xref: /illumos-gate/usr/src/uts/intel/io/vmm/amd/amdvi_hw.c (revision 1d0ec46fafb49266ae79840a692bb48af60ade70)
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 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/bus.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/malloc.h>
37 #include <sys/pcpu.h>
38 #include <sys/rman.h>
39 #include <sys/sysctl.h>
40 
41 #include <dev/pci/pcivar.h>
42 #include <dev/pci/pcireg.h>
43 
44 #include <machine/resource.h>
45 #include <machine/vmm.h>
46 #include <machine/vmparam.h>
47 #include <machine/pci_cfgreg.h>
48 
49 #include "ivhd_if.h"
50 #include "pcib_if.h"
51 
52 #include "io/iommu.h"
53 #include "amdvi_priv.h"
54 
55 SYSCTL_DECL(_hw_vmm);
56 SYSCTL_NODE(_hw_vmm, OID_AUTO, amdvi, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
57     NULL);
58 
59 #define MOD_INC(a, s, m) (((a) + (s)) % ((m) * (s)))
60 #define MOD_DEC(a, s, m) (((a) - (s)) % ((m) * (s)))
61 
62 /* Print RID or device ID in PCI string format. */
63 #define RID2PCI_STR(d) PCI_RID2BUS(d), PCI_RID2SLOT(d), PCI_RID2FUNC(d)
64 
65 static void amdvi_dump_cmds(struct amdvi_softc *softc, int count);
66 static void amdvi_print_dev_cap(struct amdvi_softc *softc);
67 
68 MALLOC_DEFINE(M_AMDVI, "amdvi", "amdvi");
69 
70 extern device_t *ivhd_devs;
71 
72 extern int ivhd_count;
73 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, count, CTLFLAG_RDTUN, &ivhd_count,
74     0, NULL);
75 
76 static int amdvi_enable_user = 0;
77 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, enable, CTLFLAG_RDTUN,
78     &amdvi_enable_user, 0, NULL);
79 TUNABLE_INT("hw.vmm.amdvi_enable", &amdvi_enable_user);
80 
81 #ifdef AMDVI_ATS_ENABLE
82 /* XXX: ATS is not tested. */
83 static int amdvi_enable_iotlb = 1;
84 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, iotlb_enabled, CTLFLAG_RDTUN,
85     &amdvi_enable_iotlb, 0, NULL);
86 TUNABLE_INT("hw.vmm.enable_iotlb", &amdvi_enable_iotlb);
87 #endif
88 
89 static int amdvi_host_ptp = 1;	/* Use page tables for host. */
90 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, host_ptp, CTLFLAG_RDTUN,
91     &amdvi_host_ptp, 0, NULL);
92 TUNABLE_INT("hw.vmm.amdvi.host_ptp", &amdvi_host_ptp);
93 
94 /* Page table level used <= supported by h/w[v1=7]. */
95 int amdvi_ptp_level = 4;
96 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, ptp_level, CTLFLAG_RDTUN,
97     &amdvi_ptp_level, 0, NULL);
98 TUNABLE_INT("hw.vmm.amdvi.ptp_level", &amdvi_ptp_level);
99 
100 /* Disable fault event reporting. */
101 static int amdvi_disable_io_fault = 0;
102 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, disable_io_fault, CTLFLAG_RDTUN,
103     &amdvi_disable_io_fault, 0, NULL);
104 TUNABLE_INT("hw.vmm.amdvi.disable_io_fault", &amdvi_disable_io_fault);
105 
106 static uint32_t amdvi_dom_id = 0;	/* 0 is reserved for host. */
107 SYSCTL_UINT(_hw_vmm_amdvi, OID_AUTO, domain_id, CTLFLAG_RD,
108     &amdvi_dom_id, 0, NULL);
109 /*
110  * Device table entry.
111  * Bus(256) x Dev(32) x Fun(8) x DTE(256 bits or 32 bytes).
112  *	= 256 * 2 * PAGE_SIZE.
113  */
114 static struct amdvi_dte amdvi_dte[PCI_NUM_DEV_MAX] __aligned(PAGE_SIZE);
115 CTASSERT(PCI_NUM_DEV_MAX == 0x10000);
116 CTASSERT(sizeof(amdvi_dte) == 0x200000);
117 
118 static SLIST_HEAD (, amdvi_domain) dom_head;
119 
120 static inline uint32_t
121 amdvi_pci_read(struct amdvi_softc *softc, int off)
122 {
123 
124 	return (pci_cfgregread(PCI_RID2BUS(softc->pci_rid),
125 	    PCI_RID2SLOT(softc->pci_rid), PCI_RID2FUNC(softc->pci_rid),
126 	    off, 4));
127 }
128 
129 #ifdef AMDVI_ATS_ENABLE
130 /* XXX: Should be in pci.c */
131 /*
132  * Check if device has ATS capability and its enabled.
133  * If ATS is absent or disabled, return (-1), otherwise ATS
134  * queue length.
135  */
136 static int
137 amdvi_find_ats_qlen(uint16_t devid)
138 {
139 	device_t dev;
140 	uint32_t off, cap;
141 	int qlen = -1;
142 
143 	dev = pci_find_bsf(PCI_RID2BUS(devid), PCI_RID2SLOT(devid),
144 			   PCI_RID2FUNC(devid));
145 
146 	if (!dev) {
147 		return (-1);
148 	}
149 #define PCIM_ATS_EN	BIT(31)
150 
151 	if (pci_find_extcap(dev, PCIZ_ATS, &off) == 0) {
152 		cap = pci_read_config(dev, off + 4, 4);
153 		qlen = (cap & 0x1F);
154 		qlen = qlen ? qlen : 32;
155 		printf("AMD-Vi: PCI device %d.%d.%d ATS %s qlen=%d\n",
156 		       RID2PCI_STR(devid),
157 		       (cap & PCIM_ATS_EN) ? "enabled" : "Disabled",
158 		       qlen);
159 		qlen = (cap & PCIM_ATS_EN) ? qlen : -1;
160 	}
161 
162 	return (qlen);
163 }
164 
165 /*
166  * Check if an endpoint device support device IOTLB or ATS.
167  */
168 static inline bool
169 amdvi_dev_support_iotlb(struct amdvi_softc *softc, uint16_t devid)
170 {
171 	struct ivhd_dev_cfg *cfg;
172 	int qlen, i;
173 	bool pci_ats, ivhd_ats;
174 
175 	qlen = amdvi_find_ats_qlen(devid);
176 	if (qlen < 0)
177 		return (false);
178 
179 	KASSERT(softc, ("softc is NULL"));
180 	cfg = softc->dev_cfg;
181 
182 	ivhd_ats = false;
183 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
184 		if ((cfg->start_id <= devid) && (cfg->end_id >= devid)) {
185 			ivhd_ats = cfg->enable_ats;
186 			break;
187 		}
188 		cfg++;
189 	}
190 
191 	pci_ats = (qlen < 0) ? false : true;
192 	if (pci_ats != ivhd_ats)
193 		device_printf(softc->dev,
194 		    "BIOS bug: mismatch in ATS setting for %d.%d.%d,"
195 		    "ATS inv qlen = %d\n", RID2PCI_STR(devid), qlen);
196 
197 	/* Ignore IVRS setting and respect PCI setting. */
198 	return (pci_ats);
199 }
200 #endif
201 
202 /* Enable IOTLB support for IOMMU if its supported. */
203 static inline void
204 amdvi_hw_enable_iotlb(struct amdvi_softc *softc)
205 {
206 #ifndef AMDVI_ATS_ENABLE
207 	softc->iotlb = false;
208 #else
209 	bool supported;
210 
211 	supported = (softc->ivhd_flag & IVHD_FLAG_IOTLB) ? true : false;
212 
213 	if (softc->pci_cap & AMDVI_PCI_CAP_IOTLB) {
214 		if (!supported)
215 			device_printf(softc->dev, "IOTLB disabled by BIOS.\n");
216 
217 		if (supported && !amdvi_enable_iotlb) {
218 			device_printf(softc->dev, "IOTLB disabled by user.\n");
219 			supported = false;
220 		}
221 	} else
222 		supported = false;
223 
224 	softc->iotlb = supported;
225 
226 #endif
227 }
228 
229 static int
230 amdvi_init_cmd(struct amdvi_softc *softc)
231 {
232 	struct amdvi_ctrl *ctrl = softc->ctrl;
233 
234 	ctrl->cmd.len = 8;	/* Use 256 command buffer entries. */
235 	softc->cmd_max = 1 << ctrl->cmd.len;
236 
237 	softc->cmd = malloc(sizeof(struct amdvi_cmd) *
238 	    softc->cmd_max, M_AMDVI, M_WAITOK | M_ZERO);
239 
240 	if ((uintptr_t)softc->cmd & PAGE_MASK)
241 		panic("AMDVi: Command buffer not aligned on page boundary.");
242 
243 	ctrl->cmd.base = vtophys(softc->cmd) / PAGE_SIZE;
244 	/*
245 	 * XXX: Reset the h/w pointers in case IOMMU is restarting,
246 	 * h/w doesn't clear these pointers based on empirical data.
247 	 */
248 	ctrl->cmd_tail = 0;
249 	ctrl->cmd_head = 0;
250 
251 	return (0);
252 }
253 
254 /*
255  * Note: Update tail pointer after we have written the command since tail
256  * pointer update cause h/w to execute new commands, see section 3.3
257  * of AMD IOMMU spec ver 2.0.
258  */
259 /* Get the command tail pointer w/o updating it. */
260 static struct amdvi_cmd *
261 amdvi_get_cmd_tail(struct amdvi_softc *softc)
262 {
263 	struct amdvi_ctrl *ctrl;
264 	struct amdvi_cmd *tail;
265 
266 	KASSERT(softc, ("softc is NULL"));
267 	KASSERT(softc->cmd != NULL, ("cmd is NULL"));
268 
269 	ctrl = softc->ctrl;
270 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
271 
272 	tail = (struct amdvi_cmd *)((uint8_t *)softc->cmd +
273 	    ctrl->cmd_tail);
274 
275 	return (tail);
276 }
277 
278 /*
279  * Update the command tail pointer which will start command execution.
280  */
281 static void
282 amdvi_update_cmd_tail(struct amdvi_softc *softc)
283 {
284 	struct amdvi_ctrl *ctrl;
285 	int size;
286 
287 	size = sizeof(struct amdvi_cmd);
288 	KASSERT(softc->cmd != NULL, ("cmd is NULL"));
289 
290 	ctrl = softc->ctrl;
291 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
292 
293 	ctrl->cmd_tail = MOD_INC(ctrl->cmd_tail, size, softc->cmd_max);
294 	softc->total_cmd++;
295 
296 #ifdef AMDVI_DEBUG_CMD
297 	device_printf(softc->dev, "cmd_tail: %s Tail:0x%x, Head:0x%x.\n",
298 	    ctrl->cmd_tail,
299 	    ctrl->cmd_head);
300 #endif
301 
302 }
303 
304 /*
305  * Various commands supported by IOMMU.
306  */
307 
308 /* Completion wait command. */
309 static void
310 amdvi_cmd_cmp(struct amdvi_softc *softc, const uint64_t data)
311 {
312 	struct amdvi_cmd *cmd;
313 	uint64_t pa;
314 
315 	cmd = amdvi_get_cmd_tail(softc);
316 	KASSERT(cmd != NULL, ("Cmd is NULL"));
317 
318 	pa = vtophys(&softc->cmp_data);
319 	cmd->opcode = AMDVI_CMP_WAIT_OPCODE;
320 	cmd->word0 = (pa & 0xFFFFFFF8) | AMDVI_CMP_WAIT_STORE;
321 	cmd->word1 = (pa >> 32) & 0xFFFFF;
322 	cmd->addr = data;
323 
324 	amdvi_update_cmd_tail(softc);
325 }
326 
327 /* Invalidate device table entry. */
328 static void
329 amdvi_cmd_inv_dte(struct amdvi_softc *softc, uint16_t devid)
330 {
331 	struct amdvi_cmd *cmd;
332 
333 	cmd = amdvi_get_cmd_tail(softc);
334 	KASSERT(cmd != NULL, ("Cmd is NULL"));
335 	cmd->opcode = AMDVI_INVD_DTE_OPCODE;
336 	cmd->word0 = devid;
337 	amdvi_update_cmd_tail(softc);
338 #ifdef AMDVI_DEBUG_CMD
339 	device_printf(softc->dev, "Invalidated DTE:0x%x\n", devid);
340 #endif
341 }
342 
343 /* Invalidate IOMMU page, use for invalidation of domain. */
344 static void
345 amdvi_cmd_inv_iommu_pages(struct amdvi_softc *softc, uint16_t domain_id,
346 			  uint64_t addr, bool guest_nested,
347 			  bool pde, bool page)
348 {
349 	struct amdvi_cmd *cmd;
350 
351 	cmd = amdvi_get_cmd_tail(softc);
352 	KASSERT(cmd != NULL, ("Cmd is NULL"));
353 
354 	cmd->opcode = AMDVI_INVD_PAGE_OPCODE;
355 	cmd->word1 = domain_id;
356 	/*
357 	 * Invalidate all addresses for this domain.
358 	 */
359 	cmd->addr = addr;
360 	cmd->addr |= pde ? AMDVI_INVD_PAGE_PDE : 0;
361 	cmd->addr |= page ? AMDVI_INVD_PAGE_S : 0;
362 
363 	amdvi_update_cmd_tail(softc);
364 }
365 
366 #ifdef AMDVI_ATS_ENABLE
367 /* Invalidate device IOTLB. */
368 static void
369 amdvi_cmd_inv_iotlb(struct amdvi_softc *softc, uint16_t devid)
370 {
371 	struct amdvi_cmd *cmd;
372 	int qlen;
373 
374 	if (!softc->iotlb)
375 		return;
376 
377 	qlen = amdvi_find_ats_qlen(devid);
378 	if (qlen < 0) {
379 		panic("AMDVI: Invalid ATS qlen(%d) for device %d.%d.%d\n",
380 		      qlen, RID2PCI_STR(devid));
381 	}
382 	cmd = amdvi_get_cmd_tail(softc);
383 	KASSERT(cmd != NULL, ("Cmd is NULL"));
384 
385 #ifdef AMDVI_DEBUG_CMD
386 	device_printf(softc->dev, "Invalidate IOTLB devID 0x%x"
387 		      " Qlen:%d\n", devid, qlen);
388 #endif
389 	cmd->opcode = AMDVI_INVD_IOTLB_OPCODE;
390 	cmd->word0 = devid;
391 	cmd->word1 = qlen;
392 	cmd->addr = AMDVI_INVD_IOTLB_ALL_ADDR |
393 		AMDVI_INVD_IOTLB_S;
394 	amdvi_update_cmd_tail(softc);
395 }
396 #endif
397 
398 #ifdef notyet				/* For Interrupt Remap. */
399 static void
400 amdvi_cmd_inv_intr_map(struct amdvi_softc *softc,
401 		       uint16_t devid)
402 {
403 	struct amdvi_cmd *cmd;
404 
405 	cmd = amdvi_get_cmd_tail(softc);
406 	KASSERT(cmd != NULL, ("Cmd is NULL"));
407 	cmd->opcode = AMDVI_INVD_INTR_OPCODE;
408 	cmd->word0 = devid;
409 	amdvi_update_cmd_tail(softc);
410 #ifdef AMDVI_DEBUG_CMD
411 	device_printf(softc->dev, "Invalidate INTR map of devID 0x%x\n", devid);
412 #endif
413 }
414 #endif
415 
416 /* Invalidate domain using INVALIDATE_IOMMU_PAGES command. */
417 static void
418 amdvi_inv_domain(struct amdvi_softc *softc, uint16_t domain_id)
419 {
420 	struct amdvi_cmd *cmd __diagused;
421 
422 	cmd = amdvi_get_cmd_tail(softc);
423 	KASSERT(cmd != NULL, ("Cmd is NULL"));
424 
425 	/*
426 	 * See section 3.3.3 of IOMMU spec rev 2.0, software note
427 	 * for invalidating domain.
428 	 */
429 	amdvi_cmd_inv_iommu_pages(softc, domain_id, AMDVI_INVD_PAGE_ALL_ADDR,
430 				false, true, true);
431 
432 #ifdef AMDVI_DEBUG_CMD
433 	device_printf(softc->dev, "Invalidate domain:0x%x\n", domain_id);
434 
435 #endif
436 }
437 
438 static	bool
439 amdvi_cmp_wait(struct amdvi_softc *softc)
440 {
441 #ifdef AMDVI_DEBUG_CMD
442 	struct amdvi_ctrl *ctrl = softc->ctrl;
443 #endif
444 	const uint64_t VERIFY = 0xA5A5;
445 	volatile uint64_t *read;
446 	int i;
447 	bool status;
448 
449 	read = &softc->cmp_data;
450 	*read = 0;
451 	amdvi_cmd_cmp(softc, VERIFY);
452 	/* Wait for h/w to update completion data. */
453 	for (i = 0; i < 100 && (*read != VERIFY); i++) {
454 		DELAY(1000);		/* 1 ms */
455 	}
456 	status = (VERIFY == softc->cmp_data) ? true : false;
457 
458 #ifdef AMDVI_DEBUG_CMD
459 	if (status)
460 		device_printf(softc->dev, "CMD completion DONE Tail:0x%x, "
461 			      "Head:0x%x, loop:%d.\n", ctrl->cmd_tail,
462 			      ctrl->cmd_head, loop);
463 #endif
464 	return (status);
465 }
466 
467 static void
468 amdvi_wait(struct amdvi_softc *softc)
469 {
470 	struct amdvi_ctrl *ctrl;
471 	int i;
472 
473 	KASSERT(softc, ("softc is NULL"));
474 
475 	ctrl = softc->ctrl;
476 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
477 	/* Don't wait if h/w is not enabled. */
478 	if ((ctrl->control & AMDVI_CTRL_EN) == 0)
479 		return;
480 
481 	for (i = 0; i < 10; i++) {
482 		if (amdvi_cmp_wait(softc))
483 			return;
484 	}
485 
486 	device_printf(softc->dev, "Error: completion failed"
487 		      " tail:0x%x, head:0x%x.\n",
488 		      ctrl->cmd_tail, ctrl->cmd_head);
489 	/* Dump the last command. */
490 	amdvi_dump_cmds(softc, 1);
491 }
492 
493 static void
494 amdvi_dump_cmds(struct amdvi_softc *softc, int count)
495 {
496 	struct amdvi_ctrl *ctrl;
497 	struct amdvi_cmd *cmd;
498 	int off, i;
499 
500 	ctrl = softc->ctrl;
501 	device_printf(softc->dev, "Dump last %d command(s):\n", count);
502 	/*
503 	 * If h/w is stuck in completion, it is the previous command,
504 	 * start dumping from previous command onward.
505 	 */
506 	off = MOD_DEC(ctrl->cmd_head, sizeof(struct amdvi_cmd),
507 	    softc->cmd_max);
508 	for (i = 0; off != ctrl->cmd_tail && i < count; i++) {
509 		cmd = (struct amdvi_cmd *)((uint8_t *)softc->cmd + off);
510 		printf("  [CMD%d, off:0x%x] opcode= 0x%x 0x%x"
511 		    " 0x%x 0x%lx\n", i, off, cmd->opcode,
512 		    cmd->word0, cmd->word1, cmd->addr);
513 		off = MOD_INC(off, sizeof(struct amdvi_cmd), softc->cmd_max);
514 	}
515 }
516 
517 static int
518 amdvi_init_event(struct amdvi_softc *softc)
519 {
520 	struct amdvi_ctrl *ctrl;
521 
522 	ctrl = softc->ctrl;
523 	ctrl->event.len = 8;
524 	softc->event_max = 1 << ctrl->event.len;
525 	softc->event = malloc(sizeof(struct amdvi_event) *
526 	    softc->event_max, M_AMDVI, M_WAITOK | M_ZERO);
527 	if ((uintptr_t)softc->event & PAGE_MASK) {
528 		device_printf(softc->dev, "Event buffer not aligned on page.");
529 		return (false);
530 	}
531 	ctrl->event.base = vtophys(softc->event) / PAGE_SIZE;
532 
533 	/* Reset the pointers. */
534 	ctrl->evt_head = 0;
535 	ctrl->evt_tail = 0;
536 
537 	return (0);
538 }
539 
540 static inline void
541 amdvi_decode_evt_flag(uint16_t flag)
542 {
543 
544 	flag &= AMDVI_EVENT_FLAG_MASK;
545 	printf(" 0x%b]\n", flag,
546 		"\020"
547 		"\001GN"
548 		"\002NX"
549 		"\003US"
550 		"\004I"
551 		"\005PR"
552 		"\006RW"
553 		"\007PE"
554 		"\010RZ"
555 		"\011TR"
556 		);
557 }
558 
559 /* See section 2.5.4 of AMD IOMMU spec ver 2.62.*/
560 static inline void
561 amdvi_decode_evt_flag_type(uint8_t type)
562 {
563 
564 	switch (AMDVI_EVENT_FLAG_TYPE(type)) {
565 	case 0:
566 		printf("RSVD\n");
567 		break;
568 	case 1:
569 		printf("Master Abort\n");
570 		break;
571 	case 2:
572 		printf("Target Abort\n");
573 		break;
574 	case 3:
575 		printf("Data Err\n");
576 		break;
577 	default:
578 		break;
579 	}
580 }
581 
582 static void
583 amdvi_decode_inv_dte_evt(uint16_t devid, uint16_t domid, uint64_t addr,
584     uint16_t flag)
585 {
586 
587 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
588 	    " Addr:0x%lx",
589 	    devid, domid, addr);
590 	amdvi_decode_evt_flag(flag);
591 }
592 
593 static void
594 amdvi_decode_pf_evt(uint16_t devid, uint16_t domid, uint64_t addr,
595     uint16_t flag)
596 {
597 
598 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
599 	    " Addr:0x%lx",
600 	    devid, domid, addr);
601 	amdvi_decode_evt_flag(flag);
602 }
603 
604 static void
605 amdvi_decode_dte_hwerr_evt(uint16_t devid, uint16_t domid,
606     uint64_t addr, uint16_t flag)
607 {
608 
609 	printf("\t[DEV_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
610 	    " Addr:0x%lx", devid, domid, addr);
611 	amdvi_decode_evt_flag(flag);
612 	amdvi_decode_evt_flag_type(flag);
613 }
614 
615 static void
616 amdvi_decode_page_hwerr_evt(uint16_t devid, uint16_t domid, uint64_t addr,
617     uint16_t flag)
618 {
619 
620 	printf("\t[PAGE_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
621 	    " Addr:0x%lx", devid, domid, addr);
622 	amdvi_decode_evt_flag(flag);
623 	amdvi_decode_evt_flag_type(AMDVI_EVENT_FLAG_TYPE(flag));
624 }
625 
626 static void
627 amdvi_decode_evt(struct amdvi_event *evt)
628 {
629 	struct amdvi_cmd *cmd;
630 
631 	switch (evt->opcode) {
632 	case AMDVI_EVENT_INVALID_DTE:
633 		amdvi_decode_inv_dte_evt(evt->devid, evt->pasid_domid,
634 		    evt->addr, evt->flag);
635 		break;
636 
637 	case AMDVI_EVENT_PFAULT:
638 		amdvi_decode_pf_evt(evt->devid, evt->pasid_domid,
639 		    evt->addr, evt->flag);
640 		break;
641 
642 	case AMDVI_EVENT_DTE_HW_ERROR:
643 		amdvi_decode_dte_hwerr_evt(evt->devid, evt->pasid_domid,
644 		    evt->addr, evt->flag);
645 		break;
646 
647 	case AMDVI_EVENT_PAGE_HW_ERROR:
648 		amdvi_decode_page_hwerr_evt(evt->devid, evt->pasid_domid,
649 		    evt->addr, evt->flag);
650 		break;
651 
652 	case AMDVI_EVENT_ILLEGAL_CMD:
653 		/* FALL THROUGH */
654 	case AMDVI_EVENT_CMD_HW_ERROR:
655 		printf("\t[%s EVT]\n", (evt->opcode == AMDVI_EVENT_ILLEGAL_CMD) ?
656 		    "ILLEGAL CMD" : "CMD HW ERR");
657 		cmd = (struct amdvi_cmd *)PHYS_TO_DMAP(evt->addr);
658 		printf("\tCMD opcode= 0x%x 0x%x 0x%x 0x%lx\n",
659 		    cmd->opcode, cmd->word0, cmd->word1, cmd->addr);
660 		break;
661 
662 	case AMDVI_EVENT_IOTLB_TIMEOUT:
663 		printf("\t[IOTLB_INV_TIMEOUT devid:0x%x addr:0x%lx]\n",
664 		    evt->devid, evt->addr);
665 		break;
666 
667 	case AMDVI_EVENT_INVALID_DTE_REQ:
668 		printf("\t[INV_DTE devid:0x%x addr:0x%lx type:0x%x tr:%d]\n",
669 		    evt->devid, evt->addr, evt->flag >> 9,
670 		    (evt->flag >> 8) & 1);
671 		break;
672 
673 	case AMDVI_EVENT_INVALID_PPR_REQ:
674 	case AMDVI_EVENT_COUNTER_ZERO:
675 		printf("AMD-Vi: v2 events.\n");
676 		break;
677 
678 	default:
679 		printf("Unsupported AMD-Vi event:%d\n", evt->opcode);
680 	}
681 }
682 
683 static void
684 amdvi_print_events(struct amdvi_softc *softc)
685 {
686 	struct amdvi_ctrl *ctrl;
687 	struct amdvi_event *event;
688 	int i, size;
689 
690 	ctrl = softc->ctrl;
691 	size = sizeof(struct amdvi_event);
692 	for (i = 0; i < softc->event_max; i++) {
693 		event = &softc->event[ctrl->evt_head / size];
694 		if (!event->opcode)
695 			break;
696 		device_printf(softc->dev, "\t[Event%d: Head:0x%x Tail:0x%x]\n",
697 		    i, ctrl->evt_head, ctrl->evt_tail);
698 		amdvi_decode_evt(event);
699 		ctrl->evt_head = MOD_INC(ctrl->evt_head, size,
700 		    softc->event_max);
701 	}
702 }
703 
704 static int
705 amdvi_init_dte(struct amdvi_softc *softc)
706 {
707 	struct amdvi_ctrl *ctrl;
708 
709 	ctrl = softc->ctrl;
710 	ctrl->dte.base = vtophys(amdvi_dte) / PAGE_SIZE;
711 	ctrl->dte.size = 0x1FF;		/* 2MB device table. */
712 
713 	return (0);
714 }
715 
716 /*
717  * Not all capabilities of IOMMU are available in ACPI IVHD flag
718  * or EFR entry, read directly from device.
719  */
720 static int
721 amdvi_print_pci_cap(device_t dev)
722 {
723 	struct amdvi_softc *softc;
724 	uint32_t off, cap;
725 
726 	softc = device_get_softc(dev);
727 	off = softc->cap_off;
728 
729 	/*
730 	 * Section 3.7.1 of IOMMU sepc rev 2.0.
731 	 * Read capability from device.
732 	 */
733 	cap = amdvi_pci_read(softc, off);
734 
735 	/* Make sure capability type[18:16] is 3. */
736 	KASSERT((((cap >> 16) & 0x7) == 0x3),
737 	    ("Not a IOMMU capability 0x%x@0x%x", cap, off));
738 
739 	softc->pci_cap = cap >> 24;
740 	device_printf(softc->dev, "PCI cap 0x%x@0x%x feature:%b\n",
741 	    cap, off, softc->pci_cap,
742 	    "\20\1IOTLB\2HT\3NPCache\4EFR\5CapExt");
743 
744 	return (0);
745 }
746 
747 static void
748 amdvi_event_intr(void *arg)
749 {
750 	struct amdvi_softc *softc;
751 	struct amdvi_ctrl *ctrl;
752 
753 	softc = (struct amdvi_softc *)arg;
754 	ctrl = softc->ctrl;
755 	device_printf(softc->dev, "EVT INTR %ld Status:0x%x"
756 	    " EVT Head:0x%x Tail:0x%x]\n", softc->event_intr_cnt++,
757 	    ctrl->status, ctrl->evt_head, ctrl->evt_tail);
758 	printf("  [CMD Total 0x%lx] Tail:0x%x, Head:0x%x.\n",
759 	    softc->total_cmd, ctrl->cmd_tail, ctrl->cmd_head);
760 
761 	amdvi_print_events(softc);
762 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
763 }
764 
765 static void
766 amdvi_free_evt_intr_res(device_t dev)
767 {
768 
769 	struct amdvi_softc *softc;
770 	device_t mmio_dev;
771 
772 	softc = device_get_softc(dev);
773 	mmio_dev = softc->pci_dev;
774 
775 	IVHD_TEARDOWN_INTR(mmio_dev);
776 }
777 
778 static bool
779 amdvi_alloc_intr_resources(struct amdvi_softc *softc)
780 {
781 	struct amdvi_ctrl *ctrl;
782 	device_t dev, mmio_dev;
783 	int err;
784 
785 	dev = softc->dev;
786 	mmio_dev = softc->pci_dev;
787 
788 	/* Clear interrupt status bits. */
789 	ctrl = softc->ctrl;
790 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
791 
792 	err = IVHD_SETUP_INTR(mmio_dev, amdvi_event_intr, softc, "fault");
793 	if (err)
794 		device_printf(dev, "Interrupt setup failed on %s\n",
795 		    device_get_nameunit(mmio_dev));
796 	return (err);
797 }
798 
799 static void
800 amdvi_print_dev_cap(struct amdvi_softc *softc)
801 {
802 	struct ivhd_dev_cfg *cfg;
803 	int i;
804 
805 	cfg = softc->dev_cfg;
806 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
807 		device_printf(softc->dev, "device [0x%x - 0x%x] "
808 		    "config:%b%s\n", cfg->start_id, cfg->end_id,
809 		    cfg->data,
810 		    "\020\001INIT\002ExtInt\003NMI"
811 		    "\007LINT0\010LINT1",
812 		    cfg->enable_ats ? "ATS enabled" : "");
813 		cfg++;
814 	}
815 }
816 
817 static int
818 amdvi_handle_sysctl(SYSCTL_HANDLER_ARGS)
819 {
820 	struct amdvi_softc *softc;
821 	int result, type, error = 0;
822 
823 	softc = (struct amdvi_softc *)arg1;
824 	type = arg2;
825 
826 	switch (type) {
827 	case 0:
828 		result = softc->ctrl->cmd_head;
829 		error = sysctl_handle_int(oidp, &result, 0,
830 		    req);
831 		break;
832 	case 1:
833 		result = softc->ctrl->cmd_tail;
834 		error = sysctl_handle_int(oidp, &result, 0,
835 		    req);
836 		break;
837 	case 2:
838 		result = softc->ctrl->evt_head;
839 		error = sysctl_handle_int(oidp, &result, 0,
840 		    req);
841 		break;
842 	case 3:
843 		result = softc->ctrl->evt_tail;
844 		error = sysctl_handle_int(oidp, &result, 0,
845 		    req);
846 		break;
847 
848 	default:
849 		device_printf(softc->dev, "Unknown sysctl:%d\n", type);
850 	}
851 
852 	return (error);
853 }
854 
855 static void
856 amdvi_add_sysctl(struct amdvi_softc *softc)
857 {
858 	struct sysctl_oid_list *child;
859 	struct sysctl_ctx_list *ctx;
860 	device_t dev;
861 
862 	dev = softc->dev;
863 	ctx = device_get_sysctl_ctx(dev);
864 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
865 
866 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "event_intr_count", CTLFLAG_RD,
867 	    &softc->event_intr_cnt, "Event interrupt count");
868 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "command_count", CTLFLAG_RD,
869 	    &softc->total_cmd, "Command submitted count");
870 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "pci_rid", CTLFLAG_RD,
871 	    &softc->pci_rid, 0, "IOMMU RID");
872 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_head",
873 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 0,
874 	    amdvi_handle_sysctl, "IU", "Command head");
875 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_tail",
876 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 1,
877 	    amdvi_handle_sysctl, "IU", "Command tail");
878 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_head",
879 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 2,
880 	    amdvi_handle_sysctl, "IU", "Command head");
881 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_tail",
882 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 3,
883 	    amdvi_handle_sysctl, "IU", "Command tail");
884 }
885 
886 int
887 amdvi_setup_hw(struct amdvi_softc *softc)
888 {
889 	device_t dev;
890 	int status;
891 
892 	dev = softc->dev;
893 
894 	amdvi_hw_enable_iotlb(softc);
895 
896 	amdvi_print_dev_cap(softc);
897 
898 	if ((status = amdvi_print_pci_cap(dev)) != 0) {
899 		device_printf(dev, "PCI capability.\n");
900 		return (status);
901 	}
902 	if ((status = amdvi_init_cmd(softc)) != 0) {
903 		device_printf(dev, "Couldn't configure command buffer.\n");
904 		return (status);
905 	}
906 	if ((status = amdvi_init_event(softc)) != 0) {
907 		device_printf(dev, "Couldn't configure event buffer.\n");
908 		return (status);
909 	}
910 	if ((status = amdvi_init_dte(softc)) != 0) {
911 		device_printf(dev, "Couldn't configure device table.\n");
912 		return (status);
913 	}
914 	if ((status = amdvi_alloc_intr_resources(softc)) != 0) {
915 		return (status);
916 	}
917 	amdvi_add_sysctl(softc);
918 	return (0);
919 }
920 
921 int
922 amdvi_teardown_hw(struct amdvi_softc *softc)
923 {
924 	device_t dev;
925 
926 	dev = softc->dev;
927 
928 	/*
929 	 * Called after disable, h/w is stopped by now, free all the resources.
930 	 */
931 	amdvi_free_evt_intr_res(dev);
932 
933 	if (softc->cmd)
934 		free(softc->cmd, M_AMDVI);
935 
936 	if (softc->event)
937 		free(softc->event, M_AMDVI);
938 
939 	return (0);
940 }
941 
942 /*********** bhyve interfaces *********************/
943 static int
944 amdvi_init(void)
945 {
946 	if (!ivhd_count) {
947 		return (EIO);
948 	}
949 	if (!amdvi_enable_user && ivhd_count) {
950 		printf("bhyve: Found %d AMD-Vi/IOMMU device(s), "
951 		    "use hw.vmm.amdvi.enable=1 to enable pass-through.\n",
952 		    ivhd_count);
953 		return (EINVAL);
954 	}
955 	return (0);
956 }
957 
958 static void
959 amdvi_cleanup(void)
960 {
961 	/* Nothing. */
962 }
963 
964 static uint16_t
965 amdvi_domainId(void)
966 {
967 
968 	/*
969 	 * If we hit maximum domain limit, rollover leaving host
970 	 * domain(0).
971 	 * XXX: make sure that this domain is not used.
972 	 */
973 	if (amdvi_dom_id == AMDVI_MAX_DOMAIN)
974 		amdvi_dom_id = 1;
975 
976 	return ((uint16_t)amdvi_dom_id++);
977 }
978 
979 static void
980 amdvi_do_inv_domain(uint16_t domain_id, bool create)
981 {
982 	struct amdvi_softc *softc;
983 	int i;
984 
985 	for (i = 0; i < ivhd_count; i++) {
986 		softc = device_get_softc(ivhd_devs[i]);
987 		KASSERT(softc, ("softc is NULL"));
988 		/*
989 		 * If not present pages are cached, invalidate page after
990 		 * creating domain.
991 		 */
992 #if 0
993 		if (create && ((softc->pci_cap & AMDVI_PCI_CAP_NPCACHE) == 0))
994 			continue;
995 #endif
996 		amdvi_inv_domain(softc, domain_id);
997 		amdvi_wait(softc);
998 	}
999 }
1000 
1001 static void *
1002 amdvi_create_domain(vm_paddr_t maxaddr)
1003 {
1004 	struct amdvi_domain *dom;
1005 
1006 	dom = malloc(sizeof(struct amdvi_domain), M_AMDVI, M_ZERO | M_WAITOK);
1007 	dom->id = amdvi_domainId();
1008 	//dom->maxaddr = maxaddr;
1009 #ifdef AMDVI_DEBUG_CMD
1010 	printf("Created domain #%d\n", dom->id);
1011 #endif
1012 	/*
1013 	 * Host domain(#0) don't create translation table.
1014 	 */
1015 	if (dom->id || amdvi_host_ptp)
1016 		dom->ptp = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1017 
1018 	dom->ptp_level = amdvi_ptp_level;
1019 
1020 	amdvi_do_inv_domain(dom->id, true);
1021 	SLIST_INSERT_HEAD(&dom_head, dom, next);
1022 
1023 	return (dom);
1024 }
1025 
1026 static void
1027 amdvi_free_ptp(uint64_t *ptp, int level)
1028 {
1029 	int i;
1030 
1031 	if (level < 1)
1032 		return;
1033 
1034 	for (i = 0; i < NPTEPG ; i++) {
1035 		if ((ptp[i] & AMDVI_PT_PRESENT) == 0)
1036 			continue;
1037 		/* XXX: Add super-page or PTE mapping > 4KB. */
1038 #ifdef notyet
1039 		/* Super-page mapping. */
1040 		if (AMDVI_PD_SUPER(ptp[i]))
1041 			continue;
1042 #endif
1043 
1044 		amdvi_free_ptp((uint64_t *)PHYS_TO_DMAP(ptp[i]
1045 		    & AMDVI_PT_MASK), level - 1);
1046 	}
1047 
1048 	free(ptp, M_AMDVI);
1049 }
1050 
1051 static void
1052 amdvi_destroy_domain(void *arg)
1053 {
1054 	struct amdvi_domain *domain;
1055 
1056 	domain = (struct amdvi_domain *)arg;
1057 	KASSERT(domain, ("domain is NULL"));
1058 #ifdef AMDVI_DEBUG_CMD
1059 	printf("Destroying domain %d\n", domain->id);
1060 #endif
1061 	if (domain->ptp)
1062 		amdvi_free_ptp(domain->ptp, domain->ptp_level);
1063 
1064 	amdvi_do_inv_domain(domain->id, false);
1065 	SLIST_REMOVE(&dom_head, domain, amdvi_domain, next);
1066 	free(domain, M_AMDVI);
1067 }
1068 
1069 static uint64_t
1070 amdvi_set_pt(uint64_t *pt, int level, vm_paddr_t gpa,
1071     vm_paddr_t hpa, uint64_t pg_size, bool create)
1072 {
1073 	uint64_t *page, pa;
1074 	int shift, index;
1075 	const int PT_SHIFT = 9;
1076 	const int PT_INDEX_MASK = (1 << PT_SHIFT) - 1;	/* Based on PT_SHIFT */
1077 
1078 	if (!pg_size)
1079 		return (0);
1080 
1081 	if (hpa & (pg_size - 1)) {
1082 		printf("HPA is not size aligned.\n");
1083 		return (0);
1084 	}
1085 	if (gpa & (pg_size - 1)) {
1086 		printf("HPA is not size aligned.\n");
1087 		return (0);
1088 	}
1089 	shift = PML4SHIFT;
1090 	while ((shift > PAGE_SHIFT) && (pg_size < (1UL << shift))) {
1091 		index = (gpa >> shift) & PT_INDEX_MASK;
1092 
1093 		if ((pt[index] == 0) && create) {
1094 			page = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1095 			pa = vtophys(page);
1096 			pt[index] = pa | AMDVI_PT_PRESENT | AMDVI_PT_RW |
1097 			    ((level - 1) << AMDVI_PD_LEVEL_SHIFT);
1098 		}
1099 #ifdef AMDVI_DEBUG_PTE
1100 		if ((gpa % 0x1000000) == 0)
1101 			printf("[level%d, shift = %d]PTE:0x%lx\n",
1102 			    level, shift, pt[index]);
1103 #endif
1104 #define PTE2PA(x)	((uint64_t)(x) & AMDVI_PT_MASK)
1105 		pa = PTE2PA(pt[index]);
1106 		pt = (uint64_t *)PHYS_TO_DMAP(pa);
1107 		shift -= PT_SHIFT;
1108 		level--;
1109 	}
1110 
1111 	/* Leaf entry. */
1112 	index = (gpa >> shift) & PT_INDEX_MASK;
1113 
1114 	if (create) {
1115 		pt[index] = hpa | AMDVI_PT_RW | AMDVI_PT_PRESENT;
1116 	} else
1117 		pt[index] = 0;
1118 
1119 #ifdef AMDVI_DEBUG_PTE
1120 	if ((gpa % 0x1000000) == 0)
1121 		printf("[Last level%d, shift = %d]PTE:0x%lx\n",
1122 		    level, shift, pt[index]);
1123 #endif
1124 	return (1ULL << shift);
1125 }
1126 
1127 static uint64_t
1128 amdvi_update_mapping(struct amdvi_domain *domain, vm_paddr_t gpa,
1129     vm_paddr_t hpa, uint64_t size, bool create)
1130 {
1131 	uint64_t mapped, *ptp, len;
1132 	int level;
1133 
1134 	KASSERT(domain, ("domain is NULL"));
1135 	level = domain->ptp_level;
1136 	KASSERT(level, ("Page table level is 0"));
1137 
1138 	ptp = domain->ptp;
1139 	KASSERT(ptp, ("PTP is NULL"));
1140 	mapped = 0;
1141 	while (mapped < size) {
1142 		len = amdvi_set_pt(ptp, level, gpa + mapped, hpa + mapped,
1143 		    PAGE_SIZE, create);
1144 		if (!len) {
1145 			printf("Error: Couldn't map HPA:0x%lx GPA:0x%lx\n",
1146 			    hpa, gpa);
1147 			return (0);
1148 		}
1149 		mapped += len;
1150 	}
1151 
1152 	return (mapped);
1153 }
1154 
1155 static uint64_t
1156 amdvi_create_mapping(void *arg, vm_paddr_t gpa, vm_paddr_t hpa,
1157     uint64_t len)
1158 {
1159 	struct amdvi_domain *domain;
1160 
1161 	domain = (struct amdvi_domain *)arg;
1162 
1163 	if (domain->id && !domain->ptp) {
1164 		printf("ptp is NULL");
1165 		return (-1);
1166 	}
1167 
1168 	/*
1169 	 * If host domain is created w/o page table, skip IOMMU page
1170 	 * table set-up.
1171 	 */
1172 	if (domain->ptp)
1173 		return (amdvi_update_mapping(domain, gpa, hpa, len, true));
1174 	else
1175 		return (len);
1176 }
1177 
1178 static uint64_t
1179 amdvi_remove_mapping(void *arg, vm_paddr_t gpa, uint64_t len)
1180 {
1181 	struct amdvi_domain *domain;
1182 
1183 	domain = (struct amdvi_domain *)arg;
1184 	/*
1185 	 * If host domain is created w/o page table, skip IOMMU page
1186 	 * table set-up.
1187 	 */
1188 	if (domain->ptp)
1189 		return (amdvi_update_mapping(domain, gpa, 0, len, false));
1190 	return
1191 	    (len);
1192 }
1193 
1194 static struct amdvi_softc *
1195 amdvi_find_iommu(uint16_t devid)
1196 {
1197 	struct amdvi_softc *softc;
1198 	int i, j;
1199 
1200 	for (i = 0; i < ivhd_count; i++) {
1201 		softc = device_get_softc(ivhd_devs[i]);
1202 		for (j = 0; j < softc->dev_cfg_cnt; j++)
1203 			if ((devid >= softc->dev_cfg[j].start_id) &&
1204 			    (devid <= softc->dev_cfg[j].end_id))
1205 				return (softc);
1206 	}
1207 
1208 	return (NULL);
1209 }
1210 
1211 /*
1212  * Set-up device table entry.
1213  * IOMMU spec Rev 2.0, section 3.2.2.2, some of the fields must
1214  * be set concurrently, e.g. read and write bits.
1215  */
1216 static void
1217 amdvi_set_dte(struct amdvi_domain *domain, struct amdvi_softc *softc,
1218     uint16_t devid, bool enable)
1219 {
1220 	struct amdvi_dte* temp;
1221 
1222 	KASSERT(domain, ("domain is NULL for pci_rid:0x%x\n", devid));
1223 	KASSERT(softc, ("softc is NULL for pci_rid:0x%x\n", devid));
1224 
1225 	temp = &amdvi_dte[devid];
1226 
1227 #ifdef AMDVI_ATS_ENABLE
1228 	/* If IOMMU and device support IOTLB, enable it. */
1229 	if (amdvi_dev_support_iotlb(softc, devid) && softc->iotlb)
1230 		temp->iotlb_enable = 1;
1231 #endif
1232 
1233 	/* Avoid duplicate I/O faults. */
1234 	temp->sup_second_io_fault = 1;
1235 	temp->sup_all_io_fault = amdvi_disable_io_fault;
1236 
1237 	temp->dt_valid = 1;
1238 	temp->domain_id = domain->id;
1239 
1240 	if (enable) {
1241 		if (domain->ptp) {
1242 			temp->pt_base = vtophys(domain->ptp) >> 12;
1243 			temp->pt_level = amdvi_ptp_level;
1244 		}
1245 		/*
1246 		 * XXX: Page table valid[TV] bit must be set even if host domain
1247 		 * page tables are not enabled.
1248 		 */
1249 		temp->pt_valid = 1;
1250 		temp->read_allow = 1;
1251 		temp->write_allow = 1;
1252 	}
1253 }
1254 
1255 static void
1256 amdvi_inv_device(struct amdvi_softc *softc, uint16_t devid)
1257 {
1258 	KASSERT(softc, ("softc is NULL"));
1259 
1260 	amdvi_cmd_inv_dte(softc, devid);
1261 #ifdef AMDVI_ATS_ENABLE
1262 	if (amdvi_dev_support_iotlb(softc, devid))
1263 		amdvi_cmd_inv_iotlb(softc, devid);
1264 #endif
1265 	amdvi_wait(softc);
1266 }
1267 
1268 static void
1269 amdvi_add_device(void *arg, uint16_t devid)
1270 {
1271 	struct amdvi_domain *domain;
1272 	struct amdvi_softc *softc;
1273 
1274 	domain = (struct amdvi_domain *)arg;
1275 	KASSERT(domain != NULL, ("domain is NULL"));
1276 #ifdef AMDVI_DEBUG_CMD
1277 	printf("Assigning device(%d.%d.%d) to domain:%d\n",
1278 	    RID2PCI_STR(devid), domain->id);
1279 #endif
1280 	softc = amdvi_find_iommu(devid);
1281 	if (softc == NULL)
1282 		return;
1283 	amdvi_set_dte(domain, softc, devid, true);
1284 	amdvi_inv_device(softc, devid);
1285 }
1286 
1287 static void
1288 amdvi_remove_device(void *arg, uint16_t devid)
1289 {
1290 	struct amdvi_domain *domain;
1291 	struct amdvi_softc *softc;
1292 
1293 	domain = (struct amdvi_domain *)arg;
1294 #ifdef AMDVI_DEBUG_CMD
1295 	printf("Remove device(0x%x) from domain:%d\n",
1296 	       devid, domain->id);
1297 #endif
1298 	softc = amdvi_find_iommu(devid);
1299 	if (softc == NULL)
1300 		return;
1301 	amdvi_set_dte(domain, softc, devid, false);
1302 	amdvi_inv_device(softc, devid);
1303 }
1304 
1305 static void
1306 amdvi_enable(void)
1307 {
1308 	struct amdvi_ctrl *ctrl;
1309 	struct amdvi_softc *softc;
1310 	uint64_t val;
1311 	int i;
1312 
1313 	for (i = 0; i < ivhd_count; i++) {
1314 		softc = device_get_softc(ivhd_devs[i]);
1315 		KASSERT(softc, ("softc is NULL\n"));
1316 		ctrl = softc->ctrl;
1317 		KASSERT(ctrl, ("ctrl is NULL\n"));
1318 
1319 		val = (	AMDVI_CTRL_EN		|
1320 			AMDVI_CTRL_CMD		|
1321 			AMDVI_CTRL_ELOG		|
1322 			AMDVI_CTRL_ELOGINT	|
1323 			AMDVI_CTRL_INV_TO_1S);
1324 
1325 		if (softc->ivhd_flag & IVHD_FLAG_COH)
1326 			val |= AMDVI_CTRL_COH;
1327 		if (softc->ivhd_flag & IVHD_FLAG_HTT)
1328 			val |= AMDVI_CTRL_HTT;
1329 		if (softc->ivhd_flag & IVHD_FLAG_RPPW)
1330 			val |= AMDVI_CTRL_RPPW;
1331 		if (softc->ivhd_flag & IVHD_FLAG_PPW)
1332 			val |= AMDVI_CTRL_PPW;
1333 		if (softc->ivhd_flag & IVHD_FLAG_ISOC)
1334 			val |= AMDVI_CTRL_ISOC;
1335 
1336 		ctrl->control = val;
1337 	}
1338 }
1339 
1340 static void
1341 amdvi_disable(void)
1342 {
1343 	struct amdvi_ctrl *ctrl;
1344 	struct amdvi_softc *softc;
1345 	int i;
1346 
1347 	for (i = 0; i < ivhd_count; i++) {
1348 		softc = device_get_softc(ivhd_devs[i]);
1349 		KASSERT(softc, ("softc is NULL\n"));
1350 		ctrl = softc->ctrl;
1351 		KASSERT(ctrl, ("ctrl is NULL\n"));
1352 
1353 		ctrl->control = 0;
1354 	}
1355 }
1356 
1357 static void
1358 amdvi_invalidate_tlb(void *arg)
1359 {
1360 	struct amdvi_domain *domain;
1361 
1362 	domain = (struct amdvi_domain *)arg;
1363 	KASSERT(domain, ("domain is NULL"));
1364 	amdvi_do_inv_domain(domain->id, false);
1365 }
1366 
1367 const struct iommu_ops iommu_ops_amd = {
1368 	.init = amdvi_init,
1369 	.cleanup = amdvi_cleanup,
1370 	.enable = amdvi_enable,
1371 	.disable = amdvi_disable,
1372 	.create_domain = amdvi_create_domain,
1373 	.destroy_domain = amdvi_destroy_domain,
1374 	.create_mapping = amdvi_create_mapping,
1375 	.remove_mapping = amdvi_remove_mapping,
1376 	.add_device = amdvi_add_device,
1377 	.remove_device = amdvi_remove_device,
1378 	.invalidate_tlb = amdvi_invalidate_tlb
1379 };
1380