xref: /openbmc/qemu/hw/i386/amd_iommu.c (revision 7200fb21)
1 /*
2  * QEMU emulation of AMD IOMMU (AMD-Vi)
3  *
4  * Copyright (C) 2011 Eduard - Gabriel Munteanu
5  * Copyright (C) 2015, 2016 David Kiarie Kahurani
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 
23 #include "qemu/osdep.h"
24 #include "hw/i386/pc.h"
25 #include "hw/pci/msi.h"
26 #include "hw/pci/pci_bus.h"
27 #include "migration/vmstate.h"
28 #include "amd_iommu.h"
29 #include "qapi/error.h"
30 #include "qemu/error-report.h"
31 #include "hw/i386/apic_internal.h"
32 #include "trace.h"
33 #include "hw/i386/apic-msidef.h"
34 
35 /* used AMD-Vi MMIO registers */
36 const char *amdvi_mmio_low[] = {
37     "AMDVI_MMIO_DEVTAB_BASE",
38     "AMDVI_MMIO_CMDBUF_BASE",
39     "AMDVI_MMIO_EVTLOG_BASE",
40     "AMDVI_MMIO_CONTROL",
41     "AMDVI_MMIO_EXCL_BASE",
42     "AMDVI_MMIO_EXCL_LIMIT",
43     "AMDVI_MMIO_EXT_FEATURES",
44     "AMDVI_MMIO_PPR_BASE",
45     "UNHANDLED"
46 };
47 const char *amdvi_mmio_high[] = {
48     "AMDVI_MMIO_COMMAND_HEAD",
49     "AMDVI_MMIO_COMMAND_TAIL",
50     "AMDVI_MMIO_EVTLOG_HEAD",
51     "AMDVI_MMIO_EVTLOG_TAIL",
52     "AMDVI_MMIO_STATUS",
53     "AMDVI_MMIO_PPR_HEAD",
54     "AMDVI_MMIO_PPR_TAIL",
55     "UNHANDLED"
56 };
57 
58 struct AMDVIAddressSpace {
59     uint8_t bus_num;            /* bus number                           */
60     uint8_t devfn;              /* device function                      */
61     AMDVIState *iommu_state;    /* AMDVI - one per machine              */
62     MemoryRegion root;          /* AMDVI Root memory map region */
63     IOMMUMemoryRegion iommu;    /* Device's address translation region  */
64     MemoryRegion iommu_ir;      /* Device's interrupt remapping region  */
65     AddressSpace as;            /* device's corresponding address space */
66 };
67 
68 /* AMDVI cache entry */
69 typedef struct AMDVIIOTLBEntry {
70     uint16_t domid;             /* assigned domain id  */
71     uint16_t devid;             /* device owning entry */
72     uint64_t perms;             /* access permissions  */
73     uint64_t translated_addr;   /* translated address  */
74     uint64_t page_mask;         /* physical page size  */
75 } AMDVIIOTLBEntry;
76 
77 /* configure MMIO registers at startup/reset */
78 static void amdvi_set_quad(AMDVIState *s, hwaddr addr, uint64_t val,
79                            uint64_t romask, uint64_t w1cmask)
80 {
81     stq_le_p(&s->mmior[addr], val);
82     stq_le_p(&s->romask[addr], romask);
83     stq_le_p(&s->w1cmask[addr], w1cmask);
84 }
85 
86 static uint16_t amdvi_readw(AMDVIState *s, hwaddr addr)
87 {
88     return lduw_le_p(&s->mmior[addr]);
89 }
90 
91 static uint32_t amdvi_readl(AMDVIState *s, hwaddr addr)
92 {
93     return ldl_le_p(&s->mmior[addr]);
94 }
95 
96 static uint64_t amdvi_readq(AMDVIState *s, hwaddr addr)
97 {
98     return ldq_le_p(&s->mmior[addr]);
99 }
100 
101 /* internal write */
102 static void amdvi_writeq_raw(AMDVIState *s, hwaddr addr, uint64_t val)
103 {
104     stq_le_p(&s->mmior[addr], val);
105 }
106 
107 /* external write */
108 static void amdvi_writew(AMDVIState *s, hwaddr addr, uint16_t val)
109 {
110     uint16_t romask = lduw_le_p(&s->romask[addr]);
111     uint16_t w1cmask = lduw_le_p(&s->w1cmask[addr]);
112     uint16_t oldval = lduw_le_p(&s->mmior[addr]);
113     stw_le_p(&s->mmior[addr],
114             ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
115 }
116 
117 static void amdvi_writel(AMDVIState *s, hwaddr addr, uint32_t val)
118 {
119     uint32_t romask = ldl_le_p(&s->romask[addr]);
120     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
121     uint32_t oldval = ldl_le_p(&s->mmior[addr]);
122     stl_le_p(&s->mmior[addr],
123             ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
124 }
125 
126 static void amdvi_writeq(AMDVIState *s, hwaddr addr, uint64_t val)
127 {
128     uint64_t romask = ldq_le_p(&s->romask[addr]);
129     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
130     uint32_t oldval = ldq_le_p(&s->mmior[addr]);
131     stq_le_p(&s->mmior[addr],
132             ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
133 }
134 
135 /* OR a 64-bit register with a 64-bit value */
136 static bool amdvi_test_mask(AMDVIState *s, hwaddr addr, uint64_t val)
137 {
138     return amdvi_readq(s, addr) | val;
139 }
140 
141 /* OR a 64-bit register with a 64-bit value storing result in the register */
142 static void amdvi_assign_orq(AMDVIState *s, hwaddr addr, uint64_t val)
143 {
144     amdvi_writeq_raw(s, addr, amdvi_readq(s, addr) | val);
145 }
146 
147 /* AND a 64-bit register with a 64-bit value storing result in the register */
148 static void amdvi_assign_andq(AMDVIState *s, hwaddr addr, uint64_t val)
149 {
150    amdvi_writeq_raw(s, addr, amdvi_readq(s, addr) & val);
151 }
152 
153 static void amdvi_generate_msi_interrupt(AMDVIState *s)
154 {
155     MSIMessage msg = {};
156     MemTxAttrs attrs = {
157         .requester_id = pci_requester_id(&s->pci.dev)
158     };
159 
160     if (msi_enabled(&s->pci.dev)) {
161         msg = msi_get_message(&s->pci.dev, 0);
162         address_space_stl_le(&address_space_memory, msg.address, msg.data,
163                              attrs, NULL);
164     }
165 }
166 
167 static void amdvi_log_event(AMDVIState *s, uint64_t *evt)
168 {
169     /* event logging not enabled */
170     if (!s->evtlog_enabled || amdvi_test_mask(s, AMDVI_MMIO_STATUS,
171         AMDVI_MMIO_STATUS_EVT_OVF)) {
172         return;
173     }
174 
175     /* event log buffer full */
176     if (s->evtlog_tail >= s->evtlog_len) {
177         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_EVT_OVF);
178         /* generate interrupt */
179         amdvi_generate_msi_interrupt(s);
180         return;
181     }
182 
183     if (dma_memory_write(&address_space_memory, s->evtlog + s->evtlog_tail,
184                          evt, AMDVI_EVENT_LEN, MEMTXATTRS_UNSPECIFIED)) {
185         trace_amdvi_evntlog_fail(s->evtlog, s->evtlog_tail);
186     }
187 
188     s->evtlog_tail += AMDVI_EVENT_LEN;
189     amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_COMP_INT);
190     amdvi_generate_msi_interrupt(s);
191 }
192 
193 static void amdvi_setevent_bits(uint64_t *buffer, uint64_t value, int start,
194                                 int length)
195 {
196     int index = start / 64, bitpos = start % 64;
197     uint64_t mask = MAKE_64BIT_MASK(start, length);
198     buffer[index] &= ~mask;
199     buffer[index] |= (value << bitpos) & mask;
200 }
201 /*
202  * AMDVi event structure
203  *    0:15   -> DeviceID
204  *    48:63  -> event type + miscellaneous info
205  *    64:127 -> related address
206  */
207 static void amdvi_encode_event(uint64_t *evt, uint16_t devid, uint64_t addr,
208                                uint16_t info)
209 {
210     evt[0] = 0;
211     evt[1] = 0;
212 
213     amdvi_setevent_bits(evt, devid, 0, 16);
214     amdvi_setevent_bits(evt, info, 48, 16);
215     amdvi_setevent_bits(evt, addr, 64, 64);
216 }
217 /* log an error encountered during a page walk
218  *
219  * @addr: virtual address in translation request
220  */
221 static void amdvi_page_fault(AMDVIState *s, uint16_t devid,
222                              hwaddr addr, uint16_t info)
223 {
224     uint64_t evt[2];
225 
226     info |= AMDVI_EVENT_IOPF_I | AMDVI_EVENT_IOPF;
227     amdvi_encode_event(evt, devid, addr, info);
228     amdvi_log_event(s, evt);
229     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
230             PCI_STATUS_SIG_TARGET_ABORT);
231 }
232 /*
233  * log a master abort accessing device table
234  *  @devtab : address of device table entry
235  *  @info : error flags
236  */
237 static void amdvi_log_devtab_error(AMDVIState *s, uint16_t devid,
238                                    hwaddr devtab, uint16_t info)
239 {
240     uint64_t evt[2];
241 
242     info |= AMDVI_EVENT_DEV_TAB_HW_ERROR;
243 
244     amdvi_encode_event(evt, devid, devtab, 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 event trying to access command buffer
250  *   @addr : address that couldn't be accessed
251  */
252 static void amdvi_log_command_error(AMDVIState *s, hwaddr addr)
253 {
254     uint64_t evt[2];
255     uint16_t info = AMDVI_EVENT_COMMAND_HW_ERROR;
256 
257     amdvi_encode_event(evt, 0, addr, info);
258     amdvi_log_event(s, evt);
259     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
260             PCI_STATUS_SIG_TARGET_ABORT);
261 }
262 /* log an illegal command event
263  *   @addr : address of illegal command
264  */
265 static void amdvi_log_illegalcom_error(AMDVIState *s, uint16_t info,
266                                        hwaddr addr)
267 {
268     uint64_t evt[2];
269 
270     info |= AMDVI_EVENT_ILLEGAL_COMMAND_ERROR;
271     amdvi_encode_event(evt, 0, addr, info);
272     amdvi_log_event(s, evt);
273 }
274 /* log an error accessing device table
275  *
276  *  @devid : device owning the table entry
277  *  @devtab : address of device table entry
278  *  @info : error flags
279  */
280 static void amdvi_log_illegaldevtab_error(AMDVIState *s, uint16_t devid,
281                                           hwaddr addr, uint16_t info)
282 {
283     uint64_t evt[2];
284 
285     info |= AMDVI_EVENT_ILLEGAL_DEVTAB_ENTRY;
286     amdvi_encode_event(evt, devid, addr, info);
287     amdvi_log_event(s, evt);
288 }
289 /* log an error accessing a PTE entry
290  * @addr : address that couldn't be accessed
291  */
292 static void amdvi_log_pagetab_error(AMDVIState *s, uint16_t devid,
293                                     hwaddr addr, uint16_t info)
294 {
295     uint64_t evt[2];
296 
297     info |= AMDVI_EVENT_PAGE_TAB_HW_ERROR;
298     amdvi_encode_event(evt, devid, addr, info);
299     amdvi_log_event(s, evt);
300     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
301              PCI_STATUS_SIG_TARGET_ABORT);
302 }
303 
304 static gboolean amdvi_uint64_equal(gconstpointer v1, gconstpointer v2)
305 {
306     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
307 }
308 
309 static guint amdvi_uint64_hash(gconstpointer v)
310 {
311     return (guint)*(const uint64_t *)v;
312 }
313 
314 static AMDVIIOTLBEntry *amdvi_iotlb_lookup(AMDVIState *s, hwaddr addr,
315                                            uint64_t devid)
316 {
317     uint64_t key = (addr >> AMDVI_PAGE_SHIFT_4K) |
318                    ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
319     return g_hash_table_lookup(s->iotlb, &key);
320 }
321 
322 static void amdvi_iotlb_reset(AMDVIState *s)
323 {
324     assert(s->iotlb);
325     trace_amdvi_iotlb_reset();
326     g_hash_table_remove_all(s->iotlb);
327 }
328 
329 static gboolean amdvi_iotlb_remove_by_devid(gpointer key, gpointer value,
330                                             gpointer user_data)
331 {
332     AMDVIIOTLBEntry *entry = (AMDVIIOTLBEntry *)value;
333     uint16_t devid = *(uint16_t *)user_data;
334     return entry->devid == devid;
335 }
336 
337 static void amdvi_iotlb_remove_page(AMDVIState *s, hwaddr addr,
338                                     uint64_t devid)
339 {
340     uint64_t key = (addr >> AMDVI_PAGE_SHIFT_4K) |
341                    ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
342     g_hash_table_remove(s->iotlb, &key);
343 }
344 
345 static void amdvi_update_iotlb(AMDVIState *s, uint16_t devid,
346                                uint64_t gpa, IOMMUTLBEntry to_cache,
347                                uint16_t domid)
348 {
349     AMDVIIOTLBEntry *entry = g_new(AMDVIIOTLBEntry, 1);
350     uint64_t *key = g_new(uint64_t, 1);
351     uint64_t gfn = gpa >> AMDVI_PAGE_SHIFT_4K;
352 
353     /* don't cache erroneous translations */
354     if (to_cache.perm != IOMMU_NONE) {
355         trace_amdvi_cache_update(domid, PCI_BUS_NUM(devid), PCI_SLOT(devid),
356                 PCI_FUNC(devid), gpa, to_cache.translated_addr);
357 
358         if (g_hash_table_size(s->iotlb) >= AMDVI_IOTLB_MAX_SIZE) {
359             amdvi_iotlb_reset(s);
360         }
361 
362         entry->domid = domid;
363         entry->perms = to_cache.perm;
364         entry->translated_addr = to_cache.translated_addr;
365         entry->page_mask = to_cache.addr_mask;
366         *key = gfn | ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
367         g_hash_table_replace(s->iotlb, key, entry);
368     }
369 }
370 
371 static void amdvi_completion_wait(AMDVIState *s, uint64_t *cmd)
372 {
373     /* pad the last 3 bits */
374     hwaddr addr = cpu_to_le64(extract64(cmd[0], 3, 49)) << 3;
375     uint64_t data = cpu_to_le64(cmd[1]);
376 
377     if (extract64(cmd[0], 52, 8)) {
378         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
379                                    s->cmdbuf + s->cmdbuf_head);
380     }
381     if (extract64(cmd[0], 0, 1)) {
382         if (dma_memory_write(&address_space_memory, addr, &data,
383                              AMDVI_COMPLETION_DATA_SIZE,
384                              MEMTXATTRS_UNSPECIFIED)) {
385             trace_amdvi_completion_wait_fail(addr);
386         }
387     }
388     /* set completion interrupt */
389     if (extract64(cmd[0], 1, 1)) {
390         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_COMP_INT);
391         /* generate interrupt */
392         amdvi_generate_msi_interrupt(s);
393     }
394     trace_amdvi_completion_wait(addr, data);
395 }
396 
397 /* log error without aborting since linux seems to be using reserved bits */
398 static void amdvi_inval_devtab_entry(AMDVIState *s, uint64_t *cmd)
399 {
400     uint16_t devid = cpu_to_le16((uint16_t)extract64(cmd[0], 0, 16));
401 
402     /* This command should invalidate internal caches of which there isn't */
403     if (extract64(cmd[0], 16, 44) || cmd[1]) {
404         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
405                                    s->cmdbuf + s->cmdbuf_head);
406     }
407     trace_amdvi_devtab_inval(PCI_BUS_NUM(devid), PCI_SLOT(devid),
408                              PCI_FUNC(devid));
409 }
410 
411 static void amdvi_complete_ppr(AMDVIState *s, uint64_t *cmd)
412 {
413     if (extract64(cmd[0], 16, 16) ||  extract64(cmd[0], 52, 8) ||
414         extract64(cmd[1], 0, 2) || extract64(cmd[1], 3, 29)
415         || extract64(cmd[1], 48, 16)) {
416         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
417                                    s->cmdbuf + s->cmdbuf_head);
418     }
419     trace_amdvi_ppr_exec();
420 }
421 
422 static void amdvi_inval_all(AMDVIState *s, uint64_t *cmd)
423 {
424     if (extract64(cmd[0], 0, 60) || cmd[1]) {
425         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
426                                    s->cmdbuf + s->cmdbuf_head);
427     }
428 
429     amdvi_iotlb_reset(s);
430     trace_amdvi_all_inval();
431 }
432 
433 static gboolean amdvi_iotlb_remove_by_domid(gpointer key, gpointer value,
434                                             gpointer user_data)
435 {
436     AMDVIIOTLBEntry *entry = (AMDVIIOTLBEntry *)value;
437     uint16_t domid = *(uint16_t *)user_data;
438     return entry->domid == domid;
439 }
440 
441 /* we don't have devid - we can't remove pages by address */
442 static void amdvi_inval_pages(AMDVIState *s, uint64_t *cmd)
443 {
444     uint16_t domid = cpu_to_le16((uint16_t)extract64(cmd[0], 32, 16));
445 
446     if (extract64(cmd[0], 20, 12) || extract64(cmd[0], 48, 12) ||
447         extract64(cmd[1], 3, 9)) {
448         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
449                                    s->cmdbuf + s->cmdbuf_head);
450     }
451 
452     g_hash_table_foreach_remove(s->iotlb, amdvi_iotlb_remove_by_domid,
453                                 &domid);
454     trace_amdvi_pages_inval(domid);
455 }
456 
457 static void amdvi_prefetch_pages(AMDVIState *s, uint64_t *cmd)
458 {
459     if (extract64(cmd[0], 16, 8) || extract64(cmd[0], 52, 8) ||
460         extract64(cmd[1], 1, 1) || extract64(cmd[1], 3, 1) ||
461         extract64(cmd[1], 5, 7)) {
462         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
463                                    s->cmdbuf + s->cmdbuf_head);
464     }
465 
466     trace_amdvi_prefetch_pages();
467 }
468 
469 static void amdvi_inval_inttable(AMDVIState *s, uint64_t *cmd)
470 {
471     if (extract64(cmd[0], 16, 44) || cmd[1]) {
472         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
473                                    s->cmdbuf + s->cmdbuf_head);
474         return;
475     }
476 
477     trace_amdvi_intr_inval();
478 }
479 
480 /* FIXME: Try to work with the specified size instead of all the pages
481  * when the S bit is on
482  */
483 static void iommu_inval_iotlb(AMDVIState *s, uint64_t *cmd)
484 {
485 
486     uint16_t devid = extract64(cmd[0], 0, 16);
487     if (extract64(cmd[1], 1, 1) || extract64(cmd[1], 3, 1) ||
488         extract64(cmd[1], 6, 6)) {
489         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
490                                    s->cmdbuf + s->cmdbuf_head);
491         return;
492     }
493 
494     if (extract64(cmd[1], 0, 1)) {
495         g_hash_table_foreach_remove(s->iotlb, amdvi_iotlb_remove_by_devid,
496                                     &devid);
497     } else {
498         amdvi_iotlb_remove_page(s, cpu_to_le64(extract64(cmd[1], 12, 52)) << 12,
499                                 cpu_to_le16(extract64(cmd[1], 0, 16)));
500     }
501     trace_amdvi_iotlb_inval();
502 }
503 
504 /* not honouring reserved bits is regarded as an illegal command */
505 static void amdvi_cmdbuf_exec(AMDVIState *s)
506 {
507     uint64_t cmd[2];
508 
509     if (dma_memory_read(&address_space_memory, s->cmdbuf + s->cmdbuf_head,
510                         cmd, AMDVI_COMMAND_SIZE, MEMTXATTRS_UNSPECIFIED)) {
511         trace_amdvi_command_read_fail(s->cmdbuf, s->cmdbuf_head);
512         amdvi_log_command_error(s, s->cmdbuf + s->cmdbuf_head);
513         return;
514     }
515 
516     switch (extract64(cmd[0], 60, 4)) {
517     case AMDVI_CMD_COMPLETION_WAIT:
518         amdvi_completion_wait(s, cmd);
519         break;
520     case AMDVI_CMD_INVAL_DEVTAB_ENTRY:
521         amdvi_inval_devtab_entry(s, cmd);
522         break;
523     case AMDVI_CMD_INVAL_AMDVI_PAGES:
524         amdvi_inval_pages(s, cmd);
525         break;
526     case AMDVI_CMD_INVAL_IOTLB_PAGES:
527         iommu_inval_iotlb(s, cmd);
528         break;
529     case AMDVI_CMD_INVAL_INTR_TABLE:
530         amdvi_inval_inttable(s, cmd);
531         break;
532     case AMDVI_CMD_PREFETCH_AMDVI_PAGES:
533         amdvi_prefetch_pages(s, cmd);
534         break;
535     case AMDVI_CMD_COMPLETE_PPR_REQUEST:
536         amdvi_complete_ppr(s, cmd);
537         break;
538     case AMDVI_CMD_INVAL_AMDVI_ALL:
539         amdvi_inval_all(s, cmd);
540         break;
541     default:
542         trace_amdvi_unhandled_command(extract64(cmd[1], 60, 4));
543         /* log illegal command */
544         amdvi_log_illegalcom_error(s, extract64(cmd[1], 60, 4),
545                                    s->cmdbuf + s->cmdbuf_head);
546     }
547 }
548 
549 static void amdvi_cmdbuf_run(AMDVIState *s)
550 {
551     if (!s->cmdbuf_enabled) {
552         trace_amdvi_command_error(amdvi_readq(s, AMDVI_MMIO_CONTROL));
553         return;
554     }
555 
556     /* check if there is work to do. */
557     while (s->cmdbuf_head != s->cmdbuf_tail) {
558         trace_amdvi_command_exec(s->cmdbuf_head, s->cmdbuf_tail, s->cmdbuf);
559         amdvi_cmdbuf_exec(s);
560         s->cmdbuf_head += AMDVI_COMMAND_SIZE;
561         amdvi_writeq_raw(s, AMDVI_MMIO_COMMAND_HEAD, s->cmdbuf_head);
562 
563         /* wrap head pointer */
564         if (s->cmdbuf_head >= s->cmdbuf_len * AMDVI_COMMAND_SIZE) {
565             s->cmdbuf_head = 0;
566         }
567     }
568 }
569 
570 static void amdvi_mmio_trace(hwaddr addr, unsigned size)
571 {
572     uint8_t index = (addr & ~0x2000) / 8;
573 
574     if ((addr & 0x2000)) {
575         /* high table */
576         index = index >= AMDVI_MMIO_REGS_HIGH ? AMDVI_MMIO_REGS_HIGH : index;
577         trace_amdvi_mmio_read(amdvi_mmio_high[index], addr, size, addr & ~0x07);
578     } else {
579         index = index >= AMDVI_MMIO_REGS_LOW ? AMDVI_MMIO_REGS_LOW : index;
580         trace_amdvi_mmio_read(amdvi_mmio_low[index], addr, size, addr & ~0x07);
581     }
582 }
583 
584 static uint64_t amdvi_mmio_read(void *opaque, hwaddr addr, unsigned size)
585 {
586     AMDVIState *s = opaque;
587 
588     uint64_t val = -1;
589     if (addr + size > AMDVI_MMIO_SIZE) {
590         trace_amdvi_mmio_read_invalid(AMDVI_MMIO_SIZE, addr, size);
591         return (uint64_t)-1;
592     }
593 
594     if (size == 2) {
595         val = amdvi_readw(s, addr);
596     } else if (size == 4) {
597         val = amdvi_readl(s, addr);
598     } else if (size == 8) {
599         val = amdvi_readq(s, addr);
600     }
601     amdvi_mmio_trace(addr, size);
602 
603     return val;
604 }
605 
606 static void amdvi_handle_control_write(AMDVIState *s)
607 {
608     unsigned long control = amdvi_readq(s, AMDVI_MMIO_CONTROL);
609     s->enabled = !!(control & AMDVI_MMIO_CONTROL_AMDVIEN);
610 
611     s->ats_enabled = !!(control & AMDVI_MMIO_CONTROL_HTTUNEN);
612     s->evtlog_enabled = s->enabled && !!(control &
613                         AMDVI_MMIO_CONTROL_EVENTLOGEN);
614 
615     s->evtlog_intr = !!(control & AMDVI_MMIO_CONTROL_EVENTINTEN);
616     s->completion_wait_intr = !!(control & AMDVI_MMIO_CONTROL_COMWAITINTEN);
617     s->cmdbuf_enabled = s->enabled && !!(control &
618                         AMDVI_MMIO_CONTROL_CMDBUFLEN);
619     s->ga_enabled = !!(control & AMDVI_MMIO_CONTROL_GAEN);
620 
621     /* update the flags depending on the control register */
622     if (s->cmdbuf_enabled) {
623         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_CMDBUF_RUN);
624     } else {
625         amdvi_assign_andq(s, AMDVI_MMIO_STATUS, ~AMDVI_MMIO_STATUS_CMDBUF_RUN);
626     }
627     if (s->evtlog_enabled) {
628         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_EVT_RUN);
629     } else {
630         amdvi_assign_andq(s, AMDVI_MMIO_STATUS, ~AMDVI_MMIO_STATUS_EVT_RUN);
631     }
632 
633     trace_amdvi_control_status(control);
634     amdvi_cmdbuf_run(s);
635 }
636 
637 static inline void amdvi_handle_devtab_write(AMDVIState *s)
638 
639 {
640     uint64_t val = amdvi_readq(s, AMDVI_MMIO_DEVICE_TABLE);
641     s->devtab = (val & AMDVI_MMIO_DEVTAB_BASE_MASK);
642 
643     /* set device table length */
644     s->devtab_len = ((val & AMDVI_MMIO_DEVTAB_SIZE_MASK) + 1 *
645                     (AMDVI_MMIO_DEVTAB_SIZE_UNIT /
646                      AMDVI_MMIO_DEVTAB_ENTRY_SIZE));
647 }
648 
649 static inline void amdvi_handle_cmdhead_write(AMDVIState *s)
650 {
651     s->cmdbuf_head = amdvi_readq(s, AMDVI_MMIO_COMMAND_HEAD)
652                      & AMDVI_MMIO_CMDBUF_HEAD_MASK;
653     amdvi_cmdbuf_run(s);
654 }
655 
656 static inline void amdvi_handle_cmdbase_write(AMDVIState *s)
657 {
658     s->cmdbuf = amdvi_readq(s, AMDVI_MMIO_COMMAND_BASE)
659                 & AMDVI_MMIO_CMDBUF_BASE_MASK;
660     s->cmdbuf_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_CMDBUF_SIZE_BYTE)
661                     & AMDVI_MMIO_CMDBUF_SIZE_MASK);
662     s->cmdbuf_head = s->cmdbuf_tail = 0;
663 }
664 
665 static inline void amdvi_handle_cmdtail_write(AMDVIState *s)
666 {
667     s->cmdbuf_tail = amdvi_readq(s, AMDVI_MMIO_COMMAND_TAIL)
668                      & AMDVI_MMIO_CMDBUF_TAIL_MASK;
669     amdvi_cmdbuf_run(s);
670 }
671 
672 static inline void amdvi_handle_excllim_write(AMDVIState *s)
673 {
674     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EXCL_LIMIT);
675     s->excl_limit = (val & AMDVI_MMIO_EXCL_LIMIT_MASK) |
676                     AMDVI_MMIO_EXCL_LIMIT_LOW;
677 }
678 
679 static inline void amdvi_handle_evtbase_write(AMDVIState *s)
680 {
681     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_BASE);
682     s->evtlog = val & AMDVI_MMIO_EVTLOG_BASE_MASK;
683     s->evtlog_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_EVTLOG_SIZE_BYTE)
684                     & AMDVI_MMIO_EVTLOG_SIZE_MASK);
685 }
686 
687 static inline void amdvi_handle_evttail_write(AMDVIState *s)
688 {
689     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_TAIL);
690     s->evtlog_tail = val & AMDVI_MMIO_EVTLOG_TAIL_MASK;
691 }
692 
693 static inline void amdvi_handle_evthead_write(AMDVIState *s)
694 {
695     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_HEAD);
696     s->evtlog_head = val & AMDVI_MMIO_EVTLOG_HEAD_MASK;
697 }
698 
699 static inline void amdvi_handle_pprbase_write(AMDVIState *s)
700 {
701     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_BASE);
702     s->ppr_log = val & AMDVI_MMIO_PPRLOG_BASE_MASK;
703     s->pprlog_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_PPRLOG_SIZE_BYTE)
704                     & AMDVI_MMIO_PPRLOG_SIZE_MASK);
705 }
706 
707 static inline void amdvi_handle_pprhead_write(AMDVIState *s)
708 {
709     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_HEAD);
710     s->pprlog_head = val & AMDVI_MMIO_PPRLOG_HEAD_MASK;
711 }
712 
713 static inline void amdvi_handle_pprtail_write(AMDVIState *s)
714 {
715     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_TAIL);
716     s->pprlog_tail = val & AMDVI_MMIO_PPRLOG_TAIL_MASK;
717 }
718 
719 /* FIXME: something might go wrong if System Software writes in chunks
720  * of one byte but linux writes in chunks of 4 bytes so currently it
721  * works correctly with linux but will definitely be busted if software
722  * reads/writes 8 bytes
723  */
724 static void amdvi_mmio_reg_write(AMDVIState *s, unsigned size, uint64_t val,
725                                  hwaddr addr)
726 {
727     if (size == 2) {
728         amdvi_writew(s, addr, val);
729     } else if (size == 4) {
730         amdvi_writel(s, addr, val);
731     } else if (size == 8) {
732         amdvi_writeq(s, addr, val);
733     }
734 }
735 
736 static void amdvi_mmio_write(void *opaque, hwaddr addr, uint64_t val,
737                              unsigned size)
738 {
739     AMDVIState *s = opaque;
740     unsigned long offset = addr & 0x07;
741 
742     if (addr + size > AMDVI_MMIO_SIZE) {
743         trace_amdvi_mmio_write("error: addr outside region: max ",
744                 (uint64_t)AMDVI_MMIO_SIZE, size, val, offset);
745         return;
746     }
747 
748     amdvi_mmio_trace(addr, size);
749     switch (addr & ~0x07) {
750     case AMDVI_MMIO_CONTROL:
751         amdvi_mmio_reg_write(s, size, val, addr);
752         amdvi_handle_control_write(s);
753         break;
754     case AMDVI_MMIO_DEVICE_TABLE:
755         amdvi_mmio_reg_write(s, size, val, addr);
756        /*  set device table address
757         *   This also suffers from inability to tell whether software
758         *   is done writing
759         */
760         if (offset || (size == 8)) {
761             amdvi_handle_devtab_write(s);
762         }
763         break;
764     case AMDVI_MMIO_COMMAND_HEAD:
765         amdvi_mmio_reg_write(s, size, val, addr);
766         amdvi_handle_cmdhead_write(s);
767         break;
768     case AMDVI_MMIO_COMMAND_BASE:
769         amdvi_mmio_reg_write(s, size, val, addr);
770         /* FIXME - make sure System Software has finished writing in case
771          * it writes in chucks less than 8 bytes in a robust way.As for
772          * now, this hacks works for the linux driver
773          */
774         if (offset || (size == 8)) {
775             amdvi_handle_cmdbase_write(s);
776         }
777         break;
778     case AMDVI_MMIO_COMMAND_TAIL:
779         amdvi_mmio_reg_write(s, size, val, addr);
780         amdvi_handle_cmdtail_write(s);
781         break;
782     case AMDVI_MMIO_EVENT_BASE:
783         amdvi_mmio_reg_write(s, size, val, addr);
784         amdvi_handle_evtbase_write(s);
785         break;
786     case AMDVI_MMIO_EVENT_HEAD:
787         amdvi_mmio_reg_write(s, size, val, addr);
788         amdvi_handle_evthead_write(s);
789         break;
790     case AMDVI_MMIO_EVENT_TAIL:
791         amdvi_mmio_reg_write(s, size, val, addr);
792         amdvi_handle_evttail_write(s);
793         break;
794     case AMDVI_MMIO_EXCL_LIMIT:
795         amdvi_mmio_reg_write(s, size, val, addr);
796         amdvi_handle_excllim_write(s);
797         break;
798         /* PPR log base - unused for now */
799     case AMDVI_MMIO_PPR_BASE:
800         amdvi_mmio_reg_write(s, size, val, addr);
801         amdvi_handle_pprbase_write(s);
802         break;
803         /* PPR log head - also unused for now */
804     case AMDVI_MMIO_PPR_HEAD:
805         amdvi_mmio_reg_write(s, size, val, addr);
806         amdvi_handle_pprhead_write(s);
807         break;
808         /* PPR log tail - unused for now */
809     case AMDVI_MMIO_PPR_TAIL:
810         amdvi_mmio_reg_write(s, size, val, addr);
811         amdvi_handle_pprtail_write(s);
812         break;
813     }
814 }
815 
816 static inline uint64_t amdvi_get_perms(uint64_t entry)
817 {
818     return (entry & (AMDVI_DEV_PERM_READ | AMDVI_DEV_PERM_WRITE)) >>
819            AMDVI_DEV_PERM_SHIFT;
820 }
821 
822 /* validate that reserved bits are honoured */
823 static bool amdvi_validate_dte(AMDVIState *s, uint16_t devid,
824                                uint64_t *dte)
825 {
826     if ((dte[0] & AMDVI_DTE_LOWER_QUAD_RESERVED)
827         || (dte[1] & AMDVI_DTE_MIDDLE_QUAD_RESERVED)
828         || (dte[2] & AMDVI_DTE_UPPER_QUAD_RESERVED) || dte[3]) {
829         amdvi_log_illegaldevtab_error(s, devid,
830                                       s->devtab +
831                                       devid * AMDVI_DEVTAB_ENTRY_SIZE, 0);
832         return false;
833     }
834 
835     return true;
836 }
837 
838 /* get a device table entry given the devid */
839 static bool amdvi_get_dte(AMDVIState *s, int devid, uint64_t *entry)
840 {
841     uint32_t offset = devid * AMDVI_DEVTAB_ENTRY_SIZE;
842 
843     if (dma_memory_read(&address_space_memory, s->devtab + offset, entry,
844                         AMDVI_DEVTAB_ENTRY_SIZE, MEMTXATTRS_UNSPECIFIED)) {
845         trace_amdvi_dte_get_fail(s->devtab, offset);
846         /* log error accessing dte */
847         amdvi_log_devtab_error(s, devid, s->devtab + offset, 0);
848         return false;
849     }
850 
851     *entry = le64_to_cpu(*entry);
852     if (!amdvi_validate_dte(s, devid, entry)) {
853         trace_amdvi_invalid_dte(entry[0]);
854         return false;
855     }
856 
857     return true;
858 }
859 
860 /* get pte translation mode */
861 static inline uint8_t get_pte_translation_mode(uint64_t pte)
862 {
863     return (pte >> AMDVI_DEV_MODE_RSHIFT) & AMDVI_DEV_MODE_MASK;
864 }
865 
866 static inline uint64_t pte_override_page_mask(uint64_t pte)
867 {
868     uint8_t page_mask = 13;
869     uint64_t addr = (pte & AMDVI_DEV_PT_ROOT_MASK) >> 12;
870     /* find the first zero bit */
871     while (addr & 1) {
872         page_mask++;
873         addr = addr >> 1;
874     }
875 
876     return ~((1ULL << page_mask) - 1);
877 }
878 
879 static inline uint64_t pte_get_page_mask(uint64_t oldlevel)
880 {
881     return ~((1UL << ((oldlevel * 9) + 3)) - 1);
882 }
883 
884 static inline uint64_t amdvi_get_pte_entry(AMDVIState *s, uint64_t pte_addr,
885                                           uint16_t devid)
886 {
887     uint64_t pte;
888 
889     if (dma_memory_read(&address_space_memory, pte_addr,
890                         &pte, sizeof(pte), MEMTXATTRS_UNSPECIFIED)) {
891         trace_amdvi_get_pte_hwerror(pte_addr);
892         amdvi_log_pagetab_error(s, devid, pte_addr, 0);
893         pte = 0;
894         return pte;
895     }
896 
897     pte = le64_to_cpu(pte);
898     return pte;
899 }
900 
901 static void amdvi_page_walk(AMDVIAddressSpace *as, uint64_t *dte,
902                             IOMMUTLBEntry *ret, unsigned perms,
903                             hwaddr addr)
904 {
905     unsigned level, present, pte_perms, oldlevel;
906     uint64_t pte = dte[0], pte_addr, page_mask;
907 
908     /* make sure the DTE has TV = 1 */
909     if (pte & AMDVI_DEV_TRANSLATION_VALID) {
910         level = get_pte_translation_mode(pte);
911         if (level >= 7) {
912             trace_amdvi_mode_invalid(level, addr);
913             return;
914         }
915         if (level == 0) {
916             goto no_remap;
917         }
918 
919         /* we are at the leaf page table or page table encodes a huge page */
920         do {
921             pte_perms = amdvi_get_perms(pte);
922             present = pte & 1;
923             if (!present || perms != (perms & pte_perms)) {
924                 amdvi_page_fault(as->iommu_state, as->devfn, addr, perms);
925                 trace_amdvi_page_fault(addr);
926                 return;
927             }
928 
929             /* go to the next lower level */
930             pte_addr = pte & AMDVI_DEV_PT_ROOT_MASK;
931             /* add offset and load pte */
932             pte_addr += ((addr >> (3 + 9 * level)) & 0x1FF) << 3;
933             pte = amdvi_get_pte_entry(as->iommu_state, pte_addr, as->devfn);
934             if (!pte) {
935                 return;
936             }
937             oldlevel = level;
938             level = get_pte_translation_mode(pte);
939         } while (level > 0 && level < 7);
940 
941         if (level == 0x7) {
942             page_mask = pte_override_page_mask(pte);
943         } else {
944             page_mask = pte_get_page_mask(oldlevel);
945         }
946 
947         /* get access permissions from pte */
948         ret->iova = addr & page_mask;
949         ret->translated_addr = (pte & AMDVI_DEV_PT_ROOT_MASK) & page_mask;
950         ret->addr_mask = ~page_mask;
951         ret->perm = amdvi_get_perms(pte);
952         return;
953     }
954 no_remap:
955     ret->iova = addr & AMDVI_PAGE_MASK_4K;
956     ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
957     ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
958     ret->perm = amdvi_get_perms(pte);
959 }
960 
961 static void amdvi_do_translate(AMDVIAddressSpace *as, hwaddr addr,
962                                bool is_write, IOMMUTLBEntry *ret)
963 {
964     AMDVIState *s = as->iommu_state;
965     uint16_t devid = PCI_BUILD_BDF(as->bus_num, as->devfn);
966     AMDVIIOTLBEntry *iotlb_entry = amdvi_iotlb_lookup(s, addr, devid);
967     uint64_t entry[4];
968 
969     if (iotlb_entry) {
970         trace_amdvi_iotlb_hit(PCI_BUS_NUM(devid), PCI_SLOT(devid),
971                 PCI_FUNC(devid), addr, iotlb_entry->translated_addr);
972         ret->iova = addr & ~iotlb_entry->page_mask;
973         ret->translated_addr = iotlb_entry->translated_addr;
974         ret->addr_mask = iotlb_entry->page_mask;
975         ret->perm = iotlb_entry->perms;
976         return;
977     }
978 
979     if (!amdvi_get_dte(s, devid, entry)) {
980         return;
981     }
982 
983     /* devices with V = 0 are not translated */
984     if (!(entry[0] & AMDVI_DEV_VALID)) {
985         goto out;
986     }
987 
988     amdvi_page_walk(as, entry, ret,
989                     is_write ? AMDVI_PERM_WRITE : AMDVI_PERM_READ, addr);
990 
991     amdvi_update_iotlb(s, devid, addr, *ret,
992                        entry[1] & AMDVI_DEV_DOMID_ID_MASK);
993     return;
994 
995 out:
996     ret->iova = addr & AMDVI_PAGE_MASK_4K;
997     ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
998     ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
999     ret->perm = IOMMU_RW;
1000 }
1001 
1002 static inline bool amdvi_is_interrupt_addr(hwaddr addr)
1003 {
1004     return addr >= AMDVI_INT_ADDR_FIRST && addr <= AMDVI_INT_ADDR_LAST;
1005 }
1006 
1007 static IOMMUTLBEntry amdvi_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
1008                                      IOMMUAccessFlags flag, int iommu_idx)
1009 {
1010     AMDVIAddressSpace *as = container_of(iommu, AMDVIAddressSpace, iommu);
1011     AMDVIState *s = as->iommu_state;
1012     IOMMUTLBEntry ret = {
1013         .target_as = &address_space_memory,
1014         .iova = addr,
1015         .translated_addr = 0,
1016         .addr_mask = ~(hwaddr)0,
1017         .perm = IOMMU_NONE
1018     };
1019 
1020     if (!s->enabled) {
1021         /* AMDVI disabled - corresponds to iommu=off not
1022          * failure to provide any parameter
1023          */
1024         ret.iova = addr & AMDVI_PAGE_MASK_4K;
1025         ret.translated_addr = addr & AMDVI_PAGE_MASK_4K;
1026         ret.addr_mask = ~AMDVI_PAGE_MASK_4K;
1027         ret.perm = IOMMU_RW;
1028         return ret;
1029     } else if (amdvi_is_interrupt_addr(addr)) {
1030         ret.iova = addr & AMDVI_PAGE_MASK_4K;
1031         ret.translated_addr = addr & AMDVI_PAGE_MASK_4K;
1032         ret.addr_mask = ~AMDVI_PAGE_MASK_4K;
1033         ret.perm = IOMMU_WO;
1034         return ret;
1035     }
1036 
1037     amdvi_do_translate(as, addr, flag & IOMMU_WO, &ret);
1038     trace_amdvi_translation_result(as->bus_num, PCI_SLOT(as->devfn),
1039             PCI_FUNC(as->devfn), addr, ret.translated_addr);
1040     return ret;
1041 }
1042 
1043 static int amdvi_get_irte(AMDVIState *s, MSIMessage *origin, uint64_t *dte,
1044                           union irte *irte, uint16_t devid)
1045 {
1046     uint64_t irte_root, offset;
1047 
1048     irte_root = dte[2] & AMDVI_IR_PHYS_ADDR_MASK;
1049     offset = (origin->data & AMDVI_IRTE_OFFSET) << 2;
1050 
1051     trace_amdvi_ir_irte(irte_root, offset);
1052 
1053     if (dma_memory_read(&address_space_memory, irte_root + offset,
1054                         irte, sizeof(*irte), MEMTXATTRS_UNSPECIFIED)) {
1055         trace_amdvi_ir_err("failed to get irte");
1056         return -AMDVI_IR_GET_IRTE;
1057     }
1058 
1059     trace_amdvi_ir_irte_val(irte->val);
1060 
1061     return 0;
1062 }
1063 
1064 static int amdvi_int_remap_legacy(AMDVIState *iommu,
1065                                   MSIMessage *origin,
1066                                   MSIMessage *translated,
1067                                   uint64_t *dte,
1068                                   X86IOMMUIrq *irq,
1069                                   uint16_t sid)
1070 {
1071     int ret;
1072     union irte irte;
1073 
1074     /* get interrupt remapping table */
1075     ret = amdvi_get_irte(iommu, origin, dte, &irte, sid);
1076     if (ret < 0) {
1077         return ret;
1078     }
1079 
1080     if (!irte.fields.valid) {
1081         trace_amdvi_ir_target_abort("RemapEn is disabled");
1082         return -AMDVI_IR_TARGET_ABORT;
1083     }
1084 
1085     if (irte.fields.guest_mode) {
1086         error_report_once("guest mode is not zero");
1087         return -AMDVI_IR_ERR;
1088     }
1089 
1090     if (irte.fields.int_type > AMDVI_IOAPIC_INT_TYPE_ARBITRATED) {
1091         error_report_once("reserved int_type");
1092         return -AMDVI_IR_ERR;
1093     }
1094 
1095     irq->delivery_mode = irte.fields.int_type;
1096     irq->vector = irte.fields.vector;
1097     irq->dest_mode = irte.fields.dm;
1098     irq->redir_hint = irte.fields.rq_eoi;
1099     irq->dest = irte.fields.destination;
1100 
1101     return 0;
1102 }
1103 
1104 static int amdvi_get_irte_ga(AMDVIState *s, MSIMessage *origin, uint64_t *dte,
1105                              struct irte_ga *irte, uint16_t devid)
1106 {
1107     uint64_t irte_root, offset;
1108 
1109     irte_root = dte[2] & AMDVI_IR_PHYS_ADDR_MASK;
1110     offset = (origin->data & AMDVI_IRTE_OFFSET) << 4;
1111     trace_amdvi_ir_irte(irte_root, offset);
1112 
1113     if (dma_memory_read(&address_space_memory, irte_root + offset,
1114                         irte, sizeof(*irte), MEMTXATTRS_UNSPECIFIED)) {
1115         trace_amdvi_ir_err("failed to get irte_ga");
1116         return -AMDVI_IR_GET_IRTE;
1117     }
1118 
1119     trace_amdvi_ir_irte_ga_val(irte->hi.val, irte->lo.val);
1120     return 0;
1121 }
1122 
1123 static int amdvi_int_remap_ga(AMDVIState *iommu,
1124                               MSIMessage *origin,
1125                               MSIMessage *translated,
1126                               uint64_t *dte,
1127                               X86IOMMUIrq *irq,
1128                               uint16_t sid)
1129 {
1130     int ret;
1131     struct irte_ga irte;
1132 
1133     /* get interrupt remapping table */
1134     ret = amdvi_get_irte_ga(iommu, origin, dte, &irte, sid);
1135     if (ret < 0) {
1136         return ret;
1137     }
1138 
1139     if (!irte.lo.fields_remap.valid) {
1140         trace_amdvi_ir_target_abort("RemapEn is disabled");
1141         return -AMDVI_IR_TARGET_ABORT;
1142     }
1143 
1144     if (irte.lo.fields_remap.guest_mode) {
1145         error_report_once("guest mode is not zero");
1146         return -AMDVI_IR_ERR;
1147     }
1148 
1149     if (irte.lo.fields_remap.int_type > AMDVI_IOAPIC_INT_TYPE_ARBITRATED) {
1150         error_report_once("reserved int_type is set");
1151         return -AMDVI_IR_ERR;
1152     }
1153 
1154     irq->delivery_mode = irte.lo.fields_remap.int_type;
1155     irq->vector = irte.hi.fields.vector;
1156     irq->dest_mode = irte.lo.fields_remap.dm;
1157     irq->redir_hint = irte.lo.fields_remap.rq_eoi;
1158     irq->dest = irte.lo.fields_remap.destination;
1159 
1160     return 0;
1161 }
1162 
1163 static int __amdvi_int_remap_msi(AMDVIState *iommu,
1164                                  MSIMessage *origin,
1165                                  MSIMessage *translated,
1166                                  uint64_t *dte,
1167                                  X86IOMMUIrq *irq,
1168                                  uint16_t sid)
1169 {
1170     int ret;
1171     uint8_t int_ctl;
1172 
1173     int_ctl = (dte[2] >> AMDVI_IR_INTCTL_SHIFT) & 3;
1174     trace_amdvi_ir_intctl(int_ctl);
1175 
1176     switch (int_ctl) {
1177     case AMDVI_IR_INTCTL_PASS:
1178         memcpy(translated, origin, sizeof(*origin));
1179         return 0;
1180     case AMDVI_IR_INTCTL_REMAP:
1181         break;
1182     case AMDVI_IR_INTCTL_ABORT:
1183         trace_amdvi_ir_target_abort("int_ctl abort");
1184         return -AMDVI_IR_TARGET_ABORT;
1185     default:
1186         trace_amdvi_ir_err("int_ctl reserved");
1187         return -AMDVI_IR_ERR;
1188     }
1189 
1190     if (iommu->ga_enabled) {
1191         ret = amdvi_int_remap_ga(iommu, origin, translated, dte, irq, sid);
1192     } else {
1193         ret = amdvi_int_remap_legacy(iommu, origin, translated, dte, irq, sid);
1194     }
1195 
1196     return ret;
1197 }
1198 
1199 /* Interrupt remapping for MSI/MSI-X entry */
1200 static int amdvi_int_remap_msi(AMDVIState *iommu,
1201                                MSIMessage *origin,
1202                                MSIMessage *translated,
1203                                uint16_t sid)
1204 {
1205     int ret = 0;
1206     uint64_t pass = 0;
1207     uint64_t dte[4] = { 0 };
1208     X86IOMMUIrq irq = { 0 };
1209     uint8_t dest_mode, delivery_mode;
1210 
1211     assert(origin && translated);
1212 
1213     /*
1214      * When IOMMU is enabled, interrupt remap request will come either from
1215      * IO-APIC or PCI device. If interrupt is from PCI device then it will
1216      * have a valid requester id but if the interrupt is from IO-APIC
1217      * then requester id will be invalid.
1218      */
1219     if (sid == X86_IOMMU_SID_INVALID) {
1220         sid = AMDVI_IOAPIC_SB_DEVID;
1221     }
1222 
1223     trace_amdvi_ir_remap_msi_req(origin->address, origin->data, sid);
1224 
1225     /* check if device table entry is set before we go further. */
1226     if (!iommu || !iommu->devtab_len) {
1227         memcpy(translated, origin, sizeof(*origin));
1228         goto out;
1229     }
1230 
1231     if (!amdvi_get_dte(iommu, sid, dte)) {
1232         return -AMDVI_IR_ERR;
1233     }
1234 
1235     /* Check if IR is enabled in DTE */
1236     if (!(dte[2] & AMDVI_IR_REMAP_ENABLE)) {
1237         memcpy(translated, origin, sizeof(*origin));
1238         goto out;
1239     }
1240 
1241     /* validate that we are configure with intremap=on */
1242     if (!x86_iommu_ir_supported(X86_IOMMU_DEVICE(iommu))) {
1243         trace_amdvi_err("Interrupt remapping is enabled in the guest but "
1244                         "not in the host. Use intremap=on to enable interrupt "
1245                         "remapping in amd-iommu.");
1246         return -AMDVI_IR_ERR;
1247     }
1248 
1249     if (origin->address < AMDVI_INT_ADDR_FIRST ||
1250         origin->address + sizeof(origin->data) > AMDVI_INT_ADDR_LAST + 1) {
1251         trace_amdvi_err("MSI is not from IOAPIC.");
1252         return -AMDVI_IR_ERR;
1253     }
1254 
1255     /*
1256      * The MSI data register [10:8] are used to get the upstream interrupt type.
1257      *
1258      * See MSI/MSI-X format:
1259      * https://pdfs.semanticscholar.org/presentation/9420/c279e942eca568157711ef5c92b800c40a79.pdf
1260      * (page 5)
1261      */
1262     delivery_mode = (origin->data >> MSI_DATA_DELIVERY_MODE_SHIFT) & 7;
1263 
1264     switch (delivery_mode) {
1265     case AMDVI_IOAPIC_INT_TYPE_FIXED:
1266     case AMDVI_IOAPIC_INT_TYPE_ARBITRATED:
1267         trace_amdvi_ir_delivery_mode("fixed/arbitrated");
1268         ret = __amdvi_int_remap_msi(iommu, origin, translated, dte, &irq, sid);
1269         if (ret < 0) {
1270             goto remap_fail;
1271         } else {
1272             /* Translate IRQ to MSI messages */
1273             x86_iommu_irq_to_msi_message(&irq, translated);
1274             goto out;
1275         }
1276         break;
1277     case AMDVI_IOAPIC_INT_TYPE_SMI:
1278         error_report("SMI is not supported!");
1279         ret = -AMDVI_IR_ERR;
1280         break;
1281     case AMDVI_IOAPIC_INT_TYPE_NMI:
1282         pass = dte[3] & AMDVI_DEV_NMI_PASS_MASK;
1283         trace_amdvi_ir_delivery_mode("nmi");
1284         break;
1285     case AMDVI_IOAPIC_INT_TYPE_INIT:
1286         pass = dte[3] & AMDVI_DEV_INT_PASS_MASK;
1287         trace_amdvi_ir_delivery_mode("init");
1288         break;
1289     case AMDVI_IOAPIC_INT_TYPE_EINT:
1290         pass = dte[3] & AMDVI_DEV_EINT_PASS_MASK;
1291         trace_amdvi_ir_delivery_mode("eint");
1292         break;
1293     default:
1294         trace_amdvi_ir_delivery_mode("unsupported delivery_mode");
1295         ret = -AMDVI_IR_ERR;
1296         break;
1297     }
1298 
1299     if (ret < 0) {
1300         goto remap_fail;
1301     }
1302 
1303     /*
1304      * The MSI address register bit[2] is used to get the destination
1305      * mode. The dest_mode 1 is valid for fixed and arbitrated interrupts
1306      * only.
1307      */
1308     dest_mode = (origin->address >> MSI_ADDR_DEST_MODE_SHIFT) & 1;
1309     if (dest_mode) {
1310         trace_amdvi_ir_err("invalid dest_mode");
1311         ret = -AMDVI_IR_ERR;
1312         goto remap_fail;
1313     }
1314 
1315     if (pass) {
1316         memcpy(translated, origin, sizeof(*origin));
1317     } else {
1318         trace_amdvi_ir_err("passthrough is not enabled");
1319         ret = -AMDVI_IR_ERR;
1320         goto remap_fail;
1321     }
1322 
1323 out:
1324     trace_amdvi_ir_remap_msi(origin->address, origin->data,
1325                              translated->address, translated->data);
1326     return 0;
1327 
1328 remap_fail:
1329     return ret;
1330 }
1331 
1332 static int amdvi_int_remap(X86IOMMUState *iommu,
1333                            MSIMessage *origin,
1334                            MSIMessage *translated,
1335                            uint16_t sid)
1336 {
1337     return amdvi_int_remap_msi(AMD_IOMMU_DEVICE(iommu), origin,
1338                                translated, sid);
1339 }
1340 
1341 static MemTxResult amdvi_mem_ir_write(void *opaque, hwaddr addr,
1342                                       uint64_t value, unsigned size,
1343                                       MemTxAttrs attrs)
1344 {
1345     int ret;
1346     MSIMessage from = { 0, 0 }, to = { 0, 0 };
1347     uint16_t sid = AMDVI_IOAPIC_SB_DEVID;
1348 
1349     from.address = (uint64_t) addr + AMDVI_INT_ADDR_FIRST;
1350     from.data = (uint32_t) value;
1351 
1352     trace_amdvi_mem_ir_write_req(addr, value, size);
1353 
1354     if (!attrs.unspecified) {
1355         /* We have explicit Source ID */
1356         sid = attrs.requester_id;
1357     }
1358 
1359     ret = amdvi_int_remap_msi(opaque, &from, &to, sid);
1360     if (ret < 0) {
1361         /* TODO: log the event using IOMMU log event interface */
1362         error_report_once("failed to remap interrupt from devid 0x%x", sid);
1363         return MEMTX_ERROR;
1364     }
1365 
1366     apic_get_class(NULL)->send_msi(&to);
1367 
1368     trace_amdvi_mem_ir_write(to.address, to.data);
1369     return MEMTX_OK;
1370 }
1371 
1372 static MemTxResult amdvi_mem_ir_read(void *opaque, hwaddr addr,
1373                                      uint64_t *data, unsigned size,
1374                                      MemTxAttrs attrs)
1375 {
1376     return MEMTX_OK;
1377 }
1378 
1379 static const MemoryRegionOps amdvi_ir_ops = {
1380     .read_with_attrs = amdvi_mem_ir_read,
1381     .write_with_attrs = amdvi_mem_ir_write,
1382     .endianness = DEVICE_LITTLE_ENDIAN,
1383     .impl = {
1384         .min_access_size = 4,
1385         .max_access_size = 4,
1386     },
1387     .valid = {
1388         .min_access_size = 4,
1389         .max_access_size = 4,
1390     }
1391 };
1392 
1393 static AddressSpace *amdvi_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
1394 {
1395     char name[128];
1396     AMDVIState *s = opaque;
1397     AMDVIAddressSpace **iommu_as, *amdvi_dev_as;
1398     int bus_num = pci_bus_num(bus);
1399 
1400     iommu_as = s->address_spaces[bus_num];
1401 
1402     /* allocate memory during the first run */
1403     if (!iommu_as) {
1404         iommu_as = g_new0(AMDVIAddressSpace *, PCI_DEVFN_MAX);
1405         s->address_spaces[bus_num] = iommu_as;
1406     }
1407 
1408     /* set up AMD-Vi region */
1409     if (!iommu_as[devfn]) {
1410         snprintf(name, sizeof(name), "amd_iommu_devfn_%d", devfn);
1411 
1412         iommu_as[devfn] = g_new0(AMDVIAddressSpace, 1);
1413         iommu_as[devfn]->bus_num = (uint8_t)bus_num;
1414         iommu_as[devfn]->devfn = (uint8_t)devfn;
1415         iommu_as[devfn]->iommu_state = s;
1416 
1417         amdvi_dev_as = iommu_as[devfn];
1418 
1419         /*
1420          * Memory region relationships looks like (Address range shows
1421          * only lower 32 bits to make it short in length...):
1422          *
1423          * |-----------------+-------------------+----------|
1424          * | Name            | Address range     | Priority |
1425          * |-----------------+-------------------+----------+
1426          * | amdvi_root      | 00000000-ffffffff |        0 |
1427          * |  amdvi_iommu    | 00000000-ffffffff |        1 |
1428          * |  amdvi_iommu_ir | fee00000-feefffff |       64 |
1429          * |-----------------+-------------------+----------|
1430          */
1431         memory_region_init_iommu(&amdvi_dev_as->iommu,
1432                                  sizeof(amdvi_dev_as->iommu),
1433                                  TYPE_AMD_IOMMU_MEMORY_REGION,
1434                                  OBJECT(s),
1435                                  "amd_iommu", UINT64_MAX);
1436         memory_region_init(&amdvi_dev_as->root, OBJECT(s),
1437                            "amdvi_root", UINT64_MAX);
1438         address_space_init(&amdvi_dev_as->as, &amdvi_dev_as->root, name);
1439         memory_region_init_io(&amdvi_dev_as->iommu_ir, OBJECT(s),
1440                               &amdvi_ir_ops, s, "amd_iommu_ir",
1441                               AMDVI_INT_ADDR_SIZE);
1442         memory_region_add_subregion_overlap(&amdvi_dev_as->root,
1443                                             AMDVI_INT_ADDR_FIRST,
1444                                             &amdvi_dev_as->iommu_ir,
1445                                             64);
1446         memory_region_add_subregion_overlap(&amdvi_dev_as->root, 0,
1447                                             MEMORY_REGION(&amdvi_dev_as->iommu),
1448                                             1);
1449     }
1450     return &iommu_as[devfn]->as;
1451 }
1452 
1453 static const PCIIOMMUOps amdvi_iommu_ops = {
1454     .get_address_space = amdvi_host_dma_iommu,
1455 };
1456 
1457 static const MemoryRegionOps mmio_mem_ops = {
1458     .read = amdvi_mmio_read,
1459     .write = amdvi_mmio_write,
1460     .endianness = DEVICE_LITTLE_ENDIAN,
1461     .impl = {
1462         .min_access_size = 1,
1463         .max_access_size = 8,
1464         .unaligned = false,
1465     },
1466     .valid = {
1467         .min_access_size = 1,
1468         .max_access_size = 8,
1469     }
1470 };
1471 
1472 static int amdvi_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
1473                                            IOMMUNotifierFlag old,
1474                                            IOMMUNotifierFlag new,
1475                                            Error **errp)
1476 {
1477     AMDVIAddressSpace *as = container_of(iommu, AMDVIAddressSpace, iommu);
1478 
1479     if (new & IOMMU_NOTIFIER_MAP) {
1480         error_setg(errp,
1481                    "device %02x.%02x.%x requires iommu notifier which is not "
1482                    "currently supported", as->bus_num, PCI_SLOT(as->devfn),
1483                    PCI_FUNC(as->devfn));
1484         return -EINVAL;
1485     }
1486     return 0;
1487 }
1488 
1489 static void amdvi_init(AMDVIState *s)
1490 {
1491     amdvi_iotlb_reset(s);
1492 
1493     s->devtab_len = 0;
1494     s->cmdbuf_len = 0;
1495     s->cmdbuf_head = 0;
1496     s->cmdbuf_tail = 0;
1497     s->evtlog_head = 0;
1498     s->evtlog_tail = 0;
1499     s->excl_enabled = false;
1500     s->excl_allow = false;
1501     s->mmio_enabled = false;
1502     s->enabled = false;
1503     s->ats_enabled = false;
1504     s->cmdbuf_enabled = false;
1505 
1506     /* reset MMIO */
1507     memset(s->mmior, 0, AMDVI_MMIO_SIZE);
1508     amdvi_set_quad(s, AMDVI_MMIO_EXT_FEATURES, AMDVI_EXT_FEATURES,
1509             0xffffffffffffffef, 0);
1510     amdvi_set_quad(s, AMDVI_MMIO_STATUS, 0, 0x98, 0x67);
1511 }
1512 
1513 static void amdvi_pci_realize(PCIDevice *pdev, Error **errp)
1514 {
1515     AMDVIPCIState *s = AMD_IOMMU_PCI(pdev);
1516     int ret;
1517 
1518     ret = pci_add_capability(pdev, AMDVI_CAPAB_ID_SEC, 0,
1519                              AMDVI_CAPAB_SIZE, errp);
1520     if (ret < 0) {
1521         return;
1522     }
1523     s->capab_offset = ret;
1524 
1525     ret = pci_add_capability(pdev, PCI_CAP_ID_MSI, 0,
1526                              AMDVI_CAPAB_REG_SIZE, errp);
1527     if (ret < 0) {
1528         return;
1529     }
1530     ret = pci_add_capability(pdev, PCI_CAP_ID_HT, 0,
1531                              AMDVI_CAPAB_REG_SIZE, errp);
1532     if (ret < 0) {
1533         return;
1534     }
1535 
1536     if (msi_init(pdev, 0, 1, true, false, errp) < 0) {
1537         return;
1538     }
1539 
1540     /* reset device ident */
1541     pci_config_set_prog_interface(pdev->config, 0);
1542 
1543     /* reset AMDVI specific capabilities, all r/o */
1544     pci_set_long(pdev->config + s->capab_offset, AMDVI_CAPAB_FEATURES);
1545     pci_set_long(pdev->config + s->capab_offset + AMDVI_CAPAB_BAR_LOW,
1546                  AMDVI_BASE_ADDR & ~(0xffff0000));
1547     pci_set_long(pdev->config + s->capab_offset + AMDVI_CAPAB_BAR_HIGH,
1548                 (AMDVI_BASE_ADDR & ~(0xffff)) >> 16);
1549     pci_set_long(pdev->config + s->capab_offset + AMDVI_CAPAB_RANGE,
1550                  0xff000000);
1551     pci_set_long(pdev->config + s->capab_offset + AMDVI_CAPAB_MISC, 0);
1552     pci_set_long(pdev->config + s->capab_offset + AMDVI_CAPAB_MISC,
1553             AMDVI_MAX_PH_ADDR | AMDVI_MAX_GVA_ADDR | AMDVI_MAX_VA_ADDR);
1554 }
1555 
1556 static void amdvi_sysbus_reset(DeviceState *dev)
1557 {
1558     AMDVIState *s = AMD_IOMMU_DEVICE(dev);
1559 
1560     msi_reset(&s->pci.dev);
1561     amdvi_init(s);
1562 }
1563 
1564 static void amdvi_sysbus_realize(DeviceState *dev, Error **errp)
1565 {
1566     AMDVIState *s = AMD_IOMMU_DEVICE(dev);
1567     MachineState *ms = MACHINE(qdev_get_machine());
1568     PCMachineState *pcms = PC_MACHINE(ms);
1569     X86MachineState *x86ms = X86_MACHINE(ms);
1570     PCIBus *bus = pcms->bus;
1571 
1572     s->iotlb = g_hash_table_new_full(amdvi_uint64_hash,
1573                                      amdvi_uint64_equal, g_free, g_free);
1574 
1575     /* This device should take care of IOMMU PCI properties */
1576     if (!qdev_realize(DEVICE(&s->pci), &bus->qbus, errp)) {
1577         return;
1578     }
1579 
1580     /* Pseudo address space under root PCI bus. */
1581     x86ms->ioapic_as = amdvi_host_dma_iommu(bus, s, AMDVI_IOAPIC_SB_DEVID);
1582 
1583     /* set up MMIO */
1584     memory_region_init_io(&s->mmio, OBJECT(s), &mmio_mem_ops, s, "amdvi-mmio",
1585                           AMDVI_MMIO_SIZE);
1586     memory_region_add_subregion(get_system_memory(), AMDVI_BASE_ADDR,
1587                                 &s->mmio);
1588     pci_setup_iommu(bus, &amdvi_iommu_ops, s);
1589     amdvi_init(s);
1590 }
1591 
1592 static const VMStateDescription vmstate_amdvi_sysbus = {
1593     .name = "amd-iommu",
1594     .unmigratable = 1
1595 };
1596 
1597 static void amdvi_sysbus_instance_init(Object *klass)
1598 {
1599     AMDVIState *s = AMD_IOMMU_DEVICE(klass);
1600 
1601     object_initialize(&s->pci, sizeof(s->pci), TYPE_AMD_IOMMU_PCI);
1602 }
1603 
1604 static void amdvi_sysbus_class_init(ObjectClass *klass, void *data)
1605 {
1606     DeviceClass *dc = DEVICE_CLASS(klass);
1607     X86IOMMUClass *dc_class = X86_IOMMU_DEVICE_CLASS(klass);
1608 
1609     dc->reset = amdvi_sysbus_reset;
1610     dc->vmsd = &vmstate_amdvi_sysbus;
1611     dc->hotpluggable = false;
1612     dc_class->realize = amdvi_sysbus_realize;
1613     dc_class->int_remap = amdvi_int_remap;
1614     /* Supported by the pc-q35-* machine types */
1615     dc->user_creatable = true;
1616     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1617     dc->desc = "AMD IOMMU (AMD-Vi) DMA Remapping device";
1618 }
1619 
1620 static const TypeInfo amdvi_sysbus = {
1621     .name = TYPE_AMD_IOMMU_DEVICE,
1622     .parent = TYPE_X86_IOMMU_DEVICE,
1623     .instance_size = sizeof(AMDVIState),
1624     .instance_init = amdvi_sysbus_instance_init,
1625     .class_init = amdvi_sysbus_class_init
1626 };
1627 
1628 static void amdvi_pci_class_init(ObjectClass *klass, void *data)
1629 {
1630     DeviceClass *dc = DEVICE_CLASS(klass);
1631     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1632 
1633     k->vendor_id = PCI_VENDOR_ID_AMD;
1634     k->class_id = 0x0806;
1635     k->realize = amdvi_pci_realize;
1636 
1637     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1638     dc->desc = "AMD IOMMU (AMD-Vi) DMA Remapping device";
1639 }
1640 
1641 static const TypeInfo amdvi_pci = {
1642     .name = TYPE_AMD_IOMMU_PCI,
1643     .parent = TYPE_PCI_DEVICE,
1644     .instance_size = sizeof(AMDVIPCIState),
1645     .class_init = amdvi_pci_class_init,
1646     .interfaces = (InterfaceInfo[]) {
1647         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
1648         { },
1649     },
1650 };
1651 
1652 static void amdvi_iommu_memory_region_class_init(ObjectClass *klass, void *data)
1653 {
1654     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
1655 
1656     imrc->translate = amdvi_translate;
1657     imrc->notify_flag_changed = amdvi_iommu_notify_flag_changed;
1658 }
1659 
1660 static const TypeInfo amdvi_iommu_memory_region_info = {
1661     .parent = TYPE_IOMMU_MEMORY_REGION,
1662     .name = TYPE_AMD_IOMMU_MEMORY_REGION,
1663     .class_init = amdvi_iommu_memory_region_class_init,
1664 };
1665 
1666 static void amdvi_register_types(void)
1667 {
1668     type_register_static(&amdvi_pci);
1669     type_register_static(&amdvi_sysbus);
1670     type_register_static(&amdvi_iommu_memory_region_info);
1671 }
1672 
1673 type_init(amdvi_register_types);
1674