xref: /openbmc/qemu/hw/i386/amd_iommu.c (revision c00506aa)
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, uint64_t val, hwaddr addr)
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)) {
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  *    55:63  -> event type + miscellaneous info
205  *    63:127 -> related address
206  */
207 static void amdvi_encode_event(uint64_t *evt, uint16_t devid, uint64_t addr,
208                                uint16_t info)
209 {
210     amdvi_setevent_bits(evt, devid, 0, 16);
211     amdvi_setevent_bits(evt, info, 55, 8);
212     amdvi_setevent_bits(evt, addr, 63, 64);
213 }
214 /* log an error encountered during a page walk
215  *
216  * @addr: virtual address in translation request
217  */
218 static void amdvi_page_fault(AMDVIState *s, uint16_t devid,
219                              hwaddr addr, uint16_t info)
220 {
221     uint64_t evt[4];
222 
223     info |= AMDVI_EVENT_IOPF_I | AMDVI_EVENT_IOPF;
224     amdvi_encode_event(evt, devid, addr, info);
225     amdvi_log_event(s, evt);
226     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
227             PCI_STATUS_SIG_TARGET_ABORT);
228 }
229 /*
230  * log a master abort accessing device table
231  *  @devtab : address of device table entry
232  *  @info : error flags
233  */
234 static void amdvi_log_devtab_error(AMDVIState *s, uint16_t devid,
235                                    hwaddr devtab, uint16_t info)
236 {
237     uint64_t evt[4];
238 
239     info |= AMDVI_EVENT_DEV_TAB_HW_ERROR;
240 
241     amdvi_encode_event(evt, devid, devtab, info);
242     amdvi_log_event(s, evt);
243     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
244             PCI_STATUS_SIG_TARGET_ABORT);
245 }
246 /* log an event trying to access command buffer
247  *   @addr : address that couldn't be accessed
248  */
249 static void amdvi_log_command_error(AMDVIState *s, hwaddr addr)
250 {
251     uint64_t evt[4], info = AMDVI_EVENT_COMMAND_HW_ERROR;
252 
253     amdvi_encode_event(evt, 0, addr, info);
254     amdvi_log_event(s, evt);
255     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
256             PCI_STATUS_SIG_TARGET_ABORT);
257 }
258 /* log an illegal comand event
259  *   @addr : address of illegal command
260  */
261 static void amdvi_log_illegalcom_error(AMDVIState *s, uint16_t info,
262                                        hwaddr addr)
263 {
264     uint64_t evt[4];
265 
266     info |= AMDVI_EVENT_ILLEGAL_COMMAND_ERROR;
267     amdvi_encode_event(evt, 0, addr, info);
268     amdvi_log_event(s, evt);
269 }
270 /* log an error accessing device table
271  *
272  *  @devid : device owning the table entry
273  *  @devtab : address of device table entry
274  *  @info : error flags
275  */
276 static void amdvi_log_illegaldevtab_error(AMDVIState *s, uint16_t devid,
277                                           hwaddr addr, uint16_t info)
278 {
279     uint64_t evt[4];
280 
281     info |= AMDVI_EVENT_ILLEGAL_DEVTAB_ENTRY;
282     amdvi_encode_event(evt, devid, addr, info);
283     amdvi_log_event(s, evt);
284 }
285 /* log an error accessing a PTE entry
286  * @addr : address that couldn't be accessed
287  */
288 static void amdvi_log_pagetab_error(AMDVIState *s, uint16_t devid,
289                                     hwaddr addr, uint16_t info)
290 {
291     uint64_t evt[4];
292 
293     info |= AMDVI_EVENT_PAGE_TAB_HW_ERROR;
294     amdvi_encode_event(evt, devid, addr, info);
295     amdvi_log_event(s, evt);
296     pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
297              PCI_STATUS_SIG_TARGET_ABORT);
298 }
299 
300 static gboolean amdvi_uint64_equal(gconstpointer v1, gconstpointer v2)
301 {
302     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
303 }
304 
305 static guint amdvi_uint64_hash(gconstpointer v)
306 {
307     return (guint)*(const uint64_t *)v;
308 }
309 
310 static AMDVIIOTLBEntry *amdvi_iotlb_lookup(AMDVIState *s, hwaddr addr,
311                                            uint64_t devid)
312 {
313     uint64_t key = (addr >> AMDVI_PAGE_SHIFT_4K) |
314                    ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
315     return g_hash_table_lookup(s->iotlb, &key);
316 }
317 
318 static void amdvi_iotlb_reset(AMDVIState *s)
319 {
320     assert(s->iotlb);
321     trace_amdvi_iotlb_reset();
322     g_hash_table_remove_all(s->iotlb);
323 }
324 
325 static gboolean amdvi_iotlb_remove_by_devid(gpointer key, gpointer value,
326                                             gpointer user_data)
327 {
328     AMDVIIOTLBEntry *entry = (AMDVIIOTLBEntry *)value;
329     uint16_t devid = *(uint16_t *)user_data;
330     return entry->devid == devid;
331 }
332 
333 static void amdvi_iotlb_remove_page(AMDVIState *s, hwaddr addr,
334                                     uint64_t devid)
335 {
336     uint64_t key = (addr >> AMDVI_PAGE_SHIFT_4K) |
337                    ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
338     g_hash_table_remove(s->iotlb, &key);
339 }
340 
341 static void amdvi_update_iotlb(AMDVIState *s, uint16_t devid,
342                                uint64_t gpa, IOMMUTLBEntry to_cache,
343                                uint16_t domid)
344 {
345     AMDVIIOTLBEntry *entry = g_new(AMDVIIOTLBEntry, 1);
346     uint64_t *key = g_new(uint64_t, 1);
347     uint64_t gfn = gpa >> AMDVI_PAGE_SHIFT_4K;
348 
349     /* don't cache erroneous translations */
350     if (to_cache.perm != IOMMU_NONE) {
351         trace_amdvi_cache_update(domid, PCI_BUS_NUM(devid), PCI_SLOT(devid),
352                 PCI_FUNC(devid), gpa, to_cache.translated_addr);
353 
354         if (g_hash_table_size(s->iotlb) >= AMDVI_IOTLB_MAX_SIZE) {
355             amdvi_iotlb_reset(s);
356         }
357 
358         entry->domid = domid;
359         entry->perms = to_cache.perm;
360         entry->translated_addr = to_cache.translated_addr;
361         entry->page_mask = to_cache.addr_mask;
362         *key = gfn | ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
363         g_hash_table_replace(s->iotlb, key, entry);
364     }
365 }
366 
367 static void amdvi_completion_wait(AMDVIState *s, uint64_t *cmd)
368 {
369     /* pad the last 3 bits */
370     hwaddr addr = cpu_to_le64(extract64(cmd[0], 3, 49)) << 3;
371     uint64_t data = cpu_to_le64(cmd[1]);
372 
373     if (extract64(cmd[0], 52, 8)) {
374         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
375                                    s->cmdbuf + s->cmdbuf_head);
376     }
377     if (extract64(cmd[0], 0, 1)) {
378         if (dma_memory_write(&address_space_memory, addr, &data,
379             AMDVI_COMPLETION_DATA_SIZE)) {
380             trace_amdvi_completion_wait_fail(addr);
381         }
382     }
383     /* set completion interrupt */
384     if (extract64(cmd[0], 1, 1)) {
385         amdvi_test_mask(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_COMP_INT);
386         /* generate interrupt */
387         amdvi_generate_msi_interrupt(s);
388     }
389     trace_amdvi_completion_wait(addr, data);
390 }
391 
392 /* log error without aborting since linux seems to be using reserved bits */
393 static void amdvi_inval_devtab_entry(AMDVIState *s, uint64_t *cmd)
394 {
395     uint16_t devid = cpu_to_le16((uint16_t)extract64(cmd[0], 0, 16));
396 
397     /* This command should invalidate internal caches of which there isn't */
398     if (extract64(cmd[0], 16, 44) || cmd[1]) {
399         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
400                                    s->cmdbuf + s->cmdbuf_head);
401     }
402     trace_amdvi_devtab_inval(PCI_BUS_NUM(devid), PCI_SLOT(devid),
403                              PCI_FUNC(devid));
404 }
405 
406 static void amdvi_complete_ppr(AMDVIState *s, uint64_t *cmd)
407 {
408     if (extract64(cmd[0], 16, 16) ||  extract64(cmd[0], 52, 8) ||
409         extract64(cmd[1], 0, 2) || extract64(cmd[1], 3, 29)
410         || extract64(cmd[1], 48, 16)) {
411         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
412                                    s->cmdbuf + s->cmdbuf_head);
413     }
414     trace_amdvi_ppr_exec();
415 }
416 
417 static void amdvi_inval_all(AMDVIState *s, uint64_t *cmd)
418 {
419     if (extract64(cmd[0], 0, 60) || cmd[1]) {
420         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
421                                    s->cmdbuf + s->cmdbuf_head);
422     }
423 
424     amdvi_iotlb_reset(s);
425     trace_amdvi_all_inval();
426 }
427 
428 static gboolean amdvi_iotlb_remove_by_domid(gpointer key, gpointer value,
429                                             gpointer user_data)
430 {
431     AMDVIIOTLBEntry *entry = (AMDVIIOTLBEntry *)value;
432     uint16_t domid = *(uint16_t *)user_data;
433     return entry->domid == domid;
434 }
435 
436 /* we don't have devid - we can't remove pages by address */
437 static void amdvi_inval_pages(AMDVIState *s, uint64_t *cmd)
438 {
439     uint16_t domid = cpu_to_le16((uint16_t)extract64(cmd[0], 32, 16));
440 
441     if (extract64(cmd[0], 20, 12) || extract64(cmd[0], 48, 12) ||
442         extract64(cmd[1], 3, 9)) {
443         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
444                                    s->cmdbuf + s->cmdbuf_head);
445     }
446 
447     g_hash_table_foreach_remove(s->iotlb, amdvi_iotlb_remove_by_domid,
448                                 &domid);
449     trace_amdvi_pages_inval(domid);
450 }
451 
452 static void amdvi_prefetch_pages(AMDVIState *s, uint64_t *cmd)
453 {
454     if (extract64(cmd[0], 16, 8) || extract64(cmd[0], 52, 8) ||
455         extract64(cmd[1], 1, 1) || extract64(cmd[1], 3, 1) ||
456         extract64(cmd[1], 5, 7)) {
457         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
458                                    s->cmdbuf + s->cmdbuf_head);
459     }
460 
461     trace_amdvi_prefetch_pages();
462 }
463 
464 static void amdvi_inval_inttable(AMDVIState *s, uint64_t *cmd)
465 {
466     if (extract64(cmd[0], 16, 44) || cmd[1]) {
467         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
468                                    s->cmdbuf + s->cmdbuf_head);
469         return;
470     }
471 
472     trace_amdvi_intr_inval();
473 }
474 
475 /* FIXME: Try to work with the specified size instead of all the pages
476  * when the S bit is on
477  */
478 static void iommu_inval_iotlb(AMDVIState *s, uint64_t *cmd)
479 {
480 
481     uint16_t devid = extract64(cmd[0], 0, 16);
482     if (extract64(cmd[1], 1, 1) || extract64(cmd[1], 3, 1) ||
483         extract64(cmd[1], 6, 6)) {
484         amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
485                                    s->cmdbuf + s->cmdbuf_head);
486         return;
487     }
488 
489     if (extract64(cmd[1], 0, 1)) {
490         g_hash_table_foreach_remove(s->iotlb, amdvi_iotlb_remove_by_devid,
491                                     &devid);
492     } else {
493         amdvi_iotlb_remove_page(s, cpu_to_le64(extract64(cmd[1], 12, 52)) << 12,
494                                 cpu_to_le16(extract64(cmd[1], 0, 16)));
495     }
496     trace_amdvi_iotlb_inval();
497 }
498 
499 /* not honouring reserved bits is regarded as an illegal command */
500 static void amdvi_cmdbuf_exec(AMDVIState *s)
501 {
502     uint64_t cmd[2];
503 
504     if (dma_memory_read(&address_space_memory, s->cmdbuf + s->cmdbuf_head,
505         cmd, AMDVI_COMMAND_SIZE)) {
506         trace_amdvi_command_read_fail(s->cmdbuf, s->cmdbuf_head);
507         amdvi_log_command_error(s, s->cmdbuf + s->cmdbuf_head);
508         return;
509     }
510 
511     switch (extract64(cmd[0], 60, 4)) {
512     case AMDVI_CMD_COMPLETION_WAIT:
513         amdvi_completion_wait(s, cmd);
514         break;
515     case AMDVI_CMD_INVAL_DEVTAB_ENTRY:
516         amdvi_inval_devtab_entry(s, cmd);
517         break;
518     case AMDVI_CMD_INVAL_AMDVI_PAGES:
519         amdvi_inval_pages(s, cmd);
520         break;
521     case AMDVI_CMD_INVAL_IOTLB_PAGES:
522         iommu_inval_iotlb(s, cmd);
523         break;
524     case AMDVI_CMD_INVAL_INTR_TABLE:
525         amdvi_inval_inttable(s, cmd);
526         break;
527     case AMDVI_CMD_PREFETCH_AMDVI_PAGES:
528         amdvi_prefetch_pages(s, cmd);
529         break;
530     case AMDVI_CMD_COMPLETE_PPR_REQUEST:
531         amdvi_complete_ppr(s, cmd);
532         break;
533     case AMDVI_CMD_INVAL_AMDVI_ALL:
534         amdvi_inval_all(s, cmd);
535         break;
536     default:
537         trace_amdvi_unhandled_command(extract64(cmd[1], 60, 4));
538         /* log illegal command */
539         amdvi_log_illegalcom_error(s, extract64(cmd[1], 60, 4),
540                                    s->cmdbuf + s->cmdbuf_head);
541     }
542 }
543 
544 static void amdvi_cmdbuf_run(AMDVIState *s)
545 {
546     if (!s->cmdbuf_enabled) {
547         trace_amdvi_command_error(amdvi_readq(s, AMDVI_MMIO_CONTROL));
548         return;
549     }
550 
551     /* check if there is work to do. */
552     while (s->cmdbuf_head != s->cmdbuf_tail) {
553         trace_amdvi_command_exec(s->cmdbuf_head, s->cmdbuf_tail, s->cmdbuf);
554         amdvi_cmdbuf_exec(s);
555         s->cmdbuf_head += AMDVI_COMMAND_SIZE;
556         amdvi_writeq_raw(s, s->cmdbuf_head, AMDVI_MMIO_COMMAND_HEAD);
557 
558         /* wrap head pointer */
559         if (s->cmdbuf_head >= s->cmdbuf_len * AMDVI_COMMAND_SIZE) {
560             s->cmdbuf_head = 0;
561         }
562     }
563 }
564 
565 static void amdvi_mmio_trace(hwaddr addr, unsigned size)
566 {
567     uint8_t index = (addr & ~0x2000) / 8;
568 
569     if ((addr & 0x2000)) {
570         /* high table */
571         index = index >= AMDVI_MMIO_REGS_HIGH ? AMDVI_MMIO_REGS_HIGH : index;
572         trace_amdvi_mmio_read(amdvi_mmio_high[index], addr, size, addr & ~0x07);
573     } else {
574         index = index >= AMDVI_MMIO_REGS_LOW ? AMDVI_MMIO_REGS_LOW : index;
575         trace_amdvi_mmio_read(amdvi_mmio_low[index], addr, size, addr & ~0x07);
576     }
577 }
578 
579 static uint64_t amdvi_mmio_read(void *opaque, hwaddr addr, unsigned size)
580 {
581     AMDVIState *s = opaque;
582 
583     uint64_t val = -1;
584     if (addr + size > AMDVI_MMIO_SIZE) {
585         trace_amdvi_mmio_read_invalid(AMDVI_MMIO_SIZE, addr, size);
586         return (uint64_t)-1;
587     }
588 
589     if (size == 2) {
590         val = amdvi_readw(s, addr);
591     } else if (size == 4) {
592         val = amdvi_readl(s, addr);
593     } else if (size == 8) {
594         val = amdvi_readq(s, addr);
595     }
596     amdvi_mmio_trace(addr, size);
597 
598     return val;
599 }
600 
601 static void amdvi_handle_control_write(AMDVIState *s)
602 {
603     unsigned long control = amdvi_readq(s, AMDVI_MMIO_CONTROL);
604     s->enabled = !!(control & AMDVI_MMIO_CONTROL_AMDVIEN);
605 
606     s->ats_enabled = !!(control & AMDVI_MMIO_CONTROL_HTTUNEN);
607     s->evtlog_enabled = s->enabled && !!(control &
608                         AMDVI_MMIO_CONTROL_EVENTLOGEN);
609 
610     s->evtlog_intr = !!(control & AMDVI_MMIO_CONTROL_EVENTINTEN);
611     s->completion_wait_intr = !!(control & AMDVI_MMIO_CONTROL_COMWAITINTEN);
612     s->cmdbuf_enabled = s->enabled && !!(control &
613                         AMDVI_MMIO_CONTROL_CMDBUFLEN);
614     s->ga_enabled = !!(control & AMDVI_MMIO_CONTROL_GAEN);
615 
616     /* update the flags depending on the control register */
617     if (s->cmdbuf_enabled) {
618         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_CMDBUF_RUN);
619     } else {
620         amdvi_assign_andq(s, AMDVI_MMIO_STATUS, ~AMDVI_MMIO_STATUS_CMDBUF_RUN);
621     }
622     if (s->evtlog_enabled) {
623         amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_EVT_RUN);
624     } else {
625         amdvi_assign_andq(s, AMDVI_MMIO_STATUS, ~AMDVI_MMIO_STATUS_EVT_RUN);
626     }
627 
628     trace_amdvi_control_status(control);
629     amdvi_cmdbuf_run(s);
630 }
631 
632 static inline void amdvi_handle_devtab_write(AMDVIState *s)
633 
634 {
635     uint64_t val = amdvi_readq(s, AMDVI_MMIO_DEVICE_TABLE);
636     s->devtab = (val & AMDVI_MMIO_DEVTAB_BASE_MASK);
637 
638     /* set device table length */
639     s->devtab_len = ((val & AMDVI_MMIO_DEVTAB_SIZE_MASK) + 1 *
640                     (AMDVI_MMIO_DEVTAB_SIZE_UNIT /
641                      AMDVI_MMIO_DEVTAB_ENTRY_SIZE));
642 }
643 
644 static inline void amdvi_handle_cmdhead_write(AMDVIState *s)
645 {
646     s->cmdbuf_head = amdvi_readq(s, AMDVI_MMIO_COMMAND_HEAD)
647                      & AMDVI_MMIO_CMDBUF_HEAD_MASK;
648     amdvi_cmdbuf_run(s);
649 }
650 
651 static inline void amdvi_handle_cmdbase_write(AMDVIState *s)
652 {
653     s->cmdbuf = amdvi_readq(s, AMDVI_MMIO_COMMAND_BASE)
654                 & AMDVI_MMIO_CMDBUF_BASE_MASK;
655     s->cmdbuf_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_CMDBUF_SIZE_BYTE)
656                     & AMDVI_MMIO_CMDBUF_SIZE_MASK);
657     s->cmdbuf_head = s->cmdbuf_tail = 0;
658 }
659 
660 static inline void amdvi_handle_cmdtail_write(AMDVIState *s)
661 {
662     s->cmdbuf_tail = amdvi_readq(s, AMDVI_MMIO_COMMAND_TAIL)
663                      & AMDVI_MMIO_CMDBUF_TAIL_MASK;
664     amdvi_cmdbuf_run(s);
665 }
666 
667 static inline void amdvi_handle_excllim_write(AMDVIState *s)
668 {
669     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EXCL_LIMIT);
670     s->excl_limit = (val & AMDVI_MMIO_EXCL_LIMIT_MASK) |
671                     AMDVI_MMIO_EXCL_LIMIT_LOW;
672 }
673 
674 static inline void amdvi_handle_evtbase_write(AMDVIState *s)
675 {
676     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_BASE);
677     s->evtlog = val & AMDVI_MMIO_EVTLOG_BASE_MASK;
678     s->evtlog_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_EVTLOG_SIZE_BYTE)
679                     & AMDVI_MMIO_EVTLOG_SIZE_MASK);
680 }
681 
682 static inline void amdvi_handle_evttail_write(AMDVIState *s)
683 {
684     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_TAIL);
685     s->evtlog_tail = val & AMDVI_MMIO_EVTLOG_TAIL_MASK;
686 }
687 
688 static inline void amdvi_handle_evthead_write(AMDVIState *s)
689 {
690     uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_HEAD);
691     s->evtlog_head = val & AMDVI_MMIO_EVTLOG_HEAD_MASK;
692 }
693 
694 static inline void amdvi_handle_pprbase_write(AMDVIState *s)
695 {
696     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_BASE);
697     s->ppr_log = val & AMDVI_MMIO_PPRLOG_BASE_MASK;
698     s->pprlog_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_PPRLOG_SIZE_BYTE)
699                     & AMDVI_MMIO_PPRLOG_SIZE_MASK);
700 }
701 
702 static inline void amdvi_handle_pprhead_write(AMDVIState *s)
703 {
704     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_HEAD);
705     s->pprlog_head = val & AMDVI_MMIO_PPRLOG_HEAD_MASK;
706 }
707 
708 static inline void amdvi_handle_pprtail_write(AMDVIState *s)
709 {
710     uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_TAIL);
711     s->pprlog_tail = val & AMDVI_MMIO_PPRLOG_TAIL_MASK;
712 }
713 
714 /* FIXME: something might go wrong if System Software writes in chunks
715  * of one byte but linux writes in chunks of 4 bytes so currently it
716  * works correctly with linux but will definitely be busted if software
717  * reads/writes 8 bytes
718  */
719 static void amdvi_mmio_reg_write(AMDVIState *s, unsigned size, uint64_t val,
720                                  hwaddr addr)
721 {
722     if (size == 2) {
723         amdvi_writew(s, addr, val);
724     } else if (size == 4) {
725         amdvi_writel(s, addr, val);
726     } else if (size == 8) {
727         amdvi_writeq(s, addr, val);
728     }
729 }
730 
731 static void amdvi_mmio_write(void *opaque, hwaddr addr, uint64_t val,
732                              unsigned size)
733 {
734     AMDVIState *s = opaque;
735     unsigned long offset = addr & 0x07;
736 
737     if (addr + size > AMDVI_MMIO_SIZE) {
738         trace_amdvi_mmio_write("error: addr outside region: max ",
739                 (uint64_t)AMDVI_MMIO_SIZE, size, val, offset);
740         return;
741     }
742 
743     amdvi_mmio_trace(addr, size);
744     switch (addr & ~0x07) {
745     case AMDVI_MMIO_CONTROL:
746         amdvi_mmio_reg_write(s, size, val, addr);
747         amdvi_handle_control_write(s);
748         break;
749     case AMDVI_MMIO_DEVICE_TABLE:
750         amdvi_mmio_reg_write(s, size, val, addr);
751        /*  set device table address
752         *   This also suffers from inability to tell whether software
753         *   is done writing
754         */
755         if (offset || (size == 8)) {
756             amdvi_handle_devtab_write(s);
757         }
758         break;
759     case AMDVI_MMIO_COMMAND_HEAD:
760         amdvi_mmio_reg_write(s, size, val, addr);
761         amdvi_handle_cmdhead_write(s);
762         break;
763     case AMDVI_MMIO_COMMAND_BASE:
764         amdvi_mmio_reg_write(s, size, val, addr);
765         /* FIXME - make sure System Software has finished writing incase
766          * it writes in chucks less than 8 bytes in a robust way.As for
767          * now, this hacks works for the linux driver
768          */
769         if (offset || (size == 8)) {
770             amdvi_handle_cmdbase_write(s);
771         }
772         break;
773     case AMDVI_MMIO_COMMAND_TAIL:
774         amdvi_mmio_reg_write(s, size, val, addr);
775         amdvi_handle_cmdtail_write(s);
776         break;
777     case AMDVI_MMIO_EVENT_BASE:
778         amdvi_mmio_reg_write(s, size, val, addr);
779         amdvi_handle_evtbase_write(s);
780         break;
781     case AMDVI_MMIO_EVENT_HEAD:
782         amdvi_mmio_reg_write(s, size, val, addr);
783         amdvi_handle_evthead_write(s);
784         break;
785     case AMDVI_MMIO_EVENT_TAIL:
786         amdvi_mmio_reg_write(s, size, val, addr);
787         amdvi_handle_evttail_write(s);
788         break;
789     case AMDVI_MMIO_EXCL_LIMIT:
790         amdvi_mmio_reg_write(s, size, val, addr);
791         amdvi_handle_excllim_write(s);
792         break;
793         /* PPR log base - unused for now */
794     case AMDVI_MMIO_PPR_BASE:
795         amdvi_mmio_reg_write(s, size, val, addr);
796         amdvi_handle_pprbase_write(s);
797         break;
798         /* PPR log head - also unused for now */
799     case AMDVI_MMIO_PPR_HEAD:
800         amdvi_mmio_reg_write(s, size, val, addr);
801         amdvi_handle_pprhead_write(s);
802         break;
803         /* PPR log tail - unused for now */
804     case AMDVI_MMIO_PPR_TAIL:
805         amdvi_mmio_reg_write(s, size, val, addr);
806         amdvi_handle_pprtail_write(s);
807         break;
808     }
809 }
810 
811 static inline uint64_t amdvi_get_perms(uint64_t entry)
812 {
813     return (entry & (AMDVI_DEV_PERM_READ | AMDVI_DEV_PERM_WRITE)) >>
814            AMDVI_DEV_PERM_SHIFT;
815 }
816 
817 /* validate that reserved bits are honoured */
818 static bool amdvi_validate_dte(AMDVIState *s, uint16_t devid,
819                                uint64_t *dte)
820 {
821     if ((dte[0] & AMDVI_DTE_LOWER_QUAD_RESERVED)
822         || (dte[1] & AMDVI_DTE_MIDDLE_QUAD_RESERVED)
823         || (dte[2] & AMDVI_DTE_UPPER_QUAD_RESERVED) || dte[3]) {
824         amdvi_log_illegaldevtab_error(s, devid,
825                                       s->devtab +
826                                       devid * AMDVI_DEVTAB_ENTRY_SIZE, 0);
827         return false;
828     }
829 
830     return true;
831 }
832 
833 /* get a device table entry given the devid */
834 static bool amdvi_get_dte(AMDVIState *s, int devid, uint64_t *entry)
835 {
836     uint32_t offset = devid * AMDVI_DEVTAB_ENTRY_SIZE;
837 
838     if (dma_memory_read(&address_space_memory, s->devtab + offset, entry,
839         AMDVI_DEVTAB_ENTRY_SIZE)) {
840         trace_amdvi_dte_get_fail(s->devtab, offset);
841         /* log error accessing dte */
842         amdvi_log_devtab_error(s, devid, s->devtab + offset, 0);
843         return false;
844     }
845 
846     *entry = le64_to_cpu(*entry);
847     if (!amdvi_validate_dte(s, devid, entry)) {
848         trace_amdvi_invalid_dte(entry[0]);
849         return false;
850     }
851 
852     return true;
853 }
854 
855 /* get pte translation mode */
856 static inline uint8_t get_pte_translation_mode(uint64_t pte)
857 {
858     return (pte >> AMDVI_DEV_MODE_RSHIFT) & AMDVI_DEV_MODE_MASK;
859 }
860 
861 static inline uint64_t pte_override_page_mask(uint64_t pte)
862 {
863     uint8_t page_mask = 12;
864     uint64_t addr = (pte & AMDVI_DEV_PT_ROOT_MASK) ^ AMDVI_DEV_PT_ROOT_MASK;
865     /* find the first zero bit */
866     while (addr & 1) {
867         page_mask++;
868         addr = addr >> 1;
869     }
870 
871     return ~((1ULL << page_mask) - 1);
872 }
873 
874 static inline uint64_t pte_get_page_mask(uint64_t oldlevel)
875 {
876     return ~((1UL << ((oldlevel * 9) + 3)) - 1);
877 }
878 
879 static inline uint64_t amdvi_get_pte_entry(AMDVIState *s, uint64_t pte_addr,
880                                           uint16_t devid)
881 {
882     uint64_t pte;
883 
884     if (dma_memory_read(&address_space_memory, pte_addr, &pte, sizeof(pte))) {
885         trace_amdvi_get_pte_hwerror(pte_addr);
886         amdvi_log_pagetab_error(s, devid, pte_addr, 0);
887         pte = 0;
888         return pte;
889     }
890 
891     pte = le64_to_cpu(pte);
892     return pte;
893 }
894 
895 static void amdvi_page_walk(AMDVIAddressSpace *as, uint64_t *dte,
896                             IOMMUTLBEntry *ret, unsigned perms,
897                             hwaddr addr)
898 {
899     unsigned level, present, pte_perms, oldlevel;
900     uint64_t pte = dte[0], pte_addr, page_mask;
901 
902     /* make sure the DTE has TV = 1 */
903     if (pte & AMDVI_DEV_TRANSLATION_VALID) {
904         level = get_pte_translation_mode(pte);
905         if (level >= 7) {
906             trace_amdvi_mode_invalid(level, addr);
907             return;
908         }
909         if (level == 0) {
910             goto no_remap;
911         }
912 
913         /* we are at the leaf page table or page table encodes a huge page */
914         while (level > 0) {
915             pte_perms = amdvi_get_perms(pte);
916             present = pte & 1;
917             if (!present || perms != (perms & pte_perms)) {
918                 amdvi_page_fault(as->iommu_state, as->devfn, addr, perms);
919                 trace_amdvi_page_fault(addr);
920                 return;
921             }
922 
923             /* go to the next lower level */
924             pte_addr = pte & AMDVI_DEV_PT_ROOT_MASK;
925             /* add offset and load pte */
926             pte_addr += ((addr >> (3 + 9 * level)) & 0x1FF) << 3;
927             pte = amdvi_get_pte_entry(as->iommu_state, pte_addr, as->devfn);
928             if (!pte) {
929                 return;
930             }
931             oldlevel = level;
932             level = get_pte_translation_mode(pte);
933             if (level == 0x7) {
934                 break;
935             }
936         }
937 
938         if (level == 0x7) {
939             page_mask = pte_override_page_mask(pte);
940         } else {
941             page_mask = pte_get_page_mask(oldlevel);
942         }
943 
944         /* get access permissions from pte */
945         ret->iova = addr & page_mask;
946         ret->translated_addr = (pte & AMDVI_DEV_PT_ROOT_MASK) & page_mask;
947         ret->addr_mask = ~page_mask;
948         ret->perm = amdvi_get_perms(pte);
949         return;
950     }
951 no_remap:
952     ret->iova = addr & AMDVI_PAGE_MASK_4K;
953     ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
954     ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
955     ret->perm = amdvi_get_perms(pte);
956 }
957 
958 static void amdvi_do_translate(AMDVIAddressSpace *as, hwaddr addr,
959                                bool is_write, IOMMUTLBEntry *ret)
960 {
961     AMDVIState *s = as->iommu_state;
962     uint16_t devid = PCI_BUILD_BDF(as->bus_num, as->devfn);
963     AMDVIIOTLBEntry *iotlb_entry = amdvi_iotlb_lookup(s, addr, devid);
964     uint64_t entry[4];
965 
966     if (iotlb_entry) {
967         trace_amdvi_iotlb_hit(PCI_BUS_NUM(devid), PCI_SLOT(devid),
968                 PCI_FUNC(devid), addr, iotlb_entry->translated_addr);
969         ret->iova = addr & ~iotlb_entry->page_mask;
970         ret->translated_addr = iotlb_entry->translated_addr;
971         ret->addr_mask = iotlb_entry->page_mask;
972         ret->perm = iotlb_entry->perms;
973         return;
974     }
975 
976     if (!amdvi_get_dte(s, devid, entry)) {
977         return;
978     }
979 
980     /* devices with V = 0 are not translated */
981     if (!(entry[0] & AMDVI_DEV_VALID)) {
982         goto out;
983     }
984 
985     amdvi_page_walk(as, entry, ret,
986                     is_write ? AMDVI_PERM_WRITE : AMDVI_PERM_READ, addr);
987 
988     amdvi_update_iotlb(s, devid, addr, *ret,
989                        entry[1] & AMDVI_DEV_DOMID_ID_MASK);
990     return;
991 
992 out:
993     ret->iova = addr & AMDVI_PAGE_MASK_4K;
994     ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
995     ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
996     ret->perm = IOMMU_RW;
997 }
998 
999 static inline bool amdvi_is_interrupt_addr(hwaddr addr)
1000 {
1001     return addr >= AMDVI_INT_ADDR_FIRST && addr <= AMDVI_INT_ADDR_LAST;
1002 }
1003 
1004 static IOMMUTLBEntry amdvi_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
1005                                      IOMMUAccessFlags flag, int iommu_idx)
1006 {
1007     AMDVIAddressSpace *as = container_of(iommu, AMDVIAddressSpace, iommu);
1008     AMDVIState *s = as->iommu_state;
1009     IOMMUTLBEntry ret = {
1010         .target_as = &address_space_memory,
1011         .iova = addr,
1012         .translated_addr = 0,
1013         .addr_mask = ~(hwaddr)0,
1014         .perm = IOMMU_NONE
1015     };
1016 
1017     if (!s->enabled) {
1018         /* AMDVI disabled - corresponds to iommu=off not
1019          * failure to provide any parameter
1020          */
1021         ret.iova = addr & AMDVI_PAGE_MASK_4K;
1022         ret.translated_addr = addr & AMDVI_PAGE_MASK_4K;
1023         ret.addr_mask = ~AMDVI_PAGE_MASK_4K;
1024         ret.perm = IOMMU_RW;
1025         return ret;
1026     } else if (amdvi_is_interrupt_addr(addr)) {
1027         ret.iova = addr & AMDVI_PAGE_MASK_4K;
1028         ret.translated_addr = addr & AMDVI_PAGE_MASK_4K;
1029         ret.addr_mask = ~AMDVI_PAGE_MASK_4K;
1030         ret.perm = IOMMU_WO;
1031         return ret;
1032     }
1033 
1034     amdvi_do_translate(as, addr, flag & IOMMU_WO, &ret);
1035     trace_amdvi_translation_result(as->bus_num, PCI_SLOT(as->devfn),
1036             PCI_FUNC(as->devfn), addr, ret.translated_addr);
1037     return ret;
1038 }
1039 
1040 static int amdvi_get_irte(AMDVIState *s, MSIMessage *origin, uint64_t *dte,
1041                           union irte *irte, uint16_t devid)
1042 {
1043     uint64_t irte_root, offset;
1044 
1045     irte_root = dte[2] & AMDVI_IR_PHYS_ADDR_MASK;
1046     offset = (origin->data & AMDVI_IRTE_OFFSET) << 2;
1047 
1048     trace_amdvi_ir_irte(irte_root, offset);
1049 
1050     if (dma_memory_read(&address_space_memory, irte_root + offset,
1051                         irte, sizeof(*irte))) {
1052         trace_amdvi_ir_err("failed to get irte");
1053         return -AMDVI_IR_GET_IRTE;
1054     }
1055 
1056     trace_amdvi_ir_irte_val(irte->val);
1057 
1058     return 0;
1059 }
1060 
1061 static int amdvi_int_remap_legacy(AMDVIState *iommu,
1062                                   MSIMessage *origin,
1063                                   MSIMessage *translated,
1064                                   uint64_t *dte,
1065                                   X86IOMMUIrq *irq,
1066                                   uint16_t sid)
1067 {
1068     int ret;
1069     union irte irte;
1070 
1071     /* get interrupt remapping table */
1072     ret = amdvi_get_irte(iommu, origin, dte, &irte, sid);
1073     if (ret < 0) {
1074         return ret;
1075     }
1076 
1077     if (!irte.fields.valid) {
1078         trace_amdvi_ir_target_abort("RemapEn is disabled");
1079         return -AMDVI_IR_TARGET_ABORT;
1080     }
1081 
1082     if (irte.fields.guest_mode) {
1083         error_report_once("guest mode is not zero");
1084         return -AMDVI_IR_ERR;
1085     }
1086 
1087     if (irte.fields.int_type > AMDVI_IOAPIC_INT_TYPE_ARBITRATED) {
1088         error_report_once("reserved int_type");
1089         return -AMDVI_IR_ERR;
1090     }
1091 
1092     irq->delivery_mode = irte.fields.int_type;
1093     irq->vector = irte.fields.vector;
1094     irq->dest_mode = irte.fields.dm;
1095     irq->redir_hint = irte.fields.rq_eoi;
1096     irq->dest = irte.fields.destination;
1097 
1098     return 0;
1099 }
1100 
1101 static int amdvi_get_irte_ga(AMDVIState *s, MSIMessage *origin, uint64_t *dte,
1102                              struct irte_ga *irte, uint16_t devid)
1103 {
1104     uint64_t irte_root, offset;
1105 
1106     irte_root = dte[2] & AMDVI_IR_PHYS_ADDR_MASK;
1107     offset = (origin->data & AMDVI_IRTE_OFFSET) << 4;
1108     trace_amdvi_ir_irte(irte_root, offset);
1109 
1110     if (dma_memory_read(&address_space_memory, irte_root + offset,
1111                         irte, sizeof(*irte))) {
1112         trace_amdvi_ir_err("failed to get irte_ga");
1113         return -AMDVI_IR_GET_IRTE;
1114     }
1115 
1116     trace_amdvi_ir_irte_ga_val(irte->hi.val, irte->lo.val);
1117     return 0;
1118 }
1119 
1120 static int amdvi_int_remap_ga(AMDVIState *iommu,
1121                               MSIMessage *origin,
1122                               MSIMessage *translated,
1123                               uint64_t *dte,
1124                               X86IOMMUIrq *irq,
1125                               uint16_t sid)
1126 {
1127     int ret;
1128     struct irte_ga irte;
1129 
1130     /* get interrupt remapping table */
1131     ret = amdvi_get_irte_ga(iommu, origin, dte, &irte, sid);
1132     if (ret < 0) {
1133         return ret;
1134     }
1135 
1136     if (!irte.lo.fields_remap.valid) {
1137         trace_amdvi_ir_target_abort("RemapEn is disabled");
1138         return -AMDVI_IR_TARGET_ABORT;
1139     }
1140 
1141     if (irte.lo.fields_remap.guest_mode) {
1142         error_report_once("guest mode is not zero");
1143         return -AMDVI_IR_ERR;
1144     }
1145 
1146     if (irte.lo.fields_remap.int_type > AMDVI_IOAPIC_INT_TYPE_ARBITRATED) {
1147         error_report_once("reserved int_type is set");
1148         return -AMDVI_IR_ERR;
1149     }
1150 
1151     irq->delivery_mode = irte.lo.fields_remap.int_type;
1152     irq->vector = irte.hi.fields.vector;
1153     irq->dest_mode = irte.lo.fields_remap.dm;
1154     irq->redir_hint = irte.lo.fields_remap.rq_eoi;
1155     irq->dest = irte.lo.fields_remap.destination;
1156 
1157     return 0;
1158 }
1159 
1160 static int __amdvi_int_remap_msi(AMDVIState *iommu,
1161                                  MSIMessage *origin,
1162                                  MSIMessage *translated,
1163                                  uint64_t *dte,
1164                                  X86IOMMUIrq *irq,
1165                                  uint16_t sid)
1166 {
1167     int ret;
1168     uint8_t int_ctl;
1169 
1170     int_ctl = (dte[2] >> AMDVI_IR_INTCTL_SHIFT) & 3;
1171     trace_amdvi_ir_intctl(int_ctl);
1172 
1173     switch (int_ctl) {
1174     case AMDVI_IR_INTCTL_PASS:
1175         memcpy(translated, origin, sizeof(*origin));
1176         return 0;
1177     case AMDVI_IR_INTCTL_REMAP:
1178         break;
1179     case AMDVI_IR_INTCTL_ABORT:
1180         trace_amdvi_ir_target_abort("int_ctl abort");
1181         return -AMDVI_IR_TARGET_ABORT;
1182     default:
1183         trace_amdvi_ir_err("int_ctl reserved");
1184         return -AMDVI_IR_ERR;
1185     }
1186 
1187     if (iommu->ga_enabled) {
1188         ret = amdvi_int_remap_ga(iommu, origin, translated, dte, irq, sid);
1189     } else {
1190         ret = amdvi_int_remap_legacy(iommu, origin, translated, dte, irq, sid);
1191     }
1192 
1193     return ret;
1194 }
1195 
1196 /* Interrupt remapping for MSI/MSI-X entry */
1197 static int amdvi_int_remap_msi(AMDVIState *iommu,
1198                                MSIMessage *origin,
1199                                MSIMessage *translated,
1200                                uint16_t sid)
1201 {
1202     int ret = 0;
1203     uint64_t pass = 0;
1204     uint64_t dte[4] = { 0 };
1205     X86IOMMUIrq irq = { 0 };
1206     uint8_t dest_mode, delivery_mode;
1207 
1208     assert(origin && translated);
1209 
1210     /*
1211      * When IOMMU is enabled, interrupt remap request will come either from
1212      * IO-APIC or PCI device. If interrupt is from PCI device then it will
1213      * have a valid requester id but if the interrupt is from IO-APIC
1214      * then requester id will be invalid.
1215      */
1216     if (sid == X86_IOMMU_SID_INVALID) {
1217         sid = AMDVI_IOAPIC_SB_DEVID;
1218     }
1219 
1220     trace_amdvi_ir_remap_msi_req(origin->address, origin->data, sid);
1221 
1222     /* check if device table entry is set before we go further. */
1223     if (!iommu || !iommu->devtab_len) {
1224         memcpy(translated, origin, sizeof(*origin));
1225         goto out;
1226     }
1227 
1228     if (!amdvi_get_dte(iommu, sid, dte)) {
1229         return -AMDVI_IR_ERR;
1230     }
1231 
1232     /* Check if IR is enabled in DTE */
1233     if (!(dte[2] & AMDVI_IR_REMAP_ENABLE)) {
1234         memcpy(translated, origin, sizeof(*origin));
1235         goto out;
1236     }
1237 
1238     /* validate that we are configure with intremap=on */
1239     if (!x86_iommu_ir_supported(X86_IOMMU_DEVICE(iommu))) {
1240         trace_amdvi_err("Interrupt remapping is enabled in the guest but "
1241                         "not in the host. Use intremap=on to enable interrupt "
1242                         "remapping in amd-iommu.");
1243         return -AMDVI_IR_ERR;
1244     }
1245 
1246     if (origin->address & AMDVI_MSI_ADDR_HI_MASK) {
1247         trace_amdvi_err("MSI address high 32 bits non-zero when "
1248                         "Interrupt Remapping enabled.");
1249         return -AMDVI_IR_ERR;
1250     }
1251 
1252     if ((origin->address & AMDVI_MSI_ADDR_LO_MASK) != APIC_DEFAULT_ADDRESS) {
1253         trace_amdvi_err("MSI is not from IOAPIC.");
1254         return -AMDVI_IR_ERR;
1255     }
1256 
1257     /*
1258      * The MSI data register [10:8] are used to get the upstream interrupt type.
1259      *
1260      * See MSI/MSI-X format:
1261      * https://pdfs.semanticscholar.org/presentation/9420/c279e942eca568157711ef5c92b800c40a79.pdf
1262      * (page 5)
1263      */
1264     delivery_mode = (origin->data >> MSI_DATA_DELIVERY_MODE_SHIFT) & 7;
1265 
1266     switch (delivery_mode) {
1267     case AMDVI_IOAPIC_INT_TYPE_FIXED:
1268     case AMDVI_IOAPIC_INT_TYPE_ARBITRATED:
1269         trace_amdvi_ir_delivery_mode("fixed/arbitrated");
1270         ret = __amdvi_int_remap_msi(iommu, origin, translated, dte, &irq, sid);
1271         if (ret < 0) {
1272             goto remap_fail;
1273         } else {
1274             /* Translate IRQ to MSI messages */
1275             x86_iommu_irq_to_msi_message(&irq, translated);
1276             goto out;
1277         }
1278         break;
1279     case AMDVI_IOAPIC_INT_TYPE_SMI:
1280         error_report("SMI is not supported!");
1281         ret = -AMDVI_IR_ERR;
1282         break;
1283     case AMDVI_IOAPIC_INT_TYPE_NMI:
1284         pass = dte[3] & AMDVI_DEV_NMI_PASS_MASK;
1285         trace_amdvi_ir_delivery_mode("nmi");
1286         break;
1287     case AMDVI_IOAPIC_INT_TYPE_INIT:
1288         pass = dte[3] & AMDVI_DEV_INT_PASS_MASK;
1289         trace_amdvi_ir_delivery_mode("init");
1290         break;
1291     case AMDVI_IOAPIC_INT_TYPE_EINT:
1292         pass = dte[3] & AMDVI_DEV_EINT_PASS_MASK;
1293         trace_amdvi_ir_delivery_mode("eint");
1294         break;
1295     default:
1296         trace_amdvi_ir_delivery_mode("unsupported delivery_mode");
1297         ret = -AMDVI_IR_ERR;
1298         break;
1299     }
1300 
1301     if (ret < 0) {
1302         goto remap_fail;
1303     }
1304 
1305     /*
1306      * The MSI address register bit[2] is used to get the destination
1307      * mode. The dest_mode 1 is valid for fixed and arbitrated interrupts
1308      * only.
1309      */
1310     dest_mode = (origin->address >> MSI_ADDR_DEST_MODE_SHIFT) & 1;
1311     if (dest_mode) {
1312         trace_amdvi_ir_err("invalid dest_mode");
1313         ret = -AMDVI_IR_ERR;
1314         goto remap_fail;
1315     }
1316 
1317     if (pass) {
1318         memcpy(translated, origin, sizeof(*origin));
1319     } else {
1320         trace_amdvi_ir_err("passthrough is not enabled");
1321         ret = -AMDVI_IR_ERR;
1322         goto remap_fail;
1323     }
1324 
1325 out:
1326     trace_amdvi_ir_remap_msi(origin->address, origin->data,
1327                              translated->address, translated->data);
1328     return 0;
1329 
1330 remap_fail:
1331     return ret;
1332 }
1333 
1334 static int amdvi_int_remap(X86IOMMUState *iommu,
1335                            MSIMessage *origin,
1336                            MSIMessage *translated,
1337                            uint16_t sid)
1338 {
1339     return amdvi_int_remap_msi(AMD_IOMMU_DEVICE(iommu), origin,
1340                                translated, sid);
1341 }
1342 
1343 static MemTxResult amdvi_mem_ir_write(void *opaque, hwaddr addr,
1344                                       uint64_t value, unsigned size,
1345                                       MemTxAttrs attrs)
1346 {
1347     int ret;
1348     MSIMessage from = { 0, 0 }, to = { 0, 0 };
1349     uint16_t sid = AMDVI_IOAPIC_SB_DEVID;
1350 
1351     from.address = (uint64_t) addr + AMDVI_INT_ADDR_FIRST;
1352     from.data = (uint32_t) value;
1353 
1354     trace_amdvi_mem_ir_write_req(addr, value, size);
1355 
1356     if (!attrs.unspecified) {
1357         /* We have explicit Source ID */
1358         sid = attrs.requester_id;
1359     }
1360 
1361     ret = amdvi_int_remap_msi(opaque, &from, &to, sid);
1362     if (ret < 0) {
1363         /* TODO: log the event using IOMMU log event interface */
1364         error_report_once("failed to remap interrupt from devid 0x%x", sid);
1365         return MEMTX_ERROR;
1366     }
1367 
1368     apic_get_class()->send_msi(&to);
1369 
1370     trace_amdvi_mem_ir_write(to.address, to.data);
1371     return MEMTX_OK;
1372 }
1373 
1374 static MemTxResult amdvi_mem_ir_read(void *opaque, hwaddr addr,
1375                                      uint64_t *data, unsigned size,
1376                                      MemTxAttrs attrs)
1377 {
1378     return MEMTX_OK;
1379 }
1380 
1381 static const MemoryRegionOps amdvi_ir_ops = {
1382     .read_with_attrs = amdvi_mem_ir_read,
1383     .write_with_attrs = amdvi_mem_ir_write,
1384     .endianness = DEVICE_LITTLE_ENDIAN,
1385     .impl = {
1386         .min_access_size = 4,
1387         .max_access_size = 4,
1388     },
1389     .valid = {
1390         .min_access_size = 4,
1391         .max_access_size = 4,
1392     }
1393 };
1394 
1395 static AddressSpace *amdvi_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
1396 {
1397     char name[128];
1398     AMDVIState *s = opaque;
1399     AMDVIAddressSpace **iommu_as, *amdvi_dev_as;
1400     int bus_num = pci_bus_num(bus);
1401 
1402     iommu_as = s->address_spaces[bus_num];
1403 
1404     /* allocate memory during the first run */
1405     if (!iommu_as) {
1406         iommu_as = g_malloc0(sizeof(AMDVIAddressSpace *) * PCI_DEVFN_MAX);
1407         s->address_spaces[bus_num] = iommu_as;
1408     }
1409 
1410     /* set up AMD-Vi region */
1411     if (!iommu_as[devfn]) {
1412         snprintf(name, sizeof(name), "amd_iommu_devfn_%d", devfn);
1413 
1414         iommu_as[devfn] = g_malloc0(sizeof(AMDVIAddressSpace));
1415         iommu_as[devfn]->bus_num = (uint8_t)bus_num;
1416         iommu_as[devfn]->devfn = (uint8_t)devfn;
1417         iommu_as[devfn]->iommu_state = s;
1418 
1419         amdvi_dev_as = iommu_as[devfn];
1420 
1421         /*
1422          * Memory region relationships looks like (Address range shows
1423          * only lower 32 bits to make it short in length...):
1424          *
1425          * |-----------------+-------------------+----------|
1426          * | Name            | Address range     | Priority |
1427          * |-----------------+-------------------+----------+
1428          * | amdvi_root      | 00000000-ffffffff |        0 |
1429          * |  amdvi_iommu    | 00000000-ffffffff |        1 |
1430          * |  amdvi_iommu_ir | fee00000-feefffff |       64 |
1431          * |-----------------+-------------------+----------|
1432          */
1433         memory_region_init_iommu(&amdvi_dev_as->iommu,
1434                                  sizeof(amdvi_dev_as->iommu),
1435                                  TYPE_AMD_IOMMU_MEMORY_REGION,
1436                                  OBJECT(s),
1437                                  "amd_iommu", UINT64_MAX);
1438         memory_region_init(&amdvi_dev_as->root, OBJECT(s),
1439                            "amdvi_root", UINT64_MAX);
1440         address_space_init(&amdvi_dev_as->as, &amdvi_dev_as->root, name);
1441         memory_region_init_io(&amdvi_dev_as->iommu_ir, OBJECT(s),
1442                               &amdvi_ir_ops, s, "amd_iommu_ir",
1443                               AMDVI_INT_ADDR_SIZE);
1444         memory_region_add_subregion_overlap(&amdvi_dev_as->root,
1445                                             AMDVI_INT_ADDR_FIRST,
1446                                             &amdvi_dev_as->iommu_ir,
1447                                             64);
1448         memory_region_add_subregion_overlap(&amdvi_dev_as->root, 0,
1449                                             MEMORY_REGION(&amdvi_dev_as->iommu),
1450                                             1);
1451     }
1452     return &iommu_as[devfn]->as;
1453 }
1454 
1455 static const MemoryRegionOps mmio_mem_ops = {
1456     .read = amdvi_mmio_read,
1457     .write = amdvi_mmio_write,
1458     .endianness = DEVICE_LITTLE_ENDIAN,
1459     .impl = {
1460         .min_access_size = 1,
1461         .max_access_size = 8,
1462         .unaligned = false,
1463     },
1464     .valid = {
1465         .min_access_size = 1,
1466         .max_access_size = 8,
1467     }
1468 };
1469 
1470 static int amdvi_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
1471                                            IOMMUNotifierFlag old,
1472                                            IOMMUNotifierFlag new,
1473                                            Error **errp)
1474 {
1475     AMDVIAddressSpace *as = container_of(iommu, AMDVIAddressSpace, iommu);
1476 
1477     if (new & IOMMU_NOTIFIER_MAP) {
1478         error_setg(errp,
1479                    "device %02x.%02x.%x requires iommu notifier which is not "
1480                    "currently supported", as->bus_num, PCI_SLOT(as->devfn),
1481                    PCI_FUNC(as->devfn));
1482         return -EINVAL;
1483     }
1484     return 0;
1485 }
1486 
1487 static void amdvi_init(AMDVIState *s)
1488 {
1489     amdvi_iotlb_reset(s);
1490 
1491     s->devtab_len = 0;
1492     s->cmdbuf_len = 0;
1493     s->cmdbuf_head = 0;
1494     s->cmdbuf_tail = 0;
1495     s->evtlog_head = 0;
1496     s->evtlog_tail = 0;
1497     s->excl_enabled = false;
1498     s->excl_allow = false;
1499     s->mmio_enabled = false;
1500     s->enabled = false;
1501     s->ats_enabled = false;
1502     s->cmdbuf_enabled = false;
1503 
1504     /* reset MMIO */
1505     memset(s->mmior, 0, AMDVI_MMIO_SIZE);
1506     amdvi_set_quad(s, AMDVI_MMIO_EXT_FEATURES, AMDVI_EXT_FEATURES,
1507             0xffffffffffffffef, 0);
1508     amdvi_set_quad(s, AMDVI_MMIO_STATUS, 0, 0x98, 0x67);
1509 
1510     /* reset device ident */
1511     pci_config_set_vendor_id(s->pci.dev.config, PCI_VENDOR_ID_AMD);
1512     pci_config_set_prog_interface(s->pci.dev.config, 00);
1513     pci_config_set_device_id(s->pci.dev.config, s->devid);
1514     pci_config_set_class(s->pci.dev.config, 0x0806);
1515 
1516     /* reset AMDVI specific capabilities, all r/o */
1517     pci_set_long(s->pci.dev.config + s->capab_offset, AMDVI_CAPAB_FEATURES);
1518     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_BAR_LOW,
1519                  s->mmio.addr & ~(0xffff0000));
1520     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_BAR_HIGH,
1521                 (s->mmio.addr & ~(0xffff)) >> 16);
1522     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_RANGE,
1523                  0xff000000);
1524     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_MISC, 0);
1525     pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_MISC,
1526             AMDVI_MAX_PH_ADDR | AMDVI_MAX_GVA_ADDR | AMDVI_MAX_VA_ADDR);
1527 }
1528 
1529 static void amdvi_reset(DeviceState *dev)
1530 {
1531     AMDVIState *s = AMD_IOMMU_DEVICE(dev);
1532 
1533     msi_reset(&s->pci.dev);
1534     amdvi_init(s);
1535 }
1536 
1537 static void amdvi_realize(DeviceState *dev, Error **errp)
1538 {
1539     int ret = 0;
1540     AMDVIState *s = AMD_IOMMU_DEVICE(dev);
1541     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
1542     MachineState *ms = MACHINE(qdev_get_machine());
1543     PCMachineState *pcms = PC_MACHINE(ms);
1544     X86MachineState *x86ms = X86_MACHINE(ms);
1545     PCIBus *bus = pcms->bus;
1546 
1547     s->iotlb = g_hash_table_new_full(amdvi_uint64_hash,
1548                                      amdvi_uint64_equal, g_free, g_free);
1549 
1550     /* This device should take care of IOMMU PCI properties */
1551     x86_iommu->type = TYPE_AMD;
1552     if (!qdev_realize(DEVICE(&s->pci), &bus->qbus, errp)) {
1553         return;
1554     }
1555     ret = pci_add_capability(&s->pci.dev, AMDVI_CAPAB_ID_SEC, 0,
1556                                          AMDVI_CAPAB_SIZE, errp);
1557     if (ret < 0) {
1558         return;
1559     }
1560     s->capab_offset = ret;
1561 
1562     ret = pci_add_capability(&s->pci.dev, PCI_CAP_ID_MSI, 0,
1563                              AMDVI_CAPAB_REG_SIZE, errp);
1564     if (ret < 0) {
1565         return;
1566     }
1567     ret = pci_add_capability(&s->pci.dev, PCI_CAP_ID_HT, 0,
1568                              AMDVI_CAPAB_REG_SIZE, errp);
1569     if (ret < 0) {
1570         return;
1571     }
1572 
1573     /* Pseudo address space under root PCI bus. */
1574     x86ms->ioapic_as = amdvi_host_dma_iommu(bus, s, AMDVI_IOAPIC_SB_DEVID);
1575 
1576     /* set up MMIO */
1577     memory_region_init_io(&s->mmio, OBJECT(s), &mmio_mem_ops, s, "amdvi-mmio",
1578                           AMDVI_MMIO_SIZE);
1579 
1580     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->mmio);
1581     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, AMDVI_BASE_ADDR);
1582     pci_setup_iommu(bus, amdvi_host_dma_iommu, s);
1583     s->devid = object_property_get_int(OBJECT(&s->pci), "addr", &error_abort);
1584     msi_init(&s->pci.dev, 0, 1, true, false, errp);
1585     amdvi_init(s);
1586 }
1587 
1588 static const VMStateDescription vmstate_amdvi = {
1589     .name = "amd-iommu",
1590     .unmigratable = 1
1591 };
1592 
1593 static void amdvi_instance_init(Object *klass)
1594 {
1595     AMDVIState *s = AMD_IOMMU_DEVICE(klass);
1596 
1597     object_initialize(&s->pci, sizeof(s->pci), TYPE_AMD_IOMMU_PCI);
1598 }
1599 
1600 static void amdvi_class_init(ObjectClass *klass, void* data)
1601 {
1602     DeviceClass *dc = DEVICE_CLASS(klass);
1603     X86IOMMUClass *dc_class = X86_IOMMU_DEVICE_CLASS(klass);
1604 
1605     dc->reset = amdvi_reset;
1606     dc->vmsd = &vmstate_amdvi;
1607     dc->hotpluggable = false;
1608     dc_class->realize = amdvi_realize;
1609     dc_class->int_remap = amdvi_int_remap;
1610     /* Supported by the pc-q35-* machine types */
1611     dc->user_creatable = true;
1612     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1613     dc->desc = "AMD IOMMU (AMD-Vi) DMA Remapping device";
1614 }
1615 
1616 static const TypeInfo amdvi = {
1617     .name = TYPE_AMD_IOMMU_DEVICE,
1618     .parent = TYPE_X86_IOMMU_DEVICE,
1619     .instance_size = sizeof(AMDVIState),
1620     .instance_init = amdvi_instance_init,
1621     .class_init = amdvi_class_init
1622 };
1623 
1624 static const TypeInfo amdviPCI = {
1625     .name = TYPE_AMD_IOMMU_PCI,
1626     .parent = TYPE_PCI_DEVICE,
1627     .instance_size = sizeof(AMDVIPCIState),
1628     .interfaces = (InterfaceInfo[]) {
1629         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
1630         { },
1631     },
1632 };
1633 
1634 static void amdvi_iommu_memory_region_class_init(ObjectClass *klass, void *data)
1635 {
1636     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
1637 
1638     imrc->translate = amdvi_translate;
1639     imrc->notify_flag_changed = amdvi_iommu_notify_flag_changed;
1640 }
1641 
1642 static const TypeInfo amdvi_iommu_memory_region_info = {
1643     .parent = TYPE_IOMMU_MEMORY_REGION,
1644     .name = TYPE_AMD_IOMMU_MEMORY_REGION,
1645     .class_init = amdvi_iommu_memory_region_class_init,
1646 };
1647 
1648 static void amdviPCI_register_types(void)
1649 {
1650     type_register_static(&amdviPCI);
1651     type_register_static(&amdvi);
1652     type_register_static(&amdvi_iommu_memory_region_info);
1653 }
1654 
1655 type_init(amdviPCI_register_types);
1656