xref: /freebsd/sys/arm64/iommu/smmu.c (revision 15f0b8c3)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2019-2020 Ruslan Bukin <br@bsdpad.com>
5  *
6  * This software was developed by SRI International and the University of
7  * Cambridge Computer Laboratory (Department of Computer Science and
8  * Technology) under DARPA contract HR0011-18-C-0016 ("ECATS"), as part of the
9  * DARPA SSITH research programme.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*
34  * Hardware overview.
35  *
36  * An incoming transaction from a peripheral device has an address, size,
37  * attributes and StreamID.
38  *
39  * In case of PCI-based devices, StreamID is a PCI rid.
40  *
41  * The StreamID is used to select a Stream Table Entry (STE) in a Stream table,
42  * which contains per-device configuration.
43  *
44  * Stream table is a linear or 2-level walk table (this driver supports both).
45  * Note that a linear table could occupy 1GB or more of memory depending on
46  * sid_bits value.
47  *
48  * STE is used to locate a Context Descriptor, which is a struct in memory
49  * that describes stages of translation, translation table type, pointer to
50  * level 0 of page tables, ASID, etc.
51  *
52  * Hardware supports two stages of translation: Stage1 (S1) and Stage2 (S2):
53  *  o S1 is used for the host machine traffic translation
54  *  o S2 is for a hypervisor
55  *
56  * This driver enables S1 stage with standard AArch64 page tables.
57  *
58  * Note that SMMU does not share TLB with a main CPU.
59  * Command queue is used by this driver to Invalidate SMMU TLB, STE cache.
60  *
61  * An arm64 SoC could have more than one SMMU instance.
62  * ACPI IORT table describes which SMMU unit is assigned for a particular
63  * peripheral device.
64  *
65  * Queues.
66  *
67  * Register interface and Memory-based circular buffer queues are used
68  * to inferface SMMU.
69  *
70  * These are a Command queue for commands to send to the SMMU and an Event
71  * queue for event/fault reports from the SMMU. Optionally PRI queue is
72  * designed for PCIe page requests reception.
73  *
74  * Note that not every hardware supports PRI services. For instance they were
75  * not found in Neoverse N1 SDP machine.
76  * (This drivers does not implement PRI queue.)
77  *
78  * All SMMU queues are arranged as circular buffers in memory. They are used
79  * in a producer-consumer fashion so that an output queue contains data
80  * produced by the SMMU and consumed by software.
81  * An input queue contains data produced by software, consumed by the SMMU.
82  *
83  * Interrupts.
84  *
85  * Interrupts are not required by this driver for normal operation.
86  * The standard wired interrupt is only triggered when an event comes from
87  * the SMMU, which is only in a case of errors (e.g. translation fault).
88  */
89 
90 #include "opt_platform.h"
91 #include "opt_acpi.h"
92 
93 #include <sys/cdefs.h>
94 __FBSDID("$FreeBSD$");
95 
96 #include <sys/param.h>
97 #include <sys/bitstring.h>
98 #include <sys/bus.h>
99 #include <sys/kernel.h>
100 #include <sys/malloc.h>
101 #include <sys/mutex.h>
102 #include <sys/rman.h>
103 #include <sys/lock.h>
104 #include <sys/sysctl.h>
105 #include <sys/tree.h>
106 #include <sys/taskqueue.h>
107 #include <vm/vm.h>
108 #include <vm/vm_page.h>
109 #ifdef DEV_ACPI
110 #include <contrib/dev/acpica/include/acpi.h>
111 #include <dev/acpica/acpivar.h>
112 #endif
113 #include <dev/pci/pcireg.h>
114 #include <dev/pci/pcivar.h>
115 #include <dev/iommu/iommu.h>
116 #include <arm64/iommu/iommu_pmap.h>
117 
118 #include <machine/bus.h>
119 
120 #ifdef FDT
121 #include <dev/fdt/fdt_common.h>
122 #include <dev/ofw/ofw_bus.h>
123 #include <dev/ofw/ofw_bus_subr.h>
124 #endif
125 
126 #include "iommu.h"
127 #include "iommu_if.h"
128 
129 #include "smmureg.h"
130 #include "smmuvar.h"
131 
132 #define	STRTAB_L1_SZ_SHIFT	20
133 #define	STRTAB_SPLIT		8
134 
135 #define	STRTAB_L1_DESC_L2PTR_M	(0x3fffffffffff << 6)
136 #define	STRTAB_L1_DESC_DWORDS	1
137 
138 #define	STRTAB_STE_DWORDS	8
139 
140 #define	CMDQ_ENTRY_DWORDS	2
141 #define	EVTQ_ENTRY_DWORDS	4
142 #define	PRIQ_ENTRY_DWORDS	2
143 
144 #define	CD_DWORDS		8
145 
146 #define	Q_WRP(q, p)		((p) & (1 << (q)->size_log2))
147 #define	Q_IDX(q, p)		((p) & ((1 << (q)->size_log2) - 1))
148 #define	Q_OVF(p)		((p) & (1 << 31)) /* Event queue overflowed */
149 
150 #define	SMMU_Q_ALIGN		(64 * 1024)
151 
152 #define		MAXADDR_48BIT	0xFFFFFFFFFFFFUL
153 #define		MAXADDR_52BIT	0xFFFFFFFFFFFFFUL
154 
155 static struct resource_spec smmu_spec[] = {
156 	{ SYS_RES_MEMORY, 0, RF_ACTIVE },
157 	{ SYS_RES_IRQ, 0, RF_ACTIVE },
158 	{ SYS_RES_IRQ, 1, RF_ACTIVE | RF_OPTIONAL },
159 	{ SYS_RES_IRQ, 2, RF_ACTIVE },
160 	{ SYS_RES_IRQ, 3, RF_ACTIVE },
161 	RESOURCE_SPEC_END
162 };
163 
164 MALLOC_DEFINE(M_SMMU, "SMMU", SMMU_DEVSTR);
165 
166 #define	dprintf(fmt, ...)
167 
168 struct smmu_event {
169 	int ident;
170 	char *str;
171 	char *msg;
172 };
173 
174 static struct smmu_event events[] = {
175 	{ 0x01, "F_UUT",
176 		"Unsupported Upstream Transaction."},
177 	{ 0x02, "C_BAD_STREAMID",
178 		"Transaction StreamID out of range."},
179 	{ 0x03, "F_STE_FETCH",
180 		"Fetch of STE caused external abort."},
181 	{ 0x04, "C_BAD_STE",
182 		"Used STE invalid."},
183 	{ 0x05, "F_BAD_ATS_TREQ",
184 		"Address Translation Request disallowed for a StreamID "
185 		"and a PCIe ATS Translation Request received."},
186 	{ 0x06, "F_STREAM_DISABLED",
187 		"The STE of a transaction marks non-substream transactions "
188 		"disabled."},
189 	{ 0x07, "F_TRANSL_FORBIDDEN",
190 		"An incoming PCIe transaction is marked Translated but "
191 		"SMMU bypass is disallowed for this StreamID."},
192 	{ 0x08, "C_BAD_SUBSTREAMID",
193 		"Incoming SubstreamID present, but configuration is invalid."},
194 	{ 0x09, "F_CD_FETCH",
195 		"Fetch of CD caused external abort."},
196 	{ 0x0a, "C_BAD_CD",
197 		"Fetched CD invalid."},
198 	{ 0x0b, "F_WALK_EABT",
199 		"An external abort occurred fetching (or updating) "
200 		"a translation table descriptor."},
201 	{ 0x10, "F_TRANSLATION",
202 		"Translation fault."},
203 	{ 0x11, "F_ADDR_SIZE",
204 		"Address Size fault."},
205 	{ 0x12, "F_ACCESS",
206 		"Access flag fault due to AF == 0 in a page or block TTD."},
207 	{ 0x13, "F_PERMISSION",
208 		"Permission fault occurred on page access."},
209 	{ 0x20, "F_TLB_CONFLICT",
210 		"A TLB conflict occurred because of the transaction."},
211 	{ 0x21, "F_CFG_CONFLICT",
212 		"A configuration cache conflict occurred due to "
213 		"the transaction."},
214 	{ 0x24, "E_PAGE_REQUEST",
215 		"Speculative page request hint."},
216 	{ 0x25, "F_VMS_FETCH",
217 		"Fetch of VMS caused external abort."},
218 	{ 0, NULL, NULL },
219 };
220 
221 static int
222 smmu_q_has_space(struct smmu_queue *q)
223 {
224 
225 	/*
226 	 * See 6.3.27 SMMU_CMDQ_PROD
227 	 *
228 	 * There is space in the queue for additional commands if:
229 	 *  SMMU_CMDQ_CONS.RD != SMMU_CMDQ_PROD.WR ||
230 	 *  SMMU_CMDQ_CONS.RD_WRAP == SMMU_CMDQ_PROD.WR_WRAP
231 	 */
232 
233 	if (Q_IDX(q, q->lc.cons) != Q_IDX(q, q->lc.prod) ||
234 	    Q_WRP(q, q->lc.cons) == Q_WRP(q, q->lc.prod))
235 		return (1);
236 
237 	return (0);
238 }
239 
240 static int
241 smmu_q_empty(struct smmu_queue *q)
242 {
243 
244 	if (Q_IDX(q, q->lc.cons) == Q_IDX(q, q->lc.prod) &&
245 	    Q_WRP(q, q->lc.cons) == Q_WRP(q, q->lc.prod))
246 		return (1);
247 
248 	return (0);
249 }
250 
251 static int __unused
252 smmu_q_consumed(struct smmu_queue *q, uint32_t prod)
253 {
254 
255 	if ((Q_WRP(q, q->lc.cons) == Q_WRP(q, prod)) &&
256 	    (Q_IDX(q, q->lc.cons) >= Q_IDX(q, prod)))
257 		return (1);
258 
259 	if ((Q_WRP(q, q->lc.cons) != Q_WRP(q, prod)) &&
260 	    (Q_IDX(q, q->lc.cons) <= Q_IDX(q, prod)))
261 		return (1);
262 
263 	return (0);
264 }
265 
266 static uint32_t
267 smmu_q_inc_cons(struct smmu_queue *q)
268 {
269 	uint32_t cons;
270 	uint32_t val;
271 
272 	cons = (Q_WRP(q, q->lc.cons) | Q_IDX(q, q->lc.cons)) + 1;
273 	val = (Q_OVF(q->lc.cons) | Q_WRP(q, cons) | Q_IDX(q, cons));
274 
275 	return (val);
276 }
277 
278 static uint32_t
279 smmu_q_inc_prod(struct smmu_queue *q)
280 {
281 	uint32_t prod;
282 	uint32_t val;
283 
284 	prod = (Q_WRP(q, q->lc.prod) | Q_IDX(q, q->lc.prod)) + 1;
285 	val = (Q_OVF(q->lc.prod) | Q_WRP(q, prod) | Q_IDX(q, prod));
286 
287 	return (val);
288 }
289 
290 static int
291 smmu_write_ack(struct smmu_softc *sc, uint32_t reg,
292     uint32_t reg_ack, uint32_t val)
293 {
294 	uint32_t v;
295 	int timeout;
296 
297 	timeout = 100000;
298 
299 	bus_write_4(sc->res[0], reg, val);
300 
301 	do {
302 		v = bus_read_4(sc->res[0], reg_ack);
303 		if (v == val)
304 			break;
305 	} while (timeout--);
306 
307 	if (timeout <= 0) {
308 		device_printf(sc->dev, "Failed to write reg.\n");
309 		return (-1);
310 	}
311 
312 	return (0);
313 }
314 
315 static inline int
316 ilog2(long x)
317 {
318 
319 	KASSERT(x > 0 && powerof2(x), ("%s: invalid arg %ld", __func__, x));
320 
321 	return (flsl(x) - 1);
322 }
323 
324 static int
325 smmu_init_queue(struct smmu_softc *sc, struct smmu_queue *q,
326     uint32_t prod_off, uint32_t cons_off, uint32_t dwords)
327 {
328 	int sz;
329 
330 	sz = (1 << q->size_log2) * dwords * 8;
331 
332 	/* Set up the command circular buffer */
333 	q->vaddr = contigmalloc(sz, M_SMMU,
334 	    M_WAITOK | M_ZERO, 0, (1ul << 48) - 1, SMMU_Q_ALIGN, 0);
335 	if (q->vaddr == NULL) {
336 		device_printf(sc->dev, "failed to allocate %d bytes\n", sz);
337 		return (-1);
338 	}
339 
340 	q->prod_off = prod_off;
341 	q->cons_off = cons_off;
342 	q->paddr = vtophys(q->vaddr);
343 
344 	q->base = CMDQ_BASE_RA | EVENTQ_BASE_WA | PRIQ_BASE_WA;
345 	q->base |= q->paddr & Q_BASE_ADDR_M;
346 	q->base |= q->size_log2 << Q_LOG2SIZE_S;
347 
348 	return (0);
349 }
350 
351 static int
352 smmu_init_queues(struct smmu_softc *sc)
353 {
354 	int err;
355 
356 	/* Command queue. */
357 	err = smmu_init_queue(sc, &sc->cmdq,
358 	    SMMU_CMDQ_PROD, SMMU_CMDQ_CONS, CMDQ_ENTRY_DWORDS);
359 	if (err)
360 		return (ENXIO);
361 
362 	/* Event queue. */
363 	err = smmu_init_queue(sc, &sc->evtq,
364 	    SMMU_EVENTQ_PROD, SMMU_EVENTQ_CONS, EVTQ_ENTRY_DWORDS);
365 	if (err)
366 		return (ENXIO);
367 
368 	if (!(sc->features & SMMU_FEATURE_PRI))
369 		return (0);
370 
371 	/* PRI queue. */
372 	err = smmu_init_queue(sc, &sc->priq,
373 	    SMMU_PRIQ_PROD, SMMU_PRIQ_CONS, PRIQ_ENTRY_DWORDS);
374 	if (err)
375 		return (ENXIO);
376 
377 	return (0);
378 }
379 
380 /*
381  * Dump 2LVL or linear STE.
382  */
383 static void
384 smmu_dump_ste(struct smmu_softc *sc, int sid)
385 {
386 	struct smmu_strtab *strtab;
387 	struct l1_desc *l1_desc;
388 	uint64_t *ste, *l1;
389 	int i;
390 
391 	strtab = &sc->strtab;
392 
393 	if (sc->features & SMMU_FEATURE_2_LVL_STREAM_TABLE) {
394 		i = sid >> STRTAB_SPLIT;
395 		l1 = (void *)((uint64_t)strtab->vaddr +
396 		    STRTAB_L1_DESC_DWORDS * 8 * i);
397 		device_printf(sc->dev, "L1 ste == %lx\n", l1[0]);
398 
399 		l1_desc = &strtab->l1[i];
400 		ste = l1_desc->va;
401 		if (ste == NULL) /* L2 is not initialized */
402 			return;
403 	} else {
404 		ste = (void *)((uint64_t)strtab->vaddr +
405 		    sid * (STRTAB_STE_DWORDS << 3));
406 	}
407 
408 	/* Dump L2 or linear STE. */
409 	for (i = 0; i < STRTAB_STE_DWORDS; i++)
410 		device_printf(sc->dev, "ste[%d] == %lx\n", i, ste[i]);
411 }
412 
413 static void __unused
414 smmu_dump_cd(struct smmu_softc *sc, struct smmu_cd *cd)
415 {
416 	uint64_t *vaddr;
417 	int i;
418 
419 	device_printf(sc->dev, "%s\n", __func__);
420 
421 	vaddr = cd->vaddr;
422 	for (i = 0; i < CD_DWORDS; i++)
423 		device_printf(sc->dev, "cd[%d] == %lx\n", i, vaddr[i]);
424 }
425 
426 static void
427 smmu_evtq_dequeue(struct smmu_softc *sc, uint32_t *evt)
428 {
429 	struct smmu_queue *evtq;
430 	void *entry_addr;
431 
432 	evtq = &sc->evtq;
433 
434 	evtq->lc.val = bus_read_8(sc->res[0], evtq->prod_off);
435 	entry_addr = (void *)((uint64_t)evtq->vaddr +
436 	    evtq->lc.cons * EVTQ_ENTRY_DWORDS * 8);
437 	memcpy(evt, entry_addr, EVTQ_ENTRY_DWORDS * 8);
438 	evtq->lc.cons = smmu_q_inc_cons(evtq);
439 	bus_write_4(sc->res[0], evtq->cons_off, evtq->lc.cons);
440 }
441 
442 static void
443 smmu_print_event(struct smmu_softc *sc, uint32_t *evt)
444 {
445 	struct smmu_event *ev;
446 	uintptr_t input_addr;
447 	uint8_t event_id;
448 	device_t dev;
449 	int sid;
450 	int i;
451 
452 	dev = sc->dev;
453 
454 	ev = NULL;
455 	event_id = evt[0] & 0xff;
456 	for (i = 0; events[i].ident != 0; i++) {
457 		if (events[i].ident == event_id) {
458 			ev = &events[i];
459 			break;
460 		}
461 	}
462 
463 	sid = evt[1];
464 	input_addr = evt[5];
465 	input_addr <<= 32;
466 	input_addr |= evt[4];
467 
468 	if (smmu_quirks_check(dev, sid, event_id, input_addr)) {
469 		/* The event is known. Don't print anything. */
470 		return;
471 	}
472 
473 	if (ev) {
474 		device_printf(sc->dev,
475 		    "Event %s (%s) received.\n", ev->str, ev->msg);
476 	} else
477 		device_printf(sc->dev, "Event 0x%x received\n", event_id);
478 
479 	device_printf(sc->dev, "SID %x, Input Address: %jx\n",
480 	    sid, input_addr);
481 
482 	for (i = 0; i < 8; i++)
483 		device_printf(sc->dev, "evt[%d] %x\n", i, evt[i]);
484 
485 	smmu_dump_ste(sc, sid);
486 }
487 
488 static void
489 make_cmd(struct smmu_softc *sc, uint64_t *cmd,
490     struct smmu_cmdq_entry *entry)
491 {
492 
493 	memset(cmd, 0, CMDQ_ENTRY_DWORDS * 8);
494 	cmd[0] = entry->opcode << CMD_QUEUE_OPCODE_S;
495 
496 	switch (entry->opcode) {
497 	case CMD_TLBI_NH_VA:
498 		cmd[0] |= (uint64_t)entry->tlbi.asid << TLBI_0_ASID_S;
499 		cmd[1] = entry->tlbi.addr & TLBI_1_ADDR_M;
500 		if (entry->tlbi.leaf) {
501 			/*
502 			 * Leaf flag means that only cached entries
503 			 * for the last level of translation table walk
504 			 * are required to be invalidated.
505 			 */
506 			cmd[1] |= TLBI_1_LEAF;
507 		}
508 		break;
509 	case CMD_TLBI_NH_ASID:
510 		cmd[0] |= (uint64_t)entry->tlbi.asid << TLBI_0_ASID_S;
511 		break;
512 	case CMD_TLBI_NSNH_ALL:
513 	case CMD_TLBI_NH_ALL:
514 	case CMD_TLBI_EL2_ALL:
515 		break;
516 	case CMD_CFGI_CD:
517 		cmd[0] |= ((uint64_t)entry->cfgi.ssid << CFGI_0_SSID_S);
518 		/* FALLTROUGH */
519 	case CMD_CFGI_STE:
520 		cmd[0] |= ((uint64_t)entry->cfgi.sid << CFGI_0_STE_SID_S);
521 		cmd[1] |= ((uint64_t)entry->cfgi.leaf << CFGI_1_LEAF_S);
522 		break;
523 	case CMD_CFGI_STE_RANGE:
524 		cmd[1] = (31 << CFGI_1_STE_RANGE_S);
525 		break;
526 	case CMD_SYNC:
527 		cmd[0] |= SYNC_0_MSH_IS | SYNC_0_MSIATTR_OIWB;
528 		if (entry->sync.msiaddr) {
529 			cmd[0] |= SYNC_0_CS_SIG_IRQ;
530 			cmd[1] |= (entry->sync.msiaddr & SYNC_1_MSIADDRESS_M);
531 		} else
532 			cmd[0] |= SYNC_0_CS_SIG_SEV;
533 		break;
534 	case CMD_PREFETCH_CONFIG:
535 		cmd[0] |= ((uint64_t)entry->prefetch.sid << PREFETCH_0_SID_S);
536 		break;
537 	};
538 }
539 
540 static void
541 smmu_cmdq_enqueue_cmd(struct smmu_softc *sc, struct smmu_cmdq_entry *entry)
542 {
543 	uint64_t cmd[CMDQ_ENTRY_DWORDS];
544 	struct smmu_queue *cmdq;
545 	void *entry_addr;
546 
547 	cmdq = &sc->cmdq;
548 
549 	make_cmd(sc, cmd, entry);
550 
551 	SMMU_LOCK(sc);
552 
553 	/* Ensure that a space is available. */
554 	do {
555 		cmdq->lc.cons = bus_read_4(sc->res[0], cmdq->cons_off);
556 	} while (smmu_q_has_space(cmdq) == 0);
557 
558 	/* Write the command to the current prod entry. */
559 	entry_addr = (void *)((uint64_t)cmdq->vaddr +
560 	    Q_IDX(cmdq, cmdq->lc.prod) * CMDQ_ENTRY_DWORDS * 8);
561 	memcpy(entry_addr, cmd, CMDQ_ENTRY_DWORDS * 8);
562 
563 	/* Increment prod index. */
564 	cmdq->lc.prod = smmu_q_inc_prod(cmdq);
565 	bus_write_4(sc->res[0], cmdq->prod_off, cmdq->lc.prod);
566 
567 	SMMU_UNLOCK(sc);
568 }
569 
570 static void __unused
571 smmu_poll_until_consumed(struct smmu_softc *sc, struct smmu_queue *q)
572 {
573 
574 	while (1) {
575 		q->lc.val = bus_read_8(sc->res[0], q->prod_off);
576 		if (smmu_q_empty(q))
577 			break;
578 		cpu_spinwait();
579 	}
580 }
581 
582 static int
583 smmu_sync(struct smmu_softc *sc)
584 {
585 	struct smmu_cmdq_entry cmd;
586 	struct smmu_queue *q;
587 	uint32_t *base;
588 	int timeout;
589 	int prod;
590 
591 	q = &sc->cmdq;
592 	prod = q->lc.prod;
593 
594 	/* Enqueue sync command. */
595 	cmd.opcode = CMD_SYNC;
596 	cmd.sync.msiaddr = q->paddr + Q_IDX(q, prod) * CMDQ_ENTRY_DWORDS * 8;
597 	smmu_cmdq_enqueue_cmd(sc, &cmd);
598 
599 	/* Wait for the sync completion. */
600 	base = (void *)((uint64_t)q->vaddr +
601 	    Q_IDX(q, prod) * CMDQ_ENTRY_DWORDS * 8);
602 
603 	/*
604 	 * It takes around 200 loops (6 instructions each)
605 	 * on Neoverse N1 to complete the sync.
606 	 */
607 	timeout = 10000;
608 
609 	do {
610 		if (*base == 0) {
611 			/* MSI write completed. */
612 			break;
613 		}
614 		cpu_spinwait();
615 	} while (timeout--);
616 
617 	if (timeout < 0)
618 		device_printf(sc->dev, "Failed to sync\n");
619 
620 	return (0);
621 }
622 
623 static int
624 smmu_sync_cd(struct smmu_softc *sc, int sid, int ssid, bool leaf)
625 {
626 	struct smmu_cmdq_entry cmd;
627 
628 	cmd.opcode = CMD_CFGI_CD;
629 	cmd.cfgi.sid = sid;
630 	cmd.cfgi.ssid = ssid;
631 	cmd.cfgi.leaf = leaf;
632 	smmu_cmdq_enqueue_cmd(sc, &cmd);
633 
634 	return (0);
635 }
636 
637 static void
638 smmu_invalidate_all_sid(struct smmu_softc *sc)
639 {
640 	struct smmu_cmdq_entry cmd;
641 
642 	/* Invalidate cached config */
643 	cmd.opcode = CMD_CFGI_STE_RANGE;
644 	smmu_cmdq_enqueue_cmd(sc, &cmd);
645 	smmu_sync(sc);
646 }
647 
648 static void
649 smmu_tlbi_all(struct smmu_softc *sc)
650 {
651 	struct smmu_cmdq_entry cmd;
652 
653 	/* Invalidate entire TLB */
654 	cmd.opcode = CMD_TLBI_NSNH_ALL;
655 	smmu_cmdq_enqueue_cmd(sc, &cmd);
656 	smmu_sync(sc);
657 }
658 
659 static void
660 smmu_tlbi_asid(struct smmu_softc *sc, uint16_t asid)
661 {
662 	struct smmu_cmdq_entry cmd;
663 
664 	/* Invalidate TLB for an ASID. */
665 	cmd.opcode = CMD_TLBI_NH_ASID;
666 	cmd.tlbi.asid = asid;
667 	smmu_cmdq_enqueue_cmd(sc, &cmd);
668 	smmu_sync(sc);
669 }
670 
671 static void
672 smmu_tlbi_va(struct smmu_softc *sc, vm_offset_t va, uint16_t asid)
673 {
674 	struct smmu_cmdq_entry cmd;
675 
676 	/* Invalidate specific range */
677 	cmd.opcode = CMD_TLBI_NH_VA;
678 	cmd.tlbi.asid = asid;
679 	cmd.tlbi.vmid = 0;
680 	cmd.tlbi.leaf = true; /* We change only L3. */
681 	cmd.tlbi.addr = va;
682 	smmu_cmdq_enqueue_cmd(sc, &cmd);
683 }
684 
685 static void
686 smmu_invalidate_sid(struct smmu_softc *sc, uint32_t sid)
687 {
688 	struct smmu_cmdq_entry cmd;
689 
690 	/* Invalidate cached config */
691 	cmd.opcode = CMD_CFGI_STE;
692 	cmd.cfgi.sid = sid;
693 	smmu_cmdq_enqueue_cmd(sc, &cmd);
694 	smmu_sync(sc);
695 }
696 
697 static void
698 smmu_prefetch_sid(struct smmu_softc *sc, uint32_t sid)
699 {
700 	struct smmu_cmdq_entry cmd;
701 
702 	cmd.opcode = CMD_PREFETCH_CONFIG;
703 	cmd.prefetch.sid = sid;
704 	smmu_cmdq_enqueue_cmd(sc, &cmd);
705 	smmu_sync(sc);
706 }
707 
708 /*
709  * Init STE in bypass mode. Traffic is not translated for the sid.
710  */
711 static void
712 smmu_init_ste_bypass(struct smmu_softc *sc, uint32_t sid, uint64_t *ste)
713 {
714 	uint64_t val;
715 
716 	val = STE0_VALID | STE0_CONFIG_BYPASS;
717 
718 	ste[1] = STE1_SHCFG_INCOMING | STE1_EATS_FULLATS;
719 	ste[2] = 0;
720 	ste[3] = 0;
721 	ste[4] = 0;
722 	ste[5] = 0;
723 	ste[6] = 0;
724 	ste[7] = 0;
725 
726 	smmu_invalidate_sid(sc, sid);
727 	ste[0] = val;
728 	dsb(sy);
729 	smmu_invalidate_sid(sc, sid);
730 
731 	smmu_prefetch_sid(sc, sid);
732 }
733 
734 /*
735  * Enable Stage1 (S1) translation for the sid.
736  */
737 static int
738 smmu_init_ste_s1(struct smmu_softc *sc, struct smmu_cd *cd,
739     uint32_t sid, uint64_t *ste)
740 {
741 	uint64_t val;
742 
743 	val = STE0_VALID;
744 
745 	/* S1 */
746 	ste[1] = STE1_EATS_FULLATS	|
747 		 STE1_S1CSH_IS		|
748 		 STE1_S1CIR_WBRA	|
749 		 STE1_S1COR_WBRA	|
750 		 STE1_STRW_NS_EL1;
751 	ste[2] = 0;
752 	ste[3] = 0;
753 	ste[4] = 0;
754 	ste[5] = 0;
755 	ste[6] = 0;
756 	ste[7] = 0;
757 
758 	if (sc->features & SMMU_FEATURE_STALL &&
759 	    ((sc->features & SMMU_FEATURE_STALL_FORCE) == 0))
760 		ste[1] |= STE1_S1STALLD;
761 
762 	/* Configure STE */
763 	val |= (cd->paddr & STE0_S1CONTEXTPTR_M);
764 	val |= STE0_CONFIG_S1_TRANS;
765 
766 	smmu_invalidate_sid(sc, sid);
767 
768 	/* The STE[0] has to be written in a single blast, last of all. */
769 	ste[0] = val;
770 	dsb(sy);
771 
772 	smmu_invalidate_sid(sc, sid);
773 	smmu_sync_cd(sc, sid, 0, true);
774 	smmu_invalidate_sid(sc, sid);
775 
776 	/* The sid will be used soon most likely. */
777 	smmu_prefetch_sid(sc, sid);
778 
779 	return (0);
780 }
781 
782 static uint64_t *
783 smmu_get_ste_addr(struct smmu_softc *sc, int sid)
784 {
785 	struct smmu_strtab *strtab;
786 	struct l1_desc *l1_desc;
787 	uint64_t *addr;
788 
789 	strtab = &sc->strtab;
790 
791 	if (sc->features & SMMU_FEATURE_2_LVL_STREAM_TABLE) {
792 		l1_desc = &strtab->l1[sid >> STRTAB_SPLIT];
793 		addr = l1_desc->va;
794 		addr += (sid & ((1 << STRTAB_SPLIT) - 1)) * STRTAB_STE_DWORDS;
795 	} else {
796 		addr = (void *)((uint64_t)strtab->vaddr +
797 		    STRTAB_STE_DWORDS * 8 * sid);
798 	};
799 
800 	return (addr);
801 }
802 
803 static int
804 smmu_init_ste(struct smmu_softc *sc, struct smmu_cd *cd, int sid, bool bypass)
805 {
806 	uint64_t *addr;
807 
808 	addr = smmu_get_ste_addr(sc, sid);
809 
810 	if (bypass)
811 		smmu_init_ste_bypass(sc, sid, addr);
812 	else
813 		smmu_init_ste_s1(sc, cd, sid, addr);
814 
815 	smmu_sync(sc);
816 
817 	return (0);
818 }
819 
820 static void
821 smmu_deinit_ste(struct smmu_softc *sc, int sid)
822 {
823 	uint64_t *ste;
824 
825 	ste = smmu_get_ste_addr(sc, sid);
826 	ste[0] = 0;
827 
828 	smmu_invalidate_sid(sc, sid);
829 	smmu_sync_cd(sc, sid, 0, true);
830 	smmu_invalidate_sid(sc, sid);
831 
832 	smmu_sync(sc);
833 }
834 
835 static int
836 smmu_init_cd(struct smmu_softc *sc, struct smmu_domain *domain)
837 {
838 	vm_paddr_t paddr;
839 	uint64_t *ptr;
840 	uint64_t val;
841 	vm_size_t size;
842 	struct smmu_cd *cd;
843 	pmap_t p;
844 
845 	size = 1 * (CD_DWORDS << 3);
846 
847 	p = &domain->p;
848 	cd = domain->cd = malloc(sizeof(struct smmu_cd),
849 	    M_SMMU, M_WAITOK | M_ZERO);
850 
851 	cd->vaddr = contigmalloc(size, M_SMMU,
852 	    M_WAITOK | M_ZERO,	/* flags */
853 	    0,			/* low */
854 	    (1ul << 40) - 1,	/* high */
855 	    size,		/* alignment */
856 	    0);			/* boundary */
857 	if (cd->vaddr == NULL) {
858 		device_printf(sc->dev, "Failed to allocate CD\n");
859 		return (ENXIO);
860 	}
861 
862 	cd->size = size;
863 	cd->paddr = vtophys(cd->vaddr);
864 
865 	ptr = cd->vaddr;
866 
867 	val = CD0_VALID;
868 	val |= CD0_AA64;
869 	val |= CD0_R;
870 	val |= CD0_A;
871 	val |= CD0_ASET;
872 	val |= (uint64_t)domain->asid << CD0_ASID_S;
873 	val |= CD0_TG0_4KB;
874 	val |= CD0_EPD1; /* Disable TT1 */
875 	val |= ((64 - sc->ias) << CD0_T0SZ_S);
876 	val |= CD0_IPS_48BITS;
877 
878 	paddr = p->pm_l0_paddr & CD1_TTB0_M;
879 	KASSERT(paddr == p->pm_l0_paddr, ("bad allocation 1"));
880 
881 	ptr[1] = paddr;
882 	ptr[2] = 0;
883 	ptr[3] = MAIR_ATTR(MAIR_DEVICE_nGnRnE, VM_MEMATTR_DEVICE)	|
884 		 MAIR_ATTR(MAIR_NORMAL_NC, VM_MEMATTR_UNCACHEABLE)	|
885 		 MAIR_ATTR(MAIR_NORMAL_WB, VM_MEMATTR_WRITE_BACK)	|
886 		 MAIR_ATTR(MAIR_NORMAL_WT, VM_MEMATTR_WRITE_THROUGH);
887 
888 	/* Install the CD. */
889 	ptr[0] = val;
890 
891 	return (0);
892 }
893 
894 static int
895 smmu_init_strtab_linear(struct smmu_softc *sc)
896 {
897 	struct smmu_strtab *strtab;
898 	vm_paddr_t base;
899 	uint32_t size;
900 	uint64_t reg;
901 
902 	strtab = &sc->strtab;
903 	strtab->num_l1_entries = (1 << sc->sid_bits);
904 
905 	size = strtab->num_l1_entries * (STRTAB_STE_DWORDS << 3);
906 
907 	if (bootverbose)
908 		device_printf(sc->dev,
909 		    "%s: linear strtab size %d, num_l1_entries %d\n",
910 		    __func__, size, strtab->num_l1_entries);
911 
912 	strtab->vaddr = contigmalloc(size, M_SMMU,
913 	    M_WAITOK | M_ZERO,	/* flags */
914 	    0,			/* low */
915 	    (1ul << 48) - 1,	/* high */
916 	    size,		/* alignment */
917 	    0);			/* boundary */
918 	if (strtab->vaddr == NULL) {
919 		device_printf(sc->dev, "failed to allocate strtab\n");
920 		return (ENXIO);
921 	}
922 
923 	reg = STRTAB_BASE_CFG_FMT_LINEAR;
924 	reg |= sc->sid_bits << STRTAB_BASE_CFG_LOG2SIZE_S;
925 	strtab->base_cfg = (uint32_t)reg;
926 
927 	base = vtophys(strtab->vaddr);
928 
929 	reg = base & STRTAB_BASE_ADDR_M;
930 	KASSERT(reg == base, ("bad allocation 2"));
931 	reg |= STRTAB_BASE_RA;
932 	strtab->base = reg;
933 
934 	return (0);
935 }
936 
937 static int
938 smmu_init_strtab_2lvl(struct smmu_softc *sc)
939 {
940 	struct smmu_strtab *strtab;
941 	vm_paddr_t base;
942 	uint64_t reg_base;
943 	uint32_t l1size;
944 	uint32_t size;
945 	uint32_t reg;
946 	int sz;
947 
948 	strtab = &sc->strtab;
949 
950 	size = STRTAB_L1_SZ_SHIFT - (ilog2(STRTAB_L1_DESC_DWORDS) + 3);
951 	size = min(size, sc->sid_bits - STRTAB_SPLIT);
952 	strtab->num_l1_entries = (1 << size);
953 	size += STRTAB_SPLIT;
954 
955 	l1size = strtab->num_l1_entries * (STRTAB_L1_DESC_DWORDS << 3);
956 
957 	if (bootverbose)
958 		device_printf(sc->dev,
959 		    "%s: size %d, l1 entries %d, l1size %d\n",
960 		    __func__, size, strtab->num_l1_entries, l1size);
961 
962 	strtab->vaddr = contigmalloc(l1size, M_SMMU,
963 	    M_WAITOK | M_ZERO,	/* flags */
964 	    0,			/* low */
965 	    (1ul << 48) - 1,	/* high */
966 	    l1size,		/* alignment */
967 	    0);			/* boundary */
968 	if (strtab->vaddr == NULL) {
969 		device_printf(sc->dev, "Failed to allocate 2lvl strtab.\n");
970 		return (ENOMEM);
971 	}
972 
973 	sz = strtab->num_l1_entries * sizeof(struct l1_desc);
974 
975 	strtab->l1 = malloc(sz, M_SMMU, M_WAITOK | M_ZERO);
976 	if (strtab->l1 == NULL) {
977 		contigfree(strtab->vaddr, l1size, M_SMMU);
978 		return (ENOMEM);
979 	}
980 
981 	reg = STRTAB_BASE_CFG_FMT_2LVL;
982 	reg |= size << STRTAB_BASE_CFG_LOG2SIZE_S;
983 	reg |= STRTAB_SPLIT << STRTAB_BASE_CFG_SPLIT_S;
984 	strtab->base_cfg = (uint32_t)reg;
985 
986 	base = vtophys(strtab->vaddr);
987 
988 	reg_base = base & STRTAB_BASE_ADDR_M;
989 	KASSERT(reg_base == base, ("bad allocation 3"));
990 	reg_base |= STRTAB_BASE_RA;
991 	strtab->base = reg_base;
992 
993 	return (0);
994 }
995 
996 static int
997 smmu_init_strtab(struct smmu_softc *sc)
998 {
999 	int error;
1000 
1001 	if (sc->features & SMMU_FEATURE_2_LVL_STREAM_TABLE)
1002 		error = smmu_init_strtab_2lvl(sc);
1003 	else
1004 		error = smmu_init_strtab_linear(sc);
1005 
1006 	return (error);
1007 }
1008 
1009 static int
1010 smmu_init_l1_entry(struct smmu_softc *sc, int sid)
1011 {
1012 	struct smmu_strtab *strtab;
1013 	struct l1_desc *l1_desc;
1014 	uint64_t *addr;
1015 	uint64_t val;
1016 	size_t size;
1017 	int i;
1018 
1019 	strtab = &sc->strtab;
1020 	l1_desc = &strtab->l1[sid >> STRTAB_SPLIT];
1021 	if (l1_desc->va) {
1022 		/* Already allocated. */
1023 		return (0);
1024 	}
1025 
1026 	size = 1 << (STRTAB_SPLIT + ilog2(STRTAB_STE_DWORDS) + 3);
1027 
1028 	l1_desc->span = STRTAB_SPLIT + 1;
1029 	l1_desc->size = size;
1030 	l1_desc->va = contigmalloc(size, M_SMMU,
1031 	    M_WAITOK | M_ZERO,	/* flags */
1032 	    0,			/* low */
1033 	    (1ul << 48) - 1,	/* high */
1034 	    size,		/* alignment */
1035 	    0);			/* boundary */
1036 	if (l1_desc->va == NULL) {
1037 		device_printf(sc->dev, "failed to allocate l2 entry\n");
1038 		return (ENXIO);
1039 	}
1040 
1041 	l1_desc->pa = vtophys(l1_desc->va);
1042 
1043 	i = sid >> STRTAB_SPLIT;
1044 	addr = (void *)((uint64_t)strtab->vaddr +
1045 	    STRTAB_L1_DESC_DWORDS * 8 * i);
1046 
1047 	/* Install the L1 entry. */
1048 	val = l1_desc->pa & STRTAB_L1_DESC_L2PTR_M;
1049 	KASSERT(val == l1_desc->pa, ("bad allocation 4"));
1050 	val |= l1_desc->span;
1051 	*addr = val;
1052 
1053 	return (0);
1054 }
1055 
1056 static void __unused
1057 smmu_deinit_l1_entry(struct smmu_softc *sc, int sid)
1058 {
1059 	struct smmu_strtab *strtab;
1060 	struct l1_desc *l1_desc;
1061 	uint64_t *addr;
1062 	int i;
1063 
1064 	strtab = &sc->strtab;
1065 
1066 	i = sid >> STRTAB_SPLIT;
1067 	addr = (void *)((uint64_t)strtab->vaddr +
1068 	    STRTAB_L1_DESC_DWORDS * 8 * i);
1069 	*addr = 0;
1070 
1071 	l1_desc = &strtab->l1[sid >> STRTAB_SPLIT];
1072 	contigfree(l1_desc->va, l1_desc->size, M_SMMU);
1073 }
1074 
1075 static int
1076 smmu_disable(struct smmu_softc *sc)
1077 {
1078 	uint32_t reg;
1079 	int error;
1080 
1081 	/* Disable SMMU */
1082 	reg = bus_read_4(sc->res[0], SMMU_CR0);
1083 	reg &= ~CR0_SMMUEN;
1084 	error = smmu_write_ack(sc, SMMU_CR0, SMMU_CR0ACK, reg);
1085 	if (error)
1086 		device_printf(sc->dev, "Could not disable SMMU.\n");
1087 
1088 	return (0);
1089 }
1090 
1091 static int
1092 smmu_event_intr(void *arg)
1093 {
1094 	uint32_t evt[EVTQ_ENTRY_DWORDS * 2];
1095 	struct smmu_softc *sc;
1096 
1097 	sc = arg;
1098 
1099 	do {
1100 		smmu_evtq_dequeue(sc, evt);
1101 		smmu_print_event(sc, evt);
1102 	} while (!smmu_q_empty(&sc->evtq));
1103 
1104 	return (FILTER_HANDLED);
1105 }
1106 
1107 static int __unused
1108 smmu_sync_intr(void *arg)
1109 {
1110 	struct smmu_softc *sc;
1111 
1112 	sc = arg;
1113 
1114 	device_printf(sc->dev, "%s\n", __func__);
1115 
1116 	return (FILTER_HANDLED);
1117 }
1118 
1119 static int
1120 smmu_gerr_intr(void *arg)
1121 {
1122 	struct smmu_softc *sc;
1123 
1124 	sc = arg;
1125 
1126 	device_printf(sc->dev, "SMMU Global Error\n");
1127 
1128 	return (FILTER_HANDLED);
1129 }
1130 
1131 static int
1132 smmu_enable_interrupts(struct smmu_softc *sc)
1133 {
1134 	uint32_t reg;
1135 	int error;
1136 
1137 	/* Disable MSI. */
1138 	bus_write_8(sc->res[0], SMMU_GERROR_IRQ_CFG0, 0);
1139 	bus_write_4(sc->res[0], SMMU_GERROR_IRQ_CFG1, 0);
1140 	bus_write_4(sc->res[0], SMMU_GERROR_IRQ_CFG2, 0);
1141 
1142 	bus_write_8(sc->res[0], SMMU_EVENTQ_IRQ_CFG0, 0);
1143 	bus_write_4(sc->res[0], SMMU_EVENTQ_IRQ_CFG1, 0);
1144 	bus_write_4(sc->res[0], SMMU_EVENTQ_IRQ_CFG2, 0);
1145 
1146 	if (sc->features & CR0_PRIQEN) {
1147 		bus_write_8(sc->res[0], SMMU_PRIQ_IRQ_CFG0, 0);
1148 		bus_write_4(sc->res[0], SMMU_PRIQ_IRQ_CFG1, 0);
1149 		bus_write_4(sc->res[0], SMMU_PRIQ_IRQ_CFG2, 0);
1150 	}
1151 
1152 	/* Disable any interrupts. */
1153 	error = smmu_write_ack(sc, SMMU_IRQ_CTRL, SMMU_IRQ_CTRLACK, 0);
1154 	if (error) {
1155 		device_printf(sc->dev, "Could not disable interrupts.\n");
1156 		return (ENXIO);
1157 	}
1158 
1159 	/* Enable interrupts. */
1160 	reg = IRQ_CTRL_EVENTQ_IRQEN | IRQ_CTRL_GERROR_IRQEN;
1161 	if (sc->features & SMMU_FEATURE_PRI)
1162 		reg |= IRQ_CTRL_PRIQ_IRQEN;
1163 
1164 	error = smmu_write_ack(sc, SMMU_IRQ_CTRL, SMMU_IRQ_CTRLACK, reg);
1165 	if (error) {
1166 		device_printf(sc->dev, "Could not enable interrupts.\n");
1167 		return (ENXIO);
1168 	}
1169 
1170 	return (0);
1171 }
1172 
1173 #ifdef DEV_ACPI
1174 static void
1175 smmu_configure_intr(struct smmu_softc *sc, struct resource *res)
1176 {
1177 	struct intr_map_data_acpi *ad;
1178 	struct intr_map_data *data;
1179 
1180 	data = rman_get_virtual(res);
1181 	KASSERT(data != NULL, ("data is NULL"));
1182 
1183 	if (data->type == INTR_MAP_DATA_ACPI) {
1184 		ad = (struct intr_map_data_acpi *)data;
1185 		ad->trig = INTR_TRIGGER_EDGE;
1186 		ad->pol = INTR_POLARITY_HIGH;
1187 	}
1188 }
1189 #endif
1190 
1191 static int
1192 smmu_setup_interrupts(struct smmu_softc *sc)
1193 {
1194 	device_t dev;
1195 	int error;
1196 
1197 	dev = sc->dev;
1198 
1199 #ifdef DEV_ACPI
1200 	/*
1201 	 * Configure SMMU interrupts as EDGE triggered manually
1202 	 * as ACPI tables carries no information for that.
1203 	 */
1204 	smmu_configure_intr(sc, sc->res[1]);
1205 	/* PRIQ is not in use. */
1206 	smmu_configure_intr(sc, sc->res[3]);
1207 	smmu_configure_intr(sc, sc->res[4]);
1208 #endif
1209 
1210 	error = bus_setup_intr(dev, sc->res[1], INTR_TYPE_MISC,
1211 	    smmu_event_intr, NULL, sc, &sc->intr_cookie[0]);
1212 	if (error) {
1213 		device_printf(dev, "Couldn't setup Event interrupt handler\n");
1214 		return (ENXIO);
1215 	}
1216 
1217 	error = bus_setup_intr(dev, sc->res[4], INTR_TYPE_MISC,
1218 	    smmu_gerr_intr, NULL, sc, &sc->intr_cookie[2]);
1219 	if (error) {
1220 		device_printf(dev, "Couldn't setup Gerr interrupt handler\n");
1221 		return (ENXIO);
1222 	}
1223 
1224 	return (0);
1225 }
1226 
1227 static int
1228 smmu_reset(struct smmu_softc *sc)
1229 {
1230 	struct smmu_cmdq_entry cmd;
1231 	struct smmu_strtab *strtab;
1232 	int error;
1233 	int reg;
1234 
1235 	reg = bus_read_4(sc->res[0], SMMU_CR0);
1236 
1237 	if (reg & CR0_SMMUEN)
1238 		device_printf(sc->dev,
1239 		    "%s: Warning: SMMU is enabled\n", __func__);
1240 
1241 	error = smmu_disable(sc);
1242 	if (error)
1243 		device_printf(sc->dev,
1244 		    "%s: Could not disable SMMU.\n", __func__);
1245 
1246 	if (smmu_enable_interrupts(sc) != 0) {
1247 		device_printf(sc->dev, "Could not enable interrupts.\n");
1248 		return (ENXIO);
1249 	}
1250 
1251 	reg = CR1_TABLE_SH_IS	|
1252 	      CR1_TABLE_OC_WBC	|
1253 	      CR1_TABLE_IC_WBC	|
1254 	      CR1_QUEUE_SH_IS	|
1255 	      CR1_QUEUE_OC_WBC	|
1256 	      CR1_QUEUE_IC_WBC;
1257 	bus_write_4(sc->res[0], SMMU_CR1, reg);
1258 
1259 	reg = CR2_PTM | CR2_RECINVSID | CR2_E2H;
1260 	bus_write_4(sc->res[0], SMMU_CR2, reg);
1261 
1262 	/* Stream table. */
1263 	strtab = &sc->strtab;
1264 	bus_write_8(sc->res[0], SMMU_STRTAB_BASE, strtab->base);
1265 	bus_write_4(sc->res[0], SMMU_STRTAB_BASE_CFG, strtab->base_cfg);
1266 
1267 	/* Command queue. */
1268 	bus_write_8(sc->res[0], SMMU_CMDQ_BASE, sc->cmdq.base);
1269 	bus_write_4(sc->res[0], SMMU_CMDQ_PROD, sc->cmdq.lc.prod);
1270 	bus_write_4(sc->res[0], SMMU_CMDQ_CONS, sc->cmdq.lc.cons);
1271 
1272 	reg = CR0_CMDQEN;
1273 	error = smmu_write_ack(sc, SMMU_CR0, SMMU_CR0ACK, reg);
1274 	if (error) {
1275 		device_printf(sc->dev, "Could not enable command queue\n");
1276 		return (ENXIO);
1277 	}
1278 
1279 	/* Invalidate cached configuration. */
1280 	smmu_invalidate_all_sid(sc);
1281 
1282 	if (sc->features & SMMU_FEATURE_HYP) {
1283 		cmd.opcode = CMD_TLBI_EL2_ALL;
1284 		smmu_cmdq_enqueue_cmd(sc, &cmd);
1285 	};
1286 
1287 	/* Invalidate TLB. */
1288 	smmu_tlbi_all(sc);
1289 
1290 	/* Event queue */
1291 	bus_write_8(sc->res[0], SMMU_EVENTQ_BASE, sc->evtq.base);
1292 	bus_write_4(sc->res[0], SMMU_EVENTQ_PROD, sc->evtq.lc.prod);
1293 	bus_write_4(sc->res[0], SMMU_EVENTQ_CONS, sc->evtq.lc.cons);
1294 
1295 	reg |= CR0_EVENTQEN;
1296 	error = smmu_write_ack(sc, SMMU_CR0, SMMU_CR0ACK, reg);
1297 	if (error) {
1298 		device_printf(sc->dev, "Could not enable event queue\n");
1299 		return (ENXIO);
1300 	}
1301 
1302 	if (sc->features & SMMU_FEATURE_PRI) {
1303 		/* PRI queue */
1304 		bus_write_8(sc->res[0], SMMU_PRIQ_BASE, sc->priq.base);
1305 		bus_write_4(sc->res[0], SMMU_PRIQ_PROD, sc->priq.lc.prod);
1306 		bus_write_4(sc->res[0], SMMU_PRIQ_CONS, sc->priq.lc.cons);
1307 
1308 		reg |= CR0_PRIQEN;
1309 		error = smmu_write_ack(sc, SMMU_CR0, SMMU_CR0ACK, reg);
1310 		if (error) {
1311 			device_printf(sc->dev, "Could not enable PRI queue\n");
1312 			return (ENXIO);
1313 		}
1314 	}
1315 
1316 	if (sc->features & SMMU_FEATURE_ATS) {
1317 		reg |= CR0_ATSCHK;
1318 		error = smmu_write_ack(sc, SMMU_CR0, SMMU_CR0ACK, reg);
1319 		if (error) {
1320 			device_printf(sc->dev, "Could not enable ATS check.\n");
1321 			return (ENXIO);
1322 		}
1323 	}
1324 
1325 	reg |= CR0_SMMUEN;
1326 	error = smmu_write_ack(sc, SMMU_CR0, SMMU_CR0ACK, reg);
1327 	if (error) {
1328 		device_printf(sc->dev, "Could not enable SMMU.\n");
1329 		return (ENXIO);
1330 	}
1331 
1332 	return (0);
1333 }
1334 
1335 static int
1336 smmu_check_features(struct smmu_softc *sc)
1337 {
1338 	uint32_t reg;
1339 	uint32_t val;
1340 
1341 	sc->features = 0;
1342 
1343 	reg = bus_read_4(sc->res[0], SMMU_IDR0);
1344 
1345 	if (reg & IDR0_ST_LVL_2) {
1346 		if (bootverbose)
1347 			device_printf(sc->dev,
1348 			    "2-level stream table supported.\n");
1349 		sc->features |= SMMU_FEATURE_2_LVL_STREAM_TABLE;
1350 	}
1351 
1352 	if (reg & IDR0_CD2L) {
1353 		if (bootverbose)
1354 			device_printf(sc->dev,
1355 			    "2-level CD table supported.\n");
1356 		sc->features |= SMMU_FEATURE_2_LVL_CD;
1357 	}
1358 
1359 	switch (reg & IDR0_TTENDIAN_M) {
1360 	case IDR0_TTENDIAN_MIXED:
1361 		if (bootverbose)
1362 			device_printf(sc->dev, "Mixed endianess supported.\n");
1363 		sc->features |= SMMU_FEATURE_TT_LE;
1364 		sc->features |= SMMU_FEATURE_TT_BE;
1365 		break;
1366 	case IDR0_TTENDIAN_LITTLE:
1367 		if (bootverbose)
1368 			device_printf(sc->dev,
1369 			    "Little endian supported only.\n");
1370 		sc->features |= SMMU_FEATURE_TT_LE;
1371 		break;
1372 	case IDR0_TTENDIAN_BIG:
1373 		if (bootverbose)
1374 			device_printf(sc->dev, "Big endian supported only.\n");
1375 		sc->features |= SMMU_FEATURE_TT_BE;
1376 		break;
1377 	default:
1378 		device_printf(sc->dev, "Unsupported endianness.\n");
1379 		return (ENXIO);
1380 	}
1381 
1382 	if (reg & IDR0_SEV)
1383 		sc->features |= SMMU_FEATURE_SEV;
1384 
1385 	if (reg & IDR0_MSI) {
1386 		if (bootverbose)
1387 			device_printf(sc->dev, "MSI feature present.\n");
1388 		sc->features |= SMMU_FEATURE_MSI;
1389 	}
1390 
1391 	if (reg & IDR0_HYP) {
1392 		if (bootverbose)
1393 			device_printf(sc->dev, "HYP feature present.\n");
1394 		sc->features |= SMMU_FEATURE_HYP;
1395 	}
1396 
1397 	if (reg & IDR0_ATS)
1398 		sc->features |= SMMU_FEATURE_ATS;
1399 
1400 	if (reg & IDR0_PRI)
1401 		sc->features |= SMMU_FEATURE_PRI;
1402 
1403 	switch (reg & IDR0_STALL_MODEL_M) {
1404 	case IDR0_STALL_MODEL_FORCE:
1405 		/* Stall is forced. */
1406 		sc->features |= SMMU_FEATURE_STALL_FORCE;
1407 		/* FALLTHROUGH */
1408 	case IDR0_STALL_MODEL_STALL:
1409 		sc->features |= SMMU_FEATURE_STALL;
1410 		break;
1411 	}
1412 
1413 	/* Grab translation stages supported. */
1414 	if (reg & IDR0_S1P) {
1415 		if (bootverbose)
1416 			device_printf(sc->dev,
1417 			    "Stage 1 translation supported.\n");
1418 		sc->features |= SMMU_FEATURE_S1P;
1419 	}
1420 	if (reg & IDR0_S2P) {
1421 		if (bootverbose)
1422 			device_printf(sc->dev,
1423 			    "Stage 2 translation supported.\n");
1424 		sc->features |= SMMU_FEATURE_S2P;
1425 	}
1426 
1427 	switch (reg & IDR0_TTF_M) {
1428 	case IDR0_TTF_ALL:
1429 	case IDR0_TTF_AA64:
1430 		sc->ias = 40;
1431 		break;
1432 	default:
1433 		device_printf(sc->dev, "No AArch64 table format support.\n");
1434 		return (ENXIO);
1435 	}
1436 
1437 	if (reg & IDR0_ASID16)
1438 		sc->asid_bits = 16;
1439 	else
1440 		sc->asid_bits = 8;
1441 
1442 	if (bootverbose)
1443 		device_printf(sc->dev, "ASID bits %d\n", sc->asid_bits);
1444 
1445 	if (reg & IDR0_VMID16)
1446 		sc->vmid_bits = 16;
1447 	else
1448 		sc->vmid_bits = 8;
1449 
1450 	reg = bus_read_4(sc->res[0], SMMU_IDR1);
1451 
1452 	if (reg & (IDR1_TABLES_PRESET | IDR1_QUEUES_PRESET | IDR1_REL)) {
1453 		device_printf(sc->dev,
1454 		    "Embedded implementations not supported by this driver.\n");
1455 		return (ENXIO);
1456 	}
1457 
1458 	val = (reg & IDR1_CMDQS_M) >> IDR1_CMDQS_S;
1459 	sc->cmdq.size_log2 = val;
1460 	if (bootverbose)
1461 		device_printf(sc->dev, "CMD queue bits %d\n", val);
1462 
1463 	val = (reg & IDR1_EVENTQS_M) >> IDR1_EVENTQS_S;
1464 	sc->evtq.size_log2 = val;
1465 	if (bootverbose)
1466 		device_printf(sc->dev, "EVENT queue bits %d\n", val);
1467 
1468 	if (sc->features & SMMU_FEATURE_PRI) {
1469 		val = (reg & IDR1_PRIQS_M) >> IDR1_PRIQS_S;
1470 		sc->priq.size_log2 = val;
1471 		if (bootverbose)
1472 			device_printf(sc->dev, "PRI queue bits %d\n", val);
1473 	}
1474 
1475 	sc->ssid_bits = (reg & IDR1_SSIDSIZE_M) >> IDR1_SSIDSIZE_S;
1476 	sc->sid_bits = (reg & IDR1_SIDSIZE_M) >> IDR1_SIDSIZE_S;
1477 
1478 	if (sc->sid_bits <= STRTAB_SPLIT)
1479 		sc->features &= ~SMMU_FEATURE_2_LVL_STREAM_TABLE;
1480 
1481 	if (bootverbose) {
1482 		device_printf(sc->dev, "SSID bits %d\n", sc->ssid_bits);
1483 		device_printf(sc->dev, "SID bits %d\n", sc->sid_bits);
1484 	}
1485 
1486 	/* IDR3 */
1487 	reg = bus_read_4(sc->res[0], SMMU_IDR3);
1488 	if (reg & IDR3_RIL)
1489 		sc->features |= SMMU_FEATURE_RANGE_INV;
1490 
1491 	/* IDR5 */
1492 	reg = bus_read_4(sc->res[0], SMMU_IDR5);
1493 
1494 	switch (reg & IDR5_OAS_M) {
1495 	case IDR5_OAS_32:
1496 		sc->oas = 32;
1497 		break;
1498 	case IDR5_OAS_36:
1499 		sc->oas = 36;
1500 		break;
1501 	case IDR5_OAS_40:
1502 		sc->oas = 40;
1503 		break;
1504 	case IDR5_OAS_42:
1505 		sc->oas = 42;
1506 		break;
1507 	case IDR5_OAS_44:
1508 		sc->oas = 44;
1509 		break;
1510 	case IDR5_OAS_48:
1511 		sc->oas = 48;
1512 		break;
1513 	case IDR5_OAS_52:
1514 		sc->oas = 52;
1515 		break;
1516 	}
1517 
1518 	sc->pgsizes = 0;
1519 	if (reg & IDR5_GRAN64K)
1520 		sc->pgsizes |= 64 * 1024;
1521 	if (reg & IDR5_GRAN16K)
1522 		sc->pgsizes |= 16 * 1024;
1523 	if (reg & IDR5_GRAN4K)
1524 		sc->pgsizes |= 4 * 1024;
1525 
1526 	if ((reg & IDR5_VAX_M) == IDR5_VAX_52)
1527 		sc->features |= SMMU_FEATURE_VAX;
1528 
1529 	return (0);
1530 }
1531 
1532 static void
1533 smmu_init_asids(struct smmu_softc *sc)
1534 {
1535 
1536 	sc->asid_set_size = (1 << sc->asid_bits);
1537 	sc->asid_set = bit_alloc(sc->asid_set_size, M_SMMU, M_WAITOK);
1538 	mtx_init(&sc->asid_set_mutex, "asid set", NULL, MTX_SPIN);
1539 }
1540 
1541 static int
1542 smmu_asid_alloc(struct smmu_softc *sc, int *new_asid)
1543 {
1544 
1545 	mtx_lock_spin(&sc->asid_set_mutex);
1546 	bit_ffc(sc->asid_set, sc->asid_set_size, new_asid);
1547 	if (*new_asid == -1) {
1548 		mtx_unlock_spin(&sc->asid_set_mutex);
1549 		return (ENOMEM);
1550 	}
1551 	bit_set(sc->asid_set, *new_asid);
1552 	mtx_unlock_spin(&sc->asid_set_mutex);
1553 
1554 	return (0);
1555 }
1556 
1557 static void
1558 smmu_asid_free(struct smmu_softc *sc, int asid)
1559 {
1560 
1561 	mtx_lock_spin(&sc->asid_set_mutex);
1562 	bit_clear(sc->asid_set, asid);
1563 	mtx_unlock_spin(&sc->asid_set_mutex);
1564 }
1565 
1566 /*
1567  * Device interface.
1568  */
1569 int
1570 smmu_attach(device_t dev)
1571 {
1572 	struct smmu_softc *sc;
1573 	int error;
1574 
1575 	sc = device_get_softc(dev);
1576 	sc->dev = dev;
1577 
1578 	mtx_init(&sc->sc_mtx, device_get_nameunit(sc->dev), "smmu", MTX_DEF);
1579 
1580 	error = smmu_setup_interrupts(sc);
1581 	if (error) {
1582 		bus_release_resources(dev, smmu_spec, sc->res);
1583 		return (ENXIO);
1584 	}
1585 
1586 	error = smmu_check_features(sc);
1587 	if (error) {
1588 		device_printf(dev, "Some features are required "
1589 		    "but not supported by hardware.\n");
1590 		return (ENXIO);
1591 	}
1592 
1593 	smmu_init_asids(sc);
1594 
1595 	error = smmu_init_queues(sc);
1596 	if (error) {
1597 		device_printf(dev, "Couldn't allocate queues.\n");
1598 		return (ENXIO);
1599 	}
1600 
1601 	error = smmu_init_strtab(sc);
1602 	if (error) {
1603 		device_printf(dev, "Couldn't allocate strtab.\n");
1604 		return (ENXIO);
1605 	}
1606 
1607 	error = smmu_reset(sc);
1608 	if (error) {
1609 		device_printf(dev, "Couldn't reset SMMU.\n");
1610 		return (ENXIO);
1611 	}
1612 
1613 	return (0);
1614 }
1615 
1616 int
1617 smmu_detach(device_t dev)
1618 {
1619 	struct smmu_softc *sc;
1620 
1621 	sc = device_get_softc(dev);
1622 
1623 	bus_release_resources(dev, smmu_spec, sc->res);
1624 
1625 	return (0);
1626 }
1627 
1628 static int
1629 smmu_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
1630 {
1631 	struct smmu_softc *sc;
1632 
1633 	sc = device_get_softc(dev);
1634 
1635 	device_printf(sc->dev, "%s\n", __func__);
1636 
1637 	return (ENOENT);
1638 }
1639 
1640 static int
1641 smmu_unmap(device_t dev, struct iommu_domain *iodom,
1642     vm_offset_t va, bus_size_t size)
1643 {
1644 	struct smmu_domain *domain;
1645 	struct smmu_softc *sc;
1646 	int err;
1647 	int i;
1648 
1649 	sc = device_get_softc(dev);
1650 
1651 	domain = (struct smmu_domain *)iodom;
1652 
1653 	err = 0;
1654 
1655 	dprintf("%s: %lx, %ld, domain %d\n", __func__, va, size, domain->asid);
1656 
1657 	for (i = 0; i < size; i += PAGE_SIZE) {
1658 		if (pmap_smmu_remove(&domain->p, va) == 0) {
1659 			/* pmap entry removed, invalidate TLB. */
1660 			smmu_tlbi_va(sc, va, domain->asid);
1661 		} else {
1662 			err = ENOENT;
1663 			break;
1664 		}
1665 		va += PAGE_SIZE;
1666 	}
1667 
1668 	smmu_sync(sc);
1669 
1670 	return (err);
1671 }
1672 
1673 static int
1674 smmu_map(device_t dev, struct iommu_domain *iodom,
1675     vm_offset_t va, vm_page_t *ma, vm_size_t size,
1676     vm_prot_t prot)
1677 {
1678 	struct smmu_domain *domain;
1679 	struct smmu_softc *sc;
1680 	vm_paddr_t pa;
1681 	int error;
1682 	int i;
1683 
1684 	sc = device_get_softc(dev);
1685 
1686 	domain = (struct smmu_domain *)iodom;
1687 
1688 	dprintf("%s: %lx -> %lx, %ld, domain %d\n", __func__, va, pa, size,
1689 	    domain->asid);
1690 
1691 	for (i = 0; size > 0; size -= PAGE_SIZE) {
1692 		pa = VM_PAGE_TO_PHYS(ma[i++]);
1693 		error = pmap_smmu_enter(&domain->p, va, pa, prot, 0);
1694 		if (error)
1695 			return (error);
1696 		smmu_tlbi_va(sc, va, domain->asid);
1697 		va += PAGE_SIZE;
1698 	}
1699 
1700 	smmu_sync(sc);
1701 
1702 	return (0);
1703 }
1704 
1705 static struct iommu_domain *
1706 smmu_domain_alloc(device_t dev, struct iommu_unit *iommu)
1707 {
1708 	struct iommu_domain *iodom;
1709 	struct smmu_domain *domain;
1710 	struct smmu_unit *unit;
1711 	struct smmu_softc *sc;
1712 	int error;
1713 	int new_asid;
1714 
1715 	sc = device_get_softc(dev);
1716 
1717 	unit = (struct smmu_unit *)iommu;
1718 
1719 	domain = malloc(sizeof(*domain), M_SMMU, M_WAITOK | M_ZERO);
1720 
1721 	error = smmu_asid_alloc(sc, &new_asid);
1722 	if (error) {
1723 		free(domain, M_SMMU);
1724 		device_printf(sc->dev,
1725 		    "Could not allocate ASID for a new domain.\n");
1726 		return (NULL);
1727 	}
1728 
1729 	domain->asid = (uint16_t)new_asid;
1730 
1731 	iommu_pmap_pinit(&domain->p);
1732 	PMAP_LOCK_INIT(&domain->p);
1733 
1734 	error = smmu_init_cd(sc, domain);
1735 	if (error) {
1736 		free(domain, M_SMMU);
1737 		device_printf(sc->dev, "Could not initialize CD\n");
1738 		return (NULL);
1739 	}
1740 
1741 	smmu_tlbi_asid(sc, domain->asid);
1742 
1743 	LIST_INIT(&domain->ctx_list);
1744 
1745 	IOMMU_LOCK(iommu);
1746 	LIST_INSERT_HEAD(&unit->domain_list, domain, next);
1747 	IOMMU_UNLOCK(iommu);
1748 
1749 	iodom = &domain->iodom;
1750 
1751 	/*
1752 	 * Use 48-bit address space regardless of VAX bit
1753 	 * as we need 64k IOMMU_PAGE_SIZE for 52-bit space.
1754 	 */
1755 	iodom->end = MAXADDR_48BIT;
1756 
1757 	return (iodom);
1758 }
1759 
1760 static void
1761 smmu_domain_free(device_t dev, struct iommu_domain *iodom)
1762 {
1763 	struct smmu_domain *domain;
1764 	struct smmu_softc *sc;
1765 	struct smmu_cd *cd;
1766 
1767 	sc = device_get_softc(dev);
1768 
1769 	domain = (struct smmu_domain *)iodom;
1770 
1771 	LIST_REMOVE(domain, next);
1772 
1773 	cd = domain->cd;
1774 
1775 	iommu_pmap_remove_pages(&domain->p);
1776 	iommu_pmap_release(&domain->p);
1777 
1778 	smmu_tlbi_asid(sc, domain->asid);
1779 	smmu_asid_free(sc, domain->asid);
1780 
1781 	contigfree(cd->vaddr, cd->size, M_SMMU);
1782 	free(cd, M_SMMU);
1783 
1784 	free(domain, M_SMMU);
1785 }
1786 
1787 static int
1788 smmu_set_buswide(device_t dev, struct smmu_domain *domain,
1789     struct smmu_ctx *ctx)
1790 {
1791 	struct smmu_softc *sc;
1792 	int i;
1793 
1794 	sc = device_get_softc(dev);
1795 
1796 	for (i = 0; i < PCI_SLOTMAX; i++)
1797 		smmu_init_ste(sc, domain->cd, (ctx->sid | i), ctx->bypass);
1798 
1799 	return (0);
1800 }
1801 
1802 #ifdef DEV_ACPI
1803 static int
1804 smmu_pci_get_sid_acpi(device_t child, u_int *xref0, u_int *sid0)
1805 {
1806 	uint16_t rid;
1807 	u_int xref;
1808 	int seg;
1809 	int err;
1810 	int sid;
1811 
1812 	seg = pci_get_domain(child);
1813 	rid = pci_get_rid(child);
1814 
1815 	err = acpi_iort_map_pci_smmuv3(seg, rid, &xref, &sid);
1816 	if (err == 0) {
1817 		if (sid0)
1818 			*sid0 = sid;
1819 		if (xref0)
1820 			*xref0 = xref;
1821 	}
1822 
1823 	return (err);
1824 }
1825 #endif
1826 
1827 #ifdef FDT
1828 static int
1829 smmu_pci_get_sid_fdt(device_t child, u_int *xref0, u_int *sid0)
1830 {
1831 	struct pci_id_ofw_iommu pi;
1832 	uint64_t base, size;
1833 	phandle_t node;
1834 	u_int xref;
1835 	int err;
1836 
1837 	err = pci_get_id(child, PCI_ID_OFW_IOMMU, (uintptr_t *)&pi);
1838 	if (err == 0) {
1839 		/* Our xref is memory base address. */
1840 		node = OF_node_from_xref(pi.xref);
1841 		fdt_regsize(node, &base, &size);
1842 		xref = base;
1843 
1844 		if (sid0)
1845 			*sid0 = pi.id;
1846 		if (xref0)
1847 			*xref0 = xref;
1848 	}
1849 
1850 	return (err);
1851 }
1852 #endif
1853 
1854 static struct iommu_ctx *
1855 smmu_ctx_alloc(device_t dev, struct iommu_domain *iodom, device_t child,
1856     bool disabled)
1857 {
1858 	struct smmu_domain *domain;
1859 	struct smmu_ctx *ctx;
1860 
1861 	domain = (struct smmu_domain *)iodom;
1862 
1863 	ctx = malloc(sizeof(struct smmu_ctx), M_SMMU, M_WAITOK | M_ZERO);
1864 	ctx->dev = child;
1865 	ctx->domain = domain;
1866 	if (disabled)
1867 		ctx->bypass = true;
1868 
1869 	IOMMU_DOMAIN_LOCK(iodom);
1870 	LIST_INSERT_HEAD(&domain->ctx_list, ctx, next);
1871 	IOMMU_DOMAIN_UNLOCK(iodom);
1872 
1873 	return (&ctx->ioctx);
1874 }
1875 
1876 static int
1877 smmu_ctx_init(device_t dev, struct iommu_ctx *ioctx)
1878 {
1879 	struct smmu_domain *domain;
1880 	struct iommu_domain *iodom;
1881 	struct smmu_softc *sc;
1882 	struct smmu_ctx *ctx;
1883 	devclass_t pci_class;
1884 	u_int sid;
1885 	int err;
1886 
1887 	ctx = (struct smmu_ctx *)ioctx;
1888 
1889 	sc = device_get_softc(dev);
1890 
1891 	domain = ctx->domain;
1892 	iodom = (struct iommu_domain *)domain;
1893 
1894 	pci_class = devclass_find("pci");
1895 	if (device_get_devclass(device_get_parent(ctx->dev)) == pci_class) {
1896 #ifdef DEV_ACPI
1897 		err = smmu_pci_get_sid_acpi(ctx->dev, NULL, &sid);
1898 #else
1899 		err = smmu_pci_get_sid_fdt(ctx->dev, NULL, &sid);
1900 #endif
1901 		if (err)
1902 			return (err);
1903 
1904 		ioctx->rid = pci_get_rid(dev);
1905 		ctx->sid = sid;
1906 		ctx->vendor = pci_get_vendor(ctx->dev);
1907 		ctx->device = pci_get_device(ctx->dev);
1908 	}
1909 
1910 	if (sc->features & SMMU_FEATURE_2_LVL_STREAM_TABLE) {
1911 		err = smmu_init_l1_entry(sc, ctx->sid);
1912 		if (err)
1913 			return (err);
1914 	}
1915 
1916 	/*
1917 	 * Neoverse N1 SDP:
1918 	 * 0x800 xhci
1919 	 * 0x700 re
1920 	 * 0x600 sata
1921 	 */
1922 
1923 	smmu_init_ste(sc, domain->cd, ctx->sid, ctx->bypass);
1924 
1925 	if (device_get_devclass(device_get_parent(ctx->dev)) == pci_class)
1926 		if (iommu_is_buswide_ctx(iodom->iommu, pci_get_bus(ctx->dev)))
1927 			smmu_set_buswide(dev, domain, ctx);
1928 
1929 	return (0);
1930 }
1931 
1932 static void
1933 smmu_ctx_free(device_t dev, struct iommu_ctx *ioctx)
1934 {
1935 	struct smmu_softc *sc;
1936 	struct smmu_ctx *ctx;
1937 
1938 	IOMMU_ASSERT_LOCKED(ioctx->domain->iommu);
1939 
1940 	sc = device_get_softc(dev);
1941 	ctx = (struct smmu_ctx *)ioctx;
1942 
1943 	smmu_deinit_ste(sc, ctx->sid);
1944 
1945 	LIST_REMOVE(ctx, next);
1946 
1947 	free(ctx, M_SMMU);
1948 }
1949 
1950 struct smmu_ctx *
1951 smmu_ctx_lookup_by_sid(device_t dev, u_int sid)
1952 {
1953 	struct smmu_softc *sc;
1954 	struct smmu_domain *domain;
1955 	struct smmu_unit *unit;
1956 	struct smmu_ctx *ctx;
1957 
1958 	sc = device_get_softc(dev);
1959 
1960 	unit = &sc->unit;
1961 
1962 	LIST_FOREACH(domain, &unit->domain_list, next) {
1963 		LIST_FOREACH(ctx, &domain->ctx_list, next) {
1964 			if (ctx->sid == sid)
1965 				return (ctx);
1966 		}
1967 	}
1968 
1969 	return (NULL);
1970 }
1971 
1972 static struct iommu_ctx *
1973 smmu_ctx_lookup(device_t dev, device_t child)
1974 {
1975 	struct iommu_unit *iommu __diagused;
1976 	struct smmu_softc *sc;
1977 	struct smmu_domain *domain;
1978 	struct smmu_unit *unit;
1979 	struct smmu_ctx *ctx;
1980 
1981 	sc = device_get_softc(dev);
1982 
1983 	unit = &sc->unit;
1984 	iommu = &unit->iommu;
1985 
1986 	IOMMU_ASSERT_LOCKED(iommu);
1987 
1988 	LIST_FOREACH(domain, &unit->domain_list, next) {
1989 		IOMMU_DOMAIN_LOCK(&domain->iodom);
1990 		LIST_FOREACH(ctx, &domain->ctx_list, next) {
1991 			if (ctx->dev == child) {
1992 				IOMMU_DOMAIN_UNLOCK(&domain->iodom);
1993 				return (&ctx->ioctx);
1994 			}
1995 		}
1996 		IOMMU_DOMAIN_UNLOCK(&domain->iodom);
1997 	}
1998 
1999 	return (NULL);
2000 }
2001 
2002 static int
2003 smmu_find(device_t dev, device_t child)
2004 {
2005 	struct smmu_softc *sc;
2006 	u_int xref;
2007 	int err;
2008 
2009 	sc = device_get_softc(dev);
2010 
2011 #ifdef DEV_ACPI
2012 	err = smmu_pci_get_sid_acpi(child, &xref, NULL);
2013 #else
2014 	err = smmu_pci_get_sid_fdt(child, &xref, NULL);
2015 #endif
2016 	if (err)
2017 		return (ENOENT);
2018 
2019 	/* Check if xref is ours. */
2020 	if (xref != sc->xref)
2021 		return (EFAULT);
2022 
2023 	return (0);
2024 }
2025 
2026 #ifdef FDT
2027 static int
2028 smmu_ofw_md_data(device_t dev, struct iommu_ctx *ioctx, pcell_t *cells,
2029     int ncells)
2030 {
2031 	struct smmu_ctx *ctx;
2032 
2033 	ctx = (struct smmu_ctx *)ioctx;
2034 
2035 	if (ncells != 1)
2036 		return (-1);
2037 
2038 	ctx->sid = cells[0];
2039 
2040 	return (0);
2041 }
2042 #endif
2043 
2044 static device_method_t smmu_methods[] = {
2045 	/* Device interface */
2046 	DEVMETHOD(device_detach,	smmu_detach),
2047 
2048 	/* SMMU interface */
2049 	DEVMETHOD(iommu_find,		smmu_find),
2050 	DEVMETHOD(iommu_map,		smmu_map),
2051 	DEVMETHOD(iommu_unmap,		smmu_unmap),
2052 	DEVMETHOD(iommu_domain_alloc,	smmu_domain_alloc),
2053 	DEVMETHOD(iommu_domain_free,	smmu_domain_free),
2054 	DEVMETHOD(iommu_ctx_alloc,	smmu_ctx_alloc),
2055 	DEVMETHOD(iommu_ctx_init,	smmu_ctx_init),
2056 	DEVMETHOD(iommu_ctx_free,	smmu_ctx_free),
2057 	DEVMETHOD(iommu_ctx_lookup,	smmu_ctx_lookup),
2058 #ifdef FDT
2059 	DEVMETHOD(iommu_ofw_md_data,	smmu_ofw_md_data),
2060 #endif
2061 
2062 	/* Bus interface */
2063 	DEVMETHOD(bus_read_ivar,	smmu_read_ivar),
2064 
2065 	/* End */
2066 	DEVMETHOD_END
2067 };
2068 
2069 DEFINE_CLASS_0(smmu, smmu_driver, smmu_methods, sizeof(struct smmu_softc));
2070