi); | |
1306 | return true; | |
1307 | } | |
1308 | ||
1309 | static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc) | |
1310 | { | |
1311 | if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) || | |
1312 | (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) { | |
1313 | VTD_DPRINTF(GENERAL, "error: non-zero reserved field in Invalidation " | |
1314 | "Wait Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, | |
1315 | inv_desc->hi, inv_desc->lo); | |
1316 | return false; | |
1317 | } | |
1318 | if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) { | |
1319 | /* Status Write */ | |
1320 | uint32_t status_data = (uint32_t)(inv_desc->lo >> | |
1321 | VTD_INV_DESC_WAIT_DATA_SHIFT); | |
1322 | ||
1323 | assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF)); | |
1324 | ||
1325 | /* FIXME: need to be masked with HAW? */ | |
1326 | dma_addr_t status_addr = inv_desc->hi; | |
1327 | VTD_DPRINTF(INV, "status data 0x%x, status addr 0x%"PRIx64, | |
1328 | status_data, status_addr); | |
1329 | status_data = cpu_to_le32(status_data); | |
1330 | if (dma_memory_write(&address_space_memory, status_addr, &status_data, | |
1331 | sizeof(status_data))) { | |
1332 | VTD_DPRINTF(GENERAL, "error: fail to perform a coherent write"); | |
1333 | return false; | |
1334 | } | |
1335 | } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) { | |
1336 | /* Interrupt flag */ | |
1337 | VTD_DPRINTF(INV, "Invalidation Wait Descriptor interrupt completion"); | |
1338 | vtd_generate_completion_event(s); | |
1339 | } else { | |
1340 | VTD_DPRINTF(GENERAL, "error: invalid Invalidation Wait Descriptor: " | |
1341 | "hi 0x%"PRIx64 " lo 0x%"PRIx64, inv_desc->hi, inv_desc->lo); | |
1342 | return false; | |
1343 | } | |
1344 | return true; | |
1345 | } | |
1346 | ||
d92fa2dc LT |
1347 | static bool vtd_process_context_cache_desc(IntelIOMMUState *s, |
1348 | VTDInvDesc *inv_desc) | |
1349 | { | |
1350 | if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) { | |
1351 | VTD_DPRINTF(GENERAL, "error: non-zero reserved field in Context-cache " | |
1352 | "Invalidate Descriptor"); | |
1353 | return false; | |
1354 | } | |
1355 | switch (inv_desc->lo & VTD_INV_DESC_CC_G) { | |
1356 | case VTD_INV_DESC_CC_DOMAIN: | |
1357 | VTD_DPRINTF(INV, "domain-selective invalidation domain 0x%"PRIx16, | |
1358 | (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo)); | |
1359 | /* Fall through */ | |
1360 | case VTD_INV_DESC_CC_GLOBAL: | |
1361 | VTD_DPRINTF(INV, "global invalidation"); | |
1362 | vtd_context_global_invalidate(s); | |
1363 | break; | |
1364 | ||
1365 | case VTD_INV_DESC_CC_DEVICE: | |
1366 | vtd_context_device_invalidate(s, VTD_INV_DESC_CC_SID(inv_desc->lo), | |
1367 | VTD_INV_DESC_CC_FM(inv_desc->lo)); | |
1368 | break; | |
1369 | ||
1370 | default: | |
1371 | VTD_DPRINTF(GENERAL, "error: invalid granularity in Context-cache " | |
1372 | "Invalidate Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, | |
1373 | inv_desc->hi, inv_desc->lo); | |
1374 | return false; | |
1375 | } | |
1376 | return true; | |
1377 | } | |
1378 | ||
b5a280c0 LT |
1379 | static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc) |
1380 | { | |
1381 | uint16_t domain_id; | |
1382 | uint8_t am; | |
1383 | hwaddr addr; | |
1384 | ||
1385 | if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) || | |
1386 | (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) { | |
1387 | VTD_DPRINTF(GENERAL, "error: non-zero reserved field in IOTLB " | |
1388 | "Invalidate Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, | |
1389 | inv_desc->hi, inv_desc->lo); | |
1390 | return false; | |
1391 | } | |
1392 | ||
1393 | switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) { | |
1394 | case VTD_INV_DESC_IOTLB_GLOBAL: | |
1395 | VTD_DPRINTF(INV, "global invalidation"); | |
1396 | vtd_iotlb_global_invalidate(s); | |
1397 | break; | |
1398 | ||
1399 | case VTD_INV_DESC_IOTLB_DOMAIN: | |
1400 | domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo); | |
1401 | VTD_DPRINTF(INV, "domain-selective invalidation domain 0x%"PRIx16, | |
1402 | domain_id); | |
1403 | vtd_iotlb_domain_invalidate(s, domain_id); | |
1404 | break; | |
1405 | ||
1406 | case VTD_INV_DESC_IOTLB_PAGE: | |
1407 | domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo); | |
1408 | addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi); | |
1409 | am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi); | |
1410 | VTD_DPRINTF(INV, "page-selective invalidation domain 0x%"PRIx16 | |
1411 | " addr 0x%"PRIx64 " mask %"PRIu8, domain_id, addr, am); | |
1412 | if (am > VTD_MAMV) { | |
1413 | VTD_DPRINTF(GENERAL, "error: supported max address mask value is " | |
1414 | "%"PRIu8, (uint8_t)VTD_MAMV); | |
1415 | return false; | |
1416 | } | |
1417 | vtd_iotlb_page_invalidate(s, domain_id, addr, am); | |
1418 | break; | |
1419 | ||
1420 | default: | |
1421 | VTD_DPRINTF(GENERAL, "error: invalid granularity in IOTLB Invalidate " | |
1422 | "Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, | |
1423 | inv_desc->hi, inv_desc->lo); | |
1424 | return false; | |
1425 | } | |
1426 | return true; | |
1427 | } | |
1428 | ||
02a2cbc8 PX |
1429 | static bool vtd_process_inv_iec_desc(IntelIOMMUState *s, |
1430 | VTDInvDesc *inv_desc) | |
1431 | { | |
1432 | VTD_DPRINTF(INV, "inv ir glob %d index %d mask %d", | |
1433 | inv_desc->iec.granularity, | |
1434 | inv_desc->iec.index, | |
1435 | inv_desc->iec.index_mask); | |
1436 | ||
1437 | vtd_iec_notify_all(s, !inv_desc->iec.granularity, | |
1438 | inv_desc->iec.index, | |
1439 | inv_desc->iec.index_mask); | |
1440 | ||
1441 | return true; | |
1442 | } | |
1443 | ||
ed7b8fbc LT |
1444 | static bool vtd_process_inv_desc(IntelIOMMUState *s) |
1445 | { | |
1446 | VTDInvDesc inv_desc; | |
1447 | uint8_t desc_type; | |
1448 | ||
1449 | VTD_DPRINTF(INV, "iq head %"PRIu16, s->iq_head); | |
1450 | if (!vtd_get_inv_desc(s->iq, s->iq_head, &inv_desc)) { | |
1451 | s->iq_last_desc_type = VTD_INV_DESC_NONE; | |
1452 | return false; | |
1453 | } | |
1454 | desc_type = inv_desc.lo & VTD_INV_DESC_TYPE; | |
1455 | /* FIXME: should update at first or at last? */ | |
1456 | s->iq_last_desc_type = desc_type; | |
1457 | ||
1458 | switch (desc_type) { | |
1459 | case VTD_INV_DESC_CC: | |
1460 | VTD_DPRINTF(INV, "Context-cache Invalidate Descriptor hi 0x%"PRIx64 | |
1461 | " lo 0x%"PRIx64, inv_desc.hi, inv_desc.lo); | |
d92fa2dc LT |
1462 | if (!vtd_process_context_cache_desc(s, &inv_desc)) { |
1463 | return false; | |
1464 | } | |
ed7b8fbc LT |
1465 | break; |
1466 | ||
1467 | case VTD_INV_DESC_IOTLB: | |
1468 | VTD_DPRINTF(INV, "IOTLB Invalidate Descriptor hi 0x%"PRIx64 | |
1469 | " lo 0x%"PRIx64, inv_desc.hi, inv_desc.lo); | |
b5a280c0 LT |
1470 | if (!vtd_process_iotlb_desc(s, &inv_desc)) { |
1471 | return false; | |
1472 | } | |
ed7b8fbc LT |
1473 | break; |
1474 | ||
1475 | case VTD_INV_DESC_WAIT: | |
1476 | VTD_DPRINTF(INV, "Invalidation Wait Descriptor hi 0x%"PRIx64 | |
1477 | " lo 0x%"PRIx64, inv_desc.hi, inv_desc.lo); | |
1478 | if (!vtd_process_wait_desc(s, &inv_desc)) { | |
1479 | return false; | |
1480 | } | |
1481 | break; | |
1482 | ||
b7910472 | 1483 | case VTD_INV_DESC_IEC: |
02a2cbc8 PX |
1484 | VTD_DPRINTF(INV, "Invalidation Interrupt Entry Cache " |
1485 | "Descriptor hi 0x%"PRIx64 " lo 0x%"PRIx64, | |
1486 | inv_desc.hi, inv_desc.lo); | |
1487 | if (!vtd_process_inv_iec_desc(s, &inv_desc)) { | |
1488 | return false; | |
1489 | } | |
b7910472 PX |
1490 | break; |
1491 | ||
ed7b8fbc LT |
1492 | default: |
1493 | VTD_DPRINTF(GENERAL, "error: unkonw Invalidation Descriptor type " | |
1494 | "hi 0x%"PRIx64 " lo 0x%"PRIx64 " type %"PRIu8, | |
1495 | inv_desc.hi, inv_desc.lo, desc_type); | |
1496 | return false; | |
1497 | } | |
1498 | s->iq_head++; | |
1499 | if (s->iq_head == s->iq_size) { | |
1500 | s->iq_head = 0; | |
1501 | } | |
1502 | return true; | |
1503 | } | |
1504 | ||
1505 | /* Try to fetch and process more Invalidation Descriptors */ | |
1506 | static void vtd_fetch_inv_desc(IntelIOMMUState *s) | |
1507 | { | |
1508 | VTD_DPRINTF(INV, "fetch Invalidation Descriptors"); | |
1509 | if (s->iq_tail >= s->iq_size) { | |
1510 | /* Detects an invalid Tail pointer */ | |
1511 | VTD_DPRINTF(GENERAL, "error: iq_tail is %"PRIu16 | |
1512 | " while iq_size is %"PRIu16, s->iq_tail, s->iq_size); | |
1513 | vtd_handle_inv_queue_error(s); | |
1514 | return; | |
1515 | } | |
1516 | while (s->iq_head != s->iq_tail) { | |
1517 | if (!vtd_process_inv_desc(s)) { | |
1518 | /* Invalidation Queue Errors */ | |
1519 | vtd_handle_inv_queue_error(s); | |
1520 | break; | |
1521 | } | |
1522 | /* Must update the IQH_REG in time */ | |
1523 | vtd_set_quad_raw(s, DMAR_IQH_REG, | |
1524 | (((uint64_t)(s->iq_head)) << VTD_IQH_QH_SHIFT) & | |
1525 | VTD_IQH_QH_MASK); | |
1526 | } | |
1527 | } | |
1528 | ||
1529 | /* Handle write to Invalidation Queue Tail Register */ | |
1530 | static void vtd_handle_iqt_write(IntelIOMMUState *s) | |
1531 | { | |
1532 | uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG); | |
1533 | ||
1534 | s->iq_tail = VTD_IQT_QT(val); | |
1535 | VTD_DPRINTF(INV, "set iq tail %"PRIu16, s->iq_tail); | |
1536 | if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) { | |
1537 | /* Process Invalidation Queue here */ | |
1538 | vtd_fetch_inv_desc(s); | |
1539 | } | |
1540 | } | |
1541 | ||
1da12ec4 LT |
1542 | static void vtd_handle_fsts_write(IntelIOMMUState *s) |
1543 | { | |
1544 | uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG); | |
1545 | uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG); | |
1546 | uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE; | |
1547 | ||
1548 | if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) { | |
1549 | vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); | |
1550 | VTD_DPRINTF(FLOG, "all pending interrupt conditions serviced, clear " | |
1551 | "IP field of FECTL_REG"); | |
1552 | } | |
ed7b8fbc LT |
1553 | /* FIXME: when IQE is Clear, should we try to fetch some Invalidation |
1554 | * Descriptors if there are any when Queued Invalidation is enabled? | |
1555 | */ | |
1da12ec4 LT |
1556 | } |
1557 | ||
1558 | static void vtd_handle_fectl_write(IntelIOMMUState *s) | |
1559 | { | |
1560 | uint32_t fectl_reg; | |
1561 | /* FIXME: when software clears the IM field, check the IP field. But do we | |
1562 | * need to compare the old value and the new value to conclude that | |
1563 | * software clears the IM field? Or just check if the IM field is zero? | |
1564 | */ | |
1565 | fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG); | |
1566 | if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) { | |
1567 | vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG); | |
1568 | vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0); | |
1569 | VTD_DPRINTF(FLOG, "IM field is cleared, generate " | |
1570 | "fault event interrupt"); | |
1571 | } | |
1572 | } | |
1573 | ||
ed7b8fbc LT |
1574 | static void vtd_handle_ics_write(IntelIOMMUState *s) |
1575 | { | |
1576 | uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG); | |
1577 | uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG); | |
1578 | ||
1579 | if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) { | |
1580 | vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); | |
1581 | VTD_DPRINTF(INV, "pending completion interrupt condition serviced, " | |
1582 | "clear IP field of IECTL_REG"); | |
1583 | } | |
1584 | } | |
1585 | ||
1586 | static void vtd_handle_iectl_write(IntelIOMMUState *s) | |
1587 | { | |
1588 | uint32_t iectl_reg; | |
1589 | /* FIXME: when software clears the IM field, check the IP field. But do we | |
1590 | * need to compare the old value and the new value to conclude that | |
1591 | * software clears the IM field? Or just check if the IM field is zero? | |
1592 | */ | |
1593 | iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG); | |
1594 | if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) { | |
1595 | vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG); | |
1596 | vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0); | |
1597 | VTD_DPRINTF(INV, "IM field is cleared, generate " | |
1598 | "invalidation event interrupt"); | |
1599 | } | |
1600 | } | |
1601 | ||
1da12ec4 LT |
1602 | static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size) |
1603 | { | |
1604 | IntelIOMMUState *s = opaque; | |
1605 | uint64_t val; | |
1606 | ||
1607 | if (addr + size > DMAR_REG_SIZE) { | |
1608 | VTD_DPRINTF(GENERAL, "error: addr outside region: max 0x%"PRIx64 | |
1609 | ", got 0x%"PRIx64 " %d", | |
1610 | (uint64_t)DMAR_REG_SIZE, addr, size); | |
1611 | return (uint64_t)-1; | |
1612 | } | |
1613 | ||
1614 | switch (addr) { | |
1615 | /* Root Table Address Register, 64-bit */ | |
1616 | case DMAR_RTADDR_REG: | |
1617 | if (size == 4) { | |
1618 | val = s->root & ((1ULL << 32) - 1); | |
1619 | } else { | |
1620 | val = s->root; | |
1621 | } | |
1622 | break; | |
1623 | ||
1624 | case DMAR_RTADDR_REG_HI: | |
1625 | assert(size == 4); | |
1626 | val = s->root >> 32; | |
1627 | break; | |
1628 | ||
ed7b8fbc LT |
1629 | /* Invalidation Queue Address Register, 64-bit */ |
1630 | case DMAR_IQA_REG: | |
1631 | val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS); | |
1632 | if (size == 4) { | |
1633 | val = val & ((1ULL << 32) - 1); | |
1634 | } | |
1635 | break; | |
1636 | ||
1637 | case DMAR_IQA_REG_HI: | |
1638 | assert(size == 4); | |
1639 | val = s->iq >> 32; | |
1640 | break; | |
1641 | ||
1da12ec4 LT |
1642 | default: |
1643 | if (size == 4) { | |
1644 | val = vtd_get_long(s, addr); | |
1645 | } else { | |
1646 | val = vtd_get_quad(s, addr); | |
1647 | } | |
1648 | } | |
1649 | VTD_DPRINTF(CSR, "addr 0x%"PRIx64 " size %d val 0x%"PRIx64, | |
1650 | addr, size, val); | |
1651 | return val; | |
1652 | } | |
1653 | ||
1654 | static void vtd_mem_write(void *opaque, hwaddr addr, | |
1655 | uint64_t val, unsigned size) | |
1656 | { | |
1657 | IntelIOMMUState *s = opaque; | |
1658 | ||
1659 | if (addr + size > DMAR_REG_SIZE) { | |
1660 | VTD_DPRINTF(GENERAL, "error: addr outside region: max 0x%"PRIx64 | |
1661 | ", got 0x%"PRIx64 " %d", | |
1662 | (uint64_t)DMAR_REG_SIZE, addr, size); | |
1663 | return; | |
1664 | } | |
1665 | ||
1666 | switch (addr) { | |
1667 | /* Global Command Register, 32-bit */ | |
1668 | case DMAR_GCMD_REG: | |
1669 | VTD_DPRINTF(CSR, "DMAR_GCMD_REG write addr 0x%"PRIx64 | |
1670 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1671 | vtd_set_long(s, addr, val); | |
1672 | vtd_handle_gcmd_write(s); | |
1673 | break; | |
1674 | ||
1675 | /* Context Command Register, 64-bit */ | |
1676 | case DMAR_CCMD_REG: | |
1677 | VTD_DPRINTF(CSR, "DMAR_CCMD_REG write addr 0x%"PRIx64 | |
1678 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1679 | if (size == 4) { | |
1680 | vtd_set_long(s, addr, val); | |
1681 | } else { | |
1682 | vtd_set_quad(s, addr, val); | |
1683 | vtd_handle_ccmd_write(s); | |
1684 | } | |
1685 | break; | |
1686 | ||
1687 | case DMAR_CCMD_REG_HI: | |
1688 | VTD_DPRINTF(CSR, "DMAR_CCMD_REG_HI write addr 0x%"PRIx64 | |
1689 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1690 | assert(size == 4); | |
1691 | vtd_set_long(s, addr, val); | |
1692 | vtd_handle_ccmd_write(s); | |
1693 | break; | |
1694 | ||
1695 | /* IOTLB Invalidation Register, 64-bit */ | |
1696 | case DMAR_IOTLB_REG: | |
1697 | VTD_DPRINTF(INV, "DMAR_IOTLB_REG write addr 0x%"PRIx64 | |
1698 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1699 | if (size == 4) { | |
1700 | vtd_set_long(s, addr, val); | |
1701 | } else { | |
1702 | vtd_set_quad(s, addr, val); | |
1703 | vtd_handle_iotlb_write(s); | |
1704 | } | |
1705 | break; | |
1706 | ||
1707 | case DMAR_IOTLB_REG_HI: | |
1708 | VTD_DPRINTF(INV, "DMAR_IOTLB_REG_HI write addr 0x%"PRIx64 | |
1709 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1710 | assert(size == 4); | |
1711 | vtd_set_long(s, addr, val); | |
1712 | vtd_handle_iotlb_write(s); | |
1713 | break; | |
1714 | ||
b5a280c0 LT |
1715 | /* Invalidate Address Register, 64-bit */ |
1716 | case DMAR_IVA_REG: | |
1717 | VTD_DPRINTF(INV, "DMAR_IVA_REG write addr 0x%"PRIx64 | |
1718 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1719 | if (size == 4) { | |
1720 | vtd_set_long(s, addr, val); | |
1721 | } else { | |
1722 | vtd_set_quad(s, addr, val); | |
1723 | } | |
1724 | break; | |
1725 | ||
1726 | case DMAR_IVA_REG_HI: | |
1727 | VTD_DPRINTF(INV, "DMAR_IVA_REG_HI write addr 0x%"PRIx64 | |
1728 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1729 | assert(size == 4); | |
1730 | vtd_set_long(s, addr, val); | |
1731 | break; | |
1732 | ||
1da12ec4 LT |
1733 | /* Fault Status Register, 32-bit */ |
1734 | case DMAR_FSTS_REG: | |
1735 | VTD_DPRINTF(FLOG, "DMAR_FSTS_REG write addr 0x%"PRIx64 | |
1736 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1737 | assert(size == 4); | |
1738 | vtd_set_long(s, addr, val); | |
1739 | vtd_handle_fsts_write(s); | |
1740 | break; | |
1741 | ||
1742 | /* Fault Event Control Register, 32-bit */ | |
1743 | case DMAR_FECTL_REG: | |
1744 | VTD_DPRINTF(FLOG, "DMAR_FECTL_REG write addr 0x%"PRIx64 | |
1745 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1746 | assert(size == 4); | |
1747 | vtd_set_long(s, addr, val); | |
1748 | vtd_handle_fectl_write(s); | |
1749 | break; | |
1750 | ||
1751 | /* Fault Event Data Register, 32-bit */ | |
1752 | case DMAR_FEDATA_REG: | |
1753 | VTD_DPRINTF(FLOG, "DMAR_FEDATA_REG write addr 0x%"PRIx64 | |
1754 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1755 | assert(size == 4); | |
1756 | vtd_set_long(s, addr, val); | |
1757 | break; | |
1758 | ||
1759 | /* Fault Event Address Register, 32-bit */ | |
1760 | case DMAR_FEADDR_REG: | |
1761 | VTD_DPRINTF(FLOG, "DMAR_FEADDR_REG write addr 0x%"PRIx64 | |
1762 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1763 | assert(size == 4); | |
1764 | vtd_set_long(s, addr, val); | |
1765 | break; | |
1766 | ||
1767 | /* Fault Event Upper Address Register, 32-bit */ | |
1768 | case DMAR_FEUADDR_REG: | |
1769 | VTD_DPRINTF(FLOG, "DMAR_FEUADDR_REG write addr 0x%"PRIx64 | |
1770 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1771 | assert(size == 4); | |
1772 | vtd_set_long(s, addr, val); | |
1773 | break; | |
1774 | ||
1775 | /* Protected Memory Enable Register, 32-bit */ | |
1776 | case DMAR_PMEN_REG: | |
1777 | VTD_DPRINTF(CSR, "DMAR_PMEN_REG write addr 0x%"PRIx64 | |
1778 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1779 | assert(size == 4); | |
1780 | vtd_set_long(s, addr, val); | |
1781 | break; | |
1782 | ||
1783 | /* Root Table Address Register, 64-bit */ | |
1784 | case DMAR_RTADDR_REG: | |
1785 | VTD_DPRINTF(CSR, "DMAR_RTADDR_REG write addr 0x%"PRIx64 | |
1786 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1787 | if (size == 4) { | |
1788 | vtd_set_long(s, addr, val); | |
1789 | } else { | |
1790 | vtd_set_quad(s, addr, val); | |
1791 | } | |
1792 | break; | |
1793 | ||
1794 | case DMAR_RTADDR_REG_HI: | |
1795 | VTD_DPRINTF(CSR, "DMAR_RTADDR_REG_HI write addr 0x%"PRIx64 | |
1796 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1797 | assert(size == 4); | |
1798 | vtd_set_long(s, addr, val); | |
1799 | break; | |
1800 | ||
ed7b8fbc LT |
1801 | /* Invalidation Queue Tail Register, 64-bit */ |
1802 | case DMAR_IQT_REG: | |
1803 | VTD_DPRINTF(INV, "DMAR_IQT_REG write addr 0x%"PRIx64 | |
1804 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1805 | if (size == 4) { | |
1806 | vtd_set_long(s, addr, val); | |
1807 | } else { | |
1808 | vtd_set_quad(s, addr, val); | |
1809 | } | |
1810 | vtd_handle_iqt_write(s); | |
1811 | break; | |
1812 | ||
1813 | case DMAR_IQT_REG_HI: | |
1814 | VTD_DPRINTF(INV, "DMAR_IQT_REG_HI write addr 0x%"PRIx64 | |
1815 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1816 | assert(size == 4); | |
1817 | vtd_set_long(s, addr, val); | |
1818 | /* 19:63 of IQT_REG is RsvdZ, do nothing here */ | |
1819 | break; | |
1820 | ||
1821 | /* Invalidation Queue Address Register, 64-bit */ | |
1822 | case DMAR_IQA_REG: | |
1823 | VTD_DPRINTF(INV, "DMAR_IQA_REG write addr 0x%"PRIx64 | |
1824 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1825 | if (size == 4) { | |
1826 | vtd_set_long(s, addr, val); | |
1827 | } else { | |
1828 | vtd_set_quad(s, addr, val); | |
1829 | } | |
1830 | break; | |
1831 | ||
1832 | case DMAR_IQA_REG_HI: | |
1833 | VTD_DPRINTF(INV, "DMAR_IQA_REG_HI write addr 0x%"PRIx64 | |
1834 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1835 | assert(size == 4); | |
1836 | vtd_set_long(s, addr, val); | |
1837 | break; | |
1838 | ||
1839 | /* Invalidation Completion Status Register, 32-bit */ | |
1840 | case DMAR_ICS_REG: | |
1841 | VTD_DPRINTF(INV, "DMAR_ICS_REG write addr 0x%"PRIx64 | |
1842 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1843 | assert(size == 4); | |
1844 | vtd_set_long(s, addr, val); | |
1845 | vtd_handle_ics_write(s); | |
1846 | break; | |
1847 | ||
1848 | /* Invalidation Event Control Register, 32-bit */ | |
1849 | case DMAR_IECTL_REG: | |
1850 | VTD_DPRINTF(INV, "DMAR_IECTL_REG write addr 0x%"PRIx64 | |
1851 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1852 | assert(size == 4); | |
1853 | vtd_set_long(s, addr, val); | |
1854 | vtd_handle_iectl_write(s); | |
1855 | break; | |
1856 | ||
1857 | /* Invalidation Event Data Register, 32-bit */ | |
1858 | case DMAR_IEDATA_REG: | |
1859 | VTD_DPRINTF(INV, "DMAR_IEDATA_REG write addr 0x%"PRIx64 | |
1860 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1861 | assert(size == 4); | |
1862 | vtd_set_long(s, addr, val); | |
1863 | break; | |
1864 | ||
1865 | /* Invalidation Event Address Register, 32-bit */ | |
1866 | case DMAR_IEADDR_REG: | |
1867 | VTD_DPRINTF(INV, "DMAR_IEADDR_REG write addr 0x%"PRIx64 | |
1868 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1869 | assert(size == 4); | |
1870 | vtd_set_long(s, addr, val); | |
1871 | break; | |
1872 | ||
1873 | /* Invalidation Event Upper Address Register, 32-bit */ | |
1874 | case DMAR_IEUADDR_REG: | |
1875 | VTD_DPRINTF(INV, "DMAR_IEUADDR_REG write addr 0x%"PRIx64 | |
1876 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1877 | assert(size == 4); | |
1878 | vtd_set_long(s, addr, val); | |
1879 | break; | |
1880 | ||
1da12ec4 LT |
1881 | /* Fault Recording Registers, 128-bit */ |
1882 | case DMAR_FRCD_REG_0_0: | |
1883 | VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_0 write addr 0x%"PRIx64 | |
1884 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1885 | if (size == 4) { | |
1886 | vtd_set_long(s, addr, val); | |
1887 | } else { | |
1888 | vtd_set_quad(s, addr, val); | |
1889 | } | |
1890 | break; | |
1891 | ||
1892 | case DMAR_FRCD_REG_0_1: | |
1893 | VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_1 write addr 0x%"PRIx64 | |
1894 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1895 | assert(size == 4); | |
1896 | vtd_set_long(s, addr, val); | |
1897 | break; | |
1898 | ||
1899 | case DMAR_FRCD_REG_0_2: | |
1900 | VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_2 write addr 0x%"PRIx64 | |
1901 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1902 | if (size == 4) { | |
1903 | vtd_set_long(s, addr, val); | |
1904 | } else { | |
1905 | vtd_set_quad(s, addr, val); | |
1906 | /* May clear bit 127 (Fault), update PPF */ | |
1907 | vtd_update_fsts_ppf(s); | |
1908 | } | |
1909 | break; | |
1910 | ||
1911 | case DMAR_FRCD_REG_0_3: | |
1912 | VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_3 write addr 0x%"PRIx64 | |
1913 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1914 | assert(size == 4); | |
1915 | vtd_set_long(s, addr, val); | |
1916 | /* May clear bit 127 (Fault), update PPF */ | |
1917 | vtd_update_fsts_ppf(s); | |
1918 | break; | |
1919 | ||
a5861439 PX |
1920 | case DMAR_IRTA_REG: |
1921 | VTD_DPRINTF(IR, "DMAR_IRTA_REG write addr 0x%"PRIx64 | |
1922 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1923 | if (size == 4) { | |
1924 | vtd_set_long(s, addr, val); | |
1925 | } else { | |
1926 | vtd_set_quad(s, addr, val); | |
1927 | } | |
1928 | break; | |
1929 | ||
1930 | case DMAR_IRTA_REG_HI: | |
1931 | VTD_DPRINTF(IR, "DMAR_IRTA_REG_HI write addr 0x%"PRIx64 | |
1932 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1933 | assert(size == 4); | |
1934 | vtd_set_long(s, addr, val); | |
1935 | break; | |
1936 | ||
1da12ec4 LT |
1937 | default: |
1938 | VTD_DPRINTF(GENERAL, "error: unhandled reg write addr 0x%"PRIx64 | |
1939 | ", size %d, val 0x%"PRIx64, addr, size, val); | |
1940 | if (size == 4) { | |
1941 | vtd_set_long(s, addr, val); | |
1942 | } else { | |
1943 | vtd_set_quad(s, addr, val); | |
1944 | } | |
1945 | } | |
1946 | } | |
1947 | ||
1948 | static IOMMUTLBEntry vtd_iommu_translate(MemoryRegion *iommu, hwaddr addr, | |
1949 | bool is_write) | |
1950 | { | |
1951 | VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu); | |
1952 | IntelIOMMUState *s = vtd_as->iommu_state; | |
1da12ec4 LT |
1953 | IOMMUTLBEntry ret = { |
1954 | .target_as = &address_space_memory, | |
1955 | .iova = addr, | |
1956 | .translated_addr = 0, | |
1957 | .addr_mask = ~(hwaddr)0, | |
1958 | .perm = IOMMU_NONE, | |
1959 | }; | |
1960 | ||
1961 | if (!s->dmar_enabled) { | |
1962 | /* DMAR disabled, passthrough, use 4k-page*/ | |
1963 | ret.iova = addr & VTD_PAGE_MASK_4K; | |
1964 | ret.translated_addr = addr & VTD_PAGE_MASK_4K; | |
1965 | ret.addr_mask = ~VTD_PAGE_MASK_4K; | |
1966 | ret.perm = IOMMU_RW; | |
1967 | return ret; | |
1968 | } | |
1969 | ||
7df953bd | 1970 | vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn, addr, |
d92fa2dc | 1971 | is_write, &ret); |
1da12ec4 LT |
1972 | VTD_DPRINTF(MMU, |
1973 | "bus %"PRIu8 " slot %"PRIu8 " func %"PRIu8 " devfn %"PRIu8 | |
7df953bd | 1974 | " gpa 0x%"PRIx64 " hpa 0x%"PRIx64, pci_bus_num(vtd_as->bus), |
d92fa2dc LT |
1975 | VTD_PCI_SLOT(vtd_as->devfn), VTD_PCI_FUNC(vtd_as->devfn), |
1976 | vtd_as->devfn, addr, ret.translated_addr); | |
1da12ec4 LT |
1977 | return ret; |
1978 | } | |
1979 | ||
5bf3d319 PX |
1980 | static void vtd_iommu_notify_flag_changed(MemoryRegion *iommu, |
1981 | IOMMUNotifierFlag old, | |
1982 | IOMMUNotifierFlag new) | |
3cb3b154 AW |
1983 | { |
1984 | VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu); | |
1985 | ||
a3276f78 PX |
1986 | if (new & IOMMU_NOTIFIER_MAP) { |
1987 | error_report("Device at bus %s addr %02x.%d requires iommu " | |
1988 | "notifier which is currently not supported by " | |
1989 | "intel-iommu emulation", | |
1990 | vtd_as->bus->qbus.name, PCI_SLOT(vtd_as->devfn), | |
1991 | PCI_FUNC(vtd_as->devfn)); | |
1992 | exit(1); | |
1993 | } | |
3cb3b154 AW |
1994 | } |
1995 | ||
1da12ec4 LT |
1996 | static const VMStateDescription vtd_vmstate = { |
1997 | .name = "iommu-intel", | |
1998 | .unmigratable = 1, | |
1999 | }; | |
2000 | ||
2001 | static const MemoryRegionOps vtd_mem_ops = { | |
2002 | .read = vtd_mem_read, | |
2003 | .write = vtd_mem_write, | |
2004 | .endianness = DEVICE_LITTLE_ENDIAN, | |
2005 | .impl = { | |
2006 | .min_access_size = 4, | |
2007 | .max_access_size = 8, | |
2008 | }, | |
2009 | .valid = { | |
2010 | .min_access_size = 4, | |
2011 | .max_access_size = 8, | |
2012 | }, | |
2013 | }; | |
2014 | ||
2015 | static Property vtd_properties[] = { | |
2016 | DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0), | |
e6b6af05 RK |
2017 | DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim, |
2018 | ON_OFF_AUTO_AUTO), | |
fb506e70 | 2019 | DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false), |
1da12ec4 LT |
2020 | DEFINE_PROP_END_OF_LIST(), |
2021 | }; | |
2022 | ||
651e4cef PX |
2023 | /* Read IRTE entry with specific index */ |
2024 | static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index, | |
bc38ee10 | 2025 | VTD_IR_TableEntry *entry, uint16_t sid) |
651e4cef | 2026 | { |
ede9c94a PX |
2027 | static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \ |
2028 | {0xffff, 0xfffb, 0xfff9, 0xfff8}; | |
651e4cef | 2029 | dma_addr_t addr = 0x00; |
ede9c94a PX |
2030 | uint16_t mask, source_id; |
2031 | uint8_t bus, bus_max, bus_min; | |
651e4cef PX |
2032 | |
2033 | addr = iommu->intr_root + index * sizeof(*entry); | |
2034 | if (dma_memory_read(&address_space_memory, addr, entry, | |
2035 | sizeof(*entry))) { | |
2036 | VTD_DPRINTF(GENERAL, "error: fail to access IR root at 0x%"PRIx64 | |
2037 | " + %"PRIu16, iommu->intr_root, index); | |
2038 | return -VTD_FR_IR_ROOT_INVAL; | |
2039 | } | |
2040 | ||
bc38ee10 | 2041 | if (!entry->irte.present) { |
651e4cef PX |
2042 | VTD_DPRINTF(GENERAL, "error: present flag not set in IRTE" |
2043 | " entry index %u value 0x%"PRIx64 " 0x%"PRIx64, | |
2044 | index, le64_to_cpu(entry->data[1]), | |
2045 | le64_to_cpu(entry->data[0])); | |
2046 | return -VTD_FR_IR_ENTRY_P; | |
2047 | } | |
2048 | ||
bc38ee10 MT |
2049 | if (entry->irte.__reserved_0 || entry->irte.__reserved_1 || |
2050 | entry->irte.__reserved_2) { | |
651e4cef PX |
2051 | VTD_DPRINTF(GENERAL, "error: IRTE entry index %"PRIu16 |
2052 | " reserved fields non-zero: 0x%"PRIx64 " 0x%"PRIx64, | |
2053 | index, le64_to_cpu(entry->data[1]), | |
2054 | le64_to_cpu(entry->data[0])); | |
2055 | return -VTD_FR_IR_IRTE_RSVD; | |
2056 | } | |
2057 | ||
ede9c94a PX |
2058 | if (sid != X86_IOMMU_SID_INVALID) { |
2059 | /* Validate IRTE SID */ | |
bc38ee10 MT |
2060 | source_id = le32_to_cpu(entry->irte.source_id); |
2061 | switch (entry->irte.sid_vtype) { | |
ede9c94a PX |
2062 | case VTD_SVT_NONE: |
2063 | VTD_DPRINTF(IR, "No SID validation for IRTE index %d", index); | |
2064 | break; | |
2065 | ||
2066 | case VTD_SVT_ALL: | |
bc38ee10 | 2067 | mask = vtd_svt_mask[entry->irte.sid_q]; |
ede9c94a PX |
2068 | if ((source_id & mask) != (sid & mask)) { |
2069 | VTD_DPRINTF(GENERAL, "SID validation for IRTE index " | |
2070 | "%d failed (reqid 0x%04x sid 0x%04x)", index, | |
2071 | sid, source_id); | |
2072 | return -VTD_FR_IR_SID_ERR; | |
2073 | } | |
2074 | break; | |
2075 | ||
2076 | case VTD_SVT_BUS: | |
2077 | bus_max = source_id >> 8; | |
2078 | bus_min = source_id & 0xff; | |
2079 | bus = sid >> 8; | |
2080 | if (bus > bus_max || bus < bus_min) { | |
2081 | VTD_DPRINTF(GENERAL, "SID validation for IRTE index %d " | |
2082 | "failed (bus %d outside %d-%d)", index, bus, | |
2083 | bus_min, bus_max); | |
2084 | return -VTD_FR_IR_SID_ERR; | |
2085 | } | |
2086 | break; | |
2087 | ||
2088 | default: | |
2089 | VTD_DPRINTF(GENERAL, "Invalid SVT bits (0x%x) in IRTE index " | |
bc38ee10 | 2090 | "%d", entry->irte.sid_vtype, index); |
ede9c94a PX |
2091 | /* Take this as verification failure. */ |
2092 | return -VTD_FR_IR_SID_ERR; | |
2093 | break; | |
2094 | } | |
2095 | } | |
651e4cef PX |
2096 | |
2097 | return 0; | |
2098 | } | |
2099 | ||
2100 | /* Fetch IRQ information of specific IR index */ | |
ede9c94a PX |
2101 | static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index, |
2102 | VTDIrq *irq, uint16_t sid) | |
651e4cef | 2103 | { |
bc38ee10 | 2104 | VTD_IR_TableEntry irte = {}; |
651e4cef PX |
2105 | int ret = 0; |
2106 | ||
ede9c94a | 2107 | ret = vtd_irte_get(iommu, index, &irte, sid); |
651e4cef PX |
2108 | if (ret) { |
2109 | return ret; | |
2110 | } | |
2111 | ||
bc38ee10 MT |
2112 | irq->trigger_mode = irte.irte.trigger_mode; |
2113 | irq->vector = irte.irte.vector; | |
2114 | irq->delivery_mode = irte.irte.delivery_mode; | |
2115 | irq->dest = le32_to_cpu(irte.irte.dest_id); | |
28589311 | 2116 | if (!iommu->intr_eime) { |
651e4cef PX |
2117 | #define VTD_IR_APIC_DEST_MASK (0xff00ULL) |
2118 | #define VTD_IR_APIC_DEST_SHIFT (8) | |
28589311 JK |
2119 | irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >> |
2120 | VTD_IR_APIC_DEST_SHIFT; | |
2121 | } | |
bc38ee10 MT |
2122 | irq->dest_mode = irte.irte.dest_mode; |
2123 | irq->redir_hint = irte.irte.redir_hint; | |
651e4cef PX |
2124 | |
2125 | VTD_DPRINTF(IR, "remapping interrupt index %d: trig:%u,vec:%u," | |
2126 | "deliver:%u,dest:%u,dest_mode:%u", index, | |
2127 | irq->trigger_mode, irq->vector, irq->delivery_mode, | |
2128 | irq->dest, irq->dest_mode); | |
2129 | ||
2130 | return 0; | |
2131 | } | |
2132 | ||
2133 | /* Generate one MSI message from VTDIrq info */ | |
2134 | static void vtd_generate_msi_message(VTDIrq *irq, MSIMessage *msg_out) | |
2135 | { | |
2136 | VTD_MSIMessage msg = {}; | |
2137 | ||
2138 | /* Generate address bits */ | |
2139 | msg.dest_mode = irq->dest_mode; | |
2140 | msg.redir_hint = irq->redir_hint; | |
2141 | msg.dest = irq->dest; | |
32946019 | 2142 | msg.__addr_hi = irq->dest & 0xffffff00; |
651e4cef PX |
2143 | msg.__addr_head = cpu_to_le32(0xfee); |
2144 | /* Keep this from original MSI address bits */ | |
2145 | msg.__not_used = irq->msi_addr_last_bits; | |
2146 | ||
2147 | /* Generate data bits */ | |
2148 | msg.vector = irq->vector; | |
2149 | msg.delivery_mode = irq->delivery_mode; | |
2150 | msg.level = 1; | |
2151 | msg.trigger_mode = irq->trigger_mode; | |
2152 | ||
2153 | msg_out->address = msg.msi_addr; | |
2154 | msg_out->data = msg.msi_data; | |
2155 | } | |
2156 | ||
2157 | /* Interrupt remapping for MSI/MSI-X entry */ | |
2158 | static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu, | |
2159 | MSIMessage *origin, | |
ede9c94a PX |
2160 | MSIMessage *translated, |
2161 | uint16_t sid) | |
651e4cef PX |
2162 | { |
2163 | int ret = 0; | |
2164 | VTD_IR_MSIAddress addr; | |
2165 | uint16_t index; | |
09cd058a | 2166 | VTDIrq irq = {}; |
651e4cef PX |
2167 | |
2168 | assert(origin && translated); | |
2169 | ||
2170 | if (!iommu || !iommu->intr_enabled) { | |
2171 | goto do_not_translate; | |
2172 | } | |
2173 | ||
2174 | if (origin->address & VTD_MSI_ADDR_HI_MASK) { | |
2175 | VTD_DPRINTF(GENERAL, "error: MSI addr high 32 bits nonzero" | |
2176 | " during interrupt remapping: 0x%"PRIx32, | |
2177 | (uint32_t)((origin->address & VTD_MSI_ADDR_HI_MASK) >> \ | |
2178 | VTD_MSI_ADDR_HI_SHIFT)); | |
2179 | return -VTD_FR_IR_REQ_RSVD; | |
2180 | } | |
2181 | ||
2182 | addr.data = origin->address & VTD_MSI_ADDR_LO_MASK; | |
bc38ee10 | 2183 | if (le16_to_cpu(addr.addr.__head) != 0xfee) { |
651e4cef PX |
2184 | VTD_DPRINTF(GENERAL, "error: MSI addr low 32 bits invalid: " |
2185 | "0x%"PRIx32, addr.data); | |
2186 | return -VTD_FR_IR_REQ_RSVD; | |
2187 | } | |
2188 | ||
2189 | /* This is compatible mode. */ | |
bc38ee10 | 2190 | if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) { |
651e4cef PX |
2191 | goto do_not_translate; |
2192 | } | |
2193 | ||
bc38ee10 | 2194 | index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l); |
651e4cef PX |
2195 | |
2196 | #define VTD_IR_MSI_DATA_SUBHANDLE (0x0000ffff) | |
2197 | #define VTD_IR_MSI_DATA_RESERVED (0xffff0000) | |
2198 | ||
bc38ee10 | 2199 | if (addr.addr.sub_valid) { |
651e4cef PX |
2200 | /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */ |
2201 | index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE; | |
2202 | } | |
2203 | ||
ede9c94a | 2204 | ret = vtd_remap_irq_get(iommu, index, &irq, sid); |
651e4cef PX |
2205 | if (ret) { |
2206 | return ret; | |
2207 | } | |
2208 | ||
bc38ee10 | 2209 | if (addr.addr.sub_valid) { |
651e4cef PX |
2210 | VTD_DPRINTF(IR, "received MSI interrupt"); |
2211 | if (origin->data & VTD_IR_MSI_DATA_RESERVED) { | |
2212 | VTD_DPRINTF(GENERAL, "error: MSI data bits non-zero for " | |
2213 | "interrupt remappable entry: 0x%"PRIx32, | |
2214 | origin->data); | |
2215 | return -VTD_FR_IR_REQ_RSVD; | |
2216 | } | |
2217 | } else { | |
2218 | uint8_t vector = origin->data & 0xff; | |
dea651a9 FW |
2219 | uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1; |
2220 | ||
651e4cef PX |
2221 | VTD_DPRINTF(IR, "received IOAPIC interrupt"); |
2222 | /* IOAPIC entry vector should be aligned with IRTE vector | |
2223 | * (see vt-d spec 5.1.5.1). */ | |
2224 | if (vector != irq.vector) { | |
2225 | VTD_DPRINTF(GENERAL, "IOAPIC vector inconsistent: " | |
2226 | "entry: %d, IRTE: %d, index: %d", | |
2227 | vector, irq.vector, index); | |
2228 | } | |
dea651a9 FW |
2229 | |
2230 | /* The Trigger Mode field must match the Trigger Mode in the IRTE. | |
2231 | * (see vt-d spec 5.1.5.1). */ | |
2232 | if (trigger_mode != irq.trigger_mode) { | |
2233 | VTD_DPRINTF(GENERAL, "IOAPIC trigger mode inconsistent: " | |
2234 | "entry: %u, IRTE: %u, index: %d", | |
2235 | trigger_mode, irq.trigger_mode, index); | |
2236 | } | |
2237 | ||
651e4cef PX |
2238 | } |
2239 | ||
2240 | /* | |
2241 | * We'd better keep the last two bits, assuming that guest OS | |
2242 | * might modify it. Keep it does not hurt after all. | |
2243 | */ | |
bc38ee10 | 2244 | irq.msi_addr_last_bits = addr.addr.__not_care; |
651e4cef PX |
2245 | |
2246 | /* Translate VTDIrq to MSI message */ | |
2247 | vtd_generate_msi_message(&irq, translated); | |
2248 | ||
2249 | VTD_DPRINTF(IR, "mapping MSI 0x%"PRIx64":0x%"PRIx32 " -> " | |
2250 | "0x%"PRIx64":0x%"PRIx32, origin->address, origin->data, | |
2251 | translated->address, translated->data); | |
2252 | return 0; | |
2253 | ||
2254 | do_not_translate: | |
2255 | memcpy(translated, origin, sizeof(*origin)); | |
2256 | return 0; | |
2257 | } | |
2258 | ||
8b5ed7df PX |
2259 | static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src, |
2260 | MSIMessage *dst, uint16_t sid) | |
2261 | { | |
ede9c94a PX |
2262 | return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu), |
2263 | src, dst, sid); | |
8b5ed7df PX |
2264 | } |
2265 | ||
651e4cef PX |
2266 | static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr, |
2267 | uint64_t *data, unsigned size, | |
2268 | MemTxAttrs attrs) | |
2269 | { | |
2270 | return MEMTX_OK; | |
2271 | } | |
2272 | ||
2273 | static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr, | |
2274 | uint64_t value, unsigned size, | |
2275 | MemTxAttrs attrs) | |
2276 | { | |
2277 | int ret = 0; | |
09cd058a | 2278 | MSIMessage from = {}, to = {}; |
ede9c94a | 2279 | uint16_t sid = X86_IOMMU_SID_INVALID; |
651e4cef PX |
2280 | |
2281 | from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST; | |
2282 | from.data = (uint32_t) value; | |
2283 | ||
ede9c94a PX |
2284 | if (!attrs.unspecified) { |
2285 | /* We have explicit Source ID */ | |
2286 | sid = attrs.requester_id; | |
2287 | } | |
2288 | ||
2289 | ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid); | |
651e4cef PX |
2290 | if (ret) { |
2291 | /* TODO: report error */ | |
2292 | VTD_DPRINTF(GENERAL, "int remap fail for addr 0x%"PRIx64 | |
2293 | " data 0x%"PRIx32, from.address, from.data); | |
2294 | /* Drop this interrupt */ | |
2295 | return MEMTX_ERROR; | |
2296 | } | |
2297 | ||
2298 | VTD_DPRINTF(IR, "delivering MSI 0x%"PRIx64":0x%"PRIx32 | |
2299 | " for device sid 0x%04x", | |
2300 | to.address, to.data, sid); | |
2301 | ||
32946019 | 2302 | apic_get_class()->send_msi(&to); |
651e4cef PX |
2303 | |
2304 | return MEMTX_OK; | |
2305 | } | |
2306 | ||
2307 | static const MemoryRegionOps vtd_mem_ir_ops = { | |
2308 | .read_with_attrs = vtd_mem_ir_read, | |
2309 | .write_with_attrs = vtd_mem_ir_write, | |
2310 | .endianness = DEVICE_LITTLE_ENDIAN, | |
2311 | .impl = { | |
2312 | .min_access_size = 4, | |
2313 | .max_access_size = 4, | |
2314 | }, | |
2315 | .valid = { | |
2316 | .min_access_size = 4, | |
2317 | .max_access_size = 4, | |
2318 | }, | |
2319 | }; | |
7df953bd KO |
2320 | |
2321 | VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn) | |
2322 | { | |
2323 | uintptr_t key = (uintptr_t)bus; | |
2324 | VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key); | |
2325 | VTDAddressSpace *vtd_dev_as; | |
2326 | ||
2327 | if (!vtd_bus) { | |
2328 | /* No corresponding free() */ | |
04af0e18 PX |
2329 | vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \ |
2330 | X86_IOMMU_PCI_DEVFN_MAX); | |
7df953bd KO |
2331 | vtd_bus->bus = bus; |
2332 | key = (uintptr_t)bus; | |
2333 | g_hash_table_insert(s->vtd_as_by_busptr, &key, vtd_bus); | |
2334 | } | |
2335 | ||
2336 | vtd_dev_as = vtd_bus->dev_as[devfn]; | |
2337 | ||
2338 | if (!vtd_dev_as) { | |
2339 | vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace)); | |
2340 | ||
2341 | vtd_dev_as->bus = bus; | |
2342 | vtd_dev_as->devfn = (uint8_t)devfn; | |
2343 | vtd_dev_as->iommu_state = s; | |
2344 | vtd_dev_as->context_cache_entry.context_cache_gen = 0; | |
2345 | memory_region_init_iommu(&vtd_dev_as->iommu, OBJECT(s), | |
2346 | &s->iommu_ops, "intel_iommu", UINT64_MAX); | |
651e4cef PX |
2347 | memory_region_init_io(&vtd_dev_as->iommu_ir, OBJECT(s), |
2348 | &vtd_mem_ir_ops, s, "intel_iommu_ir", | |
2349 | VTD_INTERRUPT_ADDR_SIZE); | |
2350 | memory_region_add_subregion(&vtd_dev_as->iommu, VTD_INTERRUPT_ADDR_FIRST, | |
2351 | &vtd_dev_as->iommu_ir); | |
7df953bd KO |
2352 | address_space_init(&vtd_dev_as->as, |
2353 | &vtd_dev_as->iommu, "intel_iommu"); | |
2354 | } | |
2355 | return vtd_dev_as; | |
2356 | } | |
2357 | ||
1da12ec4 LT |
2358 | /* Do the initialization. It will also be called when reset, so pay |
2359 | * attention when adding new initialization stuff. | |
2360 | */ | |
2361 | static void vtd_init(IntelIOMMUState *s) | |
2362 | { | |
d54bd7f8 PX |
2363 | X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s); |
2364 | ||
1da12ec4 LT |
2365 | memset(s->csr, 0, DMAR_REG_SIZE); |
2366 | memset(s->wmask, 0, DMAR_REG_SIZE); | |
2367 | memset(s->w1cmask, 0, DMAR_REG_SIZE); | |
2368 | memset(s->womask, 0, DMAR_REG_SIZE); | |
2369 | ||
2370 | s->iommu_ops.translate = vtd_iommu_translate; | |
5bf3d319 | 2371 | s->iommu_ops.notify_flag_changed = vtd_iommu_notify_flag_changed; |
1da12ec4 LT |
2372 | s->root = 0; |
2373 | s->root_extended = false; | |
2374 | s->dmar_enabled = false; | |
2375 | s->iq_head = 0; | |
2376 | s->iq_tail = 0; | |
2377 | s->iq = 0; | |
2378 | s->iq_size = 0; | |
2379 | s->qi_enabled = false; | |
2380 | s->iq_last_desc_type = VTD_INV_DESC_NONE; | |
2381 | s->next_frcd_reg = 0; | |
2382 | s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND | VTD_CAP_MGAW | | |
d66b969b | 2383 | VTD_CAP_SAGAW | VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS; |
ed7b8fbc | 2384 | s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO; |
1da12ec4 | 2385 | |
d54bd7f8 | 2386 | if (x86_iommu->intr_supported) { |
e6b6af05 RK |
2387 | s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV; |
2388 | if (s->intr_eim == ON_OFF_AUTO_ON) { | |
2389 | s->ecap |= VTD_ECAP_EIM; | |
2390 | } | |
2391 | assert(s->intr_eim != ON_OFF_AUTO_AUTO); | |
d54bd7f8 PX |
2392 | } |
2393 | ||
d92fa2dc | 2394 | vtd_reset_context_cache(s); |
b5a280c0 | 2395 | vtd_reset_iotlb(s); |
d92fa2dc | 2396 | |
1da12ec4 LT |
2397 | /* Define registers with default values and bit semantics */ |
2398 | vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0); | |
2399 | vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0); | |
2400 | vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0); | |
2401 | vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0); | |
2402 | vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL); | |
2403 | vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0); | |
2404 | vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffff000ULL, 0); | |
2405 | vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0); | |
2406 | vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL); | |
2407 | ||
2408 | /* Advanced Fault Logging not supported */ | |
2409 | vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL); | |
2410 | vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0); | |
2411 | vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0); | |
2412 | vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0); | |
2413 | ||
2414 | /* Treated as RsvdZ when EIM in ECAP_REG is not supported | |
2415 | * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0); | |
2416 | */ | |
2417 | vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0); | |
2418 | ||
2419 | /* Treated as RO for implementations that PLMR and PHMR fields reported | |
2420 | * as Clear in the CAP_REG. | |
2421 | * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0); | |
2422 | */ | |
2423 | vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0); | |
2424 | ||
ed7b8fbc LT |
2425 | vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0); |
2426 | vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0); | |
2427 | vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff007ULL, 0); | |
2428 | vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL); | |
2429 | vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0); | |
2430 | vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0); | |
2431 | vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0); | |
2432 | /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */ | |
2433 | vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0); | |
2434 | ||
1da12ec4 LT |
2435 | /* IOTLB registers */ |
2436 | vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0); | |
2437 | vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0); | |
2438 | vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL); | |
2439 | ||
2440 | /* Fault Recording Registers, 128-bit */ | |
2441 | vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0); | |
2442 | vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL); | |
a5861439 PX |
2443 | |
2444 | /* | |
28589311 | 2445 | * Interrupt remapping registers. |
a5861439 | 2446 | */ |
28589311 | 2447 | vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0); |
1da12ec4 LT |
2448 | } |
2449 | ||
2450 | /* Should not reset address_spaces when reset because devices will still use | |
2451 | * the address space they got at first (won't ask the bus again). | |
2452 | */ | |
2453 | static void vtd_reset(DeviceState *dev) | |
2454 | { | |
2455 | IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev); | |
2456 | ||
2457 | VTD_DPRINTF(GENERAL, ""); | |
2458 | vtd_init(s); | |
2459 | } | |
2460 | ||
621d983a MA |
2461 | static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn) |
2462 | { | |
2463 | IntelIOMMUState *s = opaque; | |
2464 | VTDAddressSpace *vtd_as; | |
2465 | ||
04af0e18 | 2466 | assert(0 <= devfn && devfn <= X86_IOMMU_PCI_DEVFN_MAX); |
621d983a MA |
2467 | |
2468 | vtd_as = vtd_find_add_as(s, bus, devfn); | |
2469 | return &vtd_as->as; | |
2470 | } | |
2471 | ||
e6b6af05 | 2472 | static bool vtd_decide_config(IntelIOMMUState *s, Error **errp) |
6333e93c | 2473 | { |
e6b6af05 RK |
2474 | X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s); |
2475 | ||
6333e93c RK |
2476 | /* Currently Intel IOMMU IR only support "kernel-irqchip={off|split}" */ |
2477 | if (x86_iommu->intr_supported && kvm_irqchip_in_kernel() && | |
2478 | !kvm_irqchip_is_split()) { | |
2479 | error_setg(errp, "Intel Interrupt Remapping cannot work with " | |
2480 | "kernel-irqchip=on, please use 'split|off'."); | |
2481 | return false; | |
2482 | } | |
e6b6af05 RK |
2483 | if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu->intr_supported) { |
2484 | error_setg(errp, "eim=on cannot be selected without intremap=on"); | |
2485 | return false; | |
2486 | } | |
2487 | ||
2488 | if (s->intr_eim == ON_OFF_AUTO_AUTO) { | |
fb506e70 RK |
2489 | s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim) |
2490 | && x86_iommu->intr_supported ? | |
e6b6af05 RK |
2491 | ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; |
2492 | } | |
fb506e70 RK |
2493 | if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) { |
2494 | if (!kvm_irqchip_in_kernel()) { | |
2495 | error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split"); | |
2496 | return false; | |
2497 | } | |
2498 | if (!kvm_enable_x2apic()) { | |
2499 | error_setg(errp, "eim=on requires support on the KVM side" | |
2500 | "(X2APIC_API, first shipped in v4.7)"); | |
2501 | return false; | |
2502 | } | |
2503 | } | |
e6b6af05 | 2504 | |
6333e93c RK |
2505 | return true; |
2506 | } | |
2507 | ||
1da12ec4 LT |
2508 | static void vtd_realize(DeviceState *dev, Error **errp) |
2509 | { | |
cb135f59 PX |
2510 | PCMachineState *pcms = PC_MACHINE(qdev_get_machine()); |
2511 | PCIBus *bus = pcms->bus; | |
1da12ec4 | 2512 | IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev); |
4684a204 | 2513 | X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev); |
1da12ec4 LT |
2514 | |
2515 | VTD_DPRINTF(GENERAL, ""); | |
fb9f5926 | 2516 | x86_iommu->type = TYPE_INTEL; |
6333e93c | 2517 | |
e6b6af05 | 2518 | if (!vtd_decide_config(s, errp)) { |
6333e93c RK |
2519 | return; |
2520 | } | |
2521 | ||
7df953bd | 2522 | memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num)); |
1da12ec4 LT |
2523 | memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s, |
2524 | "intel_iommu", DMAR_REG_SIZE); | |
2525 | sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem); | |
b5a280c0 LT |
2526 | /* No corresponding destroy */ |
2527 | s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal, | |
2528 | g_free, g_free); | |
7df953bd KO |
2529 | s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal, |
2530 | g_free, g_free); | |
1da12ec4 | 2531 | vtd_init(s); |
621d983a MA |
2532 | sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR); |
2533 | pci_setup_iommu(bus, vtd_host_dma_iommu, dev); | |
cb135f59 PX |
2534 | /* Pseudo address space under root PCI bus. */ |
2535 | pcms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC); | |
1da12ec4 LT |
2536 | } |
2537 | ||
2538 | static void vtd_class_init(ObjectClass *klass, void *data) | |
2539 | { | |
2540 | DeviceClass *dc = DEVICE_CLASS(klass); | |
1c7955c4 | 2541 | X86IOMMUClass *x86_class = X86_IOMMU_CLASS(klass); |
1da12ec4 LT |
2542 | |
2543 | dc->reset = vtd_reset; | |
1da12ec4 LT |
2544 | dc->vmsd = &vtd_vmstate; |
2545 | dc->props = vtd_properties; | |
621d983a | 2546 | dc->hotpluggable = false; |
1c7955c4 | 2547 | x86_class->realize = vtd_realize; |
8b5ed7df | 2548 | x86_class->int_remap = vtd_int_remap; |
1da12ec4 LT |
2549 | } |
2550 | ||
2551 | static const TypeInfo vtd_info = { | |
2552 | .name = TYPE_INTEL_IOMMU_DEVICE, | |
1c7955c4 | 2553 | .parent = TYPE_X86_IOMMU_DEVICE, |
1da12ec4 LT |
2554 | .instance_size = sizeof(IntelIOMMUState), |
2555 | .class_init = vtd_class_init, | |
2556 | }; | |
2557 | ||
2558 | static void vtd_register_types(void) | |
2559 | { | |
2560 | VTD_DPRINTF(GENERAL, ""); | |
2561 | type_register_static(&vtd_info); | |
2562 | } | |
2563 | ||
2564 | type_init(vtd_register_types) |