2 * QEMU emulation of AMD IOMMU (AMD-Vi)
4 * Copyright (C) 2011 Eduard - Gabriel Munteanu
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.
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.
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/>.
20 * Cache implementation inspired by hw/i386/intel_iommu.c
22 #include "qemu/osdep.h"
23 #include "hw/i386/amd_iommu.h"
24 #include "qapi/error.h"
25 #include "qemu/error-report.h"
28 /* used AMD-Vi MMIO registers */
29 const char *amdvi_mmio_low[] = {
30 "AMDVI_MMIO_DEVTAB_BASE",
31 "AMDVI_MMIO_CMDBUF_BASE",
32 "AMDVI_MMIO_EVTLOG_BASE",
34 "AMDVI_MMIO_EXCL_BASE",
35 "AMDVI_MMIO_EXCL_LIMIT",
36 "AMDVI_MMIO_EXT_FEATURES",
37 "AMDVI_MMIO_PPR_BASE",
40 const char *amdvi_mmio_high[] = {
41 "AMDVI_MMIO_COMMAND_HEAD",
42 "AMDVI_MMIO_COMMAND_TAIL",
43 "AMDVI_MMIO_EVTLOG_HEAD",
44 "AMDVI_MMIO_EVTLOG_TAIL",
46 "AMDVI_MMIO_PPR_HEAD",
47 "AMDVI_MMIO_PPR_TAIL",
51 struct AMDVIAddressSpace {
52 uint8_t bus_num; /* bus number */
53 uint8_t devfn; /* device function */
54 AMDVIState *iommu_state; /* AMDVI - one per machine */
55 IOMMUMemoryRegion iommu; /* Device's address translation region */
56 MemoryRegion iommu_ir; /* Device's interrupt remapping region */
57 AddressSpace as; /* device's corresponding address space */
60 /* AMDVI cache entry */
61 typedef struct AMDVIIOTLBEntry {
62 uint16_t domid; /* assigned domain id */
63 uint16_t devid; /* device owning entry */
64 uint64_t perms; /* access permissions */
65 uint64_t translated_addr; /* translated address */
66 uint64_t page_mask; /* physical page size */
69 /* configure MMIO registers at startup/reset */
70 static void amdvi_set_quad(AMDVIState *s, hwaddr addr, uint64_t val,
71 uint64_t romask, uint64_t w1cmask)
73 stq_le_p(&s->mmior[addr], val);
74 stq_le_p(&s->romask[addr], romask);
75 stq_le_p(&s->w1cmask[addr], w1cmask);
78 static uint16_t amdvi_readw(AMDVIState *s, hwaddr addr)
80 return lduw_le_p(&s->mmior[addr]);
83 static uint32_t amdvi_readl(AMDVIState *s, hwaddr addr)
85 return ldl_le_p(&s->mmior[addr]);
88 static uint64_t amdvi_readq(AMDVIState *s, hwaddr addr)
90 return ldq_le_p(&s->mmior[addr]);
94 static void amdvi_writeq_raw(AMDVIState *s, uint64_t val, hwaddr addr)
96 stq_le_p(&s->mmior[addr], val);
100 static void amdvi_writew(AMDVIState *s, hwaddr addr, uint16_t val)
102 uint16_t romask = lduw_le_p(&s->romask[addr]);
103 uint16_t w1cmask = lduw_le_p(&s->w1cmask[addr]);
104 uint16_t oldval = lduw_le_p(&s->mmior[addr]);
105 stw_le_p(&s->mmior[addr],
106 ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
109 static void amdvi_writel(AMDVIState *s, hwaddr addr, uint32_t val)
111 uint32_t romask = ldl_le_p(&s->romask[addr]);
112 uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
113 uint32_t oldval = ldl_le_p(&s->mmior[addr]);
114 stl_le_p(&s->mmior[addr],
115 ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
118 static void amdvi_writeq(AMDVIState *s, hwaddr addr, uint64_t val)
120 uint64_t romask = ldq_le_p(&s->romask[addr]);
121 uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
122 uint32_t oldval = ldq_le_p(&s->mmior[addr]);
123 stq_le_p(&s->mmior[addr],
124 ((oldval & romask) | (val & ~romask)) & ~(val & w1cmask));
127 /* OR a 64-bit register with a 64-bit value */
128 static bool amdvi_test_mask(AMDVIState *s, hwaddr addr, uint64_t val)
130 return amdvi_readq(s, addr) | val;
133 /* OR a 64-bit register with a 64-bit value storing result in the register */
134 static void amdvi_assign_orq(AMDVIState *s, hwaddr addr, uint64_t val)
136 amdvi_writeq_raw(s, addr, amdvi_readq(s, addr) | val);
139 /* AND a 64-bit register with a 64-bit value storing result in the register */
140 static void amdvi_assign_andq(AMDVIState *s, hwaddr addr, uint64_t val)
142 amdvi_writeq_raw(s, addr, amdvi_readq(s, addr) & val);
145 static void amdvi_generate_msi_interrupt(AMDVIState *s)
149 .requester_id = pci_requester_id(&s->pci.dev)
152 if (msi_enabled(&s->pci.dev)) {
153 msg = msi_get_message(&s->pci.dev, 0);
154 address_space_stl_le(&address_space_memory, msg.address, msg.data,
159 static void amdvi_log_event(AMDVIState *s, uint64_t *evt)
161 /* event logging not enabled */
162 if (!s->evtlog_enabled || amdvi_test_mask(s, AMDVI_MMIO_STATUS,
163 AMDVI_MMIO_STATUS_EVT_OVF)) {
167 /* event log buffer full */
168 if (s->evtlog_tail >= s->evtlog_len) {
169 amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_EVT_OVF);
170 /* generate interrupt */
171 amdvi_generate_msi_interrupt(s);
175 if (dma_memory_write(&address_space_memory, s->evtlog + s->evtlog_tail,
176 &evt, AMDVI_EVENT_LEN)) {
177 trace_amdvi_evntlog_fail(s->evtlog, s->evtlog_tail);
180 s->evtlog_tail += AMDVI_EVENT_LEN;
181 amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_COMP_INT);
182 amdvi_generate_msi_interrupt(s);
185 static void amdvi_setevent_bits(uint64_t *buffer, uint64_t value, int start,
188 int index = start / 64, bitpos = start % 64;
189 uint64_t mask = MAKE_64BIT_MASK(start, length);
190 buffer[index] &= ~mask;
191 buffer[index] |= (value << bitpos) & mask;
194 * AMDVi event structure
196 * 55:63 -> event type + miscellaneous info
197 * 63:127 -> related address
199 static void amdvi_encode_event(uint64_t *evt, uint16_t devid, uint64_t addr,
202 amdvi_setevent_bits(evt, devid, 0, 16);
203 amdvi_setevent_bits(evt, info, 55, 8);
204 amdvi_setevent_bits(evt, addr, 63, 64);
206 /* log an error encountered during a page walk
208 * @addr: virtual address in translation request
210 static void amdvi_page_fault(AMDVIState *s, uint16_t devid,
211 hwaddr addr, uint16_t info)
215 info |= AMDVI_EVENT_IOPF_I | AMDVI_EVENT_IOPF;
216 amdvi_encode_event(evt, devid, addr, info);
217 amdvi_log_event(s, evt);
218 pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
219 PCI_STATUS_SIG_TARGET_ABORT);
222 * log a master abort accessing device table
223 * @devtab : address of device table entry
224 * @info : error flags
226 static void amdvi_log_devtab_error(AMDVIState *s, uint16_t devid,
227 hwaddr devtab, uint16_t info)
231 info |= AMDVI_EVENT_DEV_TAB_HW_ERROR;
233 amdvi_encode_event(evt, devid, devtab, info);
234 amdvi_log_event(s, evt);
235 pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
236 PCI_STATUS_SIG_TARGET_ABORT);
238 /* log an event trying to access command buffer
239 * @addr : address that couldn't be accessed
241 static void amdvi_log_command_error(AMDVIState *s, hwaddr addr)
243 uint64_t evt[4], info = AMDVI_EVENT_COMMAND_HW_ERROR;
245 amdvi_encode_event(evt, 0, addr, info);
246 amdvi_log_event(s, evt);
247 pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
248 PCI_STATUS_SIG_TARGET_ABORT);
250 /* log an illegal comand event
251 * @addr : address of illegal command
253 static void amdvi_log_illegalcom_error(AMDVIState *s, uint16_t info,
258 info |= AMDVI_EVENT_ILLEGAL_COMMAND_ERROR;
259 amdvi_encode_event(evt, 0, addr, info);
260 amdvi_log_event(s, evt);
262 /* log an error accessing device table
264 * @devid : device owning the table entry
265 * @devtab : address of device table entry
266 * @info : error flags
268 static void amdvi_log_illegaldevtab_error(AMDVIState *s, uint16_t devid,
269 hwaddr addr, uint16_t info)
273 info |= AMDVI_EVENT_ILLEGAL_DEVTAB_ENTRY;
274 amdvi_encode_event(evt, devid, addr, info);
275 amdvi_log_event(s, evt);
277 /* log an error accessing a PTE entry
278 * @addr : address that couldn't be accessed
280 static void amdvi_log_pagetab_error(AMDVIState *s, uint16_t devid,
281 hwaddr addr, uint16_t info)
285 info |= AMDVI_EVENT_PAGE_TAB_HW_ERROR;
286 amdvi_encode_event(evt, devid, addr, info);
287 amdvi_log_event(s, evt);
288 pci_word_test_and_set_mask(s->pci.dev.config + PCI_STATUS,
289 PCI_STATUS_SIG_TARGET_ABORT);
292 static gboolean amdvi_uint64_equal(gconstpointer v1, gconstpointer v2)
294 return *((const uint64_t *)v1) == *((const uint64_t *)v2);
297 static guint amdvi_uint64_hash(gconstpointer v)
299 return (guint)*(const uint64_t *)v;
302 static AMDVIIOTLBEntry *amdvi_iotlb_lookup(AMDVIState *s, hwaddr addr,
305 uint64_t key = (addr >> AMDVI_PAGE_SHIFT_4K) |
306 ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
307 return g_hash_table_lookup(s->iotlb, &key);
310 static void amdvi_iotlb_reset(AMDVIState *s)
313 trace_amdvi_iotlb_reset();
314 g_hash_table_remove_all(s->iotlb);
317 static gboolean amdvi_iotlb_remove_by_devid(gpointer key, gpointer value,
320 AMDVIIOTLBEntry *entry = (AMDVIIOTLBEntry *)value;
321 uint16_t devid = *(uint16_t *)user_data;
322 return entry->devid == devid;
325 static void amdvi_iotlb_remove_page(AMDVIState *s, hwaddr addr,
328 uint64_t key = (addr >> AMDVI_PAGE_SHIFT_4K) |
329 ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
330 g_hash_table_remove(s->iotlb, &key);
333 static void amdvi_update_iotlb(AMDVIState *s, uint16_t devid,
334 uint64_t gpa, IOMMUTLBEntry to_cache,
337 AMDVIIOTLBEntry *entry = g_new(AMDVIIOTLBEntry, 1);
338 uint64_t *key = g_new(uint64_t, 1);
339 uint64_t gfn = gpa >> AMDVI_PAGE_SHIFT_4K;
341 /* don't cache erroneous translations */
342 if (to_cache.perm != IOMMU_NONE) {
343 trace_amdvi_cache_update(domid, PCI_BUS_NUM(devid), PCI_SLOT(devid),
344 PCI_FUNC(devid), gpa, to_cache.translated_addr);
346 if (g_hash_table_size(s->iotlb) >= AMDVI_IOTLB_MAX_SIZE) {
347 amdvi_iotlb_reset(s);
350 entry->domid = domid;
351 entry->perms = to_cache.perm;
352 entry->translated_addr = to_cache.translated_addr;
353 entry->page_mask = to_cache.addr_mask;
354 *key = gfn | ((uint64_t)(devid) << AMDVI_DEVID_SHIFT);
355 g_hash_table_replace(s->iotlb, key, entry);
359 static void amdvi_completion_wait(AMDVIState *s, uint64_t *cmd)
361 /* pad the last 3 bits */
362 hwaddr addr = cpu_to_le64(extract64(cmd[0], 3, 49)) << 3;
363 uint64_t data = cpu_to_le64(cmd[1]);
365 if (extract64(cmd[0], 51, 8)) {
366 amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
367 s->cmdbuf + s->cmdbuf_head);
369 if (extract64(cmd[0], 0, 1)) {
370 if (dma_memory_write(&address_space_memory, addr, &data,
371 AMDVI_COMPLETION_DATA_SIZE)) {
372 trace_amdvi_completion_wait_fail(addr);
375 /* set completion interrupt */
376 if (extract64(cmd[0], 1, 1)) {
377 amdvi_test_mask(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_COMP_INT);
378 /* generate interrupt */
379 amdvi_generate_msi_interrupt(s);
381 trace_amdvi_completion_wait(addr, data);
384 /* log error without aborting since linux seems to be using reserved bits */
385 static void amdvi_inval_devtab_entry(AMDVIState *s, uint64_t *cmd)
387 uint16_t devid = cpu_to_le16((uint16_t)extract64(cmd[0], 0, 16));
389 /* This command should invalidate internal caches of which there isn't */
390 if (extract64(cmd[0], 15, 16) || cmd[1]) {
391 amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
392 s->cmdbuf + s->cmdbuf_head);
394 trace_amdvi_devtab_inval(PCI_BUS_NUM(devid), PCI_SLOT(devid),
398 static void amdvi_complete_ppr(AMDVIState *s, uint64_t *cmd)
400 if (extract64(cmd[0], 15, 16) || extract64(cmd[0], 19, 8) ||
401 extract64(cmd[1], 0, 2) || extract64(cmd[1], 3, 29)
402 || extract64(cmd[1], 47, 16)) {
403 amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
404 s->cmdbuf + s->cmdbuf_head);
406 trace_amdvi_ppr_exec();
409 static void amdvi_inval_all(AMDVIState *s, uint64_t *cmd)
411 if (extract64(cmd[0], 0, 60) || cmd[1]) {
412 amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
413 s->cmdbuf + s->cmdbuf_head);
416 amdvi_iotlb_reset(s);
417 trace_amdvi_all_inval();
420 static gboolean amdvi_iotlb_remove_by_domid(gpointer key, gpointer value,
423 AMDVIIOTLBEntry *entry = (AMDVIIOTLBEntry *)value;
424 uint16_t domid = *(uint16_t *)user_data;
425 return entry->domid == domid;
428 /* we don't have devid - we can't remove pages by address */
429 static void amdvi_inval_pages(AMDVIState *s, uint64_t *cmd)
431 uint16_t domid = cpu_to_le16((uint16_t)extract64(cmd[0], 32, 16));
433 if (extract64(cmd[0], 20, 12) || extract64(cmd[0], 16, 12) ||
434 extract64(cmd[0], 3, 10)) {
435 amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
436 s->cmdbuf + s->cmdbuf_head);
439 g_hash_table_foreach_remove(s->iotlb, amdvi_iotlb_remove_by_domid,
441 trace_amdvi_pages_inval(domid);
444 static void amdvi_prefetch_pages(AMDVIState *s, uint64_t *cmd)
446 if (extract64(cmd[0], 16, 8) || extract64(cmd[0], 20, 8) ||
447 extract64(cmd[1], 1, 1) || extract64(cmd[1], 3, 1) ||
448 extract64(cmd[1], 5, 7)) {
449 amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
450 s->cmdbuf + s->cmdbuf_head);
453 trace_amdvi_prefetch_pages();
456 static void amdvi_inval_inttable(AMDVIState *s, uint64_t *cmd)
458 if (extract64(cmd[0], 16, 16) || cmd[1]) {
459 amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
460 s->cmdbuf + s->cmdbuf_head);
464 trace_amdvi_intr_inval();
467 /* FIXME: Try to work with the specified size instead of all the pages
468 * when the S bit is on
470 static void iommu_inval_iotlb(AMDVIState *s, uint64_t *cmd)
473 uint16_t devid = extract64(cmd[0], 0, 16);
474 if (extract64(cmd[1], 1, 1) || extract64(cmd[1], 3, 9)) {
475 amdvi_log_illegalcom_error(s, extract64(cmd[0], 60, 4),
476 s->cmdbuf + s->cmdbuf_head);
480 if (extract64(cmd[1], 0, 1)) {
481 g_hash_table_foreach_remove(s->iotlb, amdvi_iotlb_remove_by_devid,
484 amdvi_iotlb_remove_page(s, cpu_to_le64(extract64(cmd[1], 12, 52)) << 12,
485 cpu_to_le16(extract64(cmd[1], 0, 16)));
487 trace_amdvi_iotlb_inval();
490 /* not honouring reserved bits is regarded as an illegal command */
491 static void amdvi_cmdbuf_exec(AMDVIState *s)
495 if (dma_memory_read(&address_space_memory, s->cmdbuf + s->cmdbuf_head,
496 cmd, AMDVI_COMMAND_SIZE)) {
497 trace_amdvi_command_read_fail(s->cmdbuf, s->cmdbuf_head);
498 amdvi_log_command_error(s, s->cmdbuf + s->cmdbuf_head);
502 switch (extract64(cmd[0], 60, 4)) {
503 case AMDVI_CMD_COMPLETION_WAIT:
504 amdvi_completion_wait(s, cmd);
506 case AMDVI_CMD_INVAL_DEVTAB_ENTRY:
507 amdvi_inval_devtab_entry(s, cmd);
509 case AMDVI_CMD_INVAL_AMDVI_PAGES:
510 amdvi_inval_pages(s, cmd);
512 case AMDVI_CMD_INVAL_IOTLB_PAGES:
513 iommu_inval_iotlb(s, cmd);
515 case AMDVI_CMD_INVAL_INTR_TABLE:
516 amdvi_inval_inttable(s, cmd);
518 case AMDVI_CMD_PREFETCH_AMDVI_PAGES:
519 amdvi_prefetch_pages(s, cmd);
521 case AMDVI_CMD_COMPLETE_PPR_REQUEST:
522 amdvi_complete_ppr(s, cmd);
524 case AMDVI_CMD_INVAL_AMDVI_ALL:
525 amdvi_inval_all(s, cmd);
528 trace_amdvi_unhandled_command(extract64(cmd[1], 60, 4));
529 /* log illegal command */
530 amdvi_log_illegalcom_error(s, extract64(cmd[1], 60, 4),
531 s->cmdbuf + s->cmdbuf_head);
535 static void amdvi_cmdbuf_run(AMDVIState *s)
537 if (!s->cmdbuf_enabled) {
538 trace_amdvi_command_error(amdvi_readq(s, AMDVI_MMIO_CONTROL));
542 /* check if there is work to do. */
543 while (s->cmdbuf_head != s->cmdbuf_tail) {
544 trace_amdvi_command_exec(s->cmdbuf_head, s->cmdbuf_tail, s->cmdbuf);
545 amdvi_cmdbuf_exec(s);
546 s->cmdbuf_head += AMDVI_COMMAND_SIZE;
547 amdvi_writeq_raw(s, s->cmdbuf_head, AMDVI_MMIO_COMMAND_HEAD);
549 /* wrap head pointer */
550 if (s->cmdbuf_head >= s->cmdbuf_len * AMDVI_COMMAND_SIZE) {
556 static void amdvi_mmio_trace(hwaddr addr, unsigned size)
558 uint8_t index = (addr & ~0x2000) / 8;
560 if ((addr & 0x2000)) {
562 index = index >= AMDVI_MMIO_REGS_HIGH ? AMDVI_MMIO_REGS_HIGH : index;
563 trace_amdvi_mmio_read(amdvi_mmio_high[index], addr, size, addr & ~0x07);
565 index = index >= AMDVI_MMIO_REGS_LOW ? AMDVI_MMIO_REGS_LOW : index;
566 trace_amdvi_mmio_read(amdvi_mmio_low[index], addr, size, addr & ~0x07);
570 static uint64_t amdvi_mmio_read(void *opaque, hwaddr addr, unsigned size)
572 AMDVIState *s = opaque;
575 if (addr + size > AMDVI_MMIO_SIZE) {
576 trace_amdvi_mmio_read_invalid(AMDVI_MMIO_SIZE, addr, size);
581 val = amdvi_readw(s, addr);
582 } else if (size == 4) {
583 val = amdvi_readl(s, addr);
584 } else if (size == 8) {
585 val = amdvi_readq(s, addr);
587 amdvi_mmio_trace(addr, size);
592 static void amdvi_handle_control_write(AMDVIState *s)
594 unsigned long control = amdvi_readq(s, AMDVI_MMIO_CONTROL);
595 s->enabled = !!(control & AMDVI_MMIO_CONTROL_AMDVIEN);
597 s->ats_enabled = !!(control & AMDVI_MMIO_CONTROL_HTTUNEN);
598 s->evtlog_enabled = s->enabled && !!(control &
599 AMDVI_MMIO_CONTROL_EVENTLOGEN);
601 s->evtlog_intr = !!(control & AMDVI_MMIO_CONTROL_EVENTINTEN);
602 s->completion_wait_intr = !!(control & AMDVI_MMIO_CONTROL_COMWAITINTEN);
603 s->cmdbuf_enabled = s->enabled && !!(control &
604 AMDVI_MMIO_CONTROL_CMDBUFLEN);
606 /* update the flags depending on the control register */
607 if (s->cmdbuf_enabled) {
608 amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_CMDBUF_RUN);
610 amdvi_assign_andq(s, AMDVI_MMIO_STATUS, ~AMDVI_MMIO_STATUS_CMDBUF_RUN);
612 if (s->evtlog_enabled) {
613 amdvi_assign_orq(s, AMDVI_MMIO_STATUS, AMDVI_MMIO_STATUS_EVT_RUN);
615 amdvi_assign_andq(s, AMDVI_MMIO_STATUS, ~AMDVI_MMIO_STATUS_EVT_RUN);
618 trace_amdvi_control_status(control);
622 static inline void amdvi_handle_devtab_write(AMDVIState *s)
625 uint64_t val = amdvi_readq(s, AMDVI_MMIO_DEVICE_TABLE);
626 s->devtab = (val & AMDVI_MMIO_DEVTAB_BASE_MASK);
628 /* set device table length */
629 s->devtab_len = ((val & AMDVI_MMIO_DEVTAB_SIZE_MASK) + 1 *
630 (AMDVI_MMIO_DEVTAB_SIZE_UNIT /
631 AMDVI_MMIO_DEVTAB_ENTRY_SIZE));
634 static inline void amdvi_handle_cmdhead_write(AMDVIState *s)
636 s->cmdbuf_head = amdvi_readq(s, AMDVI_MMIO_COMMAND_HEAD)
637 & AMDVI_MMIO_CMDBUF_HEAD_MASK;
641 static inline void amdvi_handle_cmdbase_write(AMDVIState *s)
643 s->cmdbuf = amdvi_readq(s, AMDVI_MMIO_COMMAND_BASE)
644 & AMDVI_MMIO_CMDBUF_BASE_MASK;
645 s->cmdbuf_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_CMDBUF_SIZE_BYTE)
646 & AMDVI_MMIO_CMDBUF_SIZE_MASK);
647 s->cmdbuf_head = s->cmdbuf_tail = 0;
650 static inline void amdvi_handle_cmdtail_write(AMDVIState *s)
652 s->cmdbuf_tail = amdvi_readq(s, AMDVI_MMIO_COMMAND_TAIL)
653 & AMDVI_MMIO_CMDBUF_TAIL_MASK;
657 static inline void amdvi_handle_excllim_write(AMDVIState *s)
659 uint64_t val = amdvi_readq(s, AMDVI_MMIO_EXCL_LIMIT);
660 s->excl_limit = (val & AMDVI_MMIO_EXCL_LIMIT_MASK) |
661 AMDVI_MMIO_EXCL_LIMIT_LOW;
664 static inline void amdvi_handle_evtbase_write(AMDVIState *s)
666 uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_BASE);
667 s->evtlog = val & AMDVI_MMIO_EVTLOG_BASE_MASK;
668 s->evtlog_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_EVTLOG_SIZE_BYTE)
669 & AMDVI_MMIO_EVTLOG_SIZE_MASK);
672 static inline void amdvi_handle_evttail_write(AMDVIState *s)
674 uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_TAIL);
675 s->evtlog_tail = val & AMDVI_MMIO_EVTLOG_TAIL_MASK;
678 static inline void amdvi_handle_evthead_write(AMDVIState *s)
680 uint64_t val = amdvi_readq(s, AMDVI_MMIO_EVENT_HEAD);
681 s->evtlog_head = val & AMDVI_MMIO_EVTLOG_HEAD_MASK;
684 static inline void amdvi_handle_pprbase_write(AMDVIState *s)
686 uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_BASE);
687 s->ppr_log = val & AMDVI_MMIO_PPRLOG_BASE_MASK;
688 s->pprlog_len = 1UL << (amdvi_readq(s, AMDVI_MMIO_PPRLOG_SIZE_BYTE)
689 & AMDVI_MMIO_PPRLOG_SIZE_MASK);
692 static inline void amdvi_handle_pprhead_write(AMDVIState *s)
694 uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_HEAD);
695 s->pprlog_head = val & AMDVI_MMIO_PPRLOG_HEAD_MASK;
698 static inline void amdvi_handle_pprtail_write(AMDVIState *s)
700 uint64_t val = amdvi_readq(s, AMDVI_MMIO_PPR_TAIL);
701 s->pprlog_tail = val & AMDVI_MMIO_PPRLOG_TAIL_MASK;
704 /* FIXME: something might go wrong if System Software writes in chunks
705 * of one byte but linux writes in chunks of 4 bytes so currently it
706 * works correctly with linux but will definitely be busted if software
707 * reads/writes 8 bytes
709 static void amdvi_mmio_reg_write(AMDVIState *s, unsigned size, uint64_t val,
713 amdvi_writew(s, addr, val);
714 } else if (size == 4) {
715 amdvi_writel(s, addr, val);
716 } else if (size == 8) {
717 amdvi_writeq(s, addr, val);
721 static void amdvi_mmio_write(void *opaque, hwaddr addr, uint64_t val,
724 AMDVIState *s = opaque;
725 unsigned long offset = addr & 0x07;
727 if (addr + size > AMDVI_MMIO_SIZE) {
728 trace_amdvi_mmio_write("error: addr outside region: max ",
729 (uint64_t)AMDVI_MMIO_SIZE, size, val, offset);
733 amdvi_mmio_trace(addr, size);
734 switch (addr & ~0x07) {
735 case AMDVI_MMIO_CONTROL:
736 amdvi_mmio_reg_write(s, size, val, addr);
737 amdvi_handle_control_write(s);
739 case AMDVI_MMIO_DEVICE_TABLE:
740 amdvi_mmio_reg_write(s, size, val, addr);
741 /* set device table address
742 * This also suffers from inability to tell whether software
745 if (offset || (size == 8)) {
746 amdvi_handle_devtab_write(s);
749 case AMDVI_MMIO_COMMAND_HEAD:
750 amdvi_mmio_reg_write(s, size, val, addr);
751 amdvi_handle_cmdhead_write(s);
753 case AMDVI_MMIO_COMMAND_BASE:
754 amdvi_mmio_reg_write(s, size, val, addr);
755 /* FIXME - make sure System Software has finished writing incase
756 * it writes in chucks less than 8 bytes in a robust way.As for
757 * now, this hacks works for the linux driver
759 if (offset || (size == 8)) {
760 amdvi_handle_cmdbase_write(s);
763 case AMDVI_MMIO_COMMAND_TAIL:
764 amdvi_mmio_reg_write(s, size, val, addr);
765 amdvi_handle_cmdtail_write(s);
767 case AMDVI_MMIO_EVENT_BASE:
768 amdvi_mmio_reg_write(s, size, val, addr);
769 amdvi_handle_evtbase_write(s);
771 case AMDVI_MMIO_EVENT_HEAD:
772 amdvi_mmio_reg_write(s, size, val, addr);
773 amdvi_handle_evthead_write(s);
775 case AMDVI_MMIO_EVENT_TAIL:
776 amdvi_mmio_reg_write(s, size, val, addr);
777 amdvi_handle_evttail_write(s);
779 case AMDVI_MMIO_EXCL_LIMIT:
780 amdvi_mmio_reg_write(s, size, val, addr);
781 amdvi_handle_excllim_write(s);
783 /* PPR log base - unused for now */
784 case AMDVI_MMIO_PPR_BASE:
785 amdvi_mmio_reg_write(s, size, val, addr);
786 amdvi_handle_pprbase_write(s);
788 /* PPR log head - also unused for now */
789 case AMDVI_MMIO_PPR_HEAD:
790 amdvi_mmio_reg_write(s, size, val, addr);
791 amdvi_handle_pprhead_write(s);
793 /* PPR log tail - unused for now */
794 case AMDVI_MMIO_PPR_TAIL:
795 amdvi_mmio_reg_write(s, size, val, addr);
796 amdvi_handle_pprtail_write(s);
801 static inline uint64_t amdvi_get_perms(uint64_t entry)
803 return (entry & (AMDVI_DEV_PERM_READ | AMDVI_DEV_PERM_WRITE)) >>
804 AMDVI_DEV_PERM_SHIFT;
807 /* a valid entry should have V = 1 and reserved bits honoured */
808 static bool amdvi_validate_dte(AMDVIState *s, uint16_t devid,
811 if ((dte[0] & AMDVI_DTE_LOWER_QUAD_RESERVED)
812 || (dte[1] & AMDVI_DTE_MIDDLE_QUAD_RESERVED)
813 || (dte[2] & AMDVI_DTE_UPPER_QUAD_RESERVED) || dte[3]) {
814 amdvi_log_illegaldevtab_error(s, devid,
816 devid * AMDVI_DEVTAB_ENTRY_SIZE, 0);
820 return dte[0] & AMDVI_DEV_VALID;
823 /* get a device table entry given the devid */
824 static bool amdvi_get_dte(AMDVIState *s, int devid, uint64_t *entry)
826 uint32_t offset = devid * AMDVI_DEVTAB_ENTRY_SIZE;
828 if (dma_memory_read(&address_space_memory, s->devtab + offset, entry,
829 AMDVI_DEVTAB_ENTRY_SIZE)) {
830 trace_amdvi_dte_get_fail(s->devtab, offset);
831 /* log error accessing dte */
832 amdvi_log_devtab_error(s, devid, s->devtab + offset, 0);
836 *entry = le64_to_cpu(*entry);
837 if (!amdvi_validate_dte(s, devid, entry)) {
838 trace_amdvi_invalid_dte(entry[0]);
845 /* get pte translation mode */
846 static inline uint8_t get_pte_translation_mode(uint64_t pte)
848 return (pte >> AMDVI_DEV_MODE_RSHIFT) & AMDVI_DEV_MODE_MASK;
851 static inline uint64_t pte_override_page_mask(uint64_t pte)
853 uint8_t page_mask = 12;
854 uint64_t addr = (pte & AMDVI_DEV_PT_ROOT_MASK) ^ AMDVI_DEV_PT_ROOT_MASK;
855 /* find the first zero bit */
861 return ~((1ULL << page_mask) - 1);
864 static inline uint64_t pte_get_page_mask(uint64_t oldlevel)
866 return ~((1UL << ((oldlevel * 9) + 3)) - 1);
869 static inline uint64_t amdvi_get_pte_entry(AMDVIState *s, uint64_t pte_addr,
874 if (dma_memory_read(&address_space_memory, pte_addr, &pte, sizeof(pte))) {
875 trace_amdvi_get_pte_hwerror(pte_addr);
876 amdvi_log_pagetab_error(s, devid, pte_addr, 0);
881 pte = le64_to_cpu(pte);
885 static void amdvi_page_walk(AMDVIAddressSpace *as, uint64_t *dte,
886 IOMMUTLBEntry *ret, unsigned perms,
889 unsigned level, present, pte_perms, oldlevel;
890 uint64_t pte = dte[0], pte_addr, page_mask;
892 /* make sure the DTE has TV = 1 */
893 if (pte & AMDVI_DEV_TRANSLATION_VALID) {
894 level = get_pte_translation_mode(pte);
896 trace_amdvi_mode_invalid(level, addr);
903 /* we are at the leaf page table or page table encodes a huge page */
905 pte_perms = amdvi_get_perms(pte);
907 if (!present || perms != (perms & pte_perms)) {
908 amdvi_page_fault(as->iommu_state, as->devfn, addr, perms);
909 trace_amdvi_page_fault(addr);
913 /* go to the next lower level */
914 pte_addr = pte & AMDVI_DEV_PT_ROOT_MASK;
915 /* add offset and load pte */
916 pte_addr += ((addr >> (3 + 9 * level)) & 0x1FF) << 3;
917 pte = amdvi_get_pte_entry(as->iommu_state, pte_addr, as->devfn);
922 level = get_pte_translation_mode(pte);
929 page_mask = pte_override_page_mask(pte);
931 page_mask = pte_get_page_mask(oldlevel);
934 /* get access permissions from pte */
935 ret->iova = addr & page_mask;
936 ret->translated_addr = (pte & AMDVI_DEV_PT_ROOT_MASK) & page_mask;
937 ret->addr_mask = ~page_mask;
938 ret->perm = amdvi_get_perms(pte);
942 ret->iova = addr & AMDVI_PAGE_MASK_4K;
943 ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
944 ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
945 ret->perm = amdvi_get_perms(pte);
948 static void amdvi_do_translate(AMDVIAddressSpace *as, hwaddr addr,
949 bool is_write, IOMMUTLBEntry *ret)
951 AMDVIState *s = as->iommu_state;
952 uint16_t devid = PCI_BUILD_BDF(as->bus_num, as->devfn);
953 AMDVIIOTLBEntry *iotlb_entry = amdvi_iotlb_lookup(s, addr, devid);
957 trace_amdvi_iotlb_hit(PCI_BUS_NUM(devid), PCI_SLOT(devid),
958 PCI_FUNC(devid), addr, iotlb_entry->translated_addr);
959 ret->iova = addr & ~iotlb_entry->page_mask;
960 ret->translated_addr = iotlb_entry->translated_addr;
961 ret->addr_mask = iotlb_entry->page_mask;
962 ret->perm = iotlb_entry->perms;
966 /* devices with V = 0 are not translated */
967 if (!amdvi_get_dte(s, devid, entry)) {
971 amdvi_page_walk(as, entry, ret,
972 is_write ? AMDVI_PERM_WRITE : AMDVI_PERM_READ, addr);
974 amdvi_update_iotlb(s, devid, addr, *ret,
975 entry[1] & AMDVI_DEV_DOMID_ID_MASK);
979 ret->iova = addr & AMDVI_PAGE_MASK_4K;
980 ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
981 ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
982 ret->perm = IOMMU_RW;
985 static inline bool amdvi_is_interrupt_addr(hwaddr addr)
987 return addr >= AMDVI_INT_ADDR_FIRST && addr <= AMDVI_INT_ADDR_LAST;
990 static IOMMUTLBEntry amdvi_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
991 IOMMUAccessFlags flag)
993 AMDVIAddressSpace *as = container_of(iommu, AMDVIAddressSpace, iommu);
994 AMDVIState *s = as->iommu_state;
995 IOMMUTLBEntry ret = {
996 .target_as = &address_space_memory,
998 .translated_addr = 0,
999 .addr_mask = ~(hwaddr)0,
1004 /* AMDVI disabled - corresponds to iommu=off not
1005 * failure to provide any parameter
1007 ret.iova = addr & AMDVI_PAGE_MASK_4K;
1008 ret.translated_addr = addr & AMDVI_PAGE_MASK_4K;
1009 ret.addr_mask = ~AMDVI_PAGE_MASK_4K;
1010 ret.perm = IOMMU_RW;
1012 } else if (amdvi_is_interrupt_addr(addr)) {
1013 ret.iova = addr & AMDVI_PAGE_MASK_4K;
1014 ret.translated_addr = addr & AMDVI_PAGE_MASK_4K;
1015 ret.addr_mask = ~AMDVI_PAGE_MASK_4K;
1016 ret.perm = IOMMU_WO;
1020 amdvi_do_translate(as, addr, flag & IOMMU_WO, &ret);
1021 trace_amdvi_translation_result(as->bus_num, PCI_SLOT(as->devfn),
1022 PCI_FUNC(as->devfn), addr, ret.translated_addr);
1026 static AddressSpace *amdvi_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
1028 AMDVIState *s = opaque;
1029 AMDVIAddressSpace **iommu_as;
1030 int bus_num = pci_bus_num(bus);
1032 iommu_as = s->address_spaces[bus_num];
1034 /* allocate memory during the first run */
1036 iommu_as = g_malloc0(sizeof(AMDVIAddressSpace *) * PCI_DEVFN_MAX);
1037 s->address_spaces[bus_num] = iommu_as;
1040 /* set up AMD-Vi region */
1041 if (!iommu_as[devfn]) {
1042 iommu_as[devfn] = g_malloc0(sizeof(AMDVIAddressSpace));
1043 iommu_as[devfn]->bus_num = (uint8_t)bus_num;
1044 iommu_as[devfn]->devfn = (uint8_t)devfn;
1045 iommu_as[devfn]->iommu_state = s;
1047 memory_region_init_iommu(&iommu_as[devfn]->iommu, OBJECT(s),
1048 &s->iommu_ops, "amd-iommu", UINT64_MAX);
1049 address_space_init(&iommu_as[devfn]->as,
1050 MEMORY_REGION(&iommu_as[devfn]->iommu),
1053 return &iommu_as[devfn]->as;
1056 static const MemoryRegionOps mmio_mem_ops = {
1057 .read = amdvi_mmio_read,
1058 .write = amdvi_mmio_write,
1059 .endianness = DEVICE_LITTLE_ENDIAN,
1061 .min_access_size = 1,
1062 .max_access_size = 8,
1066 .min_access_size = 1,
1067 .max_access_size = 8,
1071 static void amdvi_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
1072 IOMMUNotifierFlag old,
1073 IOMMUNotifierFlag new)
1075 AMDVIAddressSpace *as = container_of(iommu, AMDVIAddressSpace, iommu);
1077 if (new & IOMMU_NOTIFIER_MAP) {
1078 error_report("device %02x.%02x.%x requires iommu notifier which is not "
1079 "currently supported", as->bus_num, PCI_SLOT(as->devfn),
1080 PCI_FUNC(as->devfn));
1085 static void amdvi_init(AMDVIState *s)
1087 amdvi_iotlb_reset(s);
1089 s->iommu_ops.translate = amdvi_translate;
1090 s->iommu_ops.notify_flag_changed = amdvi_iommu_notify_flag_changed;
1097 s->excl_enabled = false;
1098 s->excl_allow = false;
1099 s->mmio_enabled = false;
1101 s->ats_enabled = false;
1102 s->cmdbuf_enabled = false;
1105 memset(s->mmior, 0, AMDVI_MMIO_SIZE);
1106 amdvi_set_quad(s, AMDVI_MMIO_EXT_FEATURES, AMDVI_EXT_FEATURES,
1107 0xffffffffffffffef, 0);
1108 amdvi_set_quad(s, AMDVI_MMIO_STATUS, 0, 0x98, 0x67);
1110 /* reset device ident */
1111 pci_config_set_vendor_id(s->pci.dev.config, PCI_VENDOR_ID_AMD);
1112 pci_config_set_prog_interface(s->pci.dev.config, 00);
1113 pci_config_set_device_id(s->pci.dev.config, s->devid);
1114 pci_config_set_class(s->pci.dev.config, 0x0806);
1116 /* reset AMDVI specific capabilities, all r/o */
1117 pci_set_long(s->pci.dev.config + s->capab_offset, AMDVI_CAPAB_FEATURES);
1118 pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_BAR_LOW,
1119 s->mmio.addr & ~(0xffff0000));
1120 pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_BAR_HIGH,
1121 (s->mmio.addr & ~(0xffff)) >> 16);
1122 pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_RANGE,
1124 pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_MISC, 0);
1125 pci_set_long(s->pci.dev.config + s->capab_offset + AMDVI_CAPAB_MISC,
1126 AMDVI_MAX_PH_ADDR | AMDVI_MAX_GVA_ADDR | AMDVI_MAX_VA_ADDR);
1129 static void amdvi_reset(DeviceState *dev)
1131 AMDVIState *s = AMD_IOMMU_DEVICE(dev);
1133 msi_reset(&s->pci.dev);
1137 static void amdvi_realize(DeviceState *dev, Error **err)
1140 AMDVIState *s = AMD_IOMMU_DEVICE(dev);
1141 X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
1142 MachineState *ms = MACHINE(qdev_get_machine());
1143 MachineClass *mc = MACHINE_GET_CLASS(ms);
1144 PCMachineState *pcms =
1145 PC_MACHINE(object_dynamic_cast(OBJECT(ms), TYPE_PC_MACHINE));
1149 error_setg(err, "Machine-type '%s' not supported by amd-iommu",
1155 s->iotlb = g_hash_table_new_full(amdvi_uint64_hash,
1156 amdvi_uint64_equal, g_free, g_free);
1158 /* This device should take care of IOMMU PCI properties */
1159 x86_iommu->type = TYPE_AMD;
1160 qdev_set_parent_bus(DEVICE(&s->pci), &bus->qbus);
1161 object_property_set_bool(OBJECT(&s->pci), true, "realized", err);
1162 ret = pci_add_capability(&s->pci.dev, AMDVI_CAPAB_ID_SEC, 0,
1163 AMDVI_CAPAB_SIZE, err);
1167 s->capab_offset = ret;
1169 ret = pci_add_capability(&s->pci.dev, PCI_CAP_ID_MSI, 0,
1170 AMDVI_CAPAB_REG_SIZE, err);
1174 ret = pci_add_capability(&s->pci.dev, PCI_CAP_ID_HT, 0,
1175 AMDVI_CAPAB_REG_SIZE, err);
1181 memory_region_init_io(&s->mmio, OBJECT(s), &mmio_mem_ops, s, "amdvi-mmio",
1184 sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->mmio);
1185 sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, AMDVI_BASE_ADDR);
1186 pci_setup_iommu(bus, amdvi_host_dma_iommu, s);
1187 s->devid = object_property_get_int(OBJECT(&s->pci), "addr", err);
1188 msi_init(&s->pci.dev, 0, 1, true, false, err);
1192 static const VMStateDescription vmstate_amdvi = {
1193 .name = "amd-iommu",
1197 static void amdvi_instance_init(Object *klass)
1199 AMDVIState *s = AMD_IOMMU_DEVICE(klass);
1201 object_initialize(&s->pci, sizeof(s->pci), TYPE_AMD_IOMMU_PCI);
1204 static void amdvi_class_init(ObjectClass *klass, void* data)
1206 DeviceClass *dc = DEVICE_CLASS(klass);
1207 X86IOMMUClass *dc_class = X86_IOMMU_CLASS(klass);
1209 dc->reset = amdvi_reset;
1210 dc->vmsd = &vmstate_amdvi;
1211 dc->hotpluggable = false;
1212 dc_class->realize = amdvi_realize;
1213 /* Supported by the pc-q35-* machine types */
1214 dc->user_creatable = true;
1217 static const TypeInfo amdvi = {
1218 .name = TYPE_AMD_IOMMU_DEVICE,
1219 .parent = TYPE_X86_IOMMU_DEVICE,
1220 .instance_size = sizeof(AMDVIState),
1221 .instance_init = amdvi_instance_init,
1222 .class_init = amdvi_class_init
1225 static const TypeInfo amdviPCI = {
1226 .name = "AMDVI-PCI",
1227 .parent = TYPE_PCI_DEVICE,
1228 .instance_size = sizeof(AMDVIPCIState),
1231 static void amdviPCI_register_types(void)
1233 type_register_static(&amdviPCI);
1234 type_register_static(&amdvi);
1237 type_init(amdviPCI_register_types);