xref: /freebsd/sys/amd64/vmm/amd/amdvi_hw.c (revision b0d29bc47dba79f6f38e67eabadfb4b32ffd9390)
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 | CTLFLAG_MPSAFE, NULL,
62     NULL);
63 
64 #define MOD_INC(a, s, m) (((a) + (s)) % ((m) * (s)))
65 #define MOD_DEC(a, s, m) (((a) - (s)) % ((m) * (s)))
66 
67 /* Print RID or device ID in PCI string format. */
68 #define RID2PCI_STR(d) PCI_RID2BUS(d), PCI_RID2SLOT(d), PCI_RID2FUNC(d)
69 
70 static void amdvi_dump_cmds(struct amdvi_softc *softc, int count);
71 static void amdvi_print_dev_cap(struct amdvi_softc *softc);
72 
73 MALLOC_DEFINE(M_AMDVI, "amdvi", "amdvi");
74 
75 extern device_t *ivhd_devs;
76 
77 extern int ivhd_count;
78 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, count, CTLFLAG_RDTUN, &ivhd_count,
79     0, NULL);
80 
81 static int amdvi_enable_user = 0;
82 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, enable, CTLFLAG_RDTUN,
83     &amdvi_enable_user, 0, NULL);
84 TUNABLE_INT("hw.vmm.amdvi_enable", &amdvi_enable_user);
85 
86 #ifdef AMDVI_ATS_ENABLE
87 /* XXX: ATS is not tested. */
88 static int amdvi_enable_iotlb = 1;
89 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, iotlb_enabled, CTLFLAG_RDTUN,
90     &amdvi_enable_iotlb, 0, NULL);
91 TUNABLE_INT("hw.vmm.enable_iotlb", &amdvi_enable_iotlb);
92 #endif
93 
94 static int amdvi_host_ptp = 1;	/* Use page tables for host. */
95 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, host_ptp, CTLFLAG_RDTUN,
96     &amdvi_host_ptp, 0, NULL);
97 TUNABLE_INT("hw.vmm.amdvi.host_ptp", &amdvi_host_ptp);
98 
99 /* Page table level used <= supported by h/w[v1=7]. */
100 int amdvi_ptp_level = 4;
101 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, ptp_level, CTLFLAG_RDTUN,
102     &amdvi_ptp_level, 0, NULL);
103 TUNABLE_INT("hw.vmm.amdvi.ptp_level", &amdvi_ptp_level);
104 
105 /* Disable fault event reporting. */
106 static int amdvi_disable_io_fault = 0;
107 SYSCTL_INT(_hw_vmm_amdvi, OID_AUTO, disable_io_fault, CTLFLAG_RDTUN,
108     &amdvi_disable_io_fault, 0, NULL);
109 TUNABLE_INT("hw.vmm.amdvi.disable_io_fault", &amdvi_disable_io_fault);
110 
111 static uint32_t amdvi_dom_id = 0;	/* 0 is reserved for host. */
112 SYSCTL_UINT(_hw_vmm_amdvi, OID_AUTO, domain_id, CTLFLAG_RD,
113     &amdvi_dom_id, 0, NULL);
114 /*
115  * Device table entry.
116  * Bus(256) x Dev(32) x Fun(8) x DTE(256 bits or 32 bytes).
117  *	= 256 * 2 * PAGE_SIZE.
118  */
119 static struct amdvi_dte amdvi_dte[PCI_NUM_DEV_MAX] __aligned(PAGE_SIZE);
120 CTASSERT(PCI_NUM_DEV_MAX == 0x10000);
121 CTASSERT(sizeof(amdvi_dte) == 0x200000);
122 
123 static SLIST_HEAD (, amdvi_domain) dom_head;
124 
125 static inline uint32_t
126 amdvi_pci_read(struct amdvi_softc *softc, int off)
127 {
128 
129 	return (pci_cfgregread(PCI_RID2BUS(softc->pci_rid),
130 	    PCI_RID2SLOT(softc->pci_rid), PCI_RID2FUNC(softc->pci_rid),
131 	    off, 4));
132 }
133 
134 #ifdef AMDVI_ATS_ENABLE
135 /* XXX: Should be in pci.c */
136 /*
137  * Check if device has ATS capability and its enabled.
138  * If ATS is absent or disabled, return (-1), otherwise ATS
139  * queue length.
140  */
141 static int
142 amdvi_find_ats_qlen(uint16_t devid)
143 {
144 	device_t dev;
145 	uint32_t off, cap;
146 	int qlen = -1;
147 
148 	dev = pci_find_bsf(PCI_RID2BUS(devid), PCI_RID2SLOT(devid),
149 			   PCI_RID2FUNC(devid));
150 
151 	if (!dev) {
152 		return (-1);
153 	}
154 #define PCIM_ATS_EN	BIT(31)
155 
156 	if (pci_find_extcap(dev, PCIZ_ATS, &off) == 0) {
157 		cap = pci_read_config(dev, off + 4, 4);
158 		qlen = (cap & 0x1F);
159 		qlen = qlen ? qlen : 32;
160 		printf("AMD-Vi: PCI device %d.%d.%d ATS %s qlen=%d\n",
161 		       RID2PCI_STR(devid),
162 		       (cap & PCIM_ATS_EN) ? "enabled" : "Disabled",
163 		       qlen);
164 		qlen = (cap & PCIM_ATS_EN) ? qlen : -1;
165 	}
166 
167 	return (qlen);
168 }
169 
170 /*
171  * Check if an endpoint device support device IOTLB or ATS.
172  */
173 static inline bool
174 amdvi_dev_support_iotlb(struct amdvi_softc *softc, uint16_t devid)
175 {
176 	struct ivhd_dev_cfg *cfg;
177 	int qlen, i;
178 	bool pci_ats, ivhd_ats;
179 
180 	qlen = amdvi_find_ats_qlen(devid);
181 	if (qlen < 0)
182 		return (false);
183 
184 	KASSERT(softc, ("softc is NULL"));
185 	cfg = softc->dev_cfg;
186 
187 	ivhd_ats = false;
188 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
189 		if ((cfg->start_id <= devid) && (cfg->end_id >= devid)) {
190 			ivhd_ats = cfg->enable_ats;
191 			break;
192 		}
193 		cfg++;
194 	}
195 
196 	pci_ats = (qlen < 0) ? false : true;
197 	if (pci_ats != ivhd_ats)
198 		device_printf(softc->dev,
199 		    "BIOS bug: mismatch in ATS setting for %d.%d.%d,"
200 		    "ATS inv qlen = %d\n", RID2PCI_STR(devid), qlen);
201 
202 	/* Ignore IVRS setting and respect PCI setting. */
203 	return (pci_ats);
204 }
205 #endif
206 
207 /* Enable IOTLB support for IOMMU if its supported. */
208 static inline void
209 amdvi_hw_enable_iotlb(struct amdvi_softc *softc)
210 {
211 #ifndef AMDVI_ATS_ENABLE
212 	softc->iotlb = false;
213 #else
214 	bool supported;
215 
216 	supported = (softc->ivhd_flag & IVHD_FLAG_IOTLB) ? true : false;
217 
218 	if (softc->pci_cap & AMDVI_PCI_CAP_IOTLB) {
219 		if (!supported)
220 			device_printf(softc->dev, "IOTLB disabled by BIOS.\n");
221 
222 		if (supported && !amdvi_enable_iotlb) {
223 			device_printf(softc->dev, "IOTLB disabled by user.\n");
224 			supported = false;
225 		}
226 	} else
227 		supported = false;
228 
229 	softc->iotlb = supported;
230 
231 #endif
232 }
233 
234 static int
235 amdvi_init_cmd(struct amdvi_softc *softc)
236 {
237 	struct amdvi_ctrl *ctrl = softc->ctrl;
238 
239 	ctrl->cmd.len = 8;	/* Use 256 command buffer entries. */
240 	softc->cmd_max = 1 << ctrl->cmd.len;
241 
242 	softc->cmd = malloc(sizeof(struct amdvi_cmd) *
243 	    softc->cmd_max, M_AMDVI, M_WAITOK | M_ZERO);
244 
245 	if ((uintptr_t)softc->cmd & PAGE_MASK)
246 		panic("AMDVi: Command buffer not aligned on page boundary.");
247 
248 	ctrl->cmd.base = vtophys(softc->cmd) / PAGE_SIZE;
249 	/*
250 	 * XXX: Reset the h/w pointers in case IOMMU is restarting,
251 	 * h/w doesn't clear these pointers based on empirical data.
252 	 */
253 	ctrl->cmd_tail = 0;
254 	ctrl->cmd_head = 0;
255 
256 	return (0);
257 }
258 
259 /*
260  * Note: Update tail pointer after we have written the command since tail
261  * pointer update cause h/w to execute new commands, see section 3.3
262  * of AMD IOMMU spec ver 2.0.
263  */
264 /* Get the command tail pointer w/o updating it. */
265 static struct amdvi_cmd *
266 amdvi_get_cmd_tail(struct amdvi_softc *softc)
267 {
268 	struct amdvi_ctrl *ctrl;
269 	struct amdvi_cmd *tail;
270 
271 	KASSERT(softc, ("softc is NULL"));
272 	KASSERT(softc->cmd != NULL, ("cmd is NULL"));
273 
274 	ctrl = softc->ctrl;
275 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
276 
277 	tail = (struct amdvi_cmd *)((uint8_t *)softc->cmd +
278 	    ctrl->cmd_tail);
279 
280 	return (tail);
281 }
282 
283 /*
284  * Update the command tail pointer which will start command execution.
285  */
286 static void
287 amdvi_update_cmd_tail(struct amdvi_softc *softc)
288 {
289 	struct amdvi_ctrl *ctrl;
290 	int size;
291 
292 	size = sizeof(struct amdvi_cmd);
293 	KASSERT(softc->cmd != NULL, ("cmd is NULL"));
294 
295 	ctrl = softc->ctrl;
296 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
297 
298 	ctrl->cmd_tail = MOD_INC(ctrl->cmd_tail, size, softc->cmd_max);
299 	softc->total_cmd++;
300 
301 #ifdef AMDVI_DEBUG_CMD
302 	device_printf(softc->dev, "cmd_tail: %s Tail:0x%x, Head:0x%x.\n",
303 	    ctrl->cmd_tail,
304 	    ctrl->cmd_head);
305 #endif
306 
307 }
308 
309 /*
310  * Various commands supported by IOMMU.
311  */
312 
313 /* Completion wait command. */
314 static void
315 amdvi_cmd_cmp(struct amdvi_softc *softc, const uint64_t data)
316 {
317 	struct amdvi_cmd *cmd;
318 	uint64_t pa;
319 
320 	cmd = amdvi_get_cmd_tail(softc);
321 	KASSERT(cmd != NULL, ("Cmd is NULL"));
322 
323 	pa = vtophys(&softc->cmp_data);
324 	cmd->opcode = AMDVI_CMP_WAIT_OPCODE;
325 	cmd->word0 = (pa & 0xFFFFFFF8) | AMDVI_CMP_WAIT_STORE;
326 	cmd->word1 = (pa >> 32) & 0xFFFFF;
327 	cmd->addr = data;
328 
329 	amdvi_update_cmd_tail(softc);
330 }
331 
332 /* Invalidate device table entry. */
333 static void
334 amdvi_cmd_inv_dte(struct amdvi_softc *softc, uint16_t devid)
335 {
336 	struct amdvi_cmd *cmd;
337 
338 	cmd = amdvi_get_cmd_tail(softc);
339 	KASSERT(cmd != NULL, ("Cmd is NULL"));
340 	cmd->opcode = AMDVI_INVD_DTE_OPCODE;
341 	cmd->word0 = devid;
342 	amdvi_update_cmd_tail(softc);
343 #ifdef AMDVI_DEBUG_CMD
344 	device_printf(softc->dev, "Invalidated DTE:0x%x\n", devid);
345 #endif
346 }
347 
348 /* Invalidate IOMMU page, use for invalidation of domain. */
349 static void
350 amdvi_cmd_inv_iommu_pages(struct amdvi_softc *softc, uint16_t domain_id,
351 			  uint64_t addr, bool guest_nested,
352 			  bool pde, bool page)
353 {
354 	struct amdvi_cmd *cmd;
355 
356 	cmd = amdvi_get_cmd_tail(softc);
357 	KASSERT(cmd != NULL, ("Cmd is NULL"));
358 
359 
360 	cmd->opcode = AMDVI_INVD_PAGE_OPCODE;
361 	cmd->word1 = domain_id;
362 	/*
363 	 * Invalidate all addresses for this domain.
364 	 */
365 	cmd->addr = addr;
366 	cmd->addr |= pde ? AMDVI_INVD_PAGE_PDE : 0;
367 	cmd->addr |= page ? AMDVI_INVD_PAGE_S : 0;
368 
369 	amdvi_update_cmd_tail(softc);
370 }
371 
372 #ifdef AMDVI_ATS_ENABLE
373 /* Invalidate device IOTLB. */
374 static void
375 amdvi_cmd_inv_iotlb(struct amdvi_softc *softc, uint16_t devid)
376 {
377 	struct amdvi_cmd *cmd;
378 	int qlen;
379 
380 	if (!softc->iotlb)
381 		return;
382 
383 	qlen = amdvi_find_ats_qlen(devid);
384 	if (qlen < 0) {
385 		panic("AMDVI: Invalid ATS qlen(%d) for device %d.%d.%d\n",
386 		      qlen, RID2PCI_STR(devid));
387 	}
388 	cmd = amdvi_get_cmd_tail(softc);
389 	KASSERT(cmd != NULL, ("Cmd is NULL"));
390 
391 #ifdef AMDVI_DEBUG_CMD
392 	device_printf(softc->dev, "Invalidate IOTLB devID 0x%x"
393 		      " Qlen:%d\n", devid, qlen);
394 #endif
395 	cmd->opcode = AMDVI_INVD_IOTLB_OPCODE;
396 	cmd->word0 = devid;
397 	cmd->word1 = qlen;
398 	cmd->addr = AMDVI_INVD_IOTLB_ALL_ADDR |
399 		AMDVI_INVD_IOTLB_S;
400 	amdvi_update_cmd_tail(softc);
401 }
402 #endif
403 
404 #ifdef notyet				/* For Interrupt Remap. */
405 static void
406 amdvi_cmd_inv_intr_map(struct amdvi_softc *softc,
407 		       uint16_t devid)
408 {
409 	struct amdvi_cmd *cmd;
410 
411 	cmd = amdvi_get_cmd_tail(softc);
412 	KASSERT(cmd != NULL, ("Cmd is NULL"));
413 	cmd->opcode = AMDVI_INVD_INTR_OPCODE;
414 	cmd->word0 = devid;
415 	amdvi_update_cmd_tail(softc);
416 #ifdef AMDVI_DEBUG_CMD
417 	device_printf(softc->dev, "Invalidate INTR map of devID 0x%x\n", devid);
418 #endif
419 }
420 #endif
421 
422 /* Invalidate domain using INVALIDATE_IOMMU_PAGES command. */
423 static void
424 amdvi_inv_domain(struct amdvi_softc *softc, uint16_t domain_id)
425 {
426 	struct amdvi_cmd *cmd;
427 
428 	cmd = amdvi_get_cmd_tail(softc);
429 	KASSERT(cmd != NULL, ("Cmd is NULL"));
430 
431 	/*
432 	 * See section 3.3.3 of IOMMU spec rev 2.0, software note
433 	 * for invalidating domain.
434 	 */
435 	amdvi_cmd_inv_iommu_pages(softc, domain_id, AMDVI_INVD_PAGE_ALL_ADDR,
436 				false, true, true);
437 
438 #ifdef AMDVI_DEBUG_CMD
439 	device_printf(softc->dev, "Invalidate domain:0x%x\n", domain_id);
440 
441 #endif
442 }
443 
444 static	bool
445 amdvi_cmp_wait(struct amdvi_softc *softc)
446 {
447 	struct amdvi_ctrl *ctrl;
448 	const uint64_t VERIFY = 0xA5A5;
449 	volatile uint64_t *read;
450 	int i;
451 	bool status;
452 
453 	ctrl = softc->ctrl;
454 	read = &softc->cmp_data;
455 	*read = 0;
456 	amdvi_cmd_cmp(softc, VERIFY);
457 	/* Wait for h/w to update completion data. */
458 	for (i = 0; i < 100 && (*read != VERIFY); i++) {
459 		DELAY(1000);		/* 1 ms */
460 	}
461 	status = (VERIFY == softc->cmp_data) ? true : false;
462 
463 #ifdef AMDVI_DEBUG_CMD
464 	if (status)
465 		device_printf(softc->dev, "CMD completion DONE Tail:0x%x, "
466 			      "Head:0x%x, loop:%d.\n", ctrl->cmd_tail,
467 			      ctrl->cmd_head, loop);
468 #endif
469 	return (status);
470 }
471 
472 static void
473 amdvi_wait(struct amdvi_softc *softc)
474 {
475 	struct amdvi_ctrl *ctrl;
476 	int i;
477 
478 	KASSERT(softc, ("softc is NULL"));
479 
480 	ctrl = softc->ctrl;
481 	KASSERT(ctrl != NULL, ("ctrl is NULL"));
482 	/* Don't wait if h/w is not enabled. */
483 	if ((ctrl->control & AMDVI_CTRL_EN) == 0)
484 		return;
485 
486 	for (i = 0; i < 10; i++) {
487 		if (amdvi_cmp_wait(softc))
488 			return;
489 	}
490 
491 	device_printf(softc->dev, "Error: completion failed"
492 		      " tail:0x%x, head:0x%x.\n",
493 		      ctrl->cmd_tail, ctrl->cmd_head);
494 	/* Dump the last command. */
495 	amdvi_dump_cmds(softc, 1);
496 }
497 
498 static void
499 amdvi_dump_cmds(struct amdvi_softc *softc, int count)
500 {
501 	struct amdvi_ctrl *ctrl;
502 	struct amdvi_cmd *cmd;
503 	int off, i;
504 
505 	ctrl = softc->ctrl;
506 	device_printf(softc->dev, "Dump last %d command(s):\n", count);
507 	/*
508 	 * If h/w is stuck in completion, it is the previous command,
509 	 * start dumping from previous command onward.
510 	 */
511 	off = MOD_DEC(ctrl->cmd_head, sizeof(struct amdvi_cmd),
512 	    softc->cmd_max);
513 	for (i = 0; off != ctrl->cmd_tail && i < count; i++) {
514 		cmd = (struct amdvi_cmd *)((uint8_t *)softc->cmd + off);
515 		printf("  [CMD%d, off:0x%x] opcode= 0x%x 0x%x"
516 		    " 0x%x 0x%lx\n", i, off, cmd->opcode,
517 		    cmd->word0, cmd->word1, cmd->addr);
518 		off = (off + sizeof(struct amdvi_cmd)) %
519 		    (softc->cmd_max * sizeof(struct amdvi_cmd));
520 	}
521 }
522 
523 static int
524 amdvi_init_event(struct amdvi_softc *softc)
525 {
526 	struct amdvi_ctrl *ctrl;
527 
528 	ctrl = softc->ctrl;
529 	ctrl->event.len = 8;
530 	softc->event_max = 1 << ctrl->event.len;
531 	softc->event = malloc(sizeof(struct amdvi_event) *
532 	    softc->event_max, M_AMDVI, M_WAITOK | M_ZERO);
533 	if ((uintptr_t)softc->event & PAGE_MASK) {
534 		device_printf(softc->dev, "Event buffer not aligned on page.");
535 		return (false);
536 	}
537 	ctrl->event.base = vtophys(softc->event) / PAGE_SIZE;
538 
539 	/* Reset the pointers. */
540 	ctrl->evt_head = 0;
541 	ctrl->evt_tail = 0;
542 
543 	return (0);
544 }
545 
546 static inline void
547 amdvi_decode_evt_flag(uint16_t flag)
548 {
549 
550 	flag &= AMDVI_EVENT_FLAG_MASK;
551 	printf(" 0x%b]\n", flag,
552 		"\020"
553 		"\001GN"
554 		"\002NX"
555 		"\003US"
556 		"\004I"
557 		"\005PR"
558 		"\006RW"
559 		"\007PE"
560 		"\010RZ"
561 		"\011TR"
562 		);
563 }
564 
565 /* See section 2.5.4 of AMD IOMMU spec ver 2.62.*/
566 static inline void
567 amdvi_decode_evt_flag_type(uint8_t type)
568 {
569 
570 	switch (AMDVI_EVENT_FLAG_TYPE(type)) {
571 	case 0:
572 		printf("RSVD\n");
573 		break;
574 	case 1:
575 		printf("Master Abort\n");
576 		break;
577 	case 2:
578 		printf("Target Abort\n");
579 		break;
580 	case 3:
581 		printf("Data Err\n");
582 		break;
583 	default:
584 		break;
585 	}
586 }
587 
588 static void
589 amdvi_decode_inv_dte_evt(uint16_t devid, uint16_t domid, uint64_t addr,
590     uint16_t flag)
591 {
592 
593 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
594 	    " Addr:0x%lx",
595 	    devid, domid, addr);
596 	amdvi_decode_evt_flag(flag);
597 }
598 
599 static void
600 amdvi_decode_pf_evt(uint16_t devid, uint16_t domid, uint64_t addr,
601     uint16_t flag)
602 {
603 
604 	printf("\t[IO_PAGE_FAULT EVT: devId:0x%x DomId:0x%x"
605 	    " Addr:0x%lx",
606 	    devid, domid, addr);
607 	amdvi_decode_evt_flag(flag);
608 }
609 
610 static void
611 amdvi_decode_dte_hwerr_evt(uint16_t devid, uint16_t domid,
612     uint64_t addr, uint16_t flag)
613 {
614 
615 	printf("\t[DEV_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
616 	    " Addr:0x%lx", devid, domid, addr);
617 	amdvi_decode_evt_flag(flag);
618 	amdvi_decode_evt_flag_type(flag);
619 }
620 
621 static void
622 amdvi_decode_page_hwerr_evt(uint16_t devid, uint16_t domid, uint64_t addr,
623     uint16_t flag)
624 {
625 
626 	printf("\t[PAGE_TAB_HW_ERR EVT: devId:0x%x DomId:0x%x"
627 	    " Addr:0x%lx", devid, domid, addr);
628 	amdvi_decode_evt_flag(flag);
629 	amdvi_decode_evt_flag_type(AMDVI_EVENT_FLAG_TYPE(flag));
630 }
631 
632 static void
633 amdvi_decode_evt(struct amdvi_event *evt)
634 {
635 	struct amdvi_cmd *cmd;
636 
637 	switch (evt->opcode) {
638 	case AMDVI_EVENT_INVALID_DTE:
639 		amdvi_decode_inv_dte_evt(evt->devid, evt->pasid_domid,
640 		    evt->addr, evt->flag);
641 		break;
642 
643 	case AMDVI_EVENT_PFAULT:
644 		amdvi_decode_pf_evt(evt->devid, evt->pasid_domid,
645 		    evt->addr, evt->flag);
646 		break;
647 
648 	case AMDVI_EVENT_DTE_HW_ERROR:
649 		amdvi_decode_dte_hwerr_evt(evt->devid, evt->pasid_domid,
650 		    evt->addr, evt->flag);
651 		break;
652 
653 	case AMDVI_EVENT_PAGE_HW_ERROR:
654 		amdvi_decode_page_hwerr_evt(evt->devid, evt->pasid_domid,
655 		    evt->addr, evt->flag);
656 		break;
657 
658 	case AMDVI_EVENT_ILLEGAL_CMD:
659 		/* FALL THROUGH */
660 	case AMDVI_EVENT_CMD_HW_ERROR:
661 		printf("\t[%s EVT]\n", (evt->opcode == AMDVI_EVENT_ILLEGAL_CMD) ?
662 		    "ILLEGAL CMD" : "CMD HW ERR");
663 		cmd = (struct amdvi_cmd *)PHYS_TO_DMAP(evt->addr);
664 		printf("\tCMD opcode= 0x%x 0x%x 0x%x 0x%lx\n",
665 		    cmd->opcode, cmd->word0, cmd->word1, cmd->addr);
666 		break;
667 
668 	case AMDVI_EVENT_IOTLB_TIMEOUT:
669 		printf("\t[IOTLB_INV_TIMEOUT devid:0x%x addr:0x%lx]\n",
670 		    evt->devid, evt->addr);
671 		break;
672 
673 	case AMDVI_EVENT_INVALID_DTE_REQ:
674 		printf("\t[INV_DTE devid:0x%x addr:0x%lx type:0x%x tr:%d]\n",
675 		    evt->devid, evt->addr, evt->flag >> 9,
676 		    (evt->flag >> 8) & 1);
677 		break;
678 
679 	case AMDVI_EVENT_INVALID_PPR_REQ:
680 	case AMDVI_EVENT_COUNTER_ZERO:
681 		printf("AMD-Vi: v2 events.\n");
682 		break;
683 
684 	default:
685 		printf("Unsupported AMD-Vi event:%d\n", evt->opcode);
686 	}
687 }
688 
689 static void
690 amdvi_print_events(struct amdvi_softc *softc)
691 {
692 	struct amdvi_ctrl *ctrl;
693 	struct amdvi_event *event;
694 	int i, size;
695 
696 	ctrl = softc->ctrl;
697 	size = sizeof(struct amdvi_event);
698 	for (i = 0; i < softc->event_max; i++) {
699 		event = &softc->event[ctrl->evt_head / size];
700 		if (!event->opcode)
701 			break;
702 		device_printf(softc->dev, "\t[Event%d: Head:0x%x Tail:0x%x]\n",
703 		    i, ctrl->evt_head, ctrl->evt_tail);
704 		amdvi_decode_evt(event);
705 		ctrl->evt_head = MOD_INC(ctrl->evt_head, size,
706 		    softc->event_max);
707 	}
708 }
709 
710 static int
711 amdvi_init_dte(struct amdvi_softc *softc)
712 {
713 	struct amdvi_ctrl *ctrl;
714 
715 	ctrl = softc->ctrl;
716 	ctrl->dte.base = vtophys(amdvi_dte) / PAGE_SIZE;
717 	ctrl->dte.size = 0x1FF;		/* 2MB device table. */
718 
719 	return (0);
720 }
721 
722 /*
723  * Not all capabilities of IOMMU are available in ACPI IVHD flag
724  * or EFR entry, read directly from device.
725  */
726 static int
727 amdvi_print_pci_cap(device_t dev)
728 {
729 	struct amdvi_softc *softc;
730 	uint32_t off, cap;
731 
732 
733 	softc = device_get_softc(dev);
734 	off = softc->cap_off;
735 
736 	/*
737 	 * Section 3.7.1 of IOMMU sepc rev 2.0.
738 	 * Read capability from device.
739 	 */
740 	cap = amdvi_pci_read(softc, off);
741 
742 	/* Make sure capability type[18:16] is 3. */
743 	KASSERT((((cap >> 16) & 0x7) == 0x3),
744 	    ("Not a IOMMU capability 0x%x@0x%x", cap, off));
745 
746 	softc->pci_cap = cap >> 24;
747 	device_printf(softc->dev, "PCI cap 0x%x@0x%x feature:%b\n",
748 	    cap, off, softc->pci_cap,
749 	    "\20\1IOTLB\2HT\3NPCache\4EFR\5CapExt");
750 
751 	return (0);
752 }
753 
754 static void
755 amdvi_event_intr(void *arg)
756 {
757 	struct amdvi_softc *softc;
758 	struct amdvi_ctrl *ctrl;
759 
760 	softc = (struct amdvi_softc *)arg;
761 	ctrl = softc->ctrl;
762 	device_printf(softc->dev, "EVT INTR %ld Status:0x%x"
763 	    " EVT Head:0x%x Tail:0x%x]\n", softc->event_intr_cnt++,
764 	    ctrl->status, ctrl->evt_head, ctrl->evt_tail);
765 	printf("  [CMD Total 0x%lx] Tail:0x%x, Head:0x%x.\n",
766 	    softc->total_cmd, ctrl->cmd_tail, ctrl->cmd_head);
767 
768 	amdvi_print_events(softc);
769 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
770 }
771 
772 static void
773 amdvi_free_evt_intr_res(device_t dev)
774 {
775 
776 	struct amdvi_softc *softc;
777 
778 	softc = device_get_softc(dev);
779 	if (softc->event_tag != NULL) {
780 		bus_teardown_intr(dev, softc->event_res, softc->event_tag);
781 	}
782 	if (softc->event_res != NULL) {
783 		bus_release_resource(dev, SYS_RES_IRQ, softc->event_rid,
784 		    softc->event_res);
785 	}
786 	bus_delete_resource(dev, SYS_RES_IRQ, softc->event_rid);
787 	PCIB_RELEASE_MSI(device_get_parent(device_get_parent(dev)),
788 	    dev, 1, &softc->event_irq);
789 }
790 
791 static bool
792 amdvi_alloc_intr_resources(struct amdvi_softc *softc)
793 {
794 	struct amdvi_ctrl *ctrl;
795 	device_t dev, pcib;
796 	device_t mmio_dev;
797 	uint64_t msi_addr;
798 	uint32_t msi_data;
799 	int err;
800 
801 	dev = softc->dev;
802 	pcib = device_get_parent(device_get_parent(dev));
803 	mmio_dev = pci_find_bsf(PCI_RID2BUS(softc->pci_rid),
804             PCI_RID2SLOT(softc->pci_rid), PCI_RID2FUNC(softc->pci_rid));
805 	if (device_is_attached(mmio_dev)) {
806 		device_printf(dev,
807 		    "warning: IOMMU device is claimed by another driver %s\n",
808 		    device_get_driver(mmio_dev)->name);
809 	}
810 
811 	softc->event_irq = -1;
812 	softc->event_rid = 0;
813 
814 	/*
815 	 * Section 3.7.1 of IOMMU rev 2.0. With MSI, there is only one
816 	 * interrupt. XXX: Enable MSI/X support.
817 	 */
818 	err = PCIB_ALLOC_MSI(pcib, dev, 1, 1, &softc->event_irq);
819 	if (err) {
820 		device_printf(dev,
821 		    "Couldn't find event MSI IRQ resource.\n");
822 		return (ENOENT);
823 	}
824 
825 	err = bus_set_resource(dev, SYS_RES_IRQ, softc->event_rid,
826 	    softc->event_irq, 1);
827 	if (err) {
828 		device_printf(dev, "Couldn't set event MSI resource.\n");
829 		return (ENXIO);
830 	}
831 
832 	softc->event_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
833 	    &softc->event_rid, RF_ACTIVE);
834 	if (!softc->event_res) {
835 		device_printf(dev,
836 		    "Unable to allocate event INTR resource.\n");
837 		return (ENOMEM);
838 	}
839 
840 	if (bus_setup_intr(dev, softc->event_res,
841 	    INTR_TYPE_MISC | INTR_MPSAFE, NULL, amdvi_event_intr,
842 	    softc, &softc->event_tag)) {
843 		device_printf(dev, "Fail to setup event intr\n");
844 		bus_release_resource(softc->dev, SYS_RES_IRQ,
845 		    softc->event_rid, softc->event_res);
846 		softc->event_res = NULL;
847 		return (ENXIO);
848 	}
849 
850 	bus_describe_intr(dev, softc->event_res, softc->event_tag,
851 	    "fault");
852 
853 	err = PCIB_MAP_MSI(pcib, dev, softc->event_irq, &msi_addr,
854 	    &msi_data);
855 	if (err) {
856 		device_printf(dev,
857 		    "Event interrupt config failed, err=%d.\n",
858 		    err);
859 		amdvi_free_evt_intr_res(softc->dev);
860 		return (err);
861 	}
862 
863 	/* Clear interrupt status bits. */
864 	ctrl = softc->ctrl;
865 	ctrl->status &= AMDVI_STATUS_EV_OF | AMDVI_STATUS_EV_INTR;
866 
867 	/* Now enable MSI interrupt. */
868 	pci_enable_msi(mmio_dev, msi_addr, msi_data);
869 	return (0);
870 }
871 
872 
873 static void
874 amdvi_print_dev_cap(struct amdvi_softc *softc)
875 {
876 	struct ivhd_dev_cfg *cfg;
877 	int i;
878 
879 	cfg = softc->dev_cfg;
880 	for (i = 0; i < softc->dev_cfg_cnt; i++) {
881 		device_printf(softc->dev, "device [0x%x - 0x%x]"
882 		    "config:%b%s\n", cfg->start_id, cfg->end_id,
883 		    cfg->data,
884 		    "\020\001INIT\002ExtInt\003NMI"
885 		    "\007LINT0\008LINT1",
886 		    cfg->enable_ats ? "ATS enabled" : "");
887 		cfg++;
888 	}
889 }
890 
891 static int
892 amdvi_handle_sysctl(SYSCTL_HANDLER_ARGS)
893 {
894 	struct amdvi_softc *softc;
895 	int result, type, error = 0;
896 
897 	softc = (struct amdvi_softc *)arg1;
898 	type = arg2;
899 
900 	switch (type) {
901 	case 0:
902 		result = softc->ctrl->cmd_head;
903 		error = sysctl_handle_int(oidp, &result, 0,
904 		    req);
905 		break;
906 	case 1:
907 		result = softc->ctrl->cmd_tail;
908 		error = sysctl_handle_int(oidp, &result, 0,
909 		    req);
910 		break;
911 	case 2:
912 		result = softc->ctrl->evt_head;
913 		error = sysctl_handle_int(oidp, &result, 0,
914 		    req);
915 		break;
916 	case 3:
917 		result = softc->ctrl->evt_tail;
918 		error = sysctl_handle_int(oidp, &result, 0,
919 		    req);
920 		break;
921 
922 	default:
923 		device_printf(softc->dev, "Unknown sysctl:%d\n", type);
924 	}
925 
926 	return (error);
927 }
928 
929 static void
930 amdvi_add_sysctl(struct amdvi_softc *softc)
931 {
932 	struct sysctl_oid_list *child;
933 	struct sysctl_ctx_list *ctx;
934 	device_t dev;
935 
936 	dev = softc->dev;
937 	ctx = device_get_sysctl_ctx(dev);
938 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
939 
940 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "event_intr_count", CTLFLAG_RD,
941 	    &softc->event_intr_cnt, "Event interrupt count");
942 	SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "command_count", CTLFLAG_RD,
943 	    &softc->total_cmd, "Command submitted count");
944 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "pci_rid", CTLFLAG_RD,
945 	    &softc->pci_rid, 0, "IOMMU RID");
946 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "start_dev_rid", CTLFLAG_RD,
947 	    &softc->start_dev_rid, 0, "Start of device under this IOMMU");
948 	SYSCTL_ADD_U16(ctx, child, OID_AUTO, "end_dev_rid", CTLFLAG_RD,
949 	    &softc->end_dev_rid, 0, "End of device under this IOMMU");
950 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_head",
951 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 0,
952 	    amdvi_handle_sysctl, "IU", "Command head");
953 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "command_tail",
954 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 1,
955 	    amdvi_handle_sysctl, "IU", "Command tail");
956 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_head",
957 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 2,
958 	    amdvi_handle_sysctl, "IU", "Command head");
959 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "event_tail",
960 	    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, softc, 3,
961 	    amdvi_handle_sysctl, "IU", "Command tail");
962 }
963 
964 int
965 amdvi_setup_hw(struct amdvi_softc *softc)
966 {
967 	device_t dev;
968 	int status;
969 
970 	dev = softc->dev;
971 
972 	amdvi_hw_enable_iotlb(softc);
973 
974 	amdvi_print_dev_cap(softc);
975 
976 	if ((status = amdvi_print_pci_cap(dev)) != 0) {
977 		device_printf(dev, "PCI capability.\n");
978 		return (status);
979 	}
980 	if ((status = amdvi_init_cmd(softc)) != 0) {
981 		device_printf(dev, "Couldn't configure command buffer.\n");
982 		return (status);
983 	}
984 	if ((status = amdvi_init_event(softc)) != 0) {
985 		device_printf(dev, "Couldn't configure event buffer.\n");
986 		return (status);
987 	}
988 	if ((status = amdvi_init_dte(softc)) != 0) {
989 		device_printf(dev, "Couldn't configure device table.\n");
990 		return (status);
991 	}
992 	if ((status = amdvi_alloc_intr_resources(softc)) != 0) {
993 		return (status);
994 	}
995 	amdvi_add_sysctl(softc);
996 	return (0);
997 }
998 
999 int
1000 amdvi_teardown_hw(struct amdvi_softc *softc)
1001 {
1002 	device_t dev;
1003 
1004 	dev = softc->dev;
1005 
1006 	/*
1007 	 * Called after disable, h/w is stopped by now, free all the resources.
1008 	 */
1009 	amdvi_free_evt_intr_res(dev);
1010 
1011 	if (softc->cmd)
1012 		free(softc->cmd, M_AMDVI);
1013 
1014 	if (softc->event)
1015 		free(softc->event, M_AMDVI);
1016 
1017 	return (0);
1018 }
1019 
1020 /*********** bhyve interfaces *********************/
1021 static int
1022 amdvi_init(void)
1023 {
1024 	if (!ivhd_count) {
1025 		return (EIO);
1026 	}
1027 	if (!amdvi_enable_user && ivhd_count) {
1028 		printf("bhyve: Found %d AMD-Vi/IOMMU device(s), "
1029 		    	"use hw.vmm.amdvi.enable=1 to enable pass-through.\n",
1030 		    ivhd_count);
1031 		return (EINVAL);
1032 	}
1033 	return (0);
1034 }
1035 
1036 static void
1037 amdvi_cleanup(void)
1038 {
1039 	/* Nothing. */
1040 }
1041 
1042 static uint16_t
1043 amdvi_domainId(void)
1044 {
1045 
1046 	/*
1047 	 * If we hit maximum domain limit, rollover leaving host
1048 	 * domain(0).
1049 	 * XXX: make sure that this domain is not used.
1050 	 */
1051 	if (amdvi_dom_id == AMDVI_MAX_DOMAIN)
1052 		amdvi_dom_id = 1;
1053 
1054 	return ((uint16_t)amdvi_dom_id++);
1055 }
1056 
1057 static void
1058 amdvi_do_inv_domain(uint16_t domain_id, bool create)
1059 {
1060 	struct amdvi_softc *softc;
1061 	int i;
1062 
1063 	for (i = 0; i < ivhd_count; i++) {
1064 		softc = device_get_softc(ivhd_devs[i]);
1065 		KASSERT(softc, ("softc is NULL"));
1066 		/*
1067 		 * If not present pages are cached, invalidate page after
1068 		 * creating domain.
1069 		 */
1070 #if 0
1071 		if (create && ((softc->pci_cap & AMDVI_PCI_CAP_NPCACHE) == 0))
1072 			continue;
1073 #endif
1074 		amdvi_inv_domain(softc, domain_id);
1075 		amdvi_wait(softc);
1076 	}
1077 }
1078 
1079 static void *
1080 amdvi_create_domain(vm_paddr_t maxaddr)
1081 {
1082 	struct amdvi_domain *dom;
1083 
1084 	dom = malloc(sizeof(struct amdvi_domain), M_AMDVI, M_ZERO | M_WAITOK);
1085 	dom->id = amdvi_domainId();
1086 	//dom->maxaddr = maxaddr;
1087 #ifdef AMDVI_DEBUG_CMD
1088 	printf("Created domain #%d\n", dom->id);
1089 #endif
1090 	/*
1091 	 * Host domain(#0) don't create translation table.
1092 	 */
1093 	if (dom->id || amdvi_host_ptp)
1094 		dom->ptp = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1095 
1096 	dom->ptp_level = amdvi_ptp_level;
1097 
1098 	amdvi_do_inv_domain(dom->id, true);
1099 	SLIST_INSERT_HEAD(&dom_head, dom, next);
1100 
1101 	return (dom);
1102 }
1103 
1104 static void
1105 amdvi_free_ptp(uint64_t *ptp, int level)
1106 {
1107 	int i;
1108 
1109 	if (level < 1)
1110 		return;
1111 
1112 	for (i = 0; i < NPTEPG ; i++) {
1113 		if ((ptp[i] & AMDVI_PT_PRESENT) == 0)
1114 			continue;
1115 		/* XXX: Add super-page or PTE mapping > 4KB. */
1116 #ifdef notyet
1117 		/* Super-page mapping. */
1118 		if (AMDVI_PD_SUPER(ptp[i]))
1119 			continue;
1120 #endif
1121 
1122 		amdvi_free_ptp((uint64_t *)PHYS_TO_DMAP(ptp[i]
1123 		    & AMDVI_PT_MASK), level - 1);
1124 
1125 	}
1126 
1127 	free(ptp, M_AMDVI);
1128 }
1129 
1130 static void
1131 amdvi_destroy_domain(void *arg)
1132 {
1133 	struct amdvi_domain *domain;
1134 
1135 	domain = (struct amdvi_domain *)arg;
1136 	KASSERT(domain, ("domain is NULL"));
1137 #ifdef AMDVI_DEBUG_CMD
1138 	printf("Destroying domain %d\n", domain->id);
1139 #endif
1140 	if (domain->ptp)
1141 		amdvi_free_ptp(domain->ptp, domain->ptp_level);
1142 
1143 	amdvi_do_inv_domain(domain->id, false);
1144 	SLIST_REMOVE(&dom_head, domain, amdvi_domain, next);
1145 	free(domain, M_AMDVI);
1146 }
1147 
1148 static uint64_t
1149 amdvi_set_pt(uint64_t *pt, int level, vm_paddr_t gpa,
1150     vm_paddr_t hpa, uint64_t pg_size, bool create)
1151 {
1152 	uint64_t *page, pa;
1153 	int shift, index;
1154 	const int PT_SHIFT = 9;
1155 	const int PT_INDEX_MASK = (1 << PT_SHIFT) - 1;	/* Based on PT_SHIFT */
1156 
1157 	if (!pg_size)
1158 		return (0);
1159 
1160 	if (hpa & (pg_size - 1)) {
1161 		printf("HPA is not size aligned.\n");
1162 		return (0);
1163 	}
1164 	if (gpa & (pg_size - 1)) {
1165 		printf("HPA is not size aligned.\n");
1166 		return (0);
1167 	}
1168 	shift = PML4SHIFT;
1169 	while ((shift > PAGE_SHIFT) && (pg_size < (1UL << shift))) {
1170 		index = (gpa >> shift) & PT_INDEX_MASK;
1171 
1172 		if ((pt[index] == 0) && create) {
1173 			page = malloc(PAGE_SIZE, M_AMDVI, M_WAITOK | M_ZERO);
1174 			pa = vtophys(page);
1175 			pt[index] = pa | AMDVI_PT_PRESENT | AMDVI_PT_RW |
1176 			    ((level - 1) << AMDVI_PD_LEVEL_SHIFT);
1177 		}
1178 #ifdef AMDVI_DEBUG_PTE
1179 		if ((gpa % 0x1000000) == 0)
1180 			printf("[level%d, shift = %d]PTE:0x%lx\n",
1181 			    level, shift, pt[index]);
1182 #endif
1183 #define PTE2PA(x)	((uint64_t)(x) & AMDVI_PT_MASK)
1184 		pa = PTE2PA(pt[index]);
1185 		pt = (uint64_t *)PHYS_TO_DMAP(pa);
1186 		shift -= PT_SHIFT;
1187 		level--;
1188 	}
1189 
1190 	/* Leaf entry. */
1191 	index = (gpa >> shift) & PT_INDEX_MASK;
1192 
1193 	if (create) {
1194 		pt[index] = hpa | AMDVI_PT_RW | AMDVI_PT_PRESENT;
1195 	} else
1196 		pt[index] = 0;
1197 
1198 #ifdef AMDVI_DEBUG_PTE
1199 	if ((gpa % 0x1000000) == 0)
1200 		printf("[Last level%d, shift = %d]PTE:0x%lx\n",
1201 		    level, shift, pt[index]);
1202 #endif
1203 	return (1ULL << shift);
1204 }
1205 
1206 static uint64_t
1207 amdvi_update_mapping(struct amdvi_domain *domain, vm_paddr_t gpa,
1208     vm_paddr_t hpa, uint64_t size, bool create)
1209 {
1210 	uint64_t mapped, *ptp, len;
1211 	int level;
1212 
1213 	KASSERT(domain, ("domain is NULL"));
1214 	level = domain->ptp_level;
1215 	KASSERT(level, ("Page table level is 0"));
1216 
1217 	ptp = domain->ptp;
1218 	KASSERT(ptp, ("PTP is NULL"));
1219 	mapped = 0;
1220 	while (mapped < size) {
1221 		len = amdvi_set_pt(ptp, level, gpa + mapped, hpa + mapped,
1222 		    PAGE_SIZE, create);
1223 		if (!len) {
1224 			printf("Error: Couldn't map HPA:0x%lx GPA:0x%lx\n",
1225 			    hpa, gpa);
1226 			return (0);
1227 		}
1228 		mapped += len;
1229 	}
1230 
1231 	return (mapped);
1232 }
1233 
1234 static uint64_t
1235 amdvi_create_mapping(void *arg, vm_paddr_t gpa, vm_paddr_t hpa,
1236     uint64_t len)
1237 {
1238 	struct amdvi_domain *domain;
1239 
1240 	domain = (struct amdvi_domain *)arg;
1241 
1242 	if (domain->id && !domain->ptp) {
1243 		printf("ptp is NULL");
1244 		return (-1);
1245 	}
1246 
1247 	/*
1248 	 * If host domain is created w/o page table, skip IOMMU page
1249 	 * table set-up.
1250 	 */
1251 	if (domain->ptp)
1252 		return (amdvi_update_mapping(domain, gpa, hpa, len, true));
1253 	else
1254 		return (len);
1255 }
1256 
1257 static uint64_t
1258 amdvi_destroy_mapping(void *arg, vm_paddr_t gpa, uint64_t len)
1259 {
1260 	struct amdvi_domain *domain;
1261 
1262 	domain = (struct amdvi_domain *)arg;
1263 	/*
1264 	 * If host domain is created w/o page table, skip IOMMU page
1265 	 * table set-up.
1266 	 */
1267 	if (domain->ptp)
1268 		return (amdvi_update_mapping(domain, gpa, 0, len, false));
1269 	return
1270 	    (len);
1271 }
1272 
1273 static struct amdvi_softc *
1274 amdvi_find_iommu(uint16_t devid)
1275 {
1276 	struct amdvi_softc *softc;
1277 	int i;
1278 
1279 	for (i = 0; i < ivhd_count; i++) {
1280 		softc = device_get_softc(ivhd_devs[i]);
1281 		if ((devid >= softc->start_dev_rid) &&
1282 		    (devid <= softc->end_dev_rid))
1283 			return (softc);
1284 	}
1285 
1286 	/*
1287 	 * XXX: BIOS bug, device not in IVRS table, assume its from first IOMMU.
1288 	 */
1289 	printf("BIOS bug device(%d.%d.%d) doesn't have IVHD entry.\n",
1290 	    RID2PCI_STR(devid));
1291 
1292 	return (device_get_softc(ivhd_devs[0]));
1293 }
1294 
1295 /*
1296  * Set-up device table entry.
1297  * IOMMU spec Rev 2.0, section 3.2.2.2, some of the fields must
1298  * be set concurrently, e.g. read and write bits.
1299  */
1300 static void
1301 amdvi_set_dte(struct amdvi_domain *domain, uint16_t devid, bool enable)
1302 {
1303 	struct amdvi_softc *softc;
1304 	struct amdvi_dte* temp;
1305 
1306 	KASSERT(domain, ("domain is NULL for pci_rid:0x%x\n", devid));
1307 
1308 	softc = amdvi_find_iommu(devid);
1309 	KASSERT(softc, ("softc is NULL for pci_rid:0x%x\n", devid));
1310 
1311 	temp = &amdvi_dte[devid];
1312 
1313 #ifdef AMDVI_ATS_ENABLE
1314 	/* If IOMMU and device support IOTLB, enable it. */
1315 	if (amdvi_dev_support_iotlb(softc, devid) && softc->iotlb)
1316 		temp->iotlb_enable = 1;
1317 #endif
1318 
1319 	/* Avoid duplicate I/O faults. */
1320 	temp->sup_second_io_fault = 1;
1321 	temp->sup_all_io_fault = amdvi_disable_io_fault;
1322 
1323 	temp->dt_valid = 1;
1324 	temp->domain_id = domain->id;
1325 
1326 	if (enable) {
1327 		if (domain->ptp) {
1328 			temp->pt_base = vtophys(domain->ptp) >> 12;
1329 			temp->pt_level = amdvi_ptp_level;
1330 		}
1331 		/*
1332 		 * XXX: Page table valid[TV] bit must be set even if host domain
1333 		 * page tables are not enabled.
1334 		 */
1335 		temp->pt_valid = 1;
1336 		temp->read_allow = 1;
1337 		temp->write_allow = 1;
1338 	}
1339 }
1340 
1341 static void
1342 amdvi_inv_device(uint16_t devid)
1343 {
1344 	struct amdvi_softc *softc;
1345 
1346 	softc = amdvi_find_iommu(devid);
1347 	KASSERT(softc, ("softc is NULL"));
1348 
1349 	amdvi_cmd_inv_dte(softc, devid);
1350 #ifdef AMDVI_ATS_ENABLE
1351 	if (amdvi_dev_support_iotlb(softc, devid))
1352 		amdvi_cmd_inv_iotlb(softc, devid);
1353 #endif
1354 	amdvi_wait(softc);
1355 }
1356 
1357 static void
1358 amdvi_add_device(void *arg, uint16_t devid)
1359 {
1360 	struct amdvi_domain *domain;
1361 
1362 	domain = (struct amdvi_domain *)arg;
1363 	KASSERT(domain != NULL, ("domain is NULL"));
1364 #ifdef AMDVI_DEBUG_CMD
1365 	printf("Assigning device(%d.%d.%d) to domain:%d\n",
1366 	    RID2PCI_STR(devid), domain->id);
1367 #endif
1368 	amdvi_set_dte(domain, devid, true);
1369 	amdvi_inv_device(devid);
1370 }
1371 
1372 static void
1373 amdvi_remove_device(void *arg, uint16_t devid)
1374 {
1375 	struct amdvi_domain *domain;
1376 
1377 	domain = (struct amdvi_domain *)arg;
1378 #ifdef AMDVI_DEBUG_CMD
1379 	printf("Remove device(0x%x) from domain:%d\n",
1380 	       devid, domain->id);
1381 #endif
1382 	amdvi_set_dte(domain, devid, false);
1383 	amdvi_inv_device(devid);
1384 }
1385 
1386 static void
1387 amdvi_enable(void)
1388 {
1389 	struct amdvi_ctrl *ctrl;
1390 	struct amdvi_softc *softc;
1391 	uint64_t val;
1392 	int i;
1393 
1394 	for (i = 0; i < ivhd_count; i++) {
1395 		softc = device_get_softc(ivhd_devs[i]);
1396 		KASSERT(softc, ("softc is NULL\n"));
1397 		ctrl = softc->ctrl;
1398 		KASSERT(ctrl, ("ctrl is NULL\n"));
1399 
1400 		val = (	AMDVI_CTRL_EN 		|
1401 			AMDVI_CTRL_CMD 		|
1402 		    	AMDVI_CTRL_ELOG 	|
1403 		    	AMDVI_CTRL_ELOGINT 	|
1404 		    	AMDVI_CTRL_INV_TO_1S);
1405 
1406 		if (softc->ivhd_flag & IVHD_FLAG_COH)
1407 			val |= AMDVI_CTRL_COH;
1408 		if (softc->ivhd_flag & IVHD_FLAG_HTT)
1409 			val |= AMDVI_CTRL_HTT;
1410 		if (softc->ivhd_flag & IVHD_FLAG_RPPW)
1411 			val |= AMDVI_CTRL_RPPW;
1412 		if (softc->ivhd_flag & IVHD_FLAG_PPW)
1413 			val |= AMDVI_CTRL_PPW;
1414 		if (softc->ivhd_flag & IVHD_FLAG_ISOC)
1415 			val |= AMDVI_CTRL_ISOC;
1416 
1417 		ctrl->control = val;
1418 	}
1419 }
1420 
1421 static void
1422 amdvi_disable(void)
1423 {
1424 	struct amdvi_ctrl *ctrl;
1425 	struct amdvi_softc *softc;
1426 	int i;
1427 
1428 	for (i = 0; i < ivhd_count; i++) {
1429 		softc = device_get_softc(ivhd_devs[i]);
1430 		KASSERT(softc, ("softc is NULL\n"));
1431 		ctrl = softc->ctrl;
1432 		KASSERT(ctrl, ("ctrl is NULL\n"));
1433 
1434 		ctrl->control = 0;
1435 	}
1436 }
1437 
1438 static void
1439 amdvi_inv_tlb(void *arg)
1440 {
1441 	struct amdvi_domain *domain;
1442 
1443 	domain = (struct amdvi_domain *)arg;
1444 	KASSERT(domain, ("domain is NULL"));
1445 	amdvi_do_inv_domain(domain->id, false);
1446 }
1447 
1448 struct iommu_ops iommu_ops_amd = {
1449 	amdvi_init,
1450 	amdvi_cleanup,
1451 	amdvi_enable,
1452 	amdvi_disable,
1453 	amdvi_create_domain,
1454 	amdvi_destroy_domain,
1455 	amdvi_create_mapping,
1456 	amdvi_destroy_mapping,
1457 	amdvi_add_device,
1458 	amdvi_remove_device,
1459 	amdvi_inv_tlb
1460 };
1461