xref: /freebsd/sys/amd64/vmm/amd/amdvi_hw.c (revision d0b2dbfa)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2016, Anish Gupta (anish@freebsd.org)
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/bus.h>
33 #include <sys/kernel.h>
34 #include <sys/module.h>
35 #include <sys/malloc.h>
36 #include <sys/pcpu.h>
37 #include <sys/rman.h>
38 #include <sys/smp.h>
39 #include <sys/sysctl.h>
40 
41 #include <vm/vm.h>
42 #include <vm/pmap.h>
43 
44 #include <dev/pci/pcivar.h>
45 #include <dev/pci/pcireg.h>
46 
47 #include <machine/resource.h>
48 #include <machine/vmm.h>
49 #include <machine/pmap.h>
50 #include <machine/vmparam.h>
51 #include <machine/pci_cfgreg.h>
52 
53 #include "ivhd_if.h"
54 #include "pcib_if.h"
55 
56 #include "io/iommu.h"
57 #include "amdvi_priv.h"
58 
59 SYSCTL_DECL(_hw_vmm);
60 SYSCTL_NODE(_hw_vmm, OID_AUTO, amdvi, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
61     NULL);
62 
63 #define MOD_INC(a, s, m) (((a) + (s)) % ((m) * (s)))
64 #define MOD_DEC(a, s, m) (((a) - (s)) % ((m) * (s)))
65 
66 /* Print RID or device ID in PCI string format. */
67 #define RID2PCI_STR(d) PCI_RID2BUS(d), PCI_RID2SLOT(d), PCI_RID2FUNC(d)
68 
69 static void amdvi_dump_cmds(struct amdvi_softc *softc, int count);
70 static void amdvi_print_dev_cap(struct amdvi_softc *softc);
71 
72 MALLOC_DEFINE(M_AMDVI, "amdvi", "amdvi");
73 
74 extern device_t *ivhd_devs;
75 
76 extern int ivhd_count;
77 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, count, CTLFLAG_RDTUN, &ivhd_count,
78     0, NULL);
79 
80 static int amdvi_enable_user = 0;
81 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, enable, CTLFLAG_RDTUN,
82     &amdvi_enable_user, 0, NULL);
83 TUNABLE_INT("hw.vmm.amdvi_enable", &amdvi_enable_user);
84 
85 #ifdef AMDVI_ATS_ENABLE
86 /* XXX: ATS is not tested. */
87 static int amdvi_enable_iotlb = 1;
88 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, iotlb_enabled, CTLFLAG_RDTUN,
89     &amdvi_enable_iotlb, 0, NULL);
90 TUNABLE_INT("hw.vmm.enable_iotlb", &amdvi_enable_iotlb);
91 #endif
92 
93 static int amdvi_host_ptp = 1;	/* Use page tables for host. */
94 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, host_ptp, CTLFLAG_RDTUN,
95     &amdvi_host_ptp, 0, NULL);
96 TUNABLE_INT("hw.vmm.amdvi.host_ptp", &amdvi_host_ptp);
97 
98 /* Page table level used <= supported by h/w[v1=7]. */
99 int amdvi_ptp_level = 4;
100 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, ptp_level, CTLFLAG_RDTUN,
101     &amdvi_ptp_level, 0, NULL);
102 TUNABLE_INT("hw.vmm.amdvi.ptp_level", &amdvi_ptp_level);
103 
104 /* Disable fault event reporting. */
105 static int amdvi_disable_io_fault = 0;
106 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, disable_io_fault, CTLFLAG_RDTUN,
107     &amdvi_disable_io_fault, 0, NULL);
108 TUNABLE_INT("hw.vmm.amdvi.disable_io_fault", &amdvi_disable_io_fault);
109 
110 static uint32_t amdvi_dom_id = 0;	/* 0 is reserved for host. */
111 SYSCTL_UINT(_hw_vmm_amdvi, OID_AUTO, domain_id, CTLFLAG_RD,
112     &amdvi_dom_id, 0, NULL);
113 /*
114  * Device table entry.
115  * Bus(256) x Dev(32) x Fun(8) x DTE(256 bits or 32 bytes).
116  *	= 256 * 2 * PAGE_SIZE.
117  */
118 static struct amdvi_dte amdvi_dte[PCI_NUM_DEV_MAX] __aligned(PAGE_SIZE);
119 CTASSERT(PCI_NUM_DEV_MAX == 0x10000);
120 CTASSERT(sizeof(amdvi_dte) == 0x200000);
121 
122 static SLIST_HEAD (, amdvi_domain) dom_head;
123 
124 static inline uint32_t
125 amdvi_pci_read(struct amdvi_softc *softc, int off)
126 {
127 
128 	return (pci_cfgregread(PCI_RID2BUS(softc->pci_rid),
129 	    PCI_RID2SLOT(softc->pci_rid), PCI_RID2FUNC(softc->pci_rid),
130 	    off, 4));
131 }
132 
133 #ifdef AMDVI_ATS_ENABLE
134 /* XXX: Should be in pci.c */
135 /*
136  * Check if device has ATS capability and its enabled.
137  * If ATS is absent or disabled, return (-1), otherwise ATS
138  * queue length.
139  */
140 static int
141 amdvi_find_ats_qlen(uint16_t devid)
142 {
143 	device_t dev;
144 	uint32_t off, cap;
145 	int qlen = -1;
146 
147 	dev = pci_find_bsf(PCI_RID2BUS(devid), PCI_RID2SLOT(devid),
148 			   PCI_RID2FUNC(devid));
149 
150 	if (!dev) {
151 		return (-1);
152 	}
153 #define PCIM_ATS_EN	BIT(31)
154 
155 	if (pci_find_extcap(dev, PCIZ_ATS, &off) == 0) {
156 		cap = pci_read_config(dev, off + 4, 4);
157 		qlen = (cap & 0x1F);
158 		qlen = qlen ? qlen : 32;
159 		printf("AMD-Vi: PCI device %d.%d.%d ATS %s qlen=%d\n",
160 		       RID2PCI_STR(devid),
161 		       (cap & PCIM_ATS_EN) ? "enabled" : "Disabled",
162 		       qlen);
163 		qlen = (cap & PCIM_ATS_EN) ? qlen : -1;
164 	}
165 
166 	return (qlen);
167 }
168 
169 /*
170  * Check if an endpoint device support device IOTLB or ATS.
171  */
172 static inline bool
173 amdvi_dev_support_iotlb(struct amdvi_softc *softc, uint16_t devid)
174 {
175 	struct ivhd_dev_cfg *cfg;
176 	int qlen, i;
177 	bool pci_ats, ivhd_ats;
178 
179 	qlen = amdvi_find_ats_qlen(devid);
180 	if (qlen < 0)
181 		return (false);
182 
183 	KASSERT(softc, ("softc is NULL"));
184 	cfg = softc->dev_cfg;
185 
186 	ivhd_ats = false;
187 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
188 		if ((cfg->start_id <= devid) && (cfg->end_id >= devid)) {
189 			ivhd_ats = cfg->enable_ats;
190 			break;
191 		}
192 		cfg++;
193 	}
194 
195 	pci_ats = (qlen < 0) ? false : true;
196 	if (pci_ats != ivhd_ats)
197 		device_printf(softc->dev,
198 		    "BIOS bug: mismatch in ATS setting for %d.%d.%d,"
199 		    "ATS inv qlen = %d\n", RID2PCI_STR(devid), qlen);
200 
201 	/* Ignore IVRS setting and respect PCI setting. */
202 	return (pci_ats);
203 }
204 #endif
205 
206 /* Enable IOTLB support for IOMMU if its supported. */
207 static inline void
208 amdvi_hw_enable_iotlb(struct amdvi_softc *softc)
209 {
210 #ifndef AMDVI_ATS_ENABLE
211 	softc->iotlb = false;
212 #else
213 	bool supported;
214 
215 	supported = (softc->ivhd_flag & IVHD_FLAG_IOTLB) ? true : false;
216 
217 	if (softc->pci_cap & AMDVI_PCI_CAP_IOTLB) {
218 		if (!supported)
219 			device_printf(softc->dev, "IOTLB disabled by BIOS.\n");
220 
221 		if (supported && !amdvi_enable_iotlb) {
222 			device_printf(softc->dev, "IOTLB disabled by user.\n");
223 			supported = false;
224 		}
225 	} else
226 		supported = false;
227 
228 	softc->iotlb = supported;
229 
230 #endif
231 }
232 
233 static int
234 amdvi_init_cmd(struct amdvi_softc *softc)
235 {
236 	struct amdvi_ctrl *ctrl = softc->ctrl;
237 
238 	ctrl->cmd.len = 8;	/* Use 256 command buffer entries. */
239 	softc->cmd_max = 1 << ctrl->cmd.len;
240 
241 	softc->cmd = malloc(sizeof(struct amdvi_cmd) *
242 	    softc->cmd_max, M_AMDVI, M_WAITOK | M_ZERO);
243 
244 	if ((uintptr_t)softc->cmd & PAGE_MASK)
245 		panic("AMDVi: Command buffer not aligned on page boundary.");
246 
247 	ctrl->cmd.base = vtophys(softc->cmd) / PAGE_SIZE;
248 	/*
249 	 * XXX: Reset the h/w pointers in case IOMMU is restarting,
250 	 * h/w doesn't clear these pointers based on empirical data.
251 	 */
252 	ctrl->cmd_tail = 0;
253 	ctrl->cmd_head = 0;
254 
255 	return (0);
256 }
257 
258 /*
259  * Note: Update tail pointer after we have written the command since tail
260  * pointer update cause h/w to execute new commands, see section 3.3
261  * of AMD IOMMU spec ver 2.0.
262  */
263 /* Get the command tail pointer w/o updating it. */
264 static struct amdvi_cmd *
265 amdvi_get_cmd_tail(struct amdvi_softc *softc)
266 {
267 	struct amdvi_ctrl *ctrl;
268 	struct amdvi_cmd *tail;
269 
270 	KASSERT(softc, ("softc is NULL"));
271 	KASSERT(softc->cmd != NULL, ("cmd is NULL"));
272 
273 	ctrl = softc->ctrl;
274 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
275 
276 	tail = (struct amdvi_cmd *)((uint8_t *)softc->cmd +
277 	    ctrl->cmd_tail);
278 
279 	return (tail);
280 }
281 
282 /*
283  * Update the command tail pointer which will start command execution.
284  */
285 static void
286 amdvi_update_cmd_tail(struct amdvi_softc *softc)
287 {
288 	struct amdvi_ctrl *ctrl;
289 	int size;
290 
291 	size = sizeof(struct amdvi_cmd);
292 	KASSERT(softc->cmd != NULL, ("cmd is NULL"));
293 
294 	ctrl = softc->ctrl;
295 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
296 
297 	ctrl->cmd_tail = MOD_INC(ctrl->cmd_tail, size, softc->cmd_max);
298 	softc->total_cmd++;
299 
300 #ifdef AMDVI_DEBUG_CMD
301 	device_printf(softc->dev, "cmd_tail: %s Tail:0x%x, Head:0x%x.\n",
302 	    ctrl->cmd_tail,
303 	    ctrl->cmd_head);
304 #endif
305 
306 }
307 
308 /*
309  * Various commands supported by IOMMU.
310  */
311 
312 /* Completion wait command. */
313 static void
314 amdvi_cmd_cmp(struct amdvi_softc *softc, const uint64_t data)
315 {
316 	struct amdvi_cmd *cmd;
317 	uint64_t pa;
318 
319 	cmd = amdvi_get_cmd_tail(softc);
320 	KASSERT(cmd != NULL, ("Cmd is NULL"));
321 
322 	pa = vtophys(&softc->cmp_data);
323 	cmd->opcode = AMDVI_CMP_WAIT_OPCODE;
324 	cmd->word0 = (pa & 0xFFFFFFF8) | AMDVI_CMP_WAIT_STORE;
325 	cmd->word1 = (pa >> 32) & 0xFFFFF;
326 	cmd->addr = data;
327 
328 	amdvi_update_cmd_tail(softc);
329 }
330 
331 /* Invalidate device table entry. */
332 static void
333 amdvi_cmd_inv_dte(struct amdvi_softc *softc, uint16_t devid)
334 {
335 	struct amdvi_cmd *cmd;
336 
337 	cmd = amdvi_get_cmd_tail(softc);
338 	KASSERT(cmd != NULL, ("Cmd is NULL"));
339 	cmd->opcode = AMDVI_INVD_DTE_OPCODE;
340 	cmd->word0 = devid;
341 	amdvi_update_cmd_tail(softc);
342 #ifdef AMDVI_DEBUG_CMD
343 	device_printf(softc->dev, "Invalidated DTE:0x%x\n", devid);
344 #endif
345 }
346 
347 /* Invalidate IOMMU page, use for invalidation of domain. */
348 static void
349 amdvi_cmd_inv_iommu_pages(struct amdvi_softc *softc, uint16_t domain_id,
350 			  uint64_t addr, bool guest_nested,
351 			  bool pde, bool page)
352 {
353 	struct amdvi_cmd *cmd;
354 
355 	cmd = amdvi_get_cmd_tail(softc);
356 	KASSERT(cmd != NULL, ("Cmd is NULL"));
357 
358 	cmd->opcode = AMDVI_INVD_PAGE_OPCODE;
359 	cmd->word1 = domain_id;
360 	/*
361 	 * Invalidate all addresses for this domain.
362 	 */
363 	cmd->addr = addr;
364 	cmd->addr |= pde ? AMDVI_INVD_PAGE_PDE : 0;
365 	cmd->addr |= page ? AMDVI_INVD_PAGE_S : 0;
366 
367 	amdvi_update_cmd_tail(softc);
368 }
369 
370 #ifdef AMDVI_ATS_ENABLE
371 /* Invalidate device IOTLB. */
372 static void
373 amdvi_cmd_inv_iotlb(struct amdvi_softc *softc, uint16_t devid)
374 {
375 	struct amdvi_cmd *cmd;
376 	int qlen;
377 
378 	if (!softc->iotlb)
379 		return;
380 
381 	qlen = amdvi_find_ats_qlen(devid);
382 	if (qlen < 0) {
383 		panic("AMDVI: Invalid ATS qlen(%d) for device %d.%d.%d\n",
384 		      qlen, RID2PCI_STR(devid));
385 	}
386 	cmd = amdvi_get_cmd_tail(softc);
387 	KASSERT(cmd != NULL, ("Cmd is NULL"));
388 
389 #ifdef AMDVI_DEBUG_CMD
390 	device_printf(softc->dev, "Invalidate IOTLB devID 0x%x"
391 		      " Qlen:%d\n", devid, qlen);
392 #endif
393 	cmd->opcode = AMDVI_INVD_IOTLB_OPCODE;
394 	cmd->word0 = devid;
395 	cmd->word1 = qlen;
396 	cmd->addr = AMDVI_INVD_IOTLB_ALL_ADDR |
397 		AMDVI_INVD_IOTLB_S;
398 	amdvi_update_cmd_tail(softc);
399 }
400 #endif
401 
402 #ifdef notyet				/* For Interrupt Remap. */
403 static void
404 amdvi_cmd_inv_intr_map(struct amdvi_softc *softc,
405 		       uint16_t devid)
406 {
407 	struct amdvi_cmd *cmd;
408 
409 	cmd = amdvi_get_cmd_tail(softc);
410 	KASSERT(cmd != NULL, ("Cmd is NULL"));
411 	cmd->opcode = AMDVI_INVD_INTR_OPCODE;
412 	cmd->word0 = devid;
413 	amdvi_update_cmd_tail(softc);
414 #ifdef AMDVI_DEBUG_CMD
415 	device_printf(softc->dev, "Invalidate INTR map of devID 0x%x\n", devid);
416 #endif
417 }
418 #endif
419 
420 /* Invalidate domain using INVALIDATE_IOMMU_PAGES command. */
421 static void
422 amdvi_inv_domain(struct amdvi_softc *softc, uint16_t domain_id)
423 {
424 	struct amdvi_cmd *cmd __diagused;
425 
426 	cmd = amdvi_get_cmd_tail(softc);
427 	KASSERT(cmd != NULL, ("Cmd is NULL"));
428 
429 	/*
430 	 * See section 3.3.3 of IOMMU spec rev 2.0, software note
431 	 * for invalidating domain.
432 	 */
433 	amdvi_cmd_inv_iommu_pages(softc, domain_id, AMDVI_INVD_PAGE_ALL_ADDR,
434 				false, true, true);
435 
436 #ifdef AMDVI_DEBUG_CMD
437 	device_printf(softc->dev, "Invalidate domain:0x%x\n", domain_id);
438 
439 #endif
440 }
441 
442 static	bool
443 amdvi_cmp_wait(struct amdvi_softc *softc)
444 {
445 #ifdef AMDVI_DEBUG_CMD
446 	struct amdvi_ctrl *ctrl = softc->ctrl;
447 #endif
448 	const uint64_t VERIFY = 0xA5A5;
449 	volatile uint64_t *read;
450 	int i;
451 	bool status;
452 
453 	read = &softc->cmp_data;
454 	*read = 0;
455 	amdvi_cmd_cmp(softc, VERIFY);
456 	/* Wait for h/w to update completion data. */
457 	for (i = 0; i < 100 && (*read != VERIFY); i++) {
458 		DELAY(1000);		/* 1 ms */
459 	}
460 	status = (VERIFY == softc->cmp_data) ? true : false;
461 
462 #ifdef AMDVI_DEBUG_CMD
463 	if (status)
464 		device_printf(softc->dev, "CMD completion DONE Tail:0x%x, "
465 			      "Head:0x%x, loop:%d.\n", ctrl->cmd_tail,
466 			      ctrl->cmd_head, loop);
467 #endif
468 	return (status);
469 }
470 
471 static void
472 amdvi_wait(struct amdvi_softc *softc)
473 {
474 	struct amdvi_ctrl *ctrl;
475 	int i;
476 
477 	KASSERT(softc, ("softc is NULL"));
478 
479 	ctrl = softc->ctrl;
480 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
481 	/* Don't wait if h/w is not enabled. */
482 	if ((ctrl->control & AMDVI_CTRL_EN) == 0)
483 		return;
484 
485 	for (i = 0; i < 10; i++) {
486 		if (amdvi_cmp_wait(softc))
487 			return;
488 	}
489 
490 	device_printf(softc->dev, "Error: completion failed"
491 		      " tail:0x%x, head:0x%x.\n",
492 		      ctrl->cmd_tail, ctrl->cmd_head);
493 	/* Dump the last command. */
494 	amdvi_dump_cmds(softc, 1);
495 }
496 
497 static void
498 amdvi_dump_cmds(struct amdvi_softc *softc, int count)
499 {
500 	struct amdvi_ctrl *ctrl;
501 	struct amdvi_cmd *cmd;
502 	int off, i;
503 
504 	ctrl = softc->ctrl;
505 	device_printf(softc->dev, "Dump last %d command(s):\n", count);
506 	/*
507 	 * If h/w is stuck in completion, it is the previous command,
508 	 * start dumping from previous command onward.
509 	 */
510 	off = MOD_DEC(ctrl->cmd_head, sizeof(struct amdvi_cmd),
511 	    softc->cmd_max);
512 	for (i = 0; off != ctrl->cmd_tail && i < count; i++) {
513 		cmd = (struct amdvi_cmd *)((uint8_t *)softc->cmd + off);
514 		printf("  [CMD%d, off:0x%x] opcode= 0x%x 0x%x"
515 		    " 0x%x 0x%lx\n", i, off, cmd->opcode,
516 		    cmd->word0, cmd->word1, cmd->addr);
517 		off = MOD_INC(off, sizeof(struct amdvi_cmd), softc->cmd_max);
518 	}
519 }
520 
521 static int
522 amdvi_init_event(struct amdvi_softc *softc)
523 {
524 	struct amdvi_ctrl *ctrl;
525 
526 	ctrl = softc->ctrl;
527 	ctrl->event.len = 8;
528 	softc->event_max = 1 << ctrl->event.len;
529 	softc->event = malloc(sizeof(struct amdvi_event) *
530 	    softc->event_max, M_AMDVI, M_WAITOK | M_ZERO);
531 	if ((uintptr_t)softc->event & PAGE_MASK) {
532 		device_printf(softc->dev, "Event buffer not aligned on page.");
533 		return (false);
534 	}
535 	ctrl->event.base = vtophys(softc->event) / PAGE_SIZE;
536 
537 	/* Reset the pointers. */
538 	ctrl->evt_head = 0;
539 	ctrl->evt_tail = 0;
540 
541 	return (0);
542 }
543 
544 static inline void
545 amdvi_decode_evt_flag(uint16_t flag)
546 {
547 
548 	flag &= AMDVI_EVENT_FLAG_MASK;
549 	printf(" 0x%b]\n", flag,
550 		"\020"
551 		"\001GN"
552 		"\002NX"
553 		"\003US"
554 		"\004I"
555 		"\005PR"
556 		"\006RW"
557 		"\007PE"
558 		"\010RZ"
559 		"\011TR"
560 		);
561 }
562 
563 /* See section 2.5.4 of AMD IOMMU spec ver 2.62.*/
564 static inline void
565 amdvi_decode_evt_flag_type(uint8_t type)
566 {
567 
568 	switch (AMDVI_EVENT_FLAG_TYPE(type)) {
569 	case 0:
570 		printf("RSVD\n");
571 		break;
572 	case 1:
573 		printf("Master Abort\n");
574 		break;
575 	case 2:
576 		printf("Target Abort\n");
577 		break;
578 	case 3:
579 		printf("Data Err\n");
580 		break;
581 	default:
582 		break;
583 	}
584 }
585 
586 static void
587 amdvi_decode_inv_dte_evt(uint16_t devid, uint16_t domid, uint64_t addr,
588     uint16_t flag)
589 {
590 
591 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
592 	    " Addr:0x%lx",
593 	    devid, domid, addr);
594 	amdvi_decode_evt_flag(flag);
595 }
596 
597 static void
598 amdvi_decode_pf_evt(uint16_t devid, uint16_t domid, uint64_t addr,
599     uint16_t flag)
600 {
601 
602 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
603 	    " Addr:0x%lx",
604 	    devid, domid, addr);
605 	amdvi_decode_evt_flag(flag);
606 }
607 
608 static void
609 amdvi_decode_dte_hwerr_evt(uint16_t devid, uint16_t domid,
610     uint64_t addr, uint16_t flag)
611 {
612 
613 	printf("\t[DEV_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
614 	    " Addr:0x%lx", devid, domid, addr);
615 	amdvi_decode_evt_flag(flag);
616 	amdvi_decode_evt_flag_type(flag);
617 }
618 
619 static void
620 amdvi_decode_page_hwerr_evt(uint16_t devid, uint16_t domid, uint64_t addr,
621     uint16_t flag)
622 {
623 
624 	printf("\t[PAGE_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
625 	    " Addr:0x%lx", devid, domid, addr);
626 	amdvi_decode_evt_flag(flag);
627 	amdvi_decode_evt_flag_type(AMDVI_EVENT_FLAG_TYPE(flag));
628 }
629 
630 static void
631 amdvi_decode_evt(struct amdvi_event *evt)
632 {
633 	struct amdvi_cmd *cmd;
634 
635 	switch (evt->opcode) {
636 	case AMDVI_EVENT_INVALID_DTE:
637 		amdvi_decode_inv_dte_evt(evt->devid, evt->pasid_domid,
638 		    evt->addr, evt->flag);
639 		break;
640 
641 	case AMDVI_EVENT_PFAULT:
642 		amdvi_decode_pf_evt(evt->devid, evt->pasid_domid,
643 		    evt->addr, evt->flag);
644 		break;
645 
646 	case AMDVI_EVENT_DTE_HW_ERROR:
647 		amdvi_decode_dte_hwerr_evt(evt->devid, evt->pasid_domid,
648 		    evt->addr, evt->flag);
649 		break;
650 
651 	case AMDVI_EVENT_PAGE_HW_ERROR:
652 		amdvi_decode_page_hwerr_evt(evt->devid, evt->pasid_domid,
653 		    evt->addr, evt->flag);
654 		break;
655 
656 	case AMDVI_EVENT_ILLEGAL_CMD:
657 		/* FALL THROUGH */
658 	case AMDVI_EVENT_CMD_HW_ERROR:
659 		printf("\t[%s EVT]\n", (evt->opcode == AMDVI_EVENT_ILLEGAL_CMD) ?
660 		    "ILLEGAL CMD" : "CMD HW ERR");
661 		cmd = (struct amdvi_cmd *)PHYS_TO_DMAP(evt->addr);
662 		printf("\tCMD opcode= 0x%x 0x%x 0x%x 0x%lx\n",
663 		    cmd->opcode, cmd->word0, cmd->word1, cmd->addr);
664 		break;
665 
666 	case AMDVI_EVENT_IOTLB_TIMEOUT:
667 		printf("\t[IOTLB_INV_TIMEOUT devid:0x%x addr:0x%lx]\n",
668 		    evt->devid, evt->addr);
669 		break;
670 
671 	case AMDVI_EVENT_INVALID_DTE_REQ:
672 		printf("\t[INV_DTE devid:0x%x addr:0x%lx type:0x%x tr:%d]\n",
673 		    evt->devid, evt->addr, evt->flag >> 9,
674 		    (evt->flag >> 8) & 1);
675 		break;
676 
677 	case AMDVI_EVENT_INVALID_PPR_REQ:
678 	case AMDVI_EVENT_COUNTER_ZERO:
679 		printf("AMD-Vi: v2 events.\n");
680 		break;
681 
682 	default:
683 		printf("Unsupported AMD-Vi event:%d\n", evt->opcode);
684 	}
685 }
686 
687 static void
688 amdvi_print_events(struct amdvi_softc *softc)
689 {
690 	struct amdvi_ctrl *ctrl;
691 	struct amdvi_event *event;
692 	int i, size;
693 
694 	ctrl = softc->ctrl;
695 	size = sizeof(struct amdvi_event);
696 	for (i = 0; i < softc->event_max; i++) {
697 		event = &softc->event[ctrl->evt_head / size];
698 		if (!event->opcode)
699 			break;
700 		device_printf(softc->dev, "\t[Event%d: Head:0x%x Tail:0x%x]\n",
701 		    i, ctrl->evt_head, ctrl->evt_tail);
702 		amdvi_decode_evt(event);
703 		ctrl->evt_head = MOD_INC(ctrl->evt_head, size,
704 		    softc->event_max);
705 	}
706 }
707 
708 static int
709 amdvi_init_dte(struct amdvi_softc *softc)
710 {
711 	struct amdvi_ctrl *ctrl;
712 
713 	ctrl = softc->ctrl;
714 	ctrl->dte.base = vtophys(amdvi_dte) / PAGE_SIZE;
715 	ctrl->dte.size = 0x1FF;		/* 2MB device table. */
716 
717 	return (0);
718 }
719 
720 /*
721  * Not all capabilities of IOMMU are available in ACPI IVHD flag
722  * or EFR entry, read directly from device.
723  */
724 static int
725 amdvi_print_pci_cap(device_t dev)
726 {
727 	struct amdvi_softc *softc;
728 	uint32_t off, cap;
729 
730 	softc = device_get_softc(dev);
731 	off = softc->cap_off;
732 
733 	/*
734 	 * Section 3.7.1 of IOMMU sepc rev 2.0.
735 	 * Read capability from device.
736 	 */
737 	cap = amdvi_pci_read(softc, off);
738 
739 	/* Make sure capability type[18:16] is 3. */
740 	KASSERT((((cap >> 16) & 0x7) == 0x3),
741 	    ("Not a IOMMU capability 0x%x@0x%x", cap, off));
742 
743 	softc->pci_cap = cap >> 24;
744 	device_printf(softc->dev, "PCI cap 0x%x@0x%x feature:%b\n",
745 	    cap, off, softc->pci_cap,
746 	    "\20\1IOTLB\2HT\3NPCache\4EFR\5CapExt");
747 
748 	return (0);
749 }
750 
751 static void
752 amdvi_event_intr(void *arg)
753 {
754 	struct amdvi_softc *softc;
755 	struct amdvi_ctrl *ctrl;
756 
757 	softc = (struct amdvi_softc *)arg;
758 	ctrl = softc->ctrl;
759 	device_printf(softc->dev, "EVT INTR %ld Status:0x%x"
760 	    " EVT Head:0x%x Tail:0x%x]\n", softc->event_intr_cnt++,
761 	    ctrl->status, ctrl->evt_head, ctrl->evt_tail);
762 	printf("  [CMD Total 0x%lx] Tail:0x%x, Head:0x%x.\n",
763 	    softc->total_cmd, ctrl->cmd_tail, ctrl->cmd_head);
764 
765 	amdvi_print_events(softc);
766 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
767 }
768 
769 static void
770 amdvi_free_evt_intr_res(device_t dev)
771 {
772 
773 	struct amdvi_softc *softc;
774 	device_t mmio_dev;
775 
776 	softc = device_get_softc(dev);
777 	mmio_dev = softc->pci_dev;
778 
779 	IVHD_TEARDOWN_INTR(mmio_dev);
780 }
781 
782 static bool
783 amdvi_alloc_intr_resources(struct amdvi_softc *softc)
784 {
785 	struct amdvi_ctrl *ctrl;
786 	device_t dev, mmio_dev;
787 	int err;
788 
789 	dev = softc->dev;
790 	mmio_dev = softc->pci_dev;
791 
792 	/* Clear interrupt status bits. */
793 	ctrl = softc->ctrl;
794 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
795 
796 	err = IVHD_SETUP_INTR(mmio_dev, amdvi_event_intr, softc, "fault");
797 	if (err)
798 		device_printf(dev, "Interrupt setup failed on %s\n",
799 		    device_get_nameunit(mmio_dev));
800 	return (err);
801 }
802 
803 static void
804 amdvi_print_dev_cap(struct amdvi_softc *softc)
805 {
806 	struct ivhd_dev_cfg *cfg;
807 	int i;
808 
809 	cfg = softc->dev_cfg;
810 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
811 		device_printf(softc->dev, "device [0x%x - 0x%x] "
812 		    "config:%b%s\n", cfg->start_id, cfg->end_id,
813 		    cfg->data,
814 		    "\020\001INIT\002ExtInt\003NMI"
815 		    "\007LINT0\010LINT1",
816 		    cfg->enable_ats ? "ATS enabled" : "");
817 		cfg++;
818 	}
819 }
820 
821 static int
822 amdvi_handle_sysctl(SYSCTL_HANDLER_ARGS)
823 {
824 	struct amdvi_softc *softc;
825 	int result, type, error = 0;
826 
827 	softc = (struct amdvi_softc *)arg1;
828 	type = arg2;
829 
830 	switch (type) {
831 	case 0:
832 		result = softc->ctrl->cmd_head;
833 		error = sysctl_handle_int(oidp, &result, 0,
834 		    req);
835 		break;
836 	case 1:
837 		result = softc->ctrl->cmd_tail;
838 		error = sysctl_handle_int(oidp, &result, 0,
839 		    req);
840 		break;
841 	case 2:
842 		result = softc->ctrl->evt_head;
843 		error = sysctl_handle_int(oidp, &result, 0,
844 		    req);
845 		break;
846 	case 3:
847 		result = softc->ctrl->evt_tail;
848 		error = sysctl_handle_int(oidp, &result, 0,
849 		    req);
850 		break;
851 
852 	default:
853 		device_printf(softc->dev, "Unknown sysctl:%d\n", type);
854 	}
855 
856 	return (error);
857 }
858 
859 static void
860 amdvi_add_sysctl(struct amdvi_softc *softc)
861 {
862 	struct sysctl_oid_list *child;
863 	struct sysctl_ctx_list *ctx;
864 	device_t dev;
865 
866 	dev = softc->dev;
867 	ctx = device_get_sysctl_ctx(dev);
868 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
869 
870 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "event_intr_count", CTLFLAG_RD,
871 	    &softc->event_intr_cnt, "Event interrupt count");
872 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "command_count", CTLFLAG_RD,
873 	    &softc->total_cmd, "Command submitted count");
874 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "pci_rid", CTLFLAG_RD,
875 	    &softc->pci_rid, 0, "IOMMU RID");
876 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_head",
877 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 0,
878 	    amdvi_handle_sysctl, "IU", "Command head");
879 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_tail",
880 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 1,
881 	    amdvi_handle_sysctl, "IU", "Command tail");
882 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_head",
883 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 2,
884 	    amdvi_handle_sysctl, "IU", "Command head");
885 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_tail",
886 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 3,
887 	    amdvi_handle_sysctl, "IU", "Command tail");
888 }
889 
890 int
891 amdvi_setup_hw(struct amdvi_softc *softc)
892 {
893 	device_t dev;
894 	int status;
895 
896 	dev = softc->dev;
897 
898 	amdvi_hw_enable_iotlb(softc);
899 
900 	amdvi_print_dev_cap(softc);
901 
902 	if ((status = amdvi_print_pci_cap(dev)) != 0) {
903 		device_printf(dev, "PCI capability.\n");
904 		return (status);
905 	}
906 	if ((status = amdvi_init_cmd(softc)) != 0) {
907 		device_printf(dev, "Couldn't configure command buffer.\n");
908 		return (status);
909 	}
910 	if ((status = amdvi_init_event(softc)) != 0) {
911 		device_printf(dev, "Couldn't configure event buffer.\n");
912 		return (status);
913 	}
914 	if ((status = amdvi_init_dte(softc)) != 0) {
915 		device_printf(dev, "Couldn't configure device table.\n");
916 		return (status);
917 	}
918 	if ((status = amdvi_alloc_intr_resources(softc)) != 0) {
919 		return (status);
920 	}
921 	amdvi_add_sysctl(softc);
922 	return (0);
923 }
924 
925 int
926 amdvi_teardown_hw(struct amdvi_softc *softc)
927 {
928 	device_t dev;
929 
930 	dev = softc->dev;
931 
932 	/*
933 	 * Called after disable, h/w is stopped by now, free all the resources.
934 	 */
935 	amdvi_free_evt_intr_res(dev);
936 
937 	if (softc->cmd)
938 		free(softc->cmd, M_AMDVI);
939 
940 	if (softc->event)
941 		free(softc->event, M_AMDVI);
942 
943 	return (0);
944 }
945 
946 /*********** bhyve interfaces *********************/
947 static int
948 amdvi_init(void)
949 {
950 	if (!ivhd_count) {
951 		return (EIO);
952 	}
953 	if (!amdvi_enable_user && ivhd_count) {
954 		printf("bhyve: Found %d AMD-Vi/IOMMU device(s), "
955 		    	"use hw.vmm.amdvi.enable=1 to enable pass-through.\n",
956 		    ivhd_count);
957 		return (EINVAL);
958 	}
959 	return (0);
960 }
961 
962 static void
963 amdvi_cleanup(void)
964 {
965 	/* Nothing. */
966 }
967 
968 static uint16_t
969 amdvi_domainId(void)
970 {
971 
972 	/*
973 	 * If we hit maximum domain limit, rollover leaving host
974 	 * domain(0).
975 	 * XXX: make sure that this domain is not used.
976 	 */
977 	if (amdvi_dom_id == AMDVI_MAX_DOMAIN)
978 		amdvi_dom_id = 1;
979 
980 	return ((uint16_t)amdvi_dom_id++);
981 }
982 
983 static void
984 amdvi_do_inv_domain(uint16_t domain_id, bool create)
985 {
986 	struct amdvi_softc *softc;
987 	int i;
988 
989 	for (i = 0; i < ivhd_count; i++) {
990 		softc = device_get_softc(ivhd_devs[i]);
991 		KASSERT(softc, ("softc is NULL"));
992 		/*
993 		 * If not present pages are cached, invalidate page after
994 		 * creating domain.
995 		 */
996 #if 0
997 		if (create && ((softc->pci_cap & AMDVI_PCI_CAP_NPCACHE) == 0))
998 			continue;
999 #endif
1000 		amdvi_inv_domain(softc, domain_id);
1001 		amdvi_wait(softc);
1002 	}
1003 }
1004 
1005 static void *
1006 amdvi_create_domain(vm_paddr_t maxaddr)
1007 {
1008 	struct amdvi_domain *dom;
1009 
1010 	dom = malloc(sizeof(struct amdvi_domain), M_AMDVI, M_ZERO | M_WAITOK);
1011 	dom->id = amdvi_domainId();
1012 	//dom->maxaddr = maxaddr;
1013 #ifdef AMDVI_DEBUG_CMD
1014 	printf("Created domain #%d\n", dom->id);
1015 #endif
1016 	/*
1017 	 * Host domain(#0) don't create translation table.
1018 	 */
1019 	if (dom->id || amdvi_host_ptp)
1020 		dom->ptp = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1021 
1022 	dom->ptp_level = amdvi_ptp_level;
1023 
1024 	amdvi_do_inv_domain(dom->id, true);
1025 	SLIST_INSERT_HEAD(&dom_head, dom, next);
1026 
1027 	return (dom);
1028 }
1029 
1030 static void
1031 amdvi_free_ptp(uint64_t *ptp, int level)
1032 {
1033 	int i;
1034 
1035 	if (level < 1)
1036 		return;
1037 
1038 	for (i = 0; i < NPTEPG ; i++) {
1039 		if ((ptp[i] & AMDVI_PT_PRESENT) == 0)
1040 			continue;
1041 		/* XXX: Add super-page or PTE mapping > 4KB. */
1042 #ifdef notyet
1043 		/* Super-page mapping. */
1044 		if (AMDVI_PD_SUPER(ptp[i]))
1045 			continue;
1046 #endif
1047 
1048 		amdvi_free_ptp((uint64_t *)PHYS_TO_DMAP(ptp[i]
1049 		    & AMDVI_PT_MASK), level - 1);
1050 	}
1051 
1052 	free(ptp, M_AMDVI);
1053 }
1054 
1055 static void
1056 amdvi_destroy_domain(void *arg)
1057 {
1058 	struct amdvi_domain *domain;
1059 
1060 	domain = (struct amdvi_domain *)arg;
1061 	KASSERT(domain, ("domain is NULL"));
1062 #ifdef AMDVI_DEBUG_CMD
1063 	printf("Destroying domain %d\n", domain->id);
1064 #endif
1065 	if (domain->ptp)
1066 		amdvi_free_ptp(domain->ptp, domain->ptp_level);
1067 
1068 	amdvi_do_inv_domain(domain->id, false);
1069 	SLIST_REMOVE(&dom_head, domain, amdvi_domain, next);
1070 	free(domain, M_AMDVI);
1071 }
1072 
1073 static uint64_t
1074 amdvi_set_pt(uint64_t *pt, int level, vm_paddr_t gpa,
1075     vm_paddr_t hpa, uint64_t pg_size, bool create)
1076 {
1077 	uint64_t *page, pa;
1078 	int shift, index;
1079 	const int PT_SHIFT = 9;
1080 	const int PT_INDEX_MASK = (1 << PT_SHIFT) - 1;	/* Based on PT_SHIFT */
1081 
1082 	if (!pg_size)
1083 		return (0);
1084 
1085 	if (hpa & (pg_size - 1)) {
1086 		printf("HPA is not size aligned.\n");
1087 		return (0);
1088 	}
1089 	if (gpa & (pg_size - 1)) {
1090 		printf("HPA is not size aligned.\n");
1091 		return (0);
1092 	}
1093 	shift = PML4SHIFT;
1094 	while ((shift > PAGE_SHIFT) && (pg_size < (1UL << shift))) {
1095 		index = (gpa >> shift) & PT_INDEX_MASK;
1096 
1097 		if ((pt[index] == 0) && create) {
1098 			page = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1099 			pa = vtophys(page);
1100 			pt[index] = pa | AMDVI_PT_PRESENT | AMDVI_PT_RW |
1101 			    ((level - 1) << AMDVI_PD_LEVEL_SHIFT);
1102 		}
1103 #ifdef AMDVI_DEBUG_PTE
1104 		if ((gpa % 0x1000000) == 0)
1105 			printf("[level%d, shift = %d]PTE:0x%lx\n",
1106 			    level, shift, pt[index]);
1107 #endif
1108 #define PTE2PA(x)	((uint64_t)(x) & AMDVI_PT_MASK)
1109 		pa = PTE2PA(pt[index]);
1110 		pt = (uint64_t *)PHYS_TO_DMAP(pa);
1111 		shift -= PT_SHIFT;
1112 		level--;
1113 	}
1114 
1115 	/* Leaf entry. */
1116 	index = (gpa >> shift) & PT_INDEX_MASK;
1117 
1118 	if (create) {
1119 		pt[index] = hpa | AMDVI_PT_RW | AMDVI_PT_PRESENT;
1120 	} else
1121 		pt[index] = 0;
1122 
1123 #ifdef AMDVI_DEBUG_PTE
1124 	if ((gpa % 0x1000000) == 0)
1125 		printf("[Last level%d, shift = %d]PTE:0x%lx\n",
1126 		    level, shift, pt[index]);
1127 #endif
1128 	return (1ULL << shift);
1129 }
1130 
1131 static uint64_t
1132 amdvi_update_mapping(struct amdvi_domain *domain, vm_paddr_t gpa,
1133     vm_paddr_t hpa, uint64_t size, bool create)
1134 {
1135 	uint64_t mapped, *ptp, len;
1136 	int level;
1137 
1138 	KASSERT(domain, ("domain is NULL"));
1139 	level = domain->ptp_level;
1140 	KASSERT(level, ("Page table level is 0"));
1141 
1142 	ptp = domain->ptp;
1143 	KASSERT(ptp, ("PTP is NULL"));
1144 	mapped = 0;
1145 	while (mapped < size) {
1146 		len = amdvi_set_pt(ptp, level, gpa + mapped, hpa + mapped,
1147 		    PAGE_SIZE, create);
1148 		if (!len) {
1149 			printf("Error: Couldn't map HPA:0x%lx GPA:0x%lx\n",
1150 			    hpa, gpa);
1151 			return (0);
1152 		}
1153 		mapped += len;
1154 	}
1155 
1156 	return (mapped);
1157 }
1158 
1159 static uint64_t
1160 amdvi_create_mapping(void *arg, vm_paddr_t gpa, vm_paddr_t hpa,
1161     uint64_t len)
1162 {
1163 	struct amdvi_domain *domain;
1164 
1165 	domain = (struct amdvi_domain *)arg;
1166 
1167 	if (domain->id && !domain->ptp) {
1168 		printf("ptp is NULL");
1169 		return (-1);
1170 	}
1171 
1172 	/*
1173 	 * If host domain is created w/o page table, skip IOMMU page
1174 	 * table set-up.
1175 	 */
1176 	if (domain->ptp)
1177 		return (amdvi_update_mapping(domain, gpa, hpa, len, true));
1178 	else
1179 		return (len);
1180 }
1181 
1182 static uint64_t
1183 amdvi_remove_mapping(void *arg, vm_paddr_t gpa, uint64_t len)
1184 {
1185 	struct amdvi_domain *domain;
1186 
1187 	domain = (struct amdvi_domain *)arg;
1188 	/*
1189 	 * If host domain is created w/o page table, skip IOMMU page
1190 	 * table set-up.
1191 	 */
1192 	if (domain->ptp)
1193 		return (amdvi_update_mapping(domain, gpa, 0, len, false));
1194 	return
1195 	    (len);
1196 }
1197 
1198 static struct amdvi_softc *
1199 amdvi_find_iommu(uint16_t devid)
1200 {
1201 	struct amdvi_softc *softc;
1202 	int i, j;
1203 
1204 	for (i = 0; i < ivhd_count; i++) {
1205 		softc = device_get_softc(ivhd_devs[i]);
1206 		for (j = 0; j < softc->dev_cfg_cnt; j++)
1207 			if ((devid >= softc->dev_cfg[j].start_id) &&
1208 			    (devid <= softc->dev_cfg[j].end_id))
1209 				return (softc);
1210 	}
1211 
1212 	return (NULL);
1213 }
1214 
1215 /*
1216  * Set-up device table entry.
1217  * IOMMU spec Rev 2.0, section 3.2.2.2, some of the fields must
1218  * be set concurrently, e.g. read and write bits.
1219  */
1220 static void
1221 amdvi_set_dte(struct amdvi_domain *domain, struct amdvi_softc *softc,
1222     uint16_t devid, bool enable)
1223 {
1224 	struct amdvi_dte* temp;
1225 
1226 	KASSERT(domain, ("domain is NULL for pci_rid:0x%x\n", devid));
1227 	KASSERT(softc, ("softc is NULL for pci_rid:0x%x\n", devid));
1228 
1229 	temp = &amdvi_dte[devid];
1230 
1231 #ifdef AMDVI_ATS_ENABLE
1232 	/* If IOMMU and device support IOTLB, enable it. */
1233 	if (amdvi_dev_support_iotlb(softc, devid) && softc->iotlb)
1234 		temp->iotlb_enable = 1;
1235 #endif
1236 
1237 	/* Avoid duplicate I/O faults. */
1238 	temp->sup_second_io_fault = 1;
1239 	temp->sup_all_io_fault = amdvi_disable_io_fault;
1240 
1241 	temp->dt_valid = 1;
1242 	temp->domain_id = domain->id;
1243 
1244 	if (enable) {
1245 		if (domain->ptp) {
1246 			temp->pt_base = vtophys(domain->ptp) >> 12;
1247 			temp->pt_level = amdvi_ptp_level;
1248 		}
1249 		/*
1250 		 * XXX: Page table valid[TV] bit must be set even if host domain
1251 		 * page tables are not enabled.
1252 		 */
1253 		temp->pt_valid = 1;
1254 		temp->read_allow = 1;
1255 		temp->write_allow = 1;
1256 	}
1257 }
1258 
1259 static void
1260 amdvi_inv_device(struct amdvi_softc *softc, uint16_t devid)
1261 {
1262 	KASSERT(softc, ("softc is NULL"));
1263 
1264 	amdvi_cmd_inv_dte(softc, devid);
1265 #ifdef AMDVI_ATS_ENABLE
1266 	if (amdvi_dev_support_iotlb(softc, devid))
1267 		amdvi_cmd_inv_iotlb(softc, devid);
1268 #endif
1269 	amdvi_wait(softc);
1270 }
1271 
1272 static void
1273 amdvi_add_device(void *arg, uint16_t devid)
1274 {
1275 	struct amdvi_domain *domain;
1276 	struct amdvi_softc *softc;
1277 
1278 	domain = (struct amdvi_domain *)arg;
1279 	KASSERT(domain != NULL, ("domain is NULL"));
1280 #ifdef AMDVI_DEBUG_CMD
1281 	printf("Assigning device(%d.%d.%d) to domain:%d\n",
1282 	    RID2PCI_STR(devid), domain->id);
1283 #endif
1284 	softc = amdvi_find_iommu(devid);
1285 	if (softc == NULL)
1286 		return;
1287 	amdvi_set_dte(domain, softc, devid, true);
1288 	amdvi_inv_device(softc, devid);
1289 }
1290 
1291 static void
1292 amdvi_remove_device(void *arg, uint16_t devid)
1293 {
1294 	struct amdvi_domain *domain;
1295 	struct amdvi_softc *softc;
1296 
1297 	domain = (struct amdvi_domain *)arg;
1298 #ifdef AMDVI_DEBUG_CMD
1299 	printf("Remove device(0x%x) from domain:%d\n",
1300 	       devid, domain->id);
1301 #endif
1302 	softc = amdvi_find_iommu(devid);
1303 	if (softc == NULL)
1304 		return;
1305 	amdvi_set_dte(domain, softc, devid, false);
1306 	amdvi_inv_device(softc, devid);
1307 }
1308 
1309 static void
1310 amdvi_enable(void)
1311 {
1312 	struct amdvi_ctrl *ctrl;
1313 	struct amdvi_softc *softc;
1314 	uint64_t val;
1315 	int i;
1316 
1317 	for (i = 0; i < ivhd_count; i++) {
1318 		softc = device_get_softc(ivhd_devs[i]);
1319 		KASSERT(softc, ("softc is NULL\n"));
1320 		ctrl = softc->ctrl;
1321 		KASSERT(ctrl, ("ctrl is NULL\n"));
1322 
1323 		val = (	AMDVI_CTRL_EN 		|
1324 			AMDVI_CTRL_CMD 		|
1325 		    	AMDVI_CTRL_ELOG 	|
1326 		    	AMDVI_CTRL_ELOGINT 	|
1327 		    	AMDVI_CTRL_INV_TO_1S);
1328 
1329 		if (softc->ivhd_flag & IVHD_FLAG_COH)
1330 			val |= AMDVI_CTRL_COH;
1331 		if (softc->ivhd_flag & IVHD_FLAG_HTT)
1332 			val |= AMDVI_CTRL_HTT;
1333 		if (softc->ivhd_flag & IVHD_FLAG_RPPW)
1334 			val |= AMDVI_CTRL_RPPW;
1335 		if (softc->ivhd_flag & IVHD_FLAG_PPW)
1336 			val |= AMDVI_CTRL_PPW;
1337 		if (softc->ivhd_flag & IVHD_FLAG_ISOC)
1338 			val |= AMDVI_CTRL_ISOC;
1339 
1340 		ctrl->control = val;
1341 	}
1342 }
1343 
1344 static void
1345 amdvi_disable(void)
1346 {
1347 	struct amdvi_ctrl *ctrl;
1348 	struct amdvi_softc *softc;
1349 	int i;
1350 
1351 	for (i = 0; i < ivhd_count; i++) {
1352 		softc = device_get_softc(ivhd_devs[i]);
1353 		KASSERT(softc, ("softc is NULL\n"));
1354 		ctrl = softc->ctrl;
1355 		KASSERT(ctrl, ("ctrl is NULL\n"));
1356 
1357 		ctrl->control = 0;
1358 	}
1359 }
1360 
1361 static void
1362 amdvi_invalidate_tlb(void *arg)
1363 {
1364 	struct amdvi_domain *domain;
1365 
1366 	domain = (struct amdvi_domain *)arg;
1367 	KASSERT(domain, ("domain is NULL"));
1368 	amdvi_do_inv_domain(domain->id, false);
1369 }
1370 
1371 const struct iommu_ops iommu_ops_amd = {
1372 	.init = amdvi_init,
1373 	.cleanup = amdvi_cleanup,
1374 	.enable = amdvi_enable,
1375 	.disable = amdvi_disable,
1376 	.create_domain = amdvi_create_domain,
1377 	.destroy_domain = amdvi_destroy_domain,
1378 	.create_mapping = amdvi_create_mapping,
1379 	.remove_mapping = amdvi_remove_mapping,
1380 	.add_device = amdvi_add_device,
1381 	.remove_device = amdvi_remove_device,
1382 	.invalidate_tlb = amdvi_invalidate_tlb
1383 };
1384