xref: /freebsd/sys/amd64/vmm/amd/amdvi_hw.c (revision 2f513db72b034fd5ef7f080b11be5c711c15186a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/bus.h>
35 #include <sys/kernel.h>
36 #include <sys/module.h>
37 #include <sys/malloc.h>
38 #include <sys/pcpu.h>
39 #include <sys/rman.h>
40 #include <sys/smp.h>
41 #include <sys/sysctl.h>
42 
43 #include <vm/vm.h>
44 #include <vm/pmap.h>
45 
46 #include <dev/pci/pcivar.h>
47 #include <dev/pci/pcireg.h>
48 
49 #include <machine/resource.h>
50 #include <machine/vmm.h>
51 #include <machine/pmap.h>
52 #include <machine/vmparam.h>
53 #include <machine/pci_cfgreg.h>
54 
55 #include "pcib_if.h"
56 
57 #include "io/iommu.h"
58 #include "amdvi_priv.h"
59 
60 SYSCTL_DECL(_hw_vmm);
61 SYSCTL_NODE(_hw_vmm, OID_AUTO, amdvi, CTLFLAG_RW, NULL, 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 
359 	cmd->opcode = AMDVI_INVD_PAGE_OPCODE;
360 	cmd->word1 = domain_id;
361 	/*
362 	 * Invalidate all addresses for this domain.
363 	 */
364 	cmd->addr = addr;
365 	cmd->addr |= pde ? AMDVI_INVD_PAGE_PDE : 0;
366 	cmd->addr |= page ? AMDVI_INVD_PAGE_S : 0;
367 
368 	amdvi_update_cmd_tail(softc);
369 }
370 
371 #ifdef AMDVI_ATS_ENABLE
372 /* Invalidate device IOTLB. */
373 static void
374 amdvi_cmd_inv_iotlb(struct amdvi_softc *softc, uint16_t devid)
375 {
376 	struct amdvi_cmd *cmd;
377 	int qlen;
378 
379 	if (!softc->iotlb)
380 		return;
381 
382 	qlen = amdvi_find_ats_qlen(devid);
383 	if (qlen < 0) {
384 		panic("AMDVI: Invalid ATS qlen(%d) for device %d.%d.%d\n",
385 		      qlen, RID2PCI_STR(devid));
386 	}
387 	cmd = amdvi_get_cmd_tail(softc);
388 	KASSERT(cmd != NULL, ("Cmd is NULL"));
389 
390 #ifdef AMDVI_DEBUG_CMD
391 	device_printf(softc->dev, "Invalidate IOTLB devID 0x%x"
392 		      " Qlen:%d\n", devid, qlen);
393 #endif
394 	cmd->opcode = AMDVI_INVD_IOTLB_OPCODE;
395 	cmd->word0 = devid;
396 	cmd->word1 = qlen;
397 	cmd->addr = AMDVI_INVD_IOTLB_ALL_ADDR |
398 		AMDVI_INVD_IOTLB_S;
399 	amdvi_update_cmd_tail(softc);
400 }
401 #endif
402 
403 #ifdef notyet				/* For Interrupt Remap. */
404 static void
405 amdvi_cmd_inv_intr_map(struct amdvi_softc *softc,
406 		       uint16_t devid)
407 {
408 	struct amdvi_cmd *cmd;
409 
410 	cmd = amdvi_get_cmd_tail(softc);
411 	KASSERT(cmd != NULL, ("Cmd is NULL"));
412 	cmd->opcode = AMDVI_INVD_INTR_OPCODE;
413 	cmd->word0 = devid;
414 	amdvi_update_cmd_tail(softc);
415 #ifdef AMDVI_DEBUG_CMD
416 	device_printf(softc->dev, "Invalidate INTR map of devID 0x%x\n", devid);
417 #endif
418 }
419 #endif
420 
421 /* Invalidate domain using INVALIDATE_IOMMU_PAGES command. */
422 static void
423 amdvi_inv_domain(struct amdvi_softc *softc, uint16_t domain_id)
424 {
425 	struct amdvi_cmd *cmd;
426 
427 	cmd = amdvi_get_cmd_tail(softc);
428 	KASSERT(cmd != NULL, ("Cmd is NULL"));
429 
430 	/*
431 	 * See section 3.3.3 of IOMMU spec rev 2.0, software note
432 	 * for invalidating domain.
433 	 */
434 	amdvi_cmd_inv_iommu_pages(softc, domain_id, AMDVI_INVD_PAGE_ALL_ADDR,
435 				false, true, true);
436 
437 #ifdef AMDVI_DEBUG_CMD
438 	device_printf(softc->dev, "Invalidate domain:0x%x\n", domain_id);
439 
440 #endif
441 }
442 
443 static	bool
444 amdvi_cmp_wait(struct amdvi_softc *softc)
445 {
446 	struct amdvi_ctrl *ctrl;
447 	const uint64_t VERIFY = 0xA5A5;
448 	volatile uint64_t *read;
449 	int i;
450 	bool status;
451 
452 	ctrl = softc->ctrl;
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 = (off + sizeof(struct amdvi_cmd)) %
518 		    (softc->cmd_max * sizeof(struct amdvi_cmd));
519 	}
520 }
521 
522 static int
523 amdvi_init_event(struct amdvi_softc *softc)
524 {
525 	struct amdvi_ctrl *ctrl;
526 
527 	ctrl = softc->ctrl;
528 	ctrl->event.len = 8;
529 	softc->event_max = 1 << ctrl->event.len;
530 	softc->event = malloc(sizeof(struct amdvi_event) *
531 	    softc->event_max, M_AMDVI, M_WAITOK | M_ZERO);
532 	if ((uintptr_t)softc->event & PAGE_MASK) {
533 		device_printf(softc->dev, "Event buffer not aligned on page.");
534 		return (false);
535 	}
536 	ctrl->event.base = vtophys(softc->event) / PAGE_SIZE;
537 
538 	/* Reset the pointers. */
539 	ctrl->evt_head = 0;
540 	ctrl->evt_tail = 0;
541 
542 	return (0);
543 }
544 
545 static inline void
546 amdvi_decode_evt_flag(uint16_t flag)
547 {
548 
549 	flag &= AMDVI_EVENT_FLAG_MASK;
550 	printf(" 0x%b]\n", flag,
551 		"\020"
552 		"\001GN"
553 		"\002NX"
554 		"\003US"
555 		"\004I"
556 		"\005PR"
557 		"\006RW"
558 		"\007PE"
559 		"\010RZ"
560 		"\011TR"
561 		);
562 }
563 
564 /* See section 2.5.4 of AMD IOMMU spec ver 2.62.*/
565 static inline void
566 amdvi_decode_evt_flag_type(uint8_t type)
567 {
568 
569 	switch (AMDVI_EVENT_FLAG_TYPE(type)) {
570 	case 0:
571 		printf("RSVD\n");
572 		break;
573 	case 1:
574 		printf("Master Abort\n");
575 		break;
576 	case 2:
577 		printf("Target Abort\n");
578 		break;
579 	case 3:
580 		printf("Data Err\n");
581 		break;
582 	default:
583 		break;
584 	}
585 }
586 
587 static void
588 amdvi_decode_inv_dte_evt(uint16_t devid, uint16_t domid, uint64_t addr,
589     uint16_t flag)
590 {
591 
592 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
593 	    " Addr:0x%lx",
594 	    devid, domid, addr);
595 	amdvi_decode_evt_flag(flag);
596 }
597 
598 static void
599 amdvi_decode_pf_evt(uint16_t devid, uint16_t domid, uint64_t addr,
600     uint16_t flag)
601 {
602 
603 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
604 	    " Addr:0x%lx",
605 	    devid, domid, addr);
606 	amdvi_decode_evt_flag(flag);
607 }
608 
609 static void
610 amdvi_decode_dte_hwerr_evt(uint16_t devid, uint16_t domid,
611     uint64_t addr, uint16_t flag)
612 {
613 
614 	printf("\t[DEV_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
615 	    " Addr:0x%lx", devid, domid, addr);
616 	amdvi_decode_evt_flag(flag);
617 	amdvi_decode_evt_flag_type(flag);
618 }
619 
620 static void
621 amdvi_decode_page_hwerr_evt(uint16_t devid, uint16_t domid, uint64_t addr,
622     uint16_t flag)
623 {
624 
625 	printf("\t[PAGE_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
626 	    " Addr:0x%lx", devid, domid, addr);
627 	amdvi_decode_evt_flag(flag);
628 	amdvi_decode_evt_flag_type(AMDVI_EVENT_FLAG_TYPE(flag));
629 }
630 
631 static void
632 amdvi_decode_evt(struct amdvi_event *evt)
633 {
634 	struct amdvi_cmd *cmd;
635 
636 	switch (evt->opcode) {
637 	case AMDVI_EVENT_INVALID_DTE:
638 		amdvi_decode_inv_dte_evt(evt->devid, evt->pasid_domid,
639 		    evt->addr, evt->flag);
640 		break;
641 
642 	case AMDVI_EVENT_PFAULT:
643 		amdvi_decode_pf_evt(evt->devid, evt->pasid_domid,
644 		    evt->addr, evt->flag);
645 		break;
646 
647 	case AMDVI_EVENT_DTE_HW_ERROR:
648 		amdvi_decode_dte_hwerr_evt(evt->devid, evt->pasid_domid,
649 		    evt->addr, evt->flag);
650 		break;
651 
652 	case AMDVI_EVENT_PAGE_HW_ERROR:
653 		amdvi_decode_page_hwerr_evt(evt->devid, evt->pasid_domid,
654 		    evt->addr, evt->flag);
655 		break;
656 
657 	case AMDVI_EVENT_ILLEGAL_CMD:
658 		/* FALL THROUGH */
659 	case AMDVI_EVENT_CMD_HW_ERROR:
660 		printf("\t[%s EVT]\n", (evt->opcode == AMDVI_EVENT_ILLEGAL_CMD) ?
661 		    "ILLEGAL CMD" : "CMD HW ERR");
662 		cmd = (struct amdvi_cmd *)PHYS_TO_DMAP(evt->addr);
663 		printf("\tCMD opcode= 0x%x 0x%x 0x%x 0x%lx\n",
664 		    cmd->opcode, cmd->word0, cmd->word1, cmd->addr);
665 		break;
666 
667 	case AMDVI_EVENT_IOTLB_TIMEOUT:
668 		printf("\t[IOTLB_INV_TIMEOUT devid:0x%x addr:0x%lx]\n",
669 		    evt->devid, evt->addr);
670 		break;
671 
672 	case AMDVI_EVENT_INVALID_DTE_REQ:
673 		printf("\t[INV_DTE devid:0x%x addr:0x%lx type:0x%x tr:%d]\n",
674 		    evt->devid, evt->addr, evt->flag >> 9,
675 		    (evt->flag >> 8) & 1);
676 		break;
677 
678 	case AMDVI_EVENT_INVALID_PPR_REQ:
679 	case AMDVI_EVENT_COUNTER_ZERO:
680 		printf("AMD-Vi: v2 events.\n");
681 		break;
682 
683 	default:
684 		printf("Unsupported AMD-Vi event:%d\n", evt->opcode);
685 	}
686 }
687 
688 static void
689 amdvi_print_events(struct amdvi_softc *softc)
690 {
691 	struct amdvi_ctrl *ctrl;
692 	struct amdvi_event *event;
693 	int i, size;
694 
695 	ctrl = softc->ctrl;
696 	size = sizeof(struct amdvi_event);
697 	for (i = 0; i < softc->event_max; i++) {
698 		event = &softc->event[ctrl->evt_head / size];
699 		if (!event->opcode)
700 			break;
701 		device_printf(softc->dev, "\t[Event%d: Head:0x%x Tail:0x%x]\n",
702 		    i, ctrl->evt_head, ctrl->evt_tail);
703 		amdvi_decode_evt(event);
704 		ctrl->evt_head = MOD_INC(ctrl->evt_head, size,
705 		    softc->event_max);
706 	}
707 }
708 
709 static int
710 amdvi_init_dte(struct amdvi_softc *softc)
711 {
712 	struct amdvi_ctrl *ctrl;
713 
714 	ctrl = softc->ctrl;
715 	ctrl->dte.base = vtophys(amdvi_dte) / PAGE_SIZE;
716 	ctrl->dte.size = 0x1FF;		/* 2MB device table. */
717 
718 	return (0);
719 }
720 
721 /*
722  * Not all capabilities of IOMMU are available in ACPI IVHD flag
723  * or EFR entry, read directly from device.
724  */
725 static int
726 amdvi_print_pci_cap(device_t dev)
727 {
728 	struct amdvi_softc *softc;
729 	uint32_t off, cap;
730 
731 
732 	softc = device_get_softc(dev);
733 	off = softc->cap_off;
734 
735 	/*
736 	 * Section 3.7.1 of IOMMU sepc rev 2.0.
737 	 * Read capability from device.
738 	 */
739 	cap = amdvi_pci_read(softc, off);
740 
741 	/* Make sure capability type[18:16] is 3. */
742 	KASSERT((((cap >> 16) & 0x7) == 0x3),
743 	    ("Not a IOMMU capability 0x%x@0x%x", cap, off));
744 
745 	softc->pci_cap = cap >> 24;
746 	device_printf(softc->dev, "PCI cap 0x%x@0x%x feature:%b\n",
747 	    cap, off, softc->pci_cap,
748 	    "\20\1IOTLB\2HT\3NPCache\4EFR\5CapExt");
749 
750 	return (0);
751 }
752 
753 static void
754 amdvi_event_intr(void *arg)
755 {
756 	struct amdvi_softc *softc;
757 	struct amdvi_ctrl *ctrl;
758 
759 	softc = (struct amdvi_softc *)arg;
760 	ctrl = softc->ctrl;
761 	device_printf(softc->dev, "EVT INTR %ld Status:0x%x"
762 	    " EVT Head:0x%x Tail:0x%x]\n", softc->event_intr_cnt++,
763 	    ctrl->status, ctrl->evt_head, ctrl->evt_tail);
764 	printf("  [CMD Total 0x%lx] Tail:0x%x, Head:0x%x.\n",
765 	    softc->total_cmd, ctrl->cmd_tail, ctrl->cmd_head);
766 
767 	amdvi_print_events(softc);
768 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
769 }
770 
771 static void
772 amdvi_free_evt_intr_res(device_t dev)
773 {
774 
775 	struct amdvi_softc *softc;
776 
777 	softc = device_get_softc(dev);
778 	if (softc->event_tag != NULL) {
779 		bus_teardown_intr(dev, softc->event_res, softc->event_tag);
780 	}
781 	if (softc->event_res != NULL) {
782 		bus_release_resource(dev, SYS_RES_IRQ, softc->event_rid,
783 		    softc->event_res);
784 	}
785 	bus_delete_resource(dev, SYS_RES_IRQ, softc->event_rid);
786 	PCIB_RELEASE_MSI(device_get_parent(device_get_parent(dev)),
787 	    dev, 1, &softc->event_irq);
788 }
789 
790 static bool
791 amdvi_alloc_intr_resources(struct amdvi_softc *softc)
792 {
793 	struct amdvi_ctrl *ctrl;
794 	device_t dev, pcib;
795 	device_t mmio_dev;
796 	uint64_t msi_addr;
797 	uint32_t msi_data;
798 	int err;
799 
800 	dev = softc->dev;
801 	pcib = device_get_parent(device_get_parent(dev));
802 	mmio_dev = pci_find_bsf(PCI_RID2BUS(softc->pci_rid),
803             PCI_RID2SLOT(softc->pci_rid), PCI_RID2FUNC(softc->pci_rid));
804 	if (device_is_attached(mmio_dev)) {
805 		device_printf(dev,
806 		    "warning: IOMMU device is claimed by another driver %s\n",
807 		    device_get_driver(mmio_dev)->name);
808 	}
809 
810 	softc->event_irq = -1;
811 	softc->event_rid = 0;
812 
813 	/*
814 	 * Section 3.7.1 of IOMMU rev 2.0. With MSI, there is only one
815 	 * interrupt. XXX: Enable MSI/X support.
816 	 */
817 	err = PCIB_ALLOC_MSI(pcib, dev, 1, 1, &softc->event_irq);
818 	if (err) {
819 		device_printf(dev,
820 		    "Couldn't find event MSI IRQ resource.\n");
821 		return (ENOENT);
822 	}
823 
824 	err = bus_set_resource(dev, SYS_RES_IRQ, softc->event_rid,
825 	    softc->event_irq, 1);
826 	if (err) {
827 		device_printf(dev, "Couldn't set event MSI resource.\n");
828 		return (ENXIO);
829 	}
830 
831 	softc->event_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
832 	    &softc->event_rid, RF_ACTIVE);
833 	if (!softc->event_res) {
834 		device_printf(dev,
835 		    "Unable to allocate event INTR resource.\n");
836 		return (ENOMEM);
837 	}
838 
839 	if (bus_setup_intr(dev, softc->event_res,
840 	    INTR_TYPE_MISC | INTR_MPSAFE, NULL, amdvi_event_intr,
841 	    softc, &softc->event_tag)) {
842 		device_printf(dev, "Fail to setup event intr\n");
843 		bus_release_resource(softc->dev, SYS_RES_IRQ,
844 		    softc->event_rid, softc->event_res);
845 		softc->event_res = NULL;
846 		return (ENXIO);
847 	}
848 
849 	bus_describe_intr(dev, softc->event_res, softc->event_tag,
850 	    "fault");
851 
852 	err = PCIB_MAP_MSI(pcib, dev, softc->event_irq, &msi_addr,
853 	    &msi_data);
854 	if (err) {
855 		device_printf(dev,
856 		    "Event interrupt config failed, err=%d.\n",
857 		    err);
858 		amdvi_free_evt_intr_res(softc->dev);
859 		return (err);
860 	}
861 
862 	/* Clear interrupt status bits. */
863 	ctrl = softc->ctrl;
864 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
865 
866 	/* Now enable MSI interrupt. */
867 	pci_enable_msi(mmio_dev, msi_addr, msi_data);
868 	return (0);
869 }
870 
871 
872 static void
873 amdvi_print_dev_cap(struct amdvi_softc *softc)
874 {
875 	struct ivhd_dev_cfg *cfg;
876 	int i;
877 
878 	cfg = softc->dev_cfg;
879 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
880 		device_printf(softc->dev, "device [0x%x - 0x%x]"
881 		    "config:%b%s\n", cfg->start_id, cfg->end_id,
882 		    cfg->data,
883 		    "\020\001INIT\002ExtInt\003NMI"
884 		    "\007LINT0\008LINT1",
885 		    cfg->enable_ats ? "ATS enabled" : "");
886 		cfg++;
887 	}
888 }
889 
890 static int
891 amdvi_handle_sysctl(SYSCTL_HANDLER_ARGS)
892 {
893 	struct amdvi_softc *softc;
894 	int result, type, error = 0;
895 
896 	softc = (struct amdvi_softc *)arg1;
897 	type = arg2;
898 
899 	switch (type) {
900 	case 0:
901 		result = softc->ctrl->cmd_head;
902 		error = sysctl_handle_int(oidp, &result, 0,
903 		    req);
904 		break;
905 	case 1:
906 		result = softc->ctrl->cmd_tail;
907 		error = sysctl_handle_int(oidp, &result, 0,
908 		    req);
909 		break;
910 	case 2:
911 		result = softc->ctrl->evt_head;
912 		error = sysctl_handle_int(oidp, &result, 0,
913 		    req);
914 		break;
915 	case 3:
916 		result = softc->ctrl->evt_tail;
917 		error = sysctl_handle_int(oidp, &result, 0,
918 		    req);
919 		break;
920 
921 	default:
922 		device_printf(softc->dev, "Unknown sysctl:%d\n", type);
923 	}
924 
925 	return (error);
926 }
927 
928 static void
929 amdvi_add_sysctl(struct amdvi_softc *softc)
930 {
931 	struct sysctl_oid_list *child;
932 	struct sysctl_ctx_list *ctx;
933 	device_t dev;
934 
935 	dev = softc->dev;
936 	ctx = device_get_sysctl_ctx(dev);
937 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
938 
939 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "event_intr_count", CTLFLAG_RD,
940 	    &softc->event_intr_cnt, "Event interrupt count");
941 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "command_count", CTLFLAG_RD,
942 	    &softc->total_cmd, "Command submitted count");
943 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "pci_rid", CTLFLAG_RD,
944 	    &softc->pci_rid, 0, "IOMMU RID");
945 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "start_dev_rid", CTLFLAG_RD,
946 	    &softc->start_dev_rid, 0, "Start of device under this IOMMU");
947 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "end_dev_rid", CTLFLAG_RD,
948 	    &softc->end_dev_rid, 0, "End of device under this IOMMU");
949 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_head",
950 	    CTLTYPE_UINT | CTLFLAG_RD, softc, 0,
951 	    amdvi_handle_sysctl, "IU", "Command head");
952 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_tail",
953 	    CTLTYPE_UINT | CTLFLAG_RD, softc, 1,
954 	    amdvi_handle_sysctl, "IU", "Command tail");
955 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_head",
956 	    CTLTYPE_UINT | CTLFLAG_RD, softc, 2,
957 	    amdvi_handle_sysctl, "IU", "Command head");
958 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_tail",
959 	    CTLTYPE_UINT | CTLFLAG_RD, softc, 3,
960 	    amdvi_handle_sysctl, "IU", "Command tail");
961 }
962 
963 int
964 amdvi_setup_hw(struct amdvi_softc *softc)
965 {
966 	device_t dev;
967 	int status;
968 
969 	dev = softc->dev;
970 
971 	amdvi_hw_enable_iotlb(softc);
972 
973 	amdvi_print_dev_cap(softc);
974 
975 	if ((status = amdvi_print_pci_cap(dev)) != 0) {
976 		device_printf(dev, "PCI capability.\n");
977 		return (status);
978 	}
979 	if ((status = amdvi_init_cmd(softc)) != 0) {
980 		device_printf(dev, "Couldn't configure command buffer.\n");
981 		return (status);
982 	}
983 	if ((status = amdvi_init_event(softc)) != 0) {
984 		device_printf(dev, "Couldn't configure event buffer.\n");
985 		return (status);
986 	}
987 	if ((status = amdvi_init_dte(softc)) != 0) {
988 		device_printf(dev, "Couldn't configure device table.\n");
989 		return (status);
990 	}
991 	if ((status = amdvi_alloc_intr_resources(softc)) != 0) {
992 		return (status);
993 	}
994 	amdvi_add_sysctl(softc);
995 	return (0);
996 }
997 
998 int
999 amdvi_teardown_hw(struct amdvi_softc *softc)
1000 {
1001 	device_t dev;
1002 
1003 	dev = softc->dev;
1004 
1005 	/*
1006 	 * Called after disable, h/w is stopped by now, free all the resources.
1007 	 */
1008 	amdvi_free_evt_intr_res(dev);
1009 
1010 	if (softc->cmd)
1011 		free(softc->cmd, M_AMDVI);
1012 
1013 	if (softc->event)
1014 		free(softc->event, M_AMDVI);
1015 
1016 	return (0);
1017 }
1018 
1019 /*********** bhyve interfaces *********************/
1020 static int
1021 amdvi_init(void)
1022 {
1023 	if (!ivhd_count) {
1024 		return (EIO);
1025 	}
1026 	if (!amdvi_enable_user && ivhd_count) {
1027 		printf("bhyve: Found %d AMD-Vi/IOMMU device(s), "
1028 		    	"use hw.vmm.amdvi.enable=1 to enable pass-through.\n",
1029 		    ivhd_count);
1030 		return (EINVAL);
1031 	}
1032 	return (0);
1033 }
1034 
1035 static void
1036 amdvi_cleanup(void)
1037 {
1038 	/* Nothing. */
1039 }
1040 
1041 static uint16_t
1042 amdvi_domainId(void)
1043 {
1044 
1045 	/*
1046 	 * If we hit maximum domain limit, rollover leaving host
1047 	 * domain(0).
1048 	 * XXX: make sure that this domain is not used.
1049 	 */
1050 	if (amdvi_dom_id == AMDVI_MAX_DOMAIN)
1051 		amdvi_dom_id = 1;
1052 
1053 	return ((uint16_t)amdvi_dom_id++);
1054 }
1055 
1056 static void
1057 amdvi_do_inv_domain(uint16_t domain_id, bool create)
1058 {
1059 	struct amdvi_softc *softc;
1060 	int i;
1061 
1062 	for (i = 0; i < ivhd_count; i++) {
1063 		softc = device_get_softc(ivhd_devs[i]);
1064 		KASSERT(softc, ("softc is NULL"));
1065 		/*
1066 		 * If not present pages are cached, invalidate page after
1067 		 * creating domain.
1068 		 */
1069 #if 0
1070 		if (create && ((softc->pci_cap & AMDVI_PCI_CAP_NPCACHE) == 0))
1071 			continue;
1072 #endif
1073 		amdvi_inv_domain(softc, domain_id);
1074 		amdvi_wait(softc);
1075 	}
1076 }
1077 
1078 static void *
1079 amdvi_create_domain(vm_paddr_t maxaddr)
1080 {
1081 	struct amdvi_domain *dom;
1082 
1083 	dom = malloc(sizeof(struct amdvi_domain), M_AMDVI, M_ZERO | M_WAITOK);
1084 	dom->id = amdvi_domainId();
1085 	//dom->maxaddr = maxaddr;
1086 #ifdef AMDVI_DEBUG_CMD
1087 	printf("Created domain #%d\n", dom->id);
1088 #endif
1089 	/*
1090 	 * Host domain(#0) don't create translation table.
1091 	 */
1092 	if (dom->id || amdvi_host_ptp)
1093 		dom->ptp = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1094 
1095 	dom->ptp_level = amdvi_ptp_level;
1096 
1097 	amdvi_do_inv_domain(dom->id, true);
1098 	SLIST_INSERT_HEAD(&dom_head, dom, next);
1099 
1100 	return (dom);
1101 }
1102 
1103 static void
1104 amdvi_free_ptp(uint64_t *ptp, int level)
1105 {
1106 	int i;
1107 
1108 	if (level < 1)
1109 		return;
1110 
1111 	for (i = 0; i < NPTEPG ; i++) {
1112 		if ((ptp[i] & AMDVI_PT_PRESENT) == 0)
1113 			continue;
1114 		/* XXX: Add super-page or PTE mapping > 4KB. */
1115 #ifdef notyet
1116 		/* Super-page mapping. */
1117 		if (AMDVI_PD_SUPER(ptp[i]))
1118 			continue;
1119 #endif
1120 
1121 		amdvi_free_ptp((uint64_t *)PHYS_TO_DMAP(ptp[i]
1122 		    & AMDVI_PT_MASK), level - 1);
1123 
1124 	}
1125 
1126 	free(ptp, M_AMDVI);
1127 }
1128 
1129 static void
1130 amdvi_destroy_domain(void *arg)
1131 {
1132 	struct amdvi_domain *domain;
1133 
1134 	domain = (struct amdvi_domain *)arg;
1135 	KASSERT(domain, ("domain is NULL"));
1136 #ifdef AMDVI_DEBUG_CMD
1137 	printf("Destroying domain %d\n", domain->id);
1138 #endif
1139 	if (domain->ptp)
1140 		amdvi_free_ptp(domain->ptp, domain->ptp_level);
1141 
1142 	amdvi_do_inv_domain(domain->id, false);
1143 	SLIST_REMOVE(&dom_head, domain, amdvi_domain, next);
1144 	free(domain, M_AMDVI);
1145 }
1146 
1147 static uint64_t
1148 amdvi_set_pt(uint64_t *pt, int level, vm_paddr_t gpa,
1149     vm_paddr_t hpa, uint64_t pg_size, bool create)
1150 {
1151 	uint64_t *page, pa;
1152 	int shift, index;
1153 	const int PT_SHIFT = 9;
1154 	const int PT_INDEX_MASK = (1 << PT_SHIFT) - 1;	/* Based on PT_SHIFT */
1155 
1156 	if (!pg_size)
1157 		return (0);
1158 
1159 	if (hpa & (pg_size - 1)) {
1160 		printf("HPA is not size aligned.\n");
1161 		return (0);
1162 	}
1163 	if (gpa & (pg_size - 1)) {
1164 		printf("HPA is not size aligned.\n");
1165 		return (0);
1166 	}
1167 	shift = PML4SHIFT;
1168 	while ((shift > PAGE_SHIFT) && (pg_size < (1UL << shift))) {
1169 		index = (gpa >> shift) & PT_INDEX_MASK;
1170 
1171 		if ((pt[index] == 0) && create) {
1172 			page = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1173 			pa = vtophys(page);
1174 			pt[index] = pa | AMDVI_PT_PRESENT | AMDVI_PT_RW |
1175 			    ((level - 1) << AMDVI_PD_LEVEL_SHIFT);
1176 		}
1177 #ifdef AMDVI_DEBUG_PTE
1178 		if ((gpa % 0x1000000) == 0)
1179 			printf("[level%d, shift = %d]PTE:0x%lx\n",
1180 			    level, shift, pt[index]);
1181 #endif
1182 #define PTE2PA(x)	((uint64_t)(x) & AMDVI_PT_MASK)
1183 		pa = PTE2PA(pt[index]);
1184 		pt = (uint64_t *)PHYS_TO_DMAP(pa);
1185 		shift -= PT_SHIFT;
1186 		level--;
1187 	}
1188 
1189 	/* Leaf entry. */
1190 	index = (gpa >> shift) & PT_INDEX_MASK;
1191 
1192 	if (create) {
1193 		pt[index] = hpa | AMDVI_PT_RW | AMDVI_PT_PRESENT;
1194 	} else
1195 		pt[index] = 0;
1196 
1197 #ifdef AMDVI_DEBUG_PTE
1198 	if ((gpa % 0x1000000) == 0)
1199 		printf("[Last level%d, shift = %d]PTE:0x%lx\n",
1200 		    level, shift, pt[index]);
1201 #endif
1202 	return (1ULL << shift);
1203 }
1204 
1205 static uint64_t
1206 amdvi_update_mapping(struct amdvi_domain *domain, vm_paddr_t gpa,
1207     vm_paddr_t hpa, uint64_t size, bool create)
1208 {
1209 	uint64_t mapped, *ptp, len;
1210 	int level;
1211 
1212 	KASSERT(domain, ("domain is NULL"));
1213 	level = domain->ptp_level;
1214 	KASSERT(level, ("Page table level is 0"));
1215 
1216 	ptp = domain->ptp;
1217 	KASSERT(ptp, ("PTP is NULL"));
1218 	mapped = 0;
1219 	while (mapped < size) {
1220 		len = amdvi_set_pt(ptp, level, gpa + mapped, hpa + mapped,
1221 		    PAGE_SIZE, create);
1222 		if (!len) {
1223 			printf("Error: Couldn't map HPA:0x%lx GPA:0x%lx\n",
1224 			    hpa, gpa);
1225 			return (0);
1226 		}
1227 		mapped += len;
1228 	}
1229 
1230 	return (mapped);
1231 }
1232 
1233 static uint64_t
1234 amdvi_create_mapping(void *arg, vm_paddr_t gpa, vm_paddr_t hpa,
1235     uint64_t len)
1236 {
1237 	struct amdvi_domain *domain;
1238 
1239 	domain = (struct amdvi_domain *)arg;
1240 
1241 	if (domain->id && !domain->ptp) {
1242 		printf("ptp is NULL");
1243 		return (-1);
1244 	}
1245 
1246 	/*
1247 	 * If host domain is created w/o page table, skip IOMMU page
1248 	 * table set-up.
1249 	 */
1250 	if (domain->ptp)
1251 		return (amdvi_update_mapping(domain, gpa, hpa, len, true));
1252 	else
1253 		return (len);
1254 }
1255 
1256 static uint64_t
1257 amdvi_destroy_mapping(void *arg, vm_paddr_t gpa, uint64_t len)
1258 {
1259 	struct amdvi_domain *domain;
1260 
1261 	domain = (struct amdvi_domain *)arg;
1262 	/*
1263 	 * If host domain is created w/o page table, skip IOMMU page
1264 	 * table set-up.
1265 	 */
1266 	if (domain->ptp)
1267 		return (amdvi_update_mapping(domain, gpa, 0, len, false));
1268 	return
1269 	    (len);
1270 }
1271 
1272 static struct amdvi_softc *
1273 amdvi_find_iommu(uint16_t devid)
1274 {
1275 	struct amdvi_softc *softc;
1276 	int i;
1277 
1278 	for (i = 0; i < ivhd_count; i++) {
1279 		softc = device_get_softc(ivhd_devs[i]);
1280 		if ((devid >= softc->start_dev_rid) &&
1281 		    (devid <= softc->end_dev_rid))
1282 			return (softc);
1283 	}
1284 
1285 	/*
1286 	 * XXX: BIOS bug, device not in IVRS table, assume its from first IOMMU.
1287 	 */
1288 	printf("BIOS bug device(%d.%d.%d) doesn't have IVHD entry.\n",
1289 	    RID2PCI_STR(devid));
1290 
1291 	return (device_get_softc(ivhd_devs[0]));
1292 }
1293 
1294 /*
1295  * Set-up device table entry.
1296  * IOMMU spec Rev 2.0, section 3.2.2.2, some of the fields must
1297  * be set concurrently, e.g. read and write bits.
1298  */
1299 static void
1300 amdvi_set_dte(struct amdvi_domain *domain, uint16_t devid, bool enable)
1301 {
1302 	struct amdvi_softc *softc;
1303 	struct amdvi_dte* temp;
1304 
1305 	KASSERT(domain, ("domain is NULL for pci_rid:0x%x\n", devid));
1306 
1307 	softc = amdvi_find_iommu(devid);
1308 	KASSERT(softc, ("softc is NULL for pci_rid:0x%x\n", devid));
1309 
1310 	temp = &amdvi_dte[devid];
1311 
1312 #ifdef AMDVI_ATS_ENABLE
1313 	/* If IOMMU and device support IOTLB, enable it. */
1314 	if (amdvi_dev_support_iotlb(softc, devid) && softc->iotlb)
1315 		temp->iotlb_enable = 1;
1316 #endif
1317 
1318 	/* Avoid duplicate I/O faults. */
1319 	temp->sup_second_io_fault = 1;
1320 	temp->sup_all_io_fault = amdvi_disable_io_fault;
1321 
1322 	temp->dt_valid = 1;
1323 	temp->domain_id = domain->id;
1324 
1325 	if (enable) {
1326 		if (domain->ptp) {
1327 			temp->pt_base = vtophys(domain->ptp) >> 12;
1328 			temp->pt_level = amdvi_ptp_level;
1329 		}
1330 		/*
1331 		 * XXX: Page table valid[TV] bit must be set even if host domain
1332 		 * page tables are not enabled.
1333 		 */
1334 		temp->pt_valid = 1;
1335 		temp->read_allow = 1;
1336 		temp->write_allow = 1;
1337 	}
1338 }
1339 
1340 static void
1341 amdvi_inv_device(uint16_t devid)
1342 {
1343 	struct amdvi_softc *softc;
1344 
1345 	softc = amdvi_find_iommu(devid);
1346 	KASSERT(softc, ("softc is NULL"));
1347 
1348 	amdvi_cmd_inv_dte(softc, devid);
1349 #ifdef AMDVI_ATS_ENABLE
1350 	if (amdvi_dev_support_iotlb(softc, devid))
1351 		amdvi_cmd_inv_iotlb(softc, devid);
1352 #endif
1353 	amdvi_wait(softc);
1354 }
1355 
1356 static void
1357 amdvi_add_device(void *arg, uint16_t devid)
1358 {
1359 	struct amdvi_domain *domain;
1360 
1361 	domain = (struct amdvi_domain *)arg;
1362 	KASSERT(domain != NULL, ("domain is NULL"));
1363 #ifdef AMDVI_DEBUG_CMD
1364 	printf("Assigning device(%d.%d.%d) to domain:%d\n",
1365 	    RID2PCI_STR(devid), domain->id);
1366 #endif
1367 	amdvi_set_dte(domain, devid, true);
1368 	amdvi_inv_device(devid);
1369 }
1370 
1371 static void
1372 amdvi_remove_device(void *arg, uint16_t devid)
1373 {
1374 	struct amdvi_domain *domain;
1375 
1376 	domain = (struct amdvi_domain *)arg;
1377 #ifdef AMDVI_DEBUG_CMD
1378 	printf("Remove device(0x%x) from domain:%d\n",
1379 	       devid, domain->id);
1380 #endif
1381 	amdvi_set_dte(domain, devid, false);
1382 	amdvi_inv_device(devid);
1383 }
1384 
1385 static void
1386 amdvi_enable(void)
1387 {
1388 	struct amdvi_ctrl *ctrl;
1389 	struct amdvi_softc *softc;
1390 	uint64_t val;
1391 	int i;
1392 
1393 	for (i = 0; i < ivhd_count; i++) {
1394 		softc = device_get_softc(ivhd_devs[i]);
1395 		KASSERT(softc, ("softc is NULL\n"));
1396 		ctrl = softc->ctrl;
1397 		KASSERT(ctrl, ("ctrl is NULL\n"));
1398 
1399 		val = (	AMDVI_CTRL_EN 		|
1400 			AMDVI_CTRL_CMD 		|
1401 		    	AMDVI_CTRL_ELOG 	|
1402 		    	AMDVI_CTRL_ELOGINT 	|
1403 		    	AMDVI_CTRL_INV_TO_1S);
1404 
1405 		if (softc->ivhd_flag & IVHD_FLAG_COH)
1406 			val |= AMDVI_CTRL_COH;
1407 		if (softc->ivhd_flag & IVHD_FLAG_HTT)
1408 			val |= AMDVI_CTRL_HTT;
1409 		if (softc->ivhd_flag & IVHD_FLAG_RPPW)
1410 			val |= AMDVI_CTRL_RPPW;
1411 		if (softc->ivhd_flag & IVHD_FLAG_PPW)
1412 			val |= AMDVI_CTRL_PPW;
1413 		if (softc->ivhd_flag & IVHD_FLAG_ISOC)
1414 			val |= AMDVI_CTRL_ISOC;
1415 
1416 		ctrl->control = val;
1417 	}
1418 }
1419 
1420 static void
1421 amdvi_disable(void)
1422 {
1423 	struct amdvi_ctrl *ctrl;
1424 	struct amdvi_softc *softc;
1425 	int i;
1426 
1427 	for (i = 0; i < ivhd_count; i++) {
1428 		softc = device_get_softc(ivhd_devs[i]);
1429 		KASSERT(softc, ("softc is NULL\n"));
1430 		ctrl = softc->ctrl;
1431 		KASSERT(ctrl, ("ctrl is NULL\n"));
1432 
1433 		ctrl->control = 0;
1434 	}
1435 }
1436 
1437 static void
1438 amdvi_inv_tlb(void *arg)
1439 {
1440 	struct amdvi_domain *domain;
1441 
1442 	domain = (struct amdvi_domain *)arg;
1443 	KASSERT(domain, ("domain is NULL"));
1444 	amdvi_do_inv_domain(domain->id, false);
1445 }
1446 
1447 struct iommu_ops iommu_ops_amd = {
1448 	amdvi_init,
1449 	amdvi_cleanup,
1450 	amdvi_enable,
1451 	amdvi_disable,
1452 	amdvi_create_domain,
1453 	amdvi_destroy_domain,
1454 	amdvi_create_mapping,
1455 	amdvi_destroy_mapping,
1456 	amdvi_add_device,
1457 	amdvi_remove_device,
1458 	amdvi_inv_tlb
1459 };
1460