xref: /openbmc/qemu/hw/i386/amd_iommu.c (revision 8779fccb)
1 /*
2  * QEMU emulation of AMD IOMMU (AMD-Vi)
3  *
4  * Copyright (C) 2011 Eduard - Gabriel Munteanu
5  * Copyright (C) 2015 David Kiarie, <davidkiarie4@gmail.com>
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation; either version 2 of the License, or
10  * (at your option) any later version.
11 
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16 
17  * You should have received a copy of the GNU General Public License along
18  * with this program; if not, see <http://www.gnu.org/licenses/>.
19  *
20  * Cache implementation inspired by hw/i386/intel_iommu.c
21  */
22 #include "qemu/osdep.h"
23 #include "hw/i386/amd_iommu.h"
24 #include "qemu/error-report.h"
25 #include "trace.h"
26 
27 /* used AMD-Vi MMIO registers */
28 const char *amdvi_mmio_low[] = {
29     "AMDVI_MMIO_DEVTAB_BASE",
30     "AMDVI_MMIO_CMDBUF_BASE",
31     "AMDVI_MMIO_EVTLOG_BASE",
32     "AMDVI_MMIO_CONTROL",
33     "AMDVI_MMIO_EXCL_BASE",
34     "AMDVI_MMIO_EXCL_LIMIT",
35     "AMDVI_MMIO_EXT_FEATURES",
36     "AMDVI_MMIO_PPR_BASE",
37     "UNHANDLED"
38 };
39 const char *amdvi_mmio_high[] = {
40     "AMDVI_MMIO_COMMAND_HEAD",
41     "AMDVI_MMIO_COMMAND_TAIL",
42     "AMDVI_MMIO_EVTLOG_HEAD",
43     "AMDVI_MMIO_EVTLOG_TAIL",
44     "AMDVI_MMIO_STATUS",
45     "AMDVI_MMIO_PPR_HEAD",
46     "AMDVI_MMIO_PPR_TAIL",
47     "UNHANDLED"
48 };
49 
50 struct AMDVIAddressSpace {
51     uint8_t bus_num;            /* bus number                           */
52     uint8_t devfn;              /* device function                      */
53     AMDVIState *iommu_state;    /* AMDVI - one per machine              */
54     MemoryRegion iommu;         /* Device's address translation region  */
55     MemoryRegion iommu_ir;      /* Device's interrupt remapping region  */
56     AddressSpace as;            /* device's corresponding address space */
57 };
58 
59 /* AMDVI cache entry */
60 typedef struct AMDVIIOTLBEntry {
61     uint16_t domid;             /* assigned domain id  */
62     uint16_t devid;             /* device owning entry */
63     uint64_t perms;             /* access permissions  */
64     uint64_t translated_addr;   /* translated address  */
65     uint64_t page_mask;         /* physical page size  */
66 } AMDVIIOTLBEntry;
67 
68 /* configure MMIO registers at startup/reset */
69 static void amdvi_set_quad(AMDVIState *s, hwaddr addr, uint64_t val,
70                            uint64_t romask, uint64_t w1cmask)
71 {
72     stq_le_p(&s->mmior[addr], val);
73     stq_le_p(&s->romask[addr], romask);
74     stq_le_p(&s->w1cmask[addr], w1cmask);
75 }
76 
77 static uint16_t amdvi_readw(AMDVIState *s, hwaddr addr)
78 {
79     return lduw_le_p(&s->mmior[addr]);
80 }
81 
82 static uint32_t amdvi_readl(AMDVIState *s, hwaddr addr)
83 {
84     return ldl_le_p(&s->mmior[addr]);
85 }
86 
87 static uint64_t amdvi_readq(AMDVIState *s, hwaddr addr)
88 {
89     return ldq_le_p(&s->mmior[addr]);
90 }
91 
92 /* internal write */
93 static void amdvi_writeq_raw(AMDVIState *s, uint64_t val, hwaddr addr)
94 {
95     stq_le_p(&s->mmior[addr], val);
96 }
97 
98 /* external write */
99 static void amdvi_writew(AMDVIState *s, hwaddr addr, uint16_t val)
100 {
101     uint16_t romask = lduw_le_p(&s->romask[addr]);
102     uint16_t w1cmask = lduw_le_p(&s->w1cmask[addr]);
103     uint16_t oldval = lduw_le_p(&s->mmior[addr]);
104     stw_le_p(&s->mmior[addr],
105             ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
106 }
107 
108 static void amdvi_writel(AMDVIState *s, hwaddr addr, uint32_t val)
109 {
110     uint32_t romask = ldl_le_p(&s->romask[addr]);
111     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
112     uint32_t oldval = ldl_le_p(&s->mmior[addr]);
113     stl_le_p(&s->mmior[addr],
114             ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
115 }
116 
117 static void amdvi_writeq(AMDVIState *s, hwaddr addr, uint64_t val)
118 {
119     uint64_t romask = ldq_le_p(&s->romask[addr]);
120     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
121     uint32_t oldval = ldq_le_p(&s->mmior[addr]);
122     stq_le_p(&s->mmior[addr],
123             ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
124 }
125 
126 /* OR a 64-bit register with a 64-bit value */
127 static bool amdvi_test_mask(AMDVIState *s, hwaddr addr, uint64_t val)
128 {
129     return amdvi_readq(s, addr) | val;
130 }
131 
132 /* OR a 64-bit register with a 64-bit value storing result in the register */
133 static void amdvi_assign_orq(AMDVIState *s, hwaddr addr, uint64_t val)
134 {
135     amdvi_writeq_raw(s, addr, amdvi_readq(s, addr) | val);
136 }
137 
138 /* AND a 64-bit register with a 64-bit value storing result in the register */
139 static void amdvi_assign_andq(AMDVIState *s, hwaddr addr, uint64_t val)
140 {
141    amdvi_writeq_raw(s, addr, amdvi_readq(s, addr) & val);
142 }
143 
144 static void amdvi_generate_msi_interrupt(AMDVIState *s)
145 {
146     MSIMessage msg = {};
147     MemTxAttrs attrs = {
148         .requester_id = pci_requester_id(&s->pci.dev)
149     };
150 
151     if (msi_enabled(&s->pci.dev)) {
152         msg = msi_get_message(&s->pci.dev, 0);
153         address_space_stl_le(&address_space_memory, msg.address, msg.data,
154                              attrs, NULL);
155     }
156 }
157 
158 static void amdvi_log_event(AMDVIState *s, uint64_t *evt)
159 {
160     /* event logging not enabled */
161     if (!s->evtlog_enabled || amdvi_test_mask(s, AMDVI_MMIO_STATUS,
162         AMDVI_MMIO_STATUS_EVT_OVF)) {
163         return;
164     }
165 
166     /* event log buffer full */
167     if (s->evtlog_tail >= s->evtlog_len) {
168         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_EVT_OVF);
169         /* generate interrupt */
170         amdvi_generate_msi_interrupt(s);
171         return;
172     }
173 
174     if (dma_memory_write(&address_space_memory, s->evtlog + s->evtlog_tail,
175         &evt, AMDVI_EVENT_LEN)) {
176         trace_amdvi_evntlog_fail(s->evtlog, s->evtlog_tail);
177     }
178 
179     s->evtlog_tail += AMDVI_EVENT_LEN;
180     amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_COMP_INT);
181     amdvi_generate_msi_interrupt(s);
182 }
183 
184 static void amdvi_setevent_bits(uint64_t *buffer, uint64_t value, int start,
185                                 int length)
186 {
187     int index = start / 64, bitpos = start % 64;
188     uint64_t mask = MAKE_64BIT_MASK(start, length);
189     buffer[index] &= ~mask;
190     buffer[index] |= (value << bitpos) & mask;
191 }
192 /*
193  * AMDVi event structure
194  *    0:15   -> DeviceID
195  *    55:63  -> event type + miscellaneous info
196  *    63:127 -> related address
197  */
198 static void amdvi_encode_event(uint64_t *evt, uint16_t devid, uint64_t addr,
199                                uint16_t info)
200 {
201     amdvi_setevent_bits(evt, devid, 0, 16);
202     amdvi_setevent_bits(evt, info, 55, 8);
203     amdvi_setevent_bits(evt, addr, 63, 64);
204 }
205 /* log an error encountered during a page walk
206  *
207  * @addr: virtual address in translation request
208  */
209 static void amdvi_page_fault(AMDVIState *s, uint16_t devid,
210                              hwaddr addr, uint16_t info)
211 {
212     uint64_t evt[4];
213 
214     info |= AMDVI_EVENT_IOPF_I | AMDVI_EVENT_IOPF;
215     amdvi_encode_event(evt, devid, addr, info);
216     amdvi_log_event(s, evt);
217     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
218             PCI_STATUS_SIG_TARGET_ABORT);
219 }
220 /*
221  * log a master abort accessing device table
222  *  @devtab : address of device table entry
223  *  @info : error flags
224  */
225 static void amdvi_log_devtab_error(AMDVIState *s, uint16_t devid,
226                                    hwaddr devtab, uint16_t info)
227 {
228     uint64_t evt[4];
229 
230     info |= AMDVI_EVENT_DEV_TAB_HW_ERROR;
231 
232     amdvi_encode_event(evt, devid, devtab, info);
233     amdvi_log_event(s, evt);
234     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
235             PCI_STATUS_SIG_TARGET_ABORT);
236 }
237 /* log an event trying to access command buffer
238  *   @addr : address that couldn't be accessed
239  */
240 static void amdvi_log_command_error(AMDVIState *s, hwaddr addr)
241 {
242     uint64_t evt[4], info = AMDVI_EVENT_COMMAND_HW_ERROR;
243 
244     amdvi_encode_event(evt, 0, addr, info);
245     amdvi_log_event(s, evt);
246     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
247             PCI_STATUS_SIG_TARGET_ABORT);
248 }
249 /* log an illegal comand event
250  *   @addr : address of illegal command
251  */
252 static void amdvi_log_illegalcom_error(AMDVIState *s, uint16_t info,
253                                        hwaddr addr)
254 {
255     uint64_t evt[4];
256 
257     info |= AMDVI_EVENT_ILLEGAL_COMMAND_ERROR;
258     amdvi_encode_event(evt, 0, addr, info);
259     amdvi_log_event(s, evt);
260 }
261 /* log an error accessing device table
262  *
263  *  @devid : device owning the table entry
264  *  @devtab : address of device table entry
265  *  @info : error flags
266  */
267 static void amdvi_log_illegaldevtab_error(AMDVIState *s, uint16_t devid,
268                                           hwaddr addr, uint16_t info)
269 {
270     uint64_t evt[4];
271 
272     info |= AMDVI_EVENT_ILLEGAL_DEVTAB_ENTRY;
273     amdvi_encode_event(evt, devid, addr, info);
274     amdvi_log_event(s, evt);
275 }
276 /* log an error accessing a PTE entry
277  * @addr : address that couldn't be accessed
278  */
279 static void amdvi_log_pagetab_error(AMDVIState *s, uint16_t devid,
280                                     hwaddr addr, uint16_t info)
281 {
282     uint64_t evt[4];
283 
284     info |= AMDVI_EVENT_PAGE_TAB_HW_ERROR;
285     amdvi_encode_event(evt, devid, addr, info);
286     amdvi_log_event(s, evt);
287     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
288              PCI_STATUS_SIG_TARGET_ABORT);
289 }
290 
291 static gboolean amdvi_uint64_equal(gconstpointer v1, gconstpointer v2)
292 {
293     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
294 }
295 
296 static guint amdvi_uint64_hash(gconstpointer v)
297 {
298     return (guint)*(const uint64_t *)v;
299 }
300 
301 static AMDVIIOTLBEntry *amdvi_iotlb_lookup(AMDVIState *s, hwaddr addr,
302                                            uint64_t devid)
303 {
304     uint64_t key = (addr >> AMDVI_PAGE_SHIFT_4K) |
305                    ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
306     return g_hash_table_lookup(s->iotlb, &key);
307 }
308 
309 static void amdvi_iotlb_reset(AMDVIState *s)
310 {
311     assert(s->iotlb);
312     trace_amdvi_iotlb_reset();
313     g_hash_table_remove_all(s->iotlb);
314 }
315 
316 static gboolean amdvi_iotlb_remove_by_devid(gpointer key, gpointer value,
317                                             gpointer user_data)
318 {
319     AMDVIIOTLBEntry *entry = (AMDVIIOTLBEntry *)value;
320     uint16_t devid = *(uint16_t *)user_data;
321     return entry->devid == devid;
322 }
323 
324 static void amdvi_iotlb_remove_page(AMDVIState *s, hwaddr addr,
325                                     uint64_t devid)
326 {
327     uint64_t key = (addr >> AMDVI_PAGE_SHIFT_4K) |
328                    ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
329     g_hash_table_remove(s->iotlb, &key);
330 }
331 
332 static void amdvi_update_iotlb(AMDVIState *s, uint16_t devid,
333                                uint64_t gpa, IOMMUTLBEntry to_cache,
334                                uint16_t domid)
335 {
336     AMDVIIOTLBEntry *entry = g_new(AMDVIIOTLBEntry, 1);
337     uint64_t *key = g_new(uint64_t, 1);
338     uint64_t gfn = gpa >> AMDVI_PAGE_SHIFT_4K;
339 
340     /* don't cache erroneous translations */
341     if (to_cache.perm != IOMMU_NONE) {
342         trace_amdvi_cache_update(domid, PCI_BUS_NUM(devid), PCI_SLOT(devid),
343                 PCI_FUNC(devid), gpa, to_cache.translated_addr);
344 
345         if (g_hash_table_size(s->iotlb) >= AMDVI_IOTLB_MAX_SIZE) {
346             amdvi_iotlb_reset(s);
347         }
348 
349         entry->domid = domid;
350         entry->perms = to_cache.perm;
351         entry->translated_addr = to_cache.translated_addr;
352         entry->page_mask = to_cache.addr_mask;
353         *key = gfn | ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
354         g_hash_table_replace(s->iotlb, key, entry);
355     }
356 }
357 
358 static void amdvi_completion_wait(AMDVIState *s, uint64_t *cmd)
359 {
360     /* pad the last 3 bits */
361     hwaddr addr = cpu_to_le64(extract64(cmd[0], 3, 49)) << 3;
362     uint64_t data = cpu_to_le64(cmd[1]);
363 
364     if (extract64(cmd[0], 51, 8)) {
365         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
366                                    s->cmdbuf + s->cmdbuf_head);
367     }
368     if (extract64(cmd[0], 0, 1)) {
369         if (dma_memory_write(&address_space_memory, addr, &data,
370             AMDVI_COMPLETION_DATA_SIZE)) {
371             trace_amdvi_completion_wait_fail(addr);
372         }
373     }
374     /* set completion interrupt */
375     if (extract64(cmd[0], 1, 1)) {
376         amdvi_test_mask(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_COMP_INT);
377         /* generate interrupt */
378         amdvi_generate_msi_interrupt(s);
379     }
380     trace_amdvi_completion_wait(addr, data);
381 }
382 
383 /* log error without aborting since linux seems to be using reserved bits */
384 static void amdvi_inval_devtab_entry(AMDVIState *s, uint64_t *cmd)
385 {
386     uint16_t devid = cpu_to_le16((uint16_t)extract64(cmd[0], 0, 16));
387 
388     /* This command should invalidate internal caches of which there isn't */
389     if (extract64(cmd[0], 15, 16) || cmd[1]) {
390         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
391                                    s->cmdbuf + s->cmdbuf_head);
392     }
393     trace_amdvi_devtab_inval(PCI_BUS_NUM(devid), PCI_SLOT(devid),
394                              PCI_FUNC(devid));
395 }
396 
397 static void amdvi_complete_ppr(AMDVIState *s, uint64_t *cmd)
398 {
399     if (extract64(cmd[0], 15, 16) ||  extract64(cmd[0], 19, 8) ||
400         extract64(cmd[1], 0, 2) || extract64(cmd[1], 3, 29)
401         || extract64(cmd[1], 47, 16)) {
402         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
403                                    s->cmdbuf + s->cmdbuf_head);
404     }
405     trace_amdvi_ppr_exec();
406 }
407 
408 static void amdvi_inval_all(AMDVIState *s, uint64_t *cmd)
409 {
410     if (extract64(cmd[0], 0, 60) || cmd[1]) {
411         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
412                                    s->cmdbuf + s->cmdbuf_head);
413     }
414 
415     amdvi_iotlb_reset(s);
416     trace_amdvi_all_inval();
417 }
418 
419 static gboolean amdvi_iotlb_remove_by_domid(gpointer key, gpointer value,
420                                             gpointer user_data)
421 {
422     AMDVIIOTLBEntry *entry = (AMDVIIOTLBEntry *)value;
423     uint16_t domid = *(uint16_t *)user_data;
424     return entry->domid == domid;
425 }
426 
427 /* we don't have devid - we can't remove pages by address */
428 static void amdvi_inval_pages(AMDVIState *s, uint64_t *cmd)
429 {
430     uint16_t domid = cpu_to_le16((uint16_t)extract64(cmd[0], 32, 16));
431 
432     if (extract64(cmd[0], 20, 12) || extract64(cmd[0], 16, 12) ||
433         extract64(cmd[0], 3, 10)) {
434         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
435                                    s->cmdbuf + s->cmdbuf_head);
436     }
437 
438     g_hash_table_foreach_remove(s->iotlb, amdvi_iotlb_remove_by_domid,
439                                 &domid);
440     trace_amdvi_pages_inval(domid);
441 }
442 
443 static void amdvi_prefetch_pages(AMDVIState *s, uint64_t *cmd)
444 {
445     if (extract64(cmd[0], 16, 8) || extract64(cmd[0], 20, 8) ||
446         extract64(cmd[1], 1, 1) || extract64(cmd[1], 3, 1) ||
447         extract64(cmd[1], 5, 7)) {
448         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
449                                    s->cmdbuf + s->cmdbuf_head);
450     }
451 
452     trace_amdvi_prefetch_pages();
453 }
454 
455 static void amdvi_inval_inttable(AMDVIState *s, uint64_t *cmd)
456 {
457     if (extract64(cmd[0], 16, 16) || cmd[1]) {
458         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
459                                    s->cmdbuf + s->cmdbuf_head);
460         return;
461     }
462 
463     trace_amdvi_intr_inval();
464 }
465 
466 /* FIXME: Try to work with the specified size instead of all the pages
467  * when the S bit is on
468  */
469 static void iommu_inval_iotlb(AMDVIState *s, uint64_t *cmd)
470 {
471 
472     uint16_t devid = extract64(cmd[0], 0, 16);
473     if (extract64(cmd[1], 1, 1) || extract64(cmd[1], 3, 9)) {
474         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
475                                    s->cmdbuf + s->cmdbuf_head);
476         return;
477     }
478 
479     if (extract64(cmd[1], 0, 1)) {
480         g_hash_table_foreach_remove(s->iotlb, amdvi_iotlb_remove_by_devid,
481                                     &devid);
482     } else {
483         amdvi_iotlb_remove_page(s, cpu_to_le64(extract64(cmd[1], 12, 52)) << 12,
484                                 cpu_to_le16(extract64(cmd[1], 0, 16)));
485     }
486     trace_amdvi_iotlb_inval();
487 }
488 
489 /* not honouring reserved bits is regarded as an illegal command */
490 static void amdvi_cmdbuf_exec(AMDVIState *s)
491 {
492     uint64_t cmd[2];
493 
494     if (dma_memory_read(&address_space_memory, s->cmdbuf + s->cmdbuf_head,
495         cmd, AMDVI_COMMAND_SIZE)) {
496         trace_amdvi_command_read_fail(s->cmdbuf, s->cmdbuf_head);
497         amdvi_log_command_error(s, s->cmdbuf + s->cmdbuf_head);
498         return;
499     }
500 
501     switch (extract64(cmd[0], 60, 4)) {
502     case AMDVI_CMD_COMPLETION_WAIT:
503         amdvi_completion_wait(s, cmd);
504         break;
505     case AMDVI_CMD_INVAL_DEVTAB_ENTRY:
506         amdvi_inval_devtab_entry(s, cmd);
507         break;
508     case AMDVI_CMD_INVAL_AMDVI_PAGES:
509         amdvi_inval_pages(s, cmd);
510         break;
511     case AMDVI_CMD_INVAL_IOTLB_PAGES:
512         iommu_inval_iotlb(s, cmd);
513         break;
514     case AMDVI_CMD_INVAL_INTR_TABLE:
515         amdvi_inval_inttable(s, cmd);
516         break;
517     case AMDVI_CMD_PREFETCH_AMDVI_PAGES:
518         amdvi_prefetch_pages(s, cmd);
519         break;
520     case AMDVI_CMD_COMPLETE_PPR_REQUEST:
521         amdvi_complete_ppr(s, cmd);
522         break;
523     case AMDVI_CMD_INVAL_AMDVI_ALL:
524         amdvi_inval_all(s, cmd);
525         break;
526     default:
527         trace_amdvi_unhandled_command(extract64(cmd[1], 60, 4));
528         /* log illegal command */
529         amdvi_log_illegalcom_error(s, extract64(cmd[1], 60, 4),
530                                    s->cmdbuf + s->cmdbuf_head);
531     }
532 }
533 
534 static void amdvi_cmdbuf_run(AMDVIState *s)
535 {
536     if (!s->cmdbuf_enabled) {
537         trace_amdvi_command_error(amdvi_readq(s, AMDVI_MMIO_CONTROL));
538         return;
539     }
540 
541     /* check if there is work to do. */
542     while (s->cmdbuf_head != s->cmdbuf_tail) {
543         trace_amdvi_command_exec(s->cmdbuf_head, s->cmdbuf_tail, s->cmdbuf);
544         amdvi_cmdbuf_exec(s);
545         s->cmdbuf_head += AMDVI_COMMAND_SIZE;
546         amdvi_writeq_raw(s, s->cmdbuf_head, AMDVI_MMIO_COMMAND_HEAD);
547 
548         /* wrap head pointer */
549         if (s->cmdbuf_head >= s->cmdbuf_len * AMDVI_COMMAND_SIZE) {
550             s->cmdbuf_head = 0;
551         }
552     }
553 }
554 
555 static void amdvi_mmio_trace(hwaddr addr, unsigned size)
556 {
557     uint8_t index = (addr & ~0x2000) / 8;
558 
559     if ((addr & 0x2000)) {
560         /* high table */
561         index = index >= AMDVI_MMIO_REGS_HIGH ? AMDVI_MMIO_REGS_HIGH : index;
562         trace_amdvi_mmio_read(amdvi_mmio_high[index], addr, size, addr & ~0x07);
563     } else {
564         index = index >= AMDVI_MMIO_REGS_LOW ? AMDVI_MMIO_REGS_LOW : index;
565         trace_amdvi_mmio_read(amdvi_mmio_low[index], addr, size, addr & ~0x07);
566     }
567 }
568 
569 static uint64_t amdvi_mmio_read(void *opaque, hwaddr addr, unsigned size)
570 {
571     AMDVIState *s = opaque;
572 
573     uint64_t val = -1;
574     if (addr + size > AMDVI_MMIO_SIZE) {
575         trace_amdvi_mmio_read("error: addr outside region: max ",
576                 (uint64_t)AMDVI_MMIO_SIZE, addr, size);
577         return (uint64_t)-1;
578     }
579 
580     if (size == 2) {
581         val = amdvi_readw(s, addr);
582     } else if (size == 4) {
583         val = amdvi_readl(s, addr);
584     } else if (size == 8) {
585         val = amdvi_readq(s, addr);
586     }
587     amdvi_mmio_trace(addr, size);
588 
589     return val;
590 }
591 
592 static void amdvi_handle_control_write(AMDVIState *s)
593 {
594     unsigned long control = amdvi_readq(s, AMDVI_MMIO_CONTROL);
595     s->enabled = !!(control & AMDVI_MMIO_CONTROL_AMDVIEN);
596 
597     s->ats_enabled = !!(control & AMDVI_MMIO_CONTROL_HTTUNEN);
598     s->evtlog_enabled = s->enabled && !!(control &
599                         AMDVI_MMIO_CONTROL_EVENTLOGEN);
600 
601     s->evtlog_intr = !!(control & AMDVI_MMIO_CONTROL_EVENTINTEN);
602     s->completion_wait_intr = !!(control & AMDVI_MMIO_CONTROL_COMWAITINTEN);
603     s->cmdbuf_enabled = s->enabled && !!(control &
604                         AMDVI_MMIO_CONTROL_CMDBUFLEN);
605 
606     /* update the flags depending on the control register */
607     if (s->cmdbuf_enabled) {
608         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_CMDBUF_RUN);
609     } else {
610         amdvi_assign_andq(s, AMDVI_MMIO_STATUS, ~AMDVI_MMIO_STATUS_CMDBUF_RUN);
611     }
612     if (s->evtlog_enabled) {
613         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_EVT_RUN);
614     } else {
615         amdvi_assign_andq(s, AMDVI_MMIO_STATUS, ~AMDVI_MMIO_STATUS_EVT_RUN);
616     }
617 
618     trace_amdvi_control_status(control);
619     amdvi_cmdbuf_run(s);
620 }
621 
622 static inline void amdvi_handle_devtab_write(AMDVIState *s)
623 
624 {
625     uint64_t val = amdvi_readq(s, AMDVI_MMIO_DEVICE_TABLE);
626     s->devtab = (val & AMDVI_MMIO_DEVTAB_BASE_MASK);
627 
628     /* set device table length */
629     s->devtab_len = ((val & AMDVI_MMIO_DEVTAB_SIZE_MASK) + 1 *
630                     (AMDVI_MMIO_DEVTAB_SIZE_UNIT /
631                      AMDVI_MMIO_DEVTAB_ENTRY_SIZE));
632 }
633 
634 static inline void amdvi_handle_cmdhead_write(AMDVIState *s)
635 {
636     s->cmdbuf_head = amdvi_readq(s, AMDVI_MMIO_COMMAND_HEAD)
637                      & AMDVI_MMIO_CMDBUF_HEAD_MASK;
638     amdvi_cmdbuf_run(s);
639 }
640 
641 static inline void amdvi_handle_cmdbase_write(AMDVIState *s)
642 {
643     s->cmdbuf = amdvi_readq(s, AMDVI_MMIO_COMMAND_BASE)
644                 & AMDVI_MMIO_CMDBUF_BASE_MASK;
645     s->cmdbuf_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_CMDBUF_SIZE_BYTE)
646                     & AMDVI_MMIO_CMDBUF_SIZE_MASK);
647     s->cmdbuf_head = s->cmdbuf_tail = 0;
648 }
649 
650 static inline void amdvi_handle_cmdtail_write(AMDVIState *s)
651 {
652     s->cmdbuf_tail = amdvi_readq(s, AMDVI_MMIO_COMMAND_TAIL)
653                      & AMDVI_MMIO_CMDBUF_TAIL_MASK;
654     amdvi_cmdbuf_run(s);
655 }
656 
657 static inline void amdvi_handle_excllim_write(AMDVIState *s)
658 {
659     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EXCL_LIMIT);
660     s->excl_limit = (val & AMDVI_MMIO_EXCL_LIMIT_MASK) |
661                     AMDVI_MMIO_EXCL_LIMIT_LOW;
662 }
663 
664 static inline void amdvi_handle_evtbase_write(AMDVIState *s)
665 {
666     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_BASE);
667     s->evtlog = val & AMDVI_MMIO_EVTLOG_BASE_MASK;
668     s->evtlog_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_EVTLOG_SIZE_BYTE)
669                     & AMDVI_MMIO_EVTLOG_SIZE_MASK);
670 }
671 
672 static inline void amdvi_handle_evttail_write(AMDVIState *s)
673 {
674     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_TAIL);
675     s->evtlog_tail = val & AMDVI_MMIO_EVTLOG_TAIL_MASK;
676 }
677 
678 static inline void amdvi_handle_evthead_write(AMDVIState *s)
679 {
680     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_HEAD);
681     s->evtlog_head = val & AMDVI_MMIO_EVTLOG_HEAD_MASK;
682 }
683 
684 static inline void amdvi_handle_pprbase_write(AMDVIState *s)
685 {
686     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_BASE);
687     s->ppr_log = val & AMDVI_MMIO_PPRLOG_BASE_MASK;
688     s->pprlog_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_PPRLOG_SIZE_BYTE)
689                     & AMDVI_MMIO_PPRLOG_SIZE_MASK);
690 }
691 
692 static inline void amdvi_handle_pprhead_write(AMDVIState *s)
693 {
694     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_HEAD);
695     s->pprlog_head = val & AMDVI_MMIO_PPRLOG_HEAD_MASK;
696 }
697 
698 static inline void amdvi_handle_pprtail_write(AMDVIState *s)
699 {
700     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_TAIL);
701     s->pprlog_tail = val & AMDVI_MMIO_PPRLOG_TAIL_MASK;
702 }
703 
704 /* FIXME: something might go wrong if System Software writes in chunks
705  * of one byte but linux writes in chunks of 4 bytes so currently it
706  * works correctly with linux but will definitely be busted if software
707  * reads/writes 8 bytes
708  */
709 static void amdvi_mmio_reg_write(AMDVIState *s, unsigned size, uint64_t val,
710                                  hwaddr addr)
711 {
712     if (size == 2) {
713         amdvi_writew(s, addr, val);
714     } else if (size == 4) {
715         amdvi_writel(s, addr, val);
716     } else if (size == 8) {
717         amdvi_writeq(s, addr, val);
718     }
719 }
720 
721 static void amdvi_mmio_write(void *opaque, hwaddr addr, uint64_t val,
722                              unsigned size)
723 {
724     AMDVIState *s = opaque;
725     unsigned long offset = addr & 0x07;
726 
727     if (addr + size > AMDVI_MMIO_SIZE) {
728         trace_amdvi_mmio_write("error: addr outside region: max ",
729                 (uint64_t)AMDVI_MMIO_SIZE, size, val, offset);
730         return;
731     }
732 
733     amdvi_mmio_trace(addr, size);
734     switch (addr & ~0x07) {
735     case AMDVI_MMIO_CONTROL:
736         amdvi_mmio_reg_write(s, size, val, addr);
737         amdvi_handle_control_write(s);
738         break;
739     case AMDVI_MMIO_DEVICE_TABLE:
740         amdvi_mmio_reg_write(s, size, val, addr);
741        /*  set device table address
742         *   This also suffers from inability to tell whether software
743         *   is done writing
744         */
745         if (offset || (size == 8)) {
746             amdvi_handle_devtab_write(s);
747         }
748         break;
749     case AMDVI_MMIO_COMMAND_HEAD:
750         amdvi_mmio_reg_write(s, size, val, addr);
751         amdvi_handle_cmdhead_write(s);
752         break;
753     case AMDVI_MMIO_COMMAND_BASE:
754         amdvi_mmio_reg_write(s, size, val, addr);
755         /* FIXME - make sure System Software has finished writing incase
756          * it writes in chucks less than 8 bytes in a robust way.As for
757          * now, this hacks works for the linux driver
758          */
759         if (offset || (size == 8)) {
760             amdvi_handle_cmdbase_write(s);
761         }
762         break;
763     case AMDVI_MMIO_COMMAND_TAIL:
764         amdvi_mmio_reg_write(s, size, val, addr);
765         amdvi_handle_cmdtail_write(s);
766         break;
767     case AMDVI_MMIO_EVENT_BASE:
768         amdvi_mmio_reg_write(s, size, val, addr);
769         amdvi_handle_evtbase_write(s);
770         break;
771     case AMDVI_MMIO_EVENT_HEAD:
772         amdvi_mmio_reg_write(s, size, val, addr);
773         amdvi_handle_evthead_write(s);
774         break;
775     case AMDVI_MMIO_EVENT_TAIL:
776         amdvi_mmio_reg_write(s, size, val, addr);
777         amdvi_handle_evttail_write(s);
778         break;
779     case AMDVI_MMIO_EXCL_LIMIT:
780         amdvi_mmio_reg_write(s, size, val, addr);
781         amdvi_handle_excllim_write(s);
782         break;
783         /* PPR log base - unused for now */
784     case AMDVI_MMIO_PPR_BASE:
785         amdvi_mmio_reg_write(s, size, val, addr);
786         amdvi_handle_pprbase_write(s);
787         break;
788         /* PPR log head - also unused for now */
789     case AMDVI_MMIO_PPR_HEAD:
790         amdvi_mmio_reg_write(s, size, val, addr);
791         amdvi_handle_pprhead_write(s);
792         break;
793         /* PPR log tail - unused for now */
794     case AMDVI_MMIO_PPR_TAIL:
795         amdvi_mmio_reg_write(s, size, val, addr);
796         amdvi_handle_pprtail_write(s);
797         break;
798     }
799 }
800 
801 static inline uint64_t amdvi_get_perms(uint64_t entry)
802 {
803     return (entry & (AMDVI_DEV_PERM_READ | AMDVI_DEV_PERM_WRITE)) >>
804            AMDVI_DEV_PERM_SHIFT;
805 }
806 
807 /* a valid entry should have V = 1 and reserved bits honoured */
808 static bool amdvi_validate_dte(AMDVIState *s, uint16_t devid,
809                                uint64_t *dte)
810 {
811     if ((dte[0] & AMDVI_DTE_LOWER_QUAD_RESERVED)
812         || (dte[1] & AMDVI_DTE_MIDDLE_QUAD_RESERVED)
813         || (dte[2] & AMDVI_DTE_UPPER_QUAD_RESERVED) || dte[3]) {
814         amdvi_log_illegaldevtab_error(s, devid,
815                                       s->devtab +
816                                       devid * AMDVI_DEVTAB_ENTRY_SIZE, 0);
817         return false;
818     }
819 
820     return dte[0] & AMDVI_DEV_VALID;
821 }
822 
823 /* get a device table entry given the devid */
824 static bool amdvi_get_dte(AMDVIState *s, int devid, uint64_t *entry)
825 {
826     uint32_t offset = devid * AMDVI_DEVTAB_ENTRY_SIZE;
827 
828     if (dma_memory_read(&address_space_memory, s->devtab + offset, entry,
829         AMDVI_DEVTAB_ENTRY_SIZE)) {
830         trace_amdvi_dte_get_fail(s->devtab, offset);
831         /* log error accessing dte */
832         amdvi_log_devtab_error(s, devid, s->devtab + offset, 0);
833         return false;
834     }
835 
836     *entry = le64_to_cpu(*entry);
837     if (!amdvi_validate_dte(s, devid, entry)) {
838         trace_amdvi_invalid_dte(entry[0]);
839         return false;
840     }
841 
842     return true;
843 }
844 
845 /* get pte translation mode */
846 static inline uint8_t get_pte_translation_mode(uint64_t pte)
847 {
848     return (pte >> AMDVI_DEV_MODE_RSHIFT) & AMDVI_DEV_MODE_MASK;
849 }
850 
851 static inline uint64_t pte_override_page_mask(uint64_t pte)
852 {
853     uint8_t page_mask = 12;
854     uint64_t addr = (pte & AMDVI_DEV_PT_ROOT_MASK) ^ AMDVI_DEV_PT_ROOT_MASK;
855     /* find the first zero bit */
856     while (addr & 1) {
857         page_mask++;
858         addr = addr >> 1;
859     }
860 
861     return ~((1ULL << page_mask) - 1);
862 }
863 
864 static inline uint64_t pte_get_page_mask(uint64_t oldlevel)
865 {
866     return ~((1UL << ((oldlevel * 9) + 3)) - 1);
867 }
868 
869 static inline uint64_t amdvi_get_pte_entry(AMDVIState *s, uint64_t pte_addr,
870                                           uint16_t devid)
871 {
872     uint64_t pte;
873 
874     if (dma_memory_read(&address_space_memory, pte_addr, &pte, sizeof(pte))) {
875         trace_amdvi_get_pte_hwerror(pte_addr);
876         amdvi_log_pagetab_error(s, devid, pte_addr, 0);
877         pte = 0;
878         return pte;
879     }
880 
881     pte = le64_to_cpu(pte);
882     return pte;
883 }
884 
885 static void amdvi_page_walk(AMDVIAddressSpace *as, uint64_t *dte,
886                             IOMMUTLBEntry *ret, unsigned perms,
887                             hwaddr addr)
888 {
889     unsigned level, present, pte_perms, oldlevel;
890     uint64_t pte = dte[0], pte_addr, page_mask;
891 
892     /* make sure the DTE has TV = 1 */
893     if (pte & AMDVI_DEV_TRANSLATION_VALID) {
894         level = get_pte_translation_mode(pte);
895         if (level >= 7) {
896             trace_amdvi_mode_invalid(level, addr);
897             return;
898         }
899         if (level == 0) {
900             goto no_remap;
901         }
902 
903         /* we are at the leaf page table or page table encodes a huge page */
904         while (level > 0) {
905             pte_perms = amdvi_get_perms(pte);
906             present = pte & 1;
907             if (!present || perms != (perms & pte_perms)) {
908                 amdvi_page_fault(as->iommu_state, as->devfn, addr, perms);
909                 trace_amdvi_page_fault(addr);
910                 return;
911             }
912 
913             /* go to the next lower level */
914             pte_addr = pte & AMDVI_DEV_PT_ROOT_MASK;
915             /* add offset and load pte */
916             pte_addr += ((addr >> (3 + 9 * level)) & 0x1FF) << 3;
917             pte = amdvi_get_pte_entry(as->iommu_state, pte_addr, as->devfn);
918             if (!pte) {
919                 return;
920             }
921             oldlevel = level;
922             level = get_pte_translation_mode(pte);
923             if (level == 0x7) {
924                 break;
925             }
926         }
927 
928         if (level == 0x7) {
929             page_mask = pte_override_page_mask(pte);
930         } else {
931             page_mask = pte_get_page_mask(oldlevel);
932         }
933 
934         /* get access permissions from pte */
935         ret->iova = addr & page_mask;
936         ret->translated_addr = (pte & AMDVI_DEV_PT_ROOT_MASK) & page_mask;
937         ret->addr_mask = ~page_mask;
938         ret->perm = amdvi_get_perms(pte);
939         return;
940     }
941 no_remap:
942     ret->iova = addr & AMDVI_PAGE_MASK_4K;
943     ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
944     ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
945     ret->perm = amdvi_get_perms(pte);
946 }
947 
948 static void amdvi_do_translate(AMDVIAddressSpace *as, hwaddr addr,
949                                bool is_write, IOMMUTLBEntry *ret)
950 {
951     AMDVIState *s = as->iommu_state;
952     uint16_t devid = PCI_BUILD_BDF(as->bus_num, as->devfn);
953     AMDVIIOTLBEntry *iotlb_entry = amdvi_iotlb_lookup(s, addr, devid);
954     uint64_t entry[4];
955 
956     if (iotlb_entry) {
957         trace_amdvi_iotlb_hit(PCI_BUS_NUM(devid), PCI_SLOT(devid),
958                 PCI_FUNC(devid), addr, iotlb_entry->translated_addr);
959         ret->iova = addr & ~iotlb_entry->page_mask;
960         ret->translated_addr = iotlb_entry->translated_addr;
961         ret->addr_mask = iotlb_entry->page_mask;
962         ret->perm = iotlb_entry->perms;
963         return;
964     }
965 
966     /* devices with V = 0 are not translated */
967     if (!amdvi_get_dte(s, devid, entry)) {
968         goto out;
969     }
970 
971     amdvi_page_walk(as, entry, ret,
972                     is_write ? AMDVI_PERM_WRITE : AMDVI_PERM_READ, addr);
973 
974     amdvi_update_iotlb(s, devid, addr, *ret,
975                        entry[1] & AMDVI_DEV_DOMID_ID_MASK);
976     return;
977 
978 out:
979     ret->iova = addr & AMDVI_PAGE_MASK_4K;
980     ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
981     ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
982     ret->perm = IOMMU_RW;
983 }
984 
985 static inline bool amdvi_is_interrupt_addr(hwaddr addr)
986 {
987     return addr >= AMDVI_INT_ADDR_FIRST && addr <= AMDVI_INT_ADDR_LAST;
988 }
989 
990 static IOMMUTLBEntry amdvi_translate(MemoryRegion *iommu, hwaddr addr,
991                                      bool is_write)
992 {
993     AMDVIAddressSpace *as = container_of(iommu, AMDVIAddressSpace, iommu);
994     AMDVIState *s = as->iommu_state;
995     IOMMUTLBEntry ret = {
996         .target_as = &address_space_memory,
997         .iova = addr,
998         .translated_addr = 0,
999         .addr_mask = ~(hwaddr)0,
1000         .perm = IOMMU_NONE
1001     };
1002 
1003     if (!s->enabled) {
1004         /* AMDVI disabled - corresponds to iommu=off not
1005          * failure to provide any parameter
1006          */
1007         ret.iova = addr & AMDVI_PAGE_MASK_4K;
1008         ret.translated_addr = addr & AMDVI_PAGE_MASK_4K;
1009         ret.addr_mask = ~AMDVI_PAGE_MASK_4K;
1010         ret.perm = IOMMU_RW;
1011         return ret;
1012     } else if (amdvi_is_interrupt_addr(addr)) {
1013         ret.iova = addr & AMDVI_PAGE_MASK_4K;
1014         ret.translated_addr = addr & AMDVI_PAGE_MASK_4K;
1015         ret.addr_mask = ~AMDVI_PAGE_MASK_4K;
1016         ret.perm = IOMMU_WO;
1017         return ret;
1018     }
1019 
1020     amdvi_do_translate(as, addr, is_write, &ret);
1021     trace_amdvi_translation_result(as->bus_num, PCI_SLOT(as->devfn),
1022             PCI_FUNC(as->devfn), addr, ret.translated_addr);
1023     return ret;
1024 }
1025 
1026 static AddressSpace *amdvi_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
1027 {
1028     AMDVIState *s = opaque;
1029     AMDVIAddressSpace **iommu_as;
1030     int bus_num = pci_bus_num(bus);
1031 
1032     iommu_as = s->address_spaces[bus_num];
1033 
1034     /* allocate memory during the first run */
1035     if (!iommu_as) {
1036         iommu_as = g_malloc0(sizeof(AMDVIAddressSpace *) * PCI_DEVFN_MAX);
1037         s->address_spaces[bus_num] = iommu_as;
1038     }
1039 
1040     /* set up AMD-Vi region */
1041     if (!iommu_as[devfn]) {
1042         iommu_as[devfn] = g_malloc0(sizeof(AMDVIAddressSpace));
1043         iommu_as[devfn]->bus_num = (uint8_t)bus_num;
1044         iommu_as[devfn]->devfn = (uint8_t)devfn;
1045         iommu_as[devfn]->iommu_state = s;
1046 
1047         memory_region_init_iommu(&iommu_as[devfn]->iommu, OBJECT(s),
1048                                  &s->iommu_ops, "amd-iommu", UINT64_MAX);
1049         address_space_init(&iommu_as[devfn]->as, &iommu_as[devfn]->iommu,
1050                            "amd-iommu");
1051     }
1052     return &iommu_as[devfn]->as;
1053 }
1054 
1055 static const MemoryRegionOps mmio_mem_ops = {
1056     .read = amdvi_mmio_read,
1057     .write = amdvi_mmio_write,
1058     .endianness = DEVICE_LITTLE_ENDIAN,
1059     .impl = {
1060         .min_access_size = 1,
1061         .max_access_size = 8,
1062         .unaligned = false,
1063     },
1064     .valid = {
1065         .min_access_size = 1,
1066         .max_access_size = 8,
1067     }
1068 };
1069 
1070 static void amdvi_iommu_notify_flag_changed(MemoryRegion *iommu,
1071                                             IOMMUNotifierFlag old,
1072                                             IOMMUNotifierFlag new)
1073 {
1074     AMDVIAddressSpace *as = container_of(iommu, AMDVIAddressSpace, iommu);
1075 
1076     if (new & IOMMU_NOTIFIER_MAP) {
1077         error_report("device %02x.%02x.%x requires iommu notifier which is not "
1078                      "currently supported", as->bus_num, PCI_SLOT(as->devfn),
1079                      PCI_FUNC(as->devfn));
1080         exit(1);
1081     }
1082 }
1083 
1084 static void amdvi_init(AMDVIState *s)
1085 {
1086     amdvi_iotlb_reset(s);
1087 
1088     s->iommu_ops.translate = amdvi_translate;
1089     s->iommu_ops.notify_flag_changed = amdvi_iommu_notify_flag_changed;
1090     s->devtab_len = 0;
1091     s->cmdbuf_len = 0;
1092     s->cmdbuf_head = 0;
1093     s->cmdbuf_tail = 0;
1094     s->evtlog_head = 0;
1095     s->evtlog_tail = 0;
1096     s->excl_enabled = false;
1097     s->excl_allow = false;
1098     s->mmio_enabled = false;
1099     s->enabled = false;
1100     s->ats_enabled = false;
1101     s->cmdbuf_enabled = false;
1102 
1103     /* reset MMIO */
1104     memset(s->mmior, 0, AMDVI_MMIO_SIZE);
1105     amdvi_set_quad(s, AMDVI_MMIO_EXT_FEATURES, AMDVI_EXT_FEATURES,
1106             0xffffffffffffffef, 0);
1107     amdvi_set_quad(s, AMDVI_MMIO_STATUS, 0, 0x98, 0x67);
1108 
1109     /* reset device ident */
1110     pci_config_set_vendor_id(s->pci.dev.config, PCI_VENDOR_ID_AMD);
1111     pci_config_set_prog_interface(s->pci.dev.config, 00);
1112     pci_config_set_device_id(s->pci.dev.config, s->devid);
1113     pci_config_set_class(s->pci.dev.config, 0x0806);
1114 
1115     /* reset AMDVI specific capabilities, all r/o */
1116     pci_set_long(s->pci.dev.config + s->capab_offset, AMDVI_CAPAB_FEATURES);
1117     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_BAR_LOW,
1118                  s->mmio.addr & ~(0xffff0000));
1119     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_BAR_HIGH,
1120                 (s->mmio.addr & ~(0xffff)) >> 16);
1121     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_RANGE,
1122                  0xff000000);
1123     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_MISC, 0);
1124     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_MISC,
1125             AMDVI_MAX_PH_ADDR | AMDVI_MAX_GVA_ADDR | AMDVI_MAX_VA_ADDR);
1126 }
1127 
1128 static void amdvi_reset(DeviceState *dev)
1129 {
1130     AMDVIState *s = AMD_IOMMU_DEVICE(dev);
1131 
1132     msi_reset(&s->pci.dev);
1133     amdvi_init(s);
1134 }
1135 
1136 static void amdvi_realize(DeviceState *dev, Error **err)
1137 {
1138     int ret = 0;
1139     AMDVIState *s = AMD_IOMMU_DEVICE(dev);
1140     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
1141     PCIBus *bus = PC_MACHINE(qdev_get_machine())->bus;
1142     s->iotlb = g_hash_table_new_full(amdvi_uint64_hash,
1143                                      amdvi_uint64_equal, g_free, g_free);
1144 
1145     /* This device should take care of IOMMU PCI properties */
1146     x86_iommu->type = TYPE_AMD;
1147     qdev_set_parent_bus(DEVICE(&s->pci), &bus->qbus);
1148     object_property_set_bool(OBJECT(&s->pci), true, "realized", err);
1149     s->capab_offset = pci_add_capability(&s->pci.dev, AMDVI_CAPAB_ID_SEC, 0,
1150                                          AMDVI_CAPAB_SIZE);
1151     assert(s->capab_offset > 0);
1152     ret = pci_add_capability(&s->pci.dev, PCI_CAP_ID_MSI, 0, AMDVI_CAPAB_REG_SIZE);
1153     assert(ret > 0);
1154     ret = pci_add_capability(&s->pci.dev, PCI_CAP_ID_HT, 0, AMDVI_CAPAB_REG_SIZE);
1155     assert(ret > 0);
1156 
1157     /* set up MMIO */
1158     memory_region_init_io(&s->mmio, OBJECT(s), &mmio_mem_ops, s, "amdvi-mmio",
1159                           AMDVI_MMIO_SIZE);
1160 
1161     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->mmio);
1162     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, AMDVI_BASE_ADDR);
1163     pci_setup_iommu(bus, amdvi_host_dma_iommu, s);
1164     s->devid = object_property_get_int(OBJECT(&s->pci), "addr", err);
1165     msi_init(&s->pci.dev, 0, 1, true, false, err);
1166     amdvi_init(s);
1167 }
1168 
1169 static const VMStateDescription vmstate_amdvi = {
1170     .name = "amd-iommu",
1171     .unmigratable = 1
1172 };
1173 
1174 static void amdvi_instance_init(Object *klass)
1175 {
1176     AMDVIState *s = AMD_IOMMU_DEVICE(klass);
1177 
1178     object_initialize(&s->pci, sizeof(s->pci), TYPE_AMD_IOMMU_PCI);
1179 }
1180 
1181 static void amdvi_class_init(ObjectClass *klass, void* data)
1182 {
1183     DeviceClass *dc = DEVICE_CLASS(klass);
1184     X86IOMMUClass *dc_class = X86_IOMMU_CLASS(klass);
1185 
1186     dc->reset = amdvi_reset;
1187     dc->vmsd = &vmstate_amdvi;
1188     dc->hotpluggable = false;
1189     dc_class->realize = amdvi_realize;
1190 }
1191 
1192 static const TypeInfo amdvi = {
1193     .name = TYPE_AMD_IOMMU_DEVICE,
1194     .parent = TYPE_X86_IOMMU_DEVICE,
1195     .instance_size = sizeof(AMDVIState),
1196     .instance_init = amdvi_instance_init,
1197     .class_init = amdvi_class_init
1198 };
1199 
1200 static const TypeInfo amdviPCI = {
1201     .name = "AMDVI-PCI",
1202     .parent = TYPE_PCI_DEVICE,
1203     .instance_size = sizeof(AMDVIPCIState),
1204 };
1205 
1206 static void amdviPCI_register_types(void)
1207 {
1208     type_register_static(&amdviPCI);
1209     type_register_static(&amdvi);
1210 }
1211 
1212 type_init(amdviPCI_register_types);
1213