xref: /freebsd/sys/amd64/vmm/amd/amdvi_hw.c (revision f126890a)
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/param.h>
30 #include <sys/systm.h>
31 #include <sys/bus.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
34 #include <sys/malloc.h>
35 #include <sys/pcpu.h>
36 #include <sys/rman.h>
37 #include <sys/smp.h>
38 #include <sys/sysctl.h>
39 
40 #include <vm/vm.h>
41 #include <vm/pmap.h>
42 
43 #include <dev/pci/pcivar.h>
44 #include <dev/pci/pcireg.h>
45 
46 #include <machine/resource.h>
47 #include <machine/vmm.h>
48 #include <machine/pmap.h>
49 #include <machine/vmparam.h>
50 #include <machine/pci_cfgreg.h>
51 
52 #include "ivhd_if.h"
53 #include "pcib_if.h"
54 
55 #include "io/iommu.h"
56 #include "amdvi_priv.h"
57 
58 SYSCTL_DECL(_hw_vmm);
59 SYSCTL_NODE(_hw_vmm, OID_AUTO, amdvi, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
60     NULL);
61 
62 #define MOD_INC(a, s, m) (((a) + (s)) % ((m) * (s)))
63 #define MOD_DEC(a, s, m) (((a) - (s)) % ((m) * (s)))
64 
65 /* Print RID or device ID in PCI string format. */
66 #define RID2PCI_STR(d) PCI_RID2BUS(d), PCI_RID2SLOT(d), PCI_RID2FUNC(d)
67 
68 static void amdvi_dump_cmds(struct amdvi_softc *softc, int count);
69 static void amdvi_print_dev_cap(struct amdvi_softc *softc);
70 
71 MALLOC_DEFINE(M_AMDVI, "amdvi", "amdvi");
72 
73 extern device_t *ivhd_devs;
74 
75 extern int ivhd_count;
76 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, count, CTLFLAG_RDTUN, &ivhd_count,
77     0, NULL);
78 
79 static int amdvi_enable_user = 0;
80 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, enable, CTLFLAG_RDTUN,
81     &amdvi_enable_user, 0, NULL);
82 TUNABLE_INT("hw.vmm.amdvi_enable", &amdvi_enable_user);
83 
84 #ifdef AMDVI_ATS_ENABLE
85 /* XXX: ATS is not tested. */
86 static int amdvi_enable_iotlb = 1;
87 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, iotlb_enabled, CTLFLAG_RDTUN,
88     &amdvi_enable_iotlb, 0, NULL);
89 TUNABLE_INT("hw.vmm.enable_iotlb", &amdvi_enable_iotlb);
90 #endif
91 
92 static int amdvi_host_ptp = 1;	/* Use page tables for host. */
93 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, host_ptp, CTLFLAG_RDTUN,
94     &amdvi_host_ptp, 0, NULL);
95 TUNABLE_INT("hw.vmm.amdvi.host_ptp", &amdvi_host_ptp);
96 
97 /* Page table level used <= supported by h/w[v1=7]. */
98 int amdvi_ptp_level = 4;
99 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, ptp_level, CTLFLAG_RDTUN,
100     &amdvi_ptp_level, 0, NULL);
101 TUNABLE_INT("hw.vmm.amdvi.ptp_level", &amdvi_ptp_level);
102 
103 /* Disable fault event reporting. */
104 static int amdvi_disable_io_fault = 0;
105 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, disable_io_fault, CTLFLAG_RDTUN,
106     &amdvi_disable_io_fault, 0, NULL);
107 TUNABLE_INT("hw.vmm.amdvi.disable_io_fault", &amdvi_disable_io_fault);
108 
109 static uint32_t amdvi_dom_id = 0;	/* 0 is reserved for host. */
110 SYSCTL_UINT(_hw_vmm_amdvi, OID_AUTO, domain_id, CTLFLAG_RD,
111     &amdvi_dom_id, 0, NULL);
112 /*
113  * Device table entry.
114  * Bus(256) x Dev(32) x Fun(8) x DTE(256 bits or 32 bytes).
115  *	= 256 * 2 * PAGE_SIZE.
116  */
117 static struct amdvi_dte amdvi_dte[PCI_NUM_DEV_MAX] __aligned(PAGE_SIZE);
118 CTASSERT(PCI_NUM_DEV_MAX == 0x10000);
119 CTASSERT(sizeof(amdvi_dte) == 0x200000);
120 
121 static SLIST_HEAD (, amdvi_domain) dom_head;
122 
123 static inline uint32_t
124 amdvi_pci_read(struct amdvi_softc *softc, int off)
125 {
126 
127 	return (pci_cfgregread(softc->pci_seg, PCI_RID2BUS(softc->pci_rid),
128 	    PCI_RID2SLOT(softc->pci_rid), PCI_RID2FUNC(softc->pci_rid),
129 	    off, 4));
130 }
131 
132 #ifdef AMDVI_ATS_ENABLE
133 /* XXX: Should be in pci.c */
134 /*
135  * Check if device has ATS capability and its enabled.
136  * If ATS is absent or disabled, return (-1), otherwise ATS
137  * queue length.
138  */
139 static int
140 amdvi_find_ats_qlen(uint16_t devid)
141 {
142 	device_t dev;
143 	uint32_t off, cap;
144 	int qlen = -1;
145 
146 	dev = pci_find_bsf(PCI_RID2BUS(devid), PCI_RID2SLOT(devid),
147 			   PCI_RID2FUNC(devid));
148 
149 	if (!dev) {
150 		return (-1);
151 	}
152 #define PCIM_ATS_EN	BIT(31)
153 
154 	if (pci_find_extcap(dev, PCIZ_ATS, &off) == 0) {
155 		cap = pci_read_config(dev, off + 4, 4);
156 		qlen = (cap & 0x1F);
157 		qlen = qlen ? qlen : 32;
158 		printf("AMD-Vi: PCI device %d.%d.%d ATS %s qlen=%d\n",
159 		       RID2PCI_STR(devid),
160 		       (cap & PCIM_ATS_EN) ? "enabled" : "Disabled",
161 		       qlen);
162 		qlen = (cap & PCIM_ATS_EN) ? qlen : -1;
163 	}
164 
165 	return (qlen);
166 }
167 
168 /*
169  * Check if an endpoint device support device IOTLB or ATS.
170  */
171 static inline bool
172 amdvi_dev_support_iotlb(struct amdvi_softc *softc, uint16_t devid)
173 {
174 	struct ivhd_dev_cfg *cfg;
175 	int qlen, i;
176 	bool pci_ats, ivhd_ats;
177 
178 	qlen = amdvi_find_ats_qlen(devid);
179 	if (qlen < 0)
180 		return (false);
181 
182 	KASSERT(softc, ("softc is NULL"));
183 	cfg = softc->dev_cfg;
184 
185 	ivhd_ats = false;
186 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
187 		if ((cfg->start_id <= devid) && (cfg->end_id >= devid)) {
188 			ivhd_ats = cfg->enable_ats;
189 			break;
190 		}
191 		cfg++;
192 	}
193 
194 	pci_ats = (qlen < 0) ? false : true;
195 	if (pci_ats != ivhd_ats)
196 		device_printf(softc->dev,
197 		    "BIOS bug: mismatch in ATS setting for %d.%d.%d,"
198 		    "ATS inv qlen = %d\n", RID2PCI_STR(devid), qlen);
199 
200 	/* Ignore IVRS setting and respect PCI setting. */
201 	return (pci_ats);
202 }
203 #endif
204 
205 /* Enable IOTLB support for IOMMU if its supported. */
206 static inline void
207 amdvi_hw_enable_iotlb(struct amdvi_softc *softc)
208 {
209 #ifndef AMDVI_ATS_ENABLE
210 	softc->iotlb = false;
211 #else
212 	bool supported;
213 
214 	supported = (softc->ivhd_flag & IVHD_FLAG_IOTLB) ? true : false;
215 
216 	if (softc->pci_cap & AMDVI_PCI_CAP_IOTLB) {
217 		if (!supported)
218 			device_printf(softc->dev, "IOTLB disabled by BIOS.\n");
219 
220 		if (supported && !amdvi_enable_iotlb) {
221 			device_printf(softc->dev, "IOTLB disabled by user.\n");
222 			supported = false;
223 		}
224 	} else
225 		supported = false;
226 
227 	softc->iotlb = supported;
228 
229 #endif
230 }
231 
232 static int
233 amdvi_init_cmd(struct amdvi_softc *softc)
234 {
235 	struct amdvi_ctrl *ctrl = softc->ctrl;
236 
237 	ctrl->cmd.len = 8;	/* Use 256 command buffer entries. */
238 	softc->cmd_max = 1 << ctrl->cmd.len;
239 
240 	softc->cmd = malloc(sizeof(struct amdvi_cmd) *
241 	    softc->cmd_max, M_AMDVI, M_WAITOK | M_ZERO);
242 
243 	if ((uintptr_t)softc->cmd & PAGE_MASK)
244 		panic("AMDVi: Command buffer not aligned on page boundary.");
245 
246 	ctrl->cmd.base = vtophys(softc->cmd) / PAGE_SIZE;
247 	/*
248 	 * XXX: Reset the h/w pointers in case IOMMU is restarting,
249 	 * h/w doesn't clear these pointers based on empirical data.
250 	 */
251 	ctrl->cmd_tail = 0;
252 	ctrl->cmd_head = 0;
253 
254 	return (0);
255 }
256 
257 /*
258  * Note: Update tail pointer after we have written the command since tail
259  * pointer update cause h/w to execute new commands, see section 3.3
260  * of AMD IOMMU spec ver 2.0.
261  */
262 /* Get the command tail pointer w/o updating it. */
263 static struct amdvi_cmd *
264 amdvi_get_cmd_tail(struct amdvi_softc *softc)
265 {
266 	struct amdvi_ctrl *ctrl;
267 	struct amdvi_cmd *tail;
268 
269 	KASSERT(softc, ("softc is NULL"));
270 	KASSERT(softc->cmd != NULL, ("cmd is NULL"));
271 
272 	ctrl = softc->ctrl;
273 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
274 
275 	tail = (struct amdvi_cmd *)((uint8_t *)softc->cmd +
276 	    ctrl->cmd_tail);
277 
278 	return (tail);
279 }
280 
281 /*
282  * Update the command tail pointer which will start command execution.
283  */
284 static void
285 amdvi_update_cmd_tail(struct amdvi_softc *softc)
286 {
287 	struct amdvi_ctrl *ctrl;
288 	int size;
289 
290 	size = sizeof(struct amdvi_cmd);
291 	KASSERT(softc->cmd != NULL, ("cmd is NULL"));
292 
293 	ctrl = softc->ctrl;
294 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
295 
296 	ctrl->cmd_tail = MOD_INC(ctrl->cmd_tail, size, softc->cmd_max);
297 	softc->total_cmd++;
298 
299 #ifdef AMDVI_DEBUG_CMD
300 	device_printf(softc->dev, "cmd_tail: %s Tail:0x%x, Head:0x%x.\n",
301 	    ctrl->cmd_tail,
302 	    ctrl->cmd_head);
303 #endif
304 
305 }
306 
307 /*
308  * Various commands supported by IOMMU.
309  */
310 
311 /* Completion wait command. */
312 static void
313 amdvi_cmd_cmp(struct amdvi_softc *softc, const uint64_t data)
314 {
315 	struct amdvi_cmd *cmd;
316 	uint64_t pa;
317 
318 	cmd = amdvi_get_cmd_tail(softc);
319 	KASSERT(cmd != NULL, ("Cmd is NULL"));
320 
321 	pa = vtophys(&softc->cmp_data);
322 	cmd->opcode = AMDVI_CMP_WAIT_OPCODE;
323 	cmd->word0 = (pa & 0xFFFFFFF8) | AMDVI_CMP_WAIT_STORE;
324 	cmd->word1 = (pa >> 32) & 0xFFFFF;
325 	cmd->addr = data;
326 
327 	amdvi_update_cmd_tail(softc);
328 }
329 
330 /* Invalidate device table entry. */
331 static void
332 amdvi_cmd_inv_dte(struct amdvi_softc *softc, uint16_t devid)
333 {
334 	struct amdvi_cmd *cmd;
335 
336 	cmd = amdvi_get_cmd_tail(softc);
337 	KASSERT(cmd != NULL, ("Cmd is NULL"));
338 	cmd->opcode = AMDVI_INVD_DTE_OPCODE;
339 	cmd->word0 = devid;
340 	amdvi_update_cmd_tail(softc);
341 #ifdef AMDVI_DEBUG_CMD
342 	device_printf(softc->dev, "Invalidated DTE:0x%x\n", devid);
343 #endif
344 }
345 
346 /* Invalidate IOMMU page, use for invalidation of domain. */
347 static void
348 amdvi_cmd_inv_iommu_pages(struct amdvi_softc *softc, uint16_t domain_id,
349 			  uint64_t addr, bool guest_nested,
350 			  bool pde, bool page)
351 {
352 	struct amdvi_cmd *cmd;
353 
354 	cmd = amdvi_get_cmd_tail(softc);
355 	KASSERT(cmd != NULL, ("Cmd is NULL"));
356 
357 	cmd->opcode = AMDVI_INVD_PAGE_OPCODE;
358 	cmd->word1 = domain_id;
359 	/*
360 	 * Invalidate all addresses for this domain.
361 	 */
362 	cmd->addr = addr;
363 	cmd->addr |= pde ? AMDVI_INVD_PAGE_PDE : 0;
364 	cmd->addr |= page ? AMDVI_INVD_PAGE_S : 0;
365 
366 	amdvi_update_cmd_tail(softc);
367 }
368 
369 #ifdef AMDVI_ATS_ENABLE
370 /* Invalidate device IOTLB. */
371 static void
372 amdvi_cmd_inv_iotlb(struct amdvi_softc *softc, uint16_t devid)
373 {
374 	struct amdvi_cmd *cmd;
375 	int qlen;
376 
377 	if (!softc->iotlb)
378 		return;
379 
380 	qlen = amdvi_find_ats_qlen(devid);
381 	if (qlen < 0) {
382 		panic("AMDVI: Invalid ATS qlen(%d) for device %d.%d.%d\n",
383 		      qlen, RID2PCI_STR(devid));
384 	}
385 	cmd = amdvi_get_cmd_tail(softc);
386 	KASSERT(cmd != NULL, ("Cmd is NULL"));
387 
388 #ifdef AMDVI_DEBUG_CMD
389 	device_printf(softc->dev, "Invalidate IOTLB devID 0x%x"
390 		      " Qlen:%d\n", devid, qlen);
391 #endif
392 	cmd->opcode = AMDVI_INVD_IOTLB_OPCODE;
393 	cmd->word0 = devid;
394 	cmd->word1 = qlen;
395 	cmd->addr = AMDVI_INVD_IOTLB_ALL_ADDR |
396 		AMDVI_INVD_IOTLB_S;
397 	amdvi_update_cmd_tail(softc);
398 }
399 #endif
400 
401 #ifdef notyet				/* For Interrupt Remap. */
402 static void
403 amdvi_cmd_inv_intr_map(struct amdvi_softc *softc,
404 		       uint16_t devid)
405 {
406 	struct amdvi_cmd *cmd;
407 
408 	cmd = amdvi_get_cmd_tail(softc);
409 	KASSERT(cmd != NULL, ("Cmd is NULL"));
410 	cmd->opcode = AMDVI_INVD_INTR_OPCODE;
411 	cmd->word0 = devid;
412 	amdvi_update_cmd_tail(softc);
413 #ifdef AMDVI_DEBUG_CMD
414 	device_printf(softc->dev, "Invalidate INTR map of devID 0x%x\n", devid);
415 #endif
416 }
417 #endif
418 
419 /* Invalidate domain using INVALIDATE_IOMMU_PAGES command. */
420 static void
421 amdvi_inv_domain(struct amdvi_softc *softc, uint16_t domain_id)
422 {
423 	struct amdvi_cmd *cmd __diagused;
424 
425 	cmd = amdvi_get_cmd_tail(softc);
426 	KASSERT(cmd != NULL, ("Cmd is NULL"));
427 
428 	/*
429 	 * See section 3.3.3 of IOMMU spec rev 2.0, software note
430 	 * for invalidating domain.
431 	 */
432 	amdvi_cmd_inv_iommu_pages(softc, domain_id, AMDVI_INVD_PAGE_ALL_ADDR,
433 				false, true, true);
434 
435 #ifdef AMDVI_DEBUG_CMD
436 	device_printf(softc->dev, "Invalidate domain:0x%x\n", domain_id);
437 
438 #endif
439 }
440 
441 static	bool
442 amdvi_cmp_wait(struct amdvi_softc *softc)
443 {
444 #ifdef AMDVI_DEBUG_CMD
445 	struct amdvi_ctrl *ctrl = softc->ctrl;
446 #endif
447 	const uint64_t VERIFY = 0xA5A5;
448 	volatile uint64_t *read;
449 	int i;
450 	bool status;
451 
452 	read = &softc->cmp_data;
453 	*read = 0;
454 	amdvi_cmd_cmp(softc, VERIFY);
455 	/* Wait for h/w to update completion data. */
456 	for (i = 0; i < 100 && (*read != VERIFY); i++) {
457 		DELAY(1000);		/* 1 ms */
458 	}
459 	status = (VERIFY == softc->cmp_data) ? true : false;
460 
461 #ifdef AMDVI_DEBUG_CMD
462 	if (status)
463 		device_printf(softc->dev, "CMD completion DONE Tail:0x%x, "
464 			      "Head:0x%x, loop:%d.\n", ctrl->cmd_tail,
465 			      ctrl->cmd_head, loop);
466 #endif
467 	return (status);
468 }
469 
470 static void
471 amdvi_wait(struct amdvi_softc *softc)
472 {
473 	struct amdvi_ctrl *ctrl;
474 	int i;
475 
476 	KASSERT(softc, ("softc is NULL"));
477 
478 	ctrl = softc->ctrl;
479 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
480 	/* Don't wait if h/w is not enabled. */
481 	if ((ctrl->control & AMDVI_CTRL_EN) == 0)
482 		return;
483 
484 	for (i = 0; i < 10; i++) {
485 		if (amdvi_cmp_wait(softc))
486 			return;
487 	}
488 
489 	device_printf(softc->dev, "Error: completion failed"
490 		      " tail:0x%x, head:0x%x.\n",
491 		      ctrl->cmd_tail, ctrl->cmd_head);
492 	/* Dump the last command. */
493 	amdvi_dump_cmds(softc, 1);
494 }
495 
496 static void
497 amdvi_dump_cmds(struct amdvi_softc *softc, int count)
498 {
499 	struct amdvi_ctrl *ctrl;
500 	struct amdvi_cmd *cmd;
501 	int off, i;
502 
503 	ctrl = softc->ctrl;
504 	device_printf(softc->dev, "Dump last %d command(s):\n", count);
505 	/*
506 	 * If h/w is stuck in completion, it is the previous command,
507 	 * start dumping from previous command onward.
508 	 */
509 	off = MOD_DEC(ctrl->cmd_head, sizeof(struct amdvi_cmd),
510 	    softc->cmd_max);
511 	for (i = 0; off != ctrl->cmd_tail && i < count; i++) {
512 		cmd = (struct amdvi_cmd *)((uint8_t *)softc->cmd + off);
513 		printf("  [CMD%d, off:0x%x] opcode= 0x%x 0x%x"
514 		    " 0x%x 0x%lx\n", i, off, cmd->opcode,
515 		    cmd->word0, cmd->word1, cmd->addr);
516 		off = MOD_INC(off, sizeof(struct amdvi_cmd), softc->cmd_max);
517 	}
518 }
519 
520 static int
521 amdvi_init_event(struct amdvi_softc *softc)
522 {
523 	struct amdvi_ctrl *ctrl;
524 
525 	ctrl = softc->ctrl;
526 	ctrl->event.len = 8;
527 	softc->event_max = 1 << ctrl->event.len;
528 	softc->event = malloc(sizeof(struct amdvi_event) *
529 	    softc->event_max, M_AMDVI, M_WAITOK | M_ZERO);
530 	if ((uintptr_t)softc->event & PAGE_MASK) {
531 		device_printf(softc->dev, "Event buffer not aligned on page.");
532 		return (false);
533 	}
534 	ctrl->event.base = vtophys(softc->event) / PAGE_SIZE;
535 
536 	/* Reset the pointers. */
537 	ctrl->evt_head = 0;
538 	ctrl->evt_tail = 0;
539 
540 	return (0);
541 }
542 
543 static inline void
544 amdvi_decode_evt_flag(uint16_t flag)
545 {
546 
547 	flag &= AMDVI_EVENT_FLAG_MASK;
548 	printf(" 0x%b]\n", flag,
549 		"\020"
550 		"\001GN"
551 		"\002NX"
552 		"\003US"
553 		"\004I"
554 		"\005PR"
555 		"\006RW"
556 		"\007PE"
557 		"\010RZ"
558 		"\011TR"
559 		);
560 }
561 
562 /* See section 2.5.4 of AMD IOMMU spec ver 2.62.*/
563 static inline void
564 amdvi_decode_evt_flag_type(uint8_t type)
565 {
566 
567 	switch (AMDVI_EVENT_FLAG_TYPE(type)) {
568 	case 0:
569 		printf("RSVD\n");
570 		break;
571 	case 1:
572 		printf("Master Abort\n");
573 		break;
574 	case 2:
575 		printf("Target Abort\n");
576 		break;
577 	case 3:
578 		printf("Data Err\n");
579 		break;
580 	default:
581 		break;
582 	}
583 }
584 
585 static void
586 amdvi_decode_inv_dte_evt(uint16_t devid, uint16_t domid, uint64_t addr,
587     uint16_t flag)
588 {
589 
590 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
591 	    " Addr:0x%lx",
592 	    devid, domid, addr);
593 	amdvi_decode_evt_flag(flag);
594 }
595 
596 static void
597 amdvi_decode_pf_evt(uint16_t devid, uint16_t domid, uint64_t addr,
598     uint16_t flag)
599 {
600 
601 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
602 	    " Addr:0x%lx",
603 	    devid, domid, addr);
604 	amdvi_decode_evt_flag(flag);
605 }
606 
607 static void
608 amdvi_decode_dte_hwerr_evt(uint16_t devid, uint16_t domid,
609     uint64_t addr, uint16_t flag)
610 {
611 
612 	printf("\t[DEV_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
613 	    " Addr:0x%lx", devid, domid, addr);
614 	amdvi_decode_evt_flag(flag);
615 	amdvi_decode_evt_flag_type(flag);
616 }
617 
618 static void
619 amdvi_decode_page_hwerr_evt(uint16_t devid, uint16_t domid, uint64_t addr,
620     uint16_t flag)
621 {
622 
623 	printf("\t[PAGE_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
624 	    " Addr:0x%lx", devid, domid, addr);
625 	amdvi_decode_evt_flag(flag);
626 	amdvi_decode_evt_flag_type(AMDVI_EVENT_FLAG_TYPE(flag));
627 }
628 
629 static void
630 amdvi_decode_evt(struct amdvi_event *evt)
631 {
632 	struct amdvi_cmd *cmd;
633 
634 	switch (evt->opcode) {
635 	case AMDVI_EVENT_INVALID_DTE:
636 		amdvi_decode_inv_dte_evt(evt->devid, evt->pasid_domid,
637 		    evt->addr, evt->flag);
638 		break;
639 
640 	case AMDVI_EVENT_PFAULT:
641 		amdvi_decode_pf_evt(evt->devid, evt->pasid_domid,
642 		    evt->addr, evt->flag);
643 		break;
644 
645 	case AMDVI_EVENT_DTE_HW_ERROR:
646 		amdvi_decode_dte_hwerr_evt(evt->devid, evt->pasid_domid,
647 		    evt->addr, evt->flag);
648 		break;
649 
650 	case AMDVI_EVENT_PAGE_HW_ERROR:
651 		amdvi_decode_page_hwerr_evt(evt->devid, evt->pasid_domid,
652 		    evt->addr, evt->flag);
653 		break;
654 
655 	case AMDVI_EVENT_ILLEGAL_CMD:
656 		/* FALL THROUGH */
657 	case AMDVI_EVENT_CMD_HW_ERROR:
658 		printf("\t[%s EVT]\n", (evt->opcode == AMDVI_EVENT_ILLEGAL_CMD) ?
659 		    "ILLEGAL CMD" : "CMD HW ERR");
660 		cmd = (struct amdvi_cmd *)PHYS_TO_DMAP(evt->addr);
661 		printf("\tCMD opcode= 0x%x 0x%x 0x%x 0x%lx\n",
662 		    cmd->opcode, cmd->word0, cmd->word1, cmd->addr);
663 		break;
664 
665 	case AMDVI_EVENT_IOTLB_TIMEOUT:
666 		printf("\t[IOTLB_INV_TIMEOUT devid:0x%x addr:0x%lx]\n",
667 		    evt->devid, evt->addr);
668 		break;
669 
670 	case AMDVI_EVENT_INVALID_DTE_REQ:
671 		printf("\t[INV_DTE devid:0x%x addr:0x%lx type:0x%x tr:%d]\n",
672 		    evt->devid, evt->addr, evt->flag >> 9,
673 		    (evt->flag >> 8) & 1);
674 		break;
675 
676 	case AMDVI_EVENT_INVALID_PPR_REQ:
677 	case AMDVI_EVENT_COUNTER_ZERO:
678 		printf("AMD-Vi: v2 events.\n");
679 		break;
680 
681 	default:
682 		printf("Unsupported AMD-Vi event:%d\n", evt->opcode);
683 	}
684 }
685 
686 static void
687 amdvi_print_events(struct amdvi_softc *softc)
688 {
689 	struct amdvi_ctrl *ctrl;
690 	struct amdvi_event *event;
691 	int i, size;
692 
693 	ctrl = softc->ctrl;
694 	size = sizeof(struct amdvi_event);
695 	for (i = 0; i < softc->event_max; i++) {
696 		event = &softc->event[ctrl->evt_head / size];
697 		if (!event->opcode)
698 			break;
699 		device_printf(softc->dev, "\t[Event%d: Head:0x%x Tail:0x%x]\n",
700 		    i, ctrl->evt_head, ctrl->evt_tail);
701 		amdvi_decode_evt(event);
702 		ctrl->evt_head = MOD_INC(ctrl->evt_head, size,
703 		    softc->event_max);
704 	}
705 }
706 
707 static int
708 amdvi_init_dte(struct amdvi_softc *softc)
709 {
710 	struct amdvi_ctrl *ctrl;
711 
712 	ctrl = softc->ctrl;
713 	ctrl->dte.base = vtophys(amdvi_dte) / PAGE_SIZE;
714 	ctrl->dte.size = 0x1FF;		/* 2MB device table. */
715 
716 	return (0);
717 }
718 
719 /*
720  * Not all capabilities of IOMMU are available in ACPI IVHD flag
721  * or EFR entry, read directly from device.
722  */
723 static int
724 amdvi_print_pci_cap(device_t dev)
725 {
726 	struct amdvi_softc *softc;
727 	uint32_t off, cap;
728 
729 	softc = device_get_softc(dev);
730 	off = softc->cap_off;
731 
732 	/*
733 	 * Section 3.7.1 of IOMMU sepc rev 2.0.
734 	 * Read capability from device.
735 	 */
736 	cap = amdvi_pci_read(softc, off);
737 
738 	/* Make sure capability type[18:16] is 3. */
739 	KASSERT((((cap >> 16) & 0x7) == 0x3),
740 	    ("Not a IOMMU capability 0x%x@0x%x", cap, off));
741 
742 	softc->pci_cap = cap >> 24;
743 	device_printf(softc->dev, "PCI cap 0x%x@0x%x feature:%b\n",
744 	    cap, off, softc->pci_cap,
745 	    "\20\1IOTLB\2HT\3NPCache\4EFR\5CapExt");
746 
747 	return (0);
748 }
749 
750 static void
751 amdvi_event_intr(void *arg)
752 {
753 	struct amdvi_softc *softc;
754 	struct amdvi_ctrl *ctrl;
755 
756 	softc = (struct amdvi_softc *)arg;
757 	ctrl = softc->ctrl;
758 	device_printf(softc->dev, "EVT INTR %ld Status:0x%x"
759 	    " EVT Head:0x%x Tail:0x%x]\n", softc->event_intr_cnt++,
760 	    ctrl->status, ctrl->evt_head, ctrl->evt_tail);
761 	printf("  [CMD Total 0x%lx] Tail:0x%x, Head:0x%x.\n",
762 	    softc->total_cmd, ctrl->cmd_tail, ctrl->cmd_head);
763 
764 	amdvi_print_events(softc);
765 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
766 }
767 
768 static void
769 amdvi_free_evt_intr_res(device_t dev)
770 {
771 
772 	struct amdvi_softc *softc;
773 	device_t mmio_dev;
774 
775 	softc = device_get_softc(dev);
776 	mmio_dev = softc->pci_dev;
777 
778 	IVHD_TEARDOWN_INTR(mmio_dev);
779 }
780 
781 static bool
782 amdvi_alloc_intr_resources(struct amdvi_softc *softc)
783 {
784 	struct amdvi_ctrl *ctrl;
785 	device_t dev, mmio_dev;
786 	int err;
787 
788 	dev = softc->dev;
789 	mmio_dev = softc->pci_dev;
790 
791 	/* Clear interrupt status bits. */
792 	ctrl = softc->ctrl;
793 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
794 
795 	err = IVHD_SETUP_INTR(mmio_dev, amdvi_event_intr, softc, "fault");
796 	if (err)
797 		device_printf(dev, "Interrupt setup failed on %s\n",
798 		    device_get_nameunit(mmio_dev));
799 	return (err);
800 }
801 
802 static void
803 amdvi_print_dev_cap(struct amdvi_softc *softc)
804 {
805 	struct ivhd_dev_cfg *cfg;
806 	int i;
807 
808 	cfg = softc->dev_cfg;
809 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
810 		device_printf(softc->dev, "device [0x%x - 0x%x] "
811 		    "config:%b%s\n", cfg->start_id, cfg->end_id,
812 		    cfg->data,
813 		    "\020\001INIT\002ExtInt\003NMI"
814 		    "\007LINT0\010LINT1",
815 		    cfg->enable_ats ? "ATS enabled" : "");
816 		cfg++;
817 	}
818 }
819 
820 static int
821 amdvi_handle_sysctl(SYSCTL_HANDLER_ARGS)
822 {
823 	struct amdvi_softc *softc;
824 	int result, type, error = 0;
825 
826 	softc = (struct amdvi_softc *)arg1;
827 	type = arg2;
828 
829 	switch (type) {
830 	case 0:
831 		result = softc->ctrl->cmd_head;
832 		error = sysctl_handle_int(oidp, &result, 0,
833 		    req);
834 		break;
835 	case 1:
836 		result = softc->ctrl->cmd_tail;
837 		error = sysctl_handle_int(oidp, &result, 0,
838 		    req);
839 		break;
840 	case 2:
841 		result = softc->ctrl->evt_head;
842 		error = sysctl_handle_int(oidp, &result, 0,
843 		    req);
844 		break;
845 	case 3:
846 		result = softc->ctrl->evt_tail;
847 		error = sysctl_handle_int(oidp, &result, 0,
848 		    req);
849 		break;
850 
851 	default:
852 		device_printf(softc->dev, "Unknown sysctl:%d\n", type);
853 	}
854 
855 	return (error);
856 }
857 
858 static void
859 amdvi_add_sysctl(struct amdvi_softc *softc)
860 {
861 	struct sysctl_oid_list *child;
862 	struct sysctl_ctx_list *ctx;
863 	device_t dev;
864 
865 	dev = softc->dev;
866 	ctx = device_get_sysctl_ctx(dev);
867 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
868 
869 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "event_intr_count", CTLFLAG_RD,
870 	    &softc->event_intr_cnt, "Event interrupt count");
871 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "command_count", CTLFLAG_RD,
872 	    &softc->total_cmd, "Command submitted count");
873 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "pci_rid", CTLFLAG_RD,
874 	    &softc->pci_rid, 0, "IOMMU RID");
875 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_head",
876 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 0,
877 	    amdvi_handle_sysctl, "IU", "Command head");
878 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_tail",
879 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 1,
880 	    amdvi_handle_sysctl, "IU", "Command tail");
881 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_head",
882 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 2,
883 	    amdvi_handle_sysctl, "IU", "Command head");
884 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_tail",
885 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 3,
886 	    amdvi_handle_sysctl, "IU", "Command tail");
887 }
888 
889 int
890 amdvi_setup_hw(struct amdvi_softc *softc)
891 {
892 	device_t dev;
893 	int status;
894 
895 	dev = softc->dev;
896 
897 	amdvi_hw_enable_iotlb(softc);
898 
899 	amdvi_print_dev_cap(softc);
900 
901 	if ((status = amdvi_print_pci_cap(dev)) != 0) {
902 		device_printf(dev, "PCI capability.\n");
903 		return (status);
904 	}
905 	if ((status = amdvi_init_cmd(softc)) != 0) {
906 		device_printf(dev, "Couldn't configure command buffer.\n");
907 		return (status);
908 	}
909 	if ((status = amdvi_init_event(softc)) != 0) {
910 		device_printf(dev, "Couldn't configure event buffer.\n");
911 		return (status);
912 	}
913 	if ((status = amdvi_init_dte(softc)) != 0) {
914 		device_printf(dev, "Couldn't configure device table.\n");
915 		return (status);
916 	}
917 	if ((status = amdvi_alloc_intr_resources(softc)) != 0) {
918 		return (status);
919 	}
920 	amdvi_add_sysctl(softc);
921 	return (0);
922 }
923 
924 int
925 amdvi_teardown_hw(struct amdvi_softc *softc)
926 {
927 	device_t dev;
928 
929 	dev = softc->dev;
930 
931 	/*
932 	 * Called after disable, h/w is stopped by now, free all the resources.
933 	 */
934 	amdvi_free_evt_intr_res(dev);
935 
936 	if (softc->cmd)
937 		free(softc->cmd, M_AMDVI);
938 
939 	if (softc->event)
940 		free(softc->event, M_AMDVI);
941 
942 	return (0);
943 }
944 
945 /*********** bhyve interfaces *********************/
946 static int
947 amdvi_init(void)
948 {
949 	if (!ivhd_count) {
950 		return (EIO);
951 	}
952 	if (!amdvi_enable_user && ivhd_count) {
953 		printf("bhyve: Found %d AMD-Vi/IOMMU device(s), "
954 		    	"use hw.vmm.amdvi.enable=1 to enable pass-through.\n",
955 		    ivhd_count);
956 		return (EINVAL);
957 	}
958 	return (0);
959 }
960 
961 static void
962 amdvi_cleanup(void)
963 {
964 	/* Nothing. */
965 }
966 
967 static uint16_t
968 amdvi_domainId(void)
969 {
970 
971 	/*
972 	 * If we hit maximum domain limit, rollover leaving host
973 	 * domain(0).
974 	 * XXX: make sure that this domain is not used.
975 	 */
976 	if (amdvi_dom_id == AMDVI_MAX_DOMAIN)
977 		amdvi_dom_id = 1;
978 
979 	return ((uint16_t)amdvi_dom_id++);
980 }
981 
982 static void
983 amdvi_do_inv_domain(uint16_t domain_id, bool create)
984 {
985 	struct amdvi_softc *softc;
986 	int i;
987 
988 	for (i = 0; i < ivhd_count; i++) {
989 		softc = device_get_softc(ivhd_devs[i]);
990 		KASSERT(softc, ("softc is NULL"));
991 		/*
992 		 * If not present pages are cached, invalidate page after
993 		 * creating domain.
994 		 */
995 #if 0
996 		if (create && ((softc->pci_cap & AMDVI_PCI_CAP_NPCACHE) == 0))
997 			continue;
998 #endif
999 		amdvi_inv_domain(softc, domain_id);
1000 		amdvi_wait(softc);
1001 	}
1002 }
1003 
1004 static void *
1005 amdvi_create_domain(vm_paddr_t maxaddr)
1006 {
1007 	struct amdvi_domain *dom;
1008 
1009 	dom = malloc(sizeof(struct amdvi_domain), M_AMDVI, M_ZERO | M_WAITOK);
1010 	dom->id = amdvi_domainId();
1011 	//dom->maxaddr = maxaddr;
1012 #ifdef AMDVI_DEBUG_CMD
1013 	printf("Created domain #%d\n", dom->id);
1014 #endif
1015 	/*
1016 	 * Host domain(#0) don't create translation table.
1017 	 */
1018 	if (dom->id || amdvi_host_ptp)
1019 		dom->ptp = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1020 
1021 	dom->ptp_level = amdvi_ptp_level;
1022 
1023 	amdvi_do_inv_domain(dom->id, true);
1024 	SLIST_INSERT_HEAD(&dom_head, dom, next);
1025 
1026 	return (dom);
1027 }
1028 
1029 static void
1030 amdvi_free_ptp(uint64_t *ptp, int level)
1031 {
1032 	int i;
1033 
1034 	if (level < 1)
1035 		return;
1036 
1037 	for (i = 0; i < NPTEPG ; i++) {
1038 		if ((ptp[i] & AMDVI_PT_PRESENT) == 0)
1039 			continue;
1040 		/* XXX: Add super-page or PTE mapping > 4KB. */
1041 #ifdef notyet
1042 		/* Super-page mapping. */
1043 		if (AMDVI_PD_SUPER(ptp[i]))
1044 			continue;
1045 #endif
1046 
1047 		amdvi_free_ptp((uint64_t *)PHYS_TO_DMAP(ptp[i]
1048 		    & AMDVI_PT_MASK), level - 1);
1049 	}
1050 
1051 	free(ptp, M_AMDVI);
1052 }
1053 
1054 static void
1055 amdvi_destroy_domain(void *arg)
1056 {
1057 	struct amdvi_domain *domain;
1058 
1059 	domain = (struct amdvi_domain *)arg;
1060 	KASSERT(domain, ("domain is NULL"));
1061 #ifdef AMDVI_DEBUG_CMD
1062 	printf("Destroying domain %d\n", domain->id);
1063 #endif
1064 	if (domain->ptp)
1065 		amdvi_free_ptp(domain->ptp, domain->ptp_level);
1066 
1067 	amdvi_do_inv_domain(domain->id, false);
1068 	SLIST_REMOVE(&dom_head, domain, amdvi_domain, next);
1069 	free(domain, M_AMDVI);
1070 }
1071 
1072 static uint64_t
1073 amdvi_set_pt(uint64_t *pt, int level, vm_paddr_t gpa,
1074     vm_paddr_t hpa, uint64_t pg_size, bool create)
1075 {
1076 	uint64_t *page, pa;
1077 	int shift, index;
1078 	const int PT_SHIFT = 9;
1079 	const int PT_INDEX_MASK = (1 << PT_SHIFT) - 1;	/* Based on PT_SHIFT */
1080 
1081 	if (!pg_size)
1082 		return (0);
1083 
1084 	if (hpa & (pg_size - 1)) {
1085 		printf("HPA is not size aligned.\n");
1086 		return (0);
1087 	}
1088 	if (gpa & (pg_size - 1)) {
1089 		printf("HPA is not size aligned.\n");
1090 		return (0);
1091 	}
1092 	shift = PML4SHIFT;
1093 	while ((shift > PAGE_SHIFT) && (pg_size < (1UL << shift))) {
1094 		index = (gpa >> shift) & PT_INDEX_MASK;
1095 
1096 		if ((pt[index] == 0) && create) {
1097 			page = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1098 			pa = vtophys(page);
1099 			pt[index] = pa | AMDVI_PT_PRESENT | AMDVI_PT_RW |
1100 			    ((level - 1) << AMDVI_PD_LEVEL_SHIFT);
1101 		}
1102 #ifdef AMDVI_DEBUG_PTE
1103 		if ((gpa % 0x1000000) == 0)
1104 			printf("[level%d, shift = %d]PTE:0x%lx\n",
1105 			    level, shift, pt[index]);
1106 #endif
1107 #define PTE2PA(x)	((uint64_t)(x) & AMDVI_PT_MASK)
1108 		pa = PTE2PA(pt[index]);
1109 		pt = (uint64_t *)PHYS_TO_DMAP(pa);
1110 		shift -= PT_SHIFT;
1111 		level--;
1112 	}
1113 
1114 	/* Leaf entry. */
1115 	index = (gpa >> shift) & PT_INDEX_MASK;
1116 
1117 	if (create) {
1118 		pt[index] = hpa | AMDVI_PT_RW | AMDVI_PT_PRESENT;
1119 	} else
1120 		pt[index] = 0;
1121 
1122 #ifdef AMDVI_DEBUG_PTE
1123 	if ((gpa % 0x1000000) == 0)
1124 		printf("[Last level%d, shift = %d]PTE:0x%lx\n",
1125 		    level, shift, pt[index]);
1126 #endif
1127 	return (1ULL << shift);
1128 }
1129 
1130 static uint64_t
1131 amdvi_update_mapping(struct amdvi_domain *domain, vm_paddr_t gpa,
1132     vm_paddr_t hpa, uint64_t size, bool create)
1133 {
1134 	uint64_t mapped, *ptp, len;
1135 	int level;
1136 
1137 	KASSERT(domain, ("domain is NULL"));
1138 	level = domain->ptp_level;
1139 	KASSERT(level, ("Page table level is 0"));
1140 
1141 	ptp = domain->ptp;
1142 	KASSERT(ptp, ("PTP is NULL"));
1143 	mapped = 0;
1144 	while (mapped < size) {
1145 		len = amdvi_set_pt(ptp, level, gpa + mapped, hpa + mapped,
1146 		    PAGE_SIZE, create);
1147 		if (!len) {
1148 			printf("Error: Couldn't map HPA:0x%lx GPA:0x%lx\n",
1149 			    hpa, gpa);
1150 			return (0);
1151 		}
1152 		mapped += len;
1153 	}
1154 
1155 	return (mapped);
1156 }
1157 
1158 static uint64_t
1159 amdvi_create_mapping(void *arg, vm_paddr_t gpa, vm_paddr_t hpa,
1160     uint64_t len)
1161 {
1162 	struct amdvi_domain *domain;
1163 
1164 	domain = (struct amdvi_domain *)arg;
1165 
1166 	if (domain->id && !domain->ptp) {
1167 		printf("ptp is NULL");
1168 		return (-1);
1169 	}
1170 
1171 	/*
1172 	 * If host domain is created w/o page table, skip IOMMU page
1173 	 * table set-up.
1174 	 */
1175 	if (domain->ptp)
1176 		return (amdvi_update_mapping(domain, gpa, hpa, len, true));
1177 	else
1178 		return (len);
1179 }
1180 
1181 static uint64_t
1182 amdvi_remove_mapping(void *arg, vm_paddr_t gpa, uint64_t len)
1183 {
1184 	struct amdvi_domain *domain;
1185 
1186 	domain = (struct amdvi_domain *)arg;
1187 	/*
1188 	 * If host domain is created w/o page table, skip IOMMU page
1189 	 * table set-up.
1190 	 */
1191 	if (domain->ptp)
1192 		return (amdvi_update_mapping(domain, gpa, 0, len, false));
1193 	return
1194 	    (len);
1195 }
1196 
1197 static struct amdvi_softc *
1198 amdvi_find_iommu(uint16_t devid)
1199 {
1200 	struct amdvi_softc *softc;
1201 	int i, j;
1202 
1203 	for (i = 0; i < ivhd_count; i++) {
1204 		softc = device_get_softc(ivhd_devs[i]);
1205 		for (j = 0; j < softc->dev_cfg_cnt; j++)
1206 			if ((devid >= softc->dev_cfg[j].start_id) &&
1207 			    (devid <= softc->dev_cfg[j].end_id))
1208 				return (softc);
1209 	}
1210 
1211 	return (NULL);
1212 }
1213 
1214 /*
1215  * Set-up device table entry.
1216  * IOMMU spec Rev 2.0, section 3.2.2.2, some of the fields must
1217  * be set concurrently, e.g. read and write bits.
1218  */
1219 static void
1220 amdvi_set_dte(struct amdvi_domain *domain, struct amdvi_softc *softc,
1221     uint16_t devid, bool enable)
1222 {
1223 	struct amdvi_dte* temp;
1224 
1225 	KASSERT(domain, ("domain is NULL for pci_rid:0x%x\n", devid));
1226 	KASSERT(softc, ("softc is NULL for pci_rid:0x%x\n", devid));
1227 
1228 	temp = &amdvi_dte[devid];
1229 
1230 #ifdef AMDVI_ATS_ENABLE
1231 	/* If IOMMU and device support IOTLB, enable it. */
1232 	if (amdvi_dev_support_iotlb(softc, devid) && softc->iotlb)
1233 		temp->iotlb_enable = 1;
1234 #endif
1235 
1236 	/* Avoid duplicate I/O faults. */
1237 	temp->sup_second_io_fault = 1;
1238 	temp->sup_all_io_fault = amdvi_disable_io_fault;
1239 
1240 	temp->dt_valid = 1;
1241 	temp->domain_id = domain->id;
1242 
1243 	if (enable) {
1244 		if (domain->ptp) {
1245 			temp->pt_base = vtophys(domain->ptp) >> 12;
1246 			temp->pt_level = amdvi_ptp_level;
1247 		}
1248 		/*
1249 		 * XXX: Page table valid[TV] bit must be set even if host domain
1250 		 * page tables are not enabled.
1251 		 */
1252 		temp->pt_valid = 1;
1253 		temp->read_allow = 1;
1254 		temp->write_allow = 1;
1255 	}
1256 }
1257 
1258 static void
1259 amdvi_inv_device(struct amdvi_softc *softc, uint16_t devid)
1260 {
1261 	KASSERT(softc, ("softc is NULL"));
1262 
1263 	amdvi_cmd_inv_dte(softc, devid);
1264 #ifdef AMDVI_ATS_ENABLE
1265 	if (amdvi_dev_support_iotlb(softc, devid))
1266 		amdvi_cmd_inv_iotlb(softc, devid);
1267 #endif
1268 	amdvi_wait(softc);
1269 }
1270 
1271 static void
1272 amdvi_add_device(void *arg, uint16_t devid)
1273 {
1274 	struct amdvi_domain *domain;
1275 	struct amdvi_softc *softc;
1276 
1277 	domain = (struct amdvi_domain *)arg;
1278 	KASSERT(domain != NULL, ("domain is NULL"));
1279 #ifdef AMDVI_DEBUG_CMD
1280 	printf("Assigning device(%d.%d.%d) to domain:%d\n",
1281 	    RID2PCI_STR(devid), domain->id);
1282 #endif
1283 	softc = amdvi_find_iommu(devid);
1284 	if (softc == NULL)
1285 		return;
1286 	amdvi_set_dte(domain, softc, devid, true);
1287 	amdvi_inv_device(softc, devid);
1288 }
1289 
1290 static void
1291 amdvi_remove_device(void *arg, uint16_t devid)
1292 {
1293 	struct amdvi_domain *domain;
1294 	struct amdvi_softc *softc;
1295 
1296 	domain = (struct amdvi_domain *)arg;
1297 #ifdef AMDVI_DEBUG_CMD
1298 	printf("Remove device(0x%x) from domain:%d\n",
1299 	       devid, domain->id);
1300 #endif
1301 	softc = amdvi_find_iommu(devid);
1302 	if (softc == NULL)
1303 		return;
1304 	amdvi_set_dte(domain, softc, devid, false);
1305 	amdvi_inv_device(softc, devid);
1306 }
1307 
1308 static void
1309 amdvi_enable(void)
1310 {
1311 	struct amdvi_ctrl *ctrl;
1312 	struct amdvi_softc *softc;
1313 	uint64_t val;
1314 	int i;
1315 
1316 	for (i = 0; i < ivhd_count; i++) {
1317 		softc = device_get_softc(ivhd_devs[i]);
1318 		KASSERT(softc, ("softc is NULL\n"));
1319 		ctrl = softc->ctrl;
1320 		KASSERT(ctrl, ("ctrl is NULL\n"));
1321 
1322 		val = (	AMDVI_CTRL_EN 		|
1323 			AMDVI_CTRL_CMD 		|
1324 		    	AMDVI_CTRL_ELOG 	|
1325 		    	AMDVI_CTRL_ELOGINT 	|
1326 		    	AMDVI_CTRL_INV_TO_1S);
1327 
1328 		if (softc->ivhd_flag & IVHD_FLAG_COH)
1329 			val |= AMDVI_CTRL_COH;
1330 		if (softc->ivhd_flag & IVHD_FLAG_HTT)
1331 			val |= AMDVI_CTRL_HTT;
1332 		if (softc->ivhd_flag & IVHD_FLAG_RPPW)
1333 			val |= AMDVI_CTRL_RPPW;
1334 		if (softc->ivhd_flag & IVHD_FLAG_PPW)
1335 			val |= AMDVI_CTRL_PPW;
1336 		if (softc->ivhd_flag & IVHD_FLAG_ISOC)
1337 			val |= AMDVI_CTRL_ISOC;
1338 
1339 		ctrl->control = val;
1340 	}
1341 }
1342 
1343 static void
1344 amdvi_disable(void)
1345 {
1346 	struct amdvi_ctrl *ctrl;
1347 	struct amdvi_softc *softc;
1348 	int i;
1349 
1350 	for (i = 0; i < ivhd_count; i++) {
1351 		softc = device_get_softc(ivhd_devs[i]);
1352 		KASSERT(softc, ("softc is NULL\n"));
1353 		ctrl = softc->ctrl;
1354 		KASSERT(ctrl, ("ctrl is NULL\n"));
1355 
1356 		ctrl->control = 0;
1357 	}
1358 }
1359 
1360 static void
1361 amdvi_invalidate_tlb(void *arg)
1362 {
1363 	struct amdvi_domain *domain;
1364 
1365 	domain = (struct amdvi_domain *)arg;
1366 	KASSERT(domain, ("domain is NULL"));
1367 	amdvi_do_inv_domain(domain->id, false);
1368 }
1369 
1370 const struct iommu_ops iommu_ops_amd = {
1371 	.init = amdvi_init,
1372 	.cleanup = amdvi_cleanup,
1373 	.enable = amdvi_enable,
1374 	.disable = amdvi_disable,
1375 	.create_domain = amdvi_create_domain,
1376 	.destroy_domain = amdvi_destroy_domain,
1377 	.create_mapping = amdvi_create_mapping,
1378 	.remove_mapping = amdvi_remove_mapping,
1379 	.add_device = amdvi_add_device,
1380 	.remove_device = amdvi_remove_device,
1381 	.invalidate_tlb = amdvi_invalidate_tlb
1382 };
1383