xref: /freebsd/sys/x86/iommu/amd_drv.c (revision 9805e5b071a10711cfcd3c7a750ac4a4d14b6e3f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2024 The FreeBSD Foundation
5  *
6  * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include "opt_acpi.h"
32 #include "opt_ddb.h"
33 
34 #include <sys/param.h>
35 #include <sys/bus.h>
36 #include <sys/domainset.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/memdesc.h>
41 #include <sys/module.h>
42 #include <sys/mutex.h>
43 #include <sys/rman.h>
44 #include <sys/rwlock.h>
45 #include <sys/smp.h>
46 #include <sys/taskqueue.h>
47 #include <sys/tree.h>
48 #include <sys/vmem.h>
49 #include <vm/vm.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_object.h>
52 #include <vm/vm_page.h>
53 #include <vm/vm_pageout.h>
54 #include <vm/vm_pager.h>
55 #include <contrib/dev/acpica/include/acpi.h>
56 #include <contrib/dev/acpica/include/accommon.h>
57 #include <dev/acpica/acpivar.h>
58 #include <dev/pci/pcireg.h>
59 #include <dev/pci/pcivar.h>
60 #include <machine/bus.h>
61 #include <machine/pci_cfgreg.h>
62 #include "pcib_if.h"
63 #include <machine/intr_machdep.h>
64 #include <machine/md_var.h>
65 #include <machine/cputypes.h>
66 #include <x86/apicreg.h>
67 #include <x86/apicvar.h>
68 #include <dev/iommu/iommu.h>
69 #include <x86/iommu/amd_reg.h>
70 #include <x86/iommu/x86_iommu.h>
71 #include <x86/iommu/amd_iommu.h>
72 
73 static int amdiommu_enable = 0;
74 
75 /*
76  * All enumerated AMD IOMMU units.
77  * Access is unlocked, the list is not modified after early
78  * single-threaded startup.
79  */
80 static TAILQ_HEAD(, amdiommu_unit) amdiommu_units =
81     TAILQ_HEAD_INITIALIZER(amdiommu_units);
82 
83 static u_int
ivrs_info_to_unit_id(UINT32 info)84 ivrs_info_to_unit_id(UINT32 info)
85 {
86 	return ((info & ACPI_IVHD_UNIT_ID_MASK) >> 8);
87 }
88 
89 typedef bool (*amdiommu_itercc_t)(void *, void *);
90 typedef bool (*amdiommu_iter40_t)(ACPI_IVRS_HARDWARE2 *, void *);
91 typedef bool (*amdiommu_iter11_t)(ACPI_IVRS_HARDWARE2 *, void *);
92 typedef bool (*amdiommu_iter10_t)(ACPI_IVRS_HARDWARE1 *, void *);
93 
94 static bool
amdiommu_ivrs_iterate_tbl_typed(amdiommu_itercc_t iter,void * arg,int type,ACPI_TABLE_IVRS * ivrs_tbl)95 amdiommu_ivrs_iterate_tbl_typed(amdiommu_itercc_t iter, void *arg,
96     int type, ACPI_TABLE_IVRS *ivrs_tbl)
97 {
98 	char *ptr, *ptrend;
99 	bool done;
100 
101 	done = false;
102 	ptr = (char *)ivrs_tbl + sizeof(*ivrs_tbl);
103 	ptrend = (char *)ivrs_tbl + ivrs_tbl->Header.Length;
104 	for (;;) {
105 		ACPI_IVRS_HEADER *ivrsh;
106 
107 		if (ptr >= ptrend)
108 			break;
109 		ivrsh = (ACPI_IVRS_HEADER *)ptr;
110 		if (ivrsh->Length <= 0) {
111 			printf("amdiommu_iterate_tbl: corrupted IVRS table, "
112 			    "length %d\n", ivrsh->Length);
113 			break;
114 		}
115 		ptr += ivrsh->Length;
116 		if (ivrsh->Type ==  type) {
117 			done = iter((void *)ivrsh, arg);
118 			if (done)
119 				break;
120 		}
121 	}
122 	return (done);
123 }
124 
125 /*
126  * Walk over IVRS, calling callback iterators following priority:
127  * 0x40, then 0x11, then 0x10 subtable.  First iterator returning true
128  * ends the walk.
129  * Returns true if any iterator returned true, otherwise false.
130  */
131 static bool
amdiommu_ivrs_iterate_tbl(amdiommu_iter40_t iter40,amdiommu_iter11_t iter11,amdiommu_iter10_t iter10,void * arg)132 amdiommu_ivrs_iterate_tbl(amdiommu_iter40_t iter40, amdiommu_iter11_t iter11,
133     amdiommu_iter10_t iter10, void *arg)
134 {
135 	ACPI_TABLE_IVRS *ivrs_tbl;
136 	ACPI_STATUS status;
137 	bool done;
138 
139 	status = AcpiGetTable(ACPI_SIG_IVRS, 1,
140 	    (ACPI_TABLE_HEADER **)&ivrs_tbl);
141 	if (ACPI_FAILURE(status))
142 		return (false);
143 	done = false;
144 	if (iter40 != NULL)
145 		done = amdiommu_ivrs_iterate_tbl_typed(
146 		    (amdiommu_itercc_t)iter40, arg,
147 		    ACPI_IVRS_TYPE_HARDWARE3, ivrs_tbl);
148 	if (!done && iter11 != NULL)
149 		done = amdiommu_ivrs_iterate_tbl_typed(
150 		    (amdiommu_itercc_t)iter11, arg, ACPI_IVRS_TYPE_HARDWARE2,
151 		    ivrs_tbl);
152 	if (!done && iter10 != NULL)
153 		done = amdiommu_ivrs_iterate_tbl_typed(
154 		    (amdiommu_itercc_t)iter10, arg, ACPI_IVRS_TYPE_HARDWARE1,
155 		    ivrs_tbl);
156 	AcpiPutTable((ACPI_TABLE_HEADER *)ivrs_tbl);
157 	return (done);
158 }
159 
160 struct ivhd_lookup_data  {
161 	struct amdiommu_unit *sc;
162 	uint16_t devid;
163 };
164 
165 static bool
ivrs_lookup_ivhd_0x40(ACPI_IVRS_HARDWARE2 * h2,void * arg)166 ivrs_lookup_ivhd_0x40(ACPI_IVRS_HARDWARE2 *h2, void *arg)
167 {
168 	struct ivhd_lookup_data *ildp;
169 
170 	KASSERT(h2->Header.Type == ACPI_IVRS_TYPE_HARDWARE2 ||
171 	    h2->Header.Type == ACPI_IVRS_TYPE_HARDWARE3,
172 	    ("Misparsed IVHD, h2 type %#x", h2->Header.Type));
173 
174 	ildp = arg;
175 	if (h2->Header.DeviceId != ildp->devid)
176 		return (false);
177 
178 	ildp->sc->unit_dom = h2->PciSegmentGroup;
179 	ildp->sc->iommu.unit = ivrs_info_to_unit_id(h2->Info);
180 	ildp->sc->efr = h2->EfrRegisterImage;
181 	return (true);
182 }
183 
184 static bool
ivrs_lookup_ivhd_0x10(ACPI_IVRS_HARDWARE1 * h1,void * arg)185 ivrs_lookup_ivhd_0x10(ACPI_IVRS_HARDWARE1 *h1, void *arg)
186 {
187 	struct ivhd_lookup_data *ildp;
188 
189 	KASSERT(h1->Header.Type == ACPI_IVRS_TYPE_HARDWARE1,
190 	    ("Misparsed IVHD, h1 type %#x", h1->Header.Type));
191 
192 	ildp = arg;
193 	if (h1->Header.DeviceId != ildp->devid)
194 		return (false);
195 
196 	ildp->sc->unit_dom = h1->PciSegmentGroup;
197 	ildp->sc->iommu.unit = ivrs_info_to_unit_id(h1->Info);
198 	return (true);
199 }
200 
201 static u_int
amdiommu_devtbl_sz(struct amdiommu_unit * sc __unused)202 amdiommu_devtbl_sz(struct amdiommu_unit *sc __unused)
203 {
204 	return (sizeof(struct amdiommu_dte) * (1 << 16));
205 }
206 
207 static void
amdiommu_free_dev_tbl(struct amdiommu_unit * sc)208 amdiommu_free_dev_tbl(struct amdiommu_unit *sc)
209 {
210 	u_int devtbl_sz;
211 
212 	devtbl_sz = amdiommu_devtbl_sz(sc);
213 	pmap_qremove((vm_offset_t)sc->dev_tbl, atop(devtbl_sz));
214 	kva_free((vm_offset_t)sc->dev_tbl, devtbl_sz);
215 	sc->dev_tbl = NULL;
216 	vm_object_deallocate(sc->devtbl_obj);
217 	sc->devtbl_obj = NULL;
218 }
219 
220 static int
amdiommu_create_dev_tbl(struct amdiommu_unit * sc)221 amdiommu_create_dev_tbl(struct amdiommu_unit *sc)
222 {
223 	vm_offset_t seg_vaddr;
224 	u_int devtbl_sz, dom, i, reclaimno, segnum_log, segnum, seg_sz;
225 	int error;
226 
227 	segnum_log = (sc->efr & AMDIOMMU_EFR_DEVTBLSEG_MASK) >>
228 	    AMDIOMMU_EFR_DEVTBLSEG_SHIFT;
229 	segnum = 1 << segnum_log;
230 
231 	devtbl_sz = amdiommu_devtbl_sz(sc);
232 	seg_sz = devtbl_sz / segnum;
233 	sc->devtbl_obj = vm_pager_allocate(OBJT_PHYS, NULL, atop(devtbl_sz),
234 	    VM_PROT_ALL, 0, NULL);
235 	if (bus_get_domain(sc->iommu.dev, &dom) == 0)
236 		sc->devtbl_obj->domain.dr_policy = DOMAINSET_PREF(dom);
237 
238 	sc->hw_ctrl &= ~AMDIOMMU_CTRL_DEVTABSEG_MASK;
239 	sc->hw_ctrl |= (uint64_t)segnum_log << ilog2(AMDIOMMU_CTRL_DEVTABSEG_2);
240 	sc->hw_ctrl |= AMDIOMMU_CTRL_COHERENT;
241 	amdiommu_write8(sc, AMDIOMMU_CTRL, sc->hw_ctrl);
242 
243 	seg_vaddr = kva_alloc(devtbl_sz);
244 	if (seg_vaddr == 0)
245 		return (ENOMEM);
246 	sc->dev_tbl = (void *)seg_vaddr;
247 
248 	for (i = 0; i < segnum; i++) {
249 		vm_page_t m;
250 		uint64_t rval;
251 		u_int reg;
252 
253 		for (reclaimno = 0; reclaimno < 3; reclaimno++) {
254 			VM_OBJECT_WLOCK(sc->devtbl_obj);
255 			m = vm_page_alloc_contig(sc->devtbl_obj,
256 			    i * atop(seg_sz),
257 			    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY,
258 			    atop(seg_sz), 0, ~0ul, IOMMU_PAGE_SIZE, 0,
259 			    VM_MEMATTR_DEFAULT);
260 			VM_OBJECT_WUNLOCK(sc->devtbl_obj);
261 			if (m != NULL)
262 				break;
263 			error = vm_page_reclaim_contig(VM_ALLOC_NORMAL,
264 			    atop(seg_sz), 0, ~0ul, IOMMU_PAGE_SIZE, 0);
265 			if (error != 0)
266 				vm_wait(sc->devtbl_obj);
267 		}
268 		if (m == NULL) {
269 			amdiommu_free_dev_tbl(sc);
270 			return (ENOMEM);
271 		}
272 
273 		rval = VM_PAGE_TO_PHYS(m) | (atop(seg_sz) - 1);
274 		for (u_int j = 0; j < atop(seg_sz);
275 		     j++, seg_vaddr += PAGE_SIZE, m++) {
276 			pmap_zero_page(m);
277 			pmap_qenter(seg_vaddr, &m, 1);
278 		}
279 		reg = i == 0 ? AMDIOMMU_DEVTAB_BASE : AMDIOMMU_DEVTAB_S1_BASE +
280 		    i - 1;
281 		amdiommu_write8(sc, reg, rval);
282 	}
283 
284 	return (0);
285 }
286 
287 static int
amdiommu_cmd_event_intr(void * arg)288 amdiommu_cmd_event_intr(void *arg)
289 {
290 	struct amdiommu_unit *unit;
291 	uint64_t status;
292 
293 	unit = arg;
294 	status = amdiommu_read8(unit, AMDIOMMU_CMDEV_STATUS);
295 	if ((status & AMDIOMMU_CMDEVS_COMWAITINT) != 0) {
296 		amdiommu_write8(unit, AMDIOMMU_CMDEV_STATUS,
297 		    AMDIOMMU_CMDEVS_COMWAITINT);
298 		taskqueue_enqueue(unit->x86c.qi_taskqueue,
299 		    &unit->x86c.qi_task);
300 	}
301 	if ((status & (AMDIOMMU_CMDEVS_EVLOGINT |
302 	    AMDIOMMU_CMDEVS_EVOVRFLW)) != 0)
303 		amdiommu_event_intr(unit, status);
304 	return (FILTER_HANDLED);
305 }
306 
307 static int
amdiommu_setup_intr(struct amdiommu_unit * sc)308 amdiommu_setup_intr(struct amdiommu_unit *sc)
309 {
310 	int error, msi_count, msix_count;
311 
312 	msi_count = pci_msi_count(sc->iommu.dev);
313 	msix_count = pci_msix_count(sc->iommu.dev);
314 	if (msi_count == 0 && msix_count == 0) {
315 		device_printf(sc->iommu.dev, "needs MSI-class intr\n");
316 		return (ENXIO);
317 	}
318 
319 #if 0
320 	/*
321 	 * XXXKIB how MSI-X is supposed to be organized for BAR-less
322 	 * function?  Practically available hardware implements only
323 	 * one IOMMU unit per function, and uses MSI.
324 	 */
325 	if (msix_count > 0) {
326 		sc->msix_table = bus_alloc_resource_any(sc->iommu.dev,
327 		    SYS_RES_MEMORY, &sc->msix_tab_rid, RF_ACTIVE);
328 		if (sc->msix_table == NULL)
329 			return (ENXIO);
330 
331 		if (sc->msix_pba_rid != sc->msix_tab_rid) {
332 			/* Separate BAR for PBA */
333 			sc->msix_pba = bus_alloc_resource_any(sc->iommu.dev,
334 			    SYS_RES_MEMORY,
335 			    &sc->msix_pba_rid, RF_ACTIVE);
336 			if (sc->msix_pba == NULL) {
337 				bus_release_resource(sc->iommu.dev,
338 				    SYS_RES_MEMORY, &sc->msix_tab_rid,
339 				    sc->msix_table);
340 				return (ENXIO);
341 			}
342 		}
343 	}
344 #endif
345 
346 	error = ENXIO;
347 	if (msix_count > 0) {
348 		error = pci_alloc_msix(sc->iommu.dev, &msix_count);
349 		if (error == 0)
350 			sc->numirqs = msix_count;
351 	}
352 	if (error != 0 && msi_count > 0) {
353 		error = pci_alloc_msi(sc->iommu.dev, &msi_count);
354 		if (error == 0)
355 			sc->numirqs = msi_count;
356 	}
357 	if (error != 0) {
358 		device_printf(sc->iommu.dev,
359 		    "Failed to allocate MSI/MSI-x (%d)\n", error);
360 		return (ENXIO);
361 	}
362 
363 	/*
364 	 * XXXKIB Spec states that MISC0.MsiNum must be zero for IOMMU
365 	 * using MSI interrupts.  But at least one BIOS programmed '2'
366 	 * there, making driver use wrong rid and causing
367 	 * command/event interrupt ignored as stray.  Try to fix it
368 	 * with dirty force by assuming MsiNum is zero for MSI.
369 	 */
370 	sc->irq_cmdev_rid = 1;
371 	if (msix_count > 0) {
372 		sc->irq_cmdev_rid += pci_read_config(sc->iommu.dev,
373 		    sc->seccap_reg + PCIR_AMDIOMMU_MISC0, 4) &
374 		    PCIM_AMDIOMMU_MISC0_MSINUM_MASK;
375 	}
376 
377 	sc->irq_cmdev = bus_alloc_resource_any(sc->iommu.dev, SYS_RES_IRQ,
378 	    &sc->irq_cmdev_rid, RF_SHAREABLE | RF_ACTIVE);
379 	if (sc->irq_cmdev == NULL) {
380 		device_printf(sc->iommu.dev,
381 		    "unable to map CMD/EV interrupt\n");
382 		return (ENXIO);
383 	}
384 	error = bus_setup_intr(sc->iommu.dev, sc->irq_cmdev,
385 	    INTR_TYPE_MISC, amdiommu_cmd_event_intr, NULL, sc,
386 	    &sc->irq_cmdev_cookie);
387 	if (error != 0) {
388 		device_printf(sc->iommu.dev,
389 		    "unable to setup interrupt (%d)\n", error);
390 		return (ENXIO);
391 	}
392 	bus_describe_intr(sc->iommu.dev, sc->irq_cmdev, sc->irq_cmdev_cookie,
393 	    "cmdev");
394 
395 	if (x2apic_mode) {
396 		AMDIOMMU_LOCK(sc);
397 		sc->hw_ctrl |= AMDIOMMU_CTRL_GA_EN | AMDIOMMU_CTRL_XT_EN;
398 		amdiommu_write8(sc, AMDIOMMU_CTRL, sc->hw_ctrl);
399 		// XXXKIB AMDIOMMU_CTRL_INTCAPXT_EN and program x2APIC_CTRL
400 		AMDIOMMU_UNLOCK(sc);
401 	}
402 
403 	return (0);
404 }
405 
406 static int
amdiommu_probe(device_t dev)407 amdiommu_probe(device_t dev)
408 {
409 	int seccap_reg;
410 	int error;
411 	uint32_t cap_h, cap_type, cap_rev;
412 
413 	if (acpi_disabled("amdiommu"))
414 		return (ENXIO);
415 	TUNABLE_INT_FETCH("hw.amdiommu.enable", &amdiommu_enable);
416 	if (!amdiommu_enable)
417 		return (ENXIO);
418 	if (pci_get_class(dev) != PCIC_BASEPERIPH ||
419 	    pci_get_subclass(dev) != PCIS_BASEPERIPH_IOMMU)
420 		return (ENXIO);
421 
422 	error = pci_find_cap(dev, PCIY_SECDEV, &seccap_reg);
423 	if (error != 0 || seccap_reg == 0)
424 		return (ENXIO);
425 
426 	cap_h = pci_read_config(dev, seccap_reg + PCIR_AMDIOMMU_CAP_HEADER,
427 	    4);
428 	cap_type = cap_h & PCIM_AMDIOMMU_CAP_TYPE_MASK;
429 	cap_rev = cap_h & PCIM_AMDIOMMU_CAP_REV_MASK;
430 	if (cap_type != PCIM_AMDIOMMU_CAP_TYPE_VAL &&
431 	    cap_rev != PCIM_AMDIOMMU_CAP_REV_VAL)
432 		return (ENXIO);
433 
434 	device_set_desc(dev, "DMA remap");
435 	return (BUS_PROBE_SPECIFIC);
436 }
437 
438 static int
amdiommu_attach(device_t dev)439 amdiommu_attach(device_t dev)
440 {
441 	struct amdiommu_unit *sc;
442 	struct ivhd_lookup_data ild;
443 	int error;
444 	uint32_t base_low, base_high;
445 	bool res;
446 
447 	sc = device_get_softc(dev);
448 	sc->iommu.dev = dev;
449 
450 	error = pci_find_cap(dev, PCIY_SECDEV, &sc->seccap_reg);
451 	if (error != 0 || sc->seccap_reg == 0)
452 		return (ENXIO);
453 
454 	base_low = pci_read_config(dev, sc->seccap_reg +
455 	    PCIR_AMDIOMMU_BASE_LOW, 4);
456 	base_high = pci_read_config(dev, sc->seccap_reg +
457 	    PCIR_AMDIOMMU_BASE_HIGH, 4);
458 	sc->mmio_base = (base_low & PCIM_AMDIOMMU_BASE_LOW_ADDRM) |
459 	    ((uint64_t)base_high << 32);
460 
461 	sc->device_id = pci_get_rid(dev);
462 	ild.sc = sc;
463 	ild.devid = sc->device_id;
464 	res = amdiommu_ivrs_iterate_tbl(ivrs_lookup_ivhd_0x40,
465 	    ivrs_lookup_ivhd_0x40, ivrs_lookup_ivhd_0x10, &ild);
466 	if (!res) {
467 		device_printf(dev, "Cannot find IVHD\n");
468 		return (ENXIO);
469 	}
470 
471 	mtx_init(&sc->iommu.lock, "amdihw", NULL, MTX_DEF);
472 	sc->domids = new_unrhdr(0, 0xffff, &sc->iommu.lock);
473 	LIST_INIT(&sc->domains);
474 	sysctl_ctx_init(&sc->iommu.sysctl_ctx);
475 
476 	sc->mmio_sz = ((sc->efr & AMDIOMMU_EFR_PC_SUP) != 0 ? 512 : 16) *
477 	    1024;
478 
479 	sc->mmio_rid = AMDIOMMU_RID;
480 	error = bus_set_resource(dev, SYS_RES_MEMORY, AMDIOMMU_RID,
481 	    sc->mmio_base, sc->mmio_sz);
482 	if (error != 0) {
483 		device_printf(dev,
484 		    "bus_set_resource %#jx-%#jx failed, error %d\n",
485 		    (uintmax_t)sc->mmio_base, (uintmax_t)sc->mmio_base +
486 		    sc->mmio_sz, error);
487 		error = ENXIO;
488 		goto errout1;
489 	}
490 	sc->mmio_res = bus_alloc_resource(dev, SYS_RES_MEMORY, &sc->mmio_rid,
491 	    sc->mmio_base, sc->mmio_base + sc->mmio_sz - 1, sc->mmio_sz,
492 	    RF_ALLOCATED | RF_ACTIVE | RF_SHAREABLE);
493 	if (sc->mmio_res == NULL) {
494 		device_printf(dev,
495 		    "bus_alloc_resource %#jx-%#jx failed\n",
496 		    (uintmax_t)sc->mmio_base, (uintmax_t)sc->mmio_base +
497 		    sc->mmio_sz);
498 		error = ENXIO;
499 		goto errout2;
500 	}
501 
502 	sc->hw_ctrl = amdiommu_read8(sc, AMDIOMMU_CTRL);
503 	if (bootverbose)
504 		device_printf(dev, "ctrl reg %#jx\n", (uintmax_t)sc->hw_ctrl);
505 	if ((sc->hw_ctrl & AMDIOMMU_CTRL_EN) != 0) {
506 		device_printf(dev, "CTRL_EN is set, bailing out\n");
507 		error = EBUSY;
508 		goto errout2;
509 	}
510 
511 	iommu_high = BUS_SPACE_MAXADDR;
512 
513 	error = amdiommu_create_dev_tbl(sc);
514 	if (error != 0)
515 		goto errout3;
516 
517 	error = amdiommu_init_cmd(sc);
518 	if (error != 0)
519 		goto errout4;
520 
521 	error = amdiommu_init_event(sc);
522 	if (error != 0)
523 		goto errout5;
524 
525 	error = amdiommu_setup_intr(sc);
526 	if (error != 0)
527 		goto errout6;
528 
529 	error = iommu_init_busdma(AMD2IOMMU(sc));
530 	if (error != 0)
531 		goto errout7;
532 
533 	error = amdiommu_init_irt(sc);
534 	if (error != 0)
535 		goto errout8;
536 
537 	/*
538 	 * Unlike DMAR, AMD IOMMU does not process command queue
539 	 * unless IOMMU is enabled.  But since non-present devtab
540 	 * entry makes IOMMU ignore transactions from corresponding
541 	 * initiator, de-facto IOMMU operations are disabled for the
542 	 * DMA and intr remapping.
543 	 */
544 	AMDIOMMU_LOCK(sc);
545 	sc->hw_ctrl |= AMDIOMMU_CTRL_EN;
546 	amdiommu_write8(sc, AMDIOMMU_CTRL, sc->hw_ctrl);
547 	if (bootverbose) {
548 		printf("amdiommu%d: enabled translation\n",
549 		    AMD2IOMMU(sc)->unit);
550 	}
551 	AMDIOMMU_UNLOCK(sc);
552 
553 	TAILQ_INSERT_TAIL(&amdiommu_units, sc, unit_next);
554 	return (0);
555 
556 errout8:
557 	iommu_fini_busdma(&sc->iommu);
558 errout7:
559 	pci_release_msi(dev);
560 errout6:
561 	amdiommu_fini_event(sc);
562 errout5:
563 	amdiommu_fini_cmd(sc);
564 errout4:
565 	amdiommu_free_dev_tbl(sc);
566 errout3:
567 	bus_release_resource(dev, SYS_RES_MEMORY, sc->mmio_rid, sc->mmio_res);
568 errout2:
569 	bus_delete_resource(dev, SYS_RES_MEMORY, sc->mmio_rid);
570 errout1:
571 	sysctl_ctx_free(&sc->iommu.sysctl_ctx);
572 	delete_unrhdr(sc->domids);
573 	mtx_destroy(&sc->iommu.lock);
574 
575 	return (error);
576 }
577 
578 static int
amdiommu_detach(device_t dev)579 amdiommu_detach(device_t dev)
580 {
581 	return (EBUSY);
582 }
583 
584 static int
amdiommu_suspend(device_t dev)585 amdiommu_suspend(device_t dev)
586 {
587 	/* XXXKIB */
588 	return (0);
589 }
590 
591 static int
amdiommu_resume(device_t dev)592 amdiommu_resume(device_t dev)
593 {
594 	/* XXXKIB */
595 	return (0);
596 }
597 
598 static device_method_t amdiommu_methods[] = {
599 	DEVMETHOD(device_probe, amdiommu_probe),
600 	DEVMETHOD(device_attach, amdiommu_attach),
601 	DEVMETHOD(device_detach, amdiommu_detach),
602 	DEVMETHOD(device_suspend, amdiommu_suspend),
603 	DEVMETHOD(device_resume, amdiommu_resume),
604 	DEVMETHOD_END
605 };
606 
607 static driver_t	amdiommu_driver = {
608 	"amdiommu",
609 	amdiommu_methods,
610 	sizeof(struct amdiommu_unit),
611 };
612 
613 EARLY_DRIVER_MODULE(amdiommu, pci, amdiommu_driver, 0, 0, BUS_PASS_SUPPORTDEV);
614 MODULE_DEPEND(amdiommu, pci, 1, 1, 1);
615 
616 static struct amdiommu_unit *
amdiommu_unit_by_device_id(u_int pci_seg,u_int device_id)617 amdiommu_unit_by_device_id(u_int pci_seg, u_int device_id)
618 {
619 	struct amdiommu_unit *unit;
620 
621 	TAILQ_FOREACH(unit, &amdiommu_units, unit_next) {
622 		if (unit->unit_dom == pci_seg && unit->device_id == device_id)
623 			return (unit);
624 	}
625 	return (NULL);
626 }
627 
628 struct ivhd_find_unit {
629 	u_int		domain;
630 	uintptr_t	rid;
631 	int		devno;
632 	enum {
633 		IFU_DEV_PCI,
634 		IFU_DEV_IOAPIC,
635 		IFU_DEV_HPET,
636 	} type;
637 	u_int		device_id;
638 	uint16_t	rid_real;
639 	uint8_t		dte;
640 	uint32_t	edte;
641 };
642 
643 static bool
amdiommu_find_unit_scan_ivrs(ACPI_IVRS_DE_HEADER * d,size_t tlen,struct ivhd_find_unit * ifu)644 amdiommu_find_unit_scan_ivrs(ACPI_IVRS_DE_HEADER *d, size_t tlen,
645     struct ivhd_find_unit *ifu)
646 {
647 	char *db, *de;
648 	size_t len;
649 
650 	for (de = (char *)d + tlen; (char *)d < de;
651 	     d = (ACPI_IVRS_DE_HEADER *)(db + len)) {
652 		db = (char *)d;
653 		if (d->Type == ACPI_IVRS_TYPE_PAD4) {
654 			len = sizeof(ACPI_IVRS_DEVICE4);
655 		} else if (d->Type == ACPI_IVRS_TYPE_ALL) {
656 			ACPI_IVRS_DEVICE4 *d4;
657 
658 			d4 = (ACPI_IVRS_DEVICE4 *)db;
659 			len = sizeof(*d4);
660 			ifu->dte = d4->Header.DataSetting;
661 		} else if (d->Type == ACPI_IVRS_TYPE_SELECT) {
662 			ACPI_IVRS_DEVICE4 *d4;
663 
664 			d4 = (ACPI_IVRS_DEVICE4 *)db;
665 			if (d4->Header.Id == ifu->rid) {
666 				ifu->dte = d4->Header.DataSetting;
667 				ifu->rid_real = ifu->rid;
668 				return (true);
669 			}
670 			len = sizeof(*d4);
671 		} else if (d->Type == ACPI_IVRS_TYPE_START) {
672 			ACPI_IVRS_DEVICE4 *d4, *d4n;
673 
674 			d4 = (ACPI_IVRS_DEVICE4 *)db;
675 			d4n = d4 + 1;
676 			if (d4n->Header.Type != ACPI_IVRS_TYPE_END) {
677 				printf("IVRS dev4 start not followed by END "
678 				    "(%#x)\n", d4n->Header.Type);
679 				return (false);
680 			}
681 			if (d4->Header.Id <= ifu->rid &&
682 			    ifu->rid <= d4n->Header.Id) {
683 				ifu->dte = d4->Header.DataSetting;
684 				ifu->rid_real = ifu->rid;
685 				return (true);
686 			}
687 			len = 2 * sizeof(*d4);
688 		} else if (d->Type == ACPI_IVRS_TYPE_PAD8) {
689 			len = sizeof(ACPI_IVRS_DEVICE8A);
690 		} else if (d->Type == ACPI_IVRS_TYPE_ALIAS_SELECT) {
691 			ACPI_IVRS_DEVICE8A *d8a;
692 
693 			d8a = (ACPI_IVRS_DEVICE8A *)db;
694 			if (d8a->Header.Id == ifu->rid) {
695 				ifu->dte = d8a->Header.DataSetting;
696 				ifu->rid_real = d8a->UsedId;
697 				return (true);
698 			}
699 			len = sizeof(*d8a);
700 		} else if (d->Type == ACPI_IVRS_TYPE_ALIAS_START) {
701 			ACPI_IVRS_DEVICE8A *d8a;
702 			ACPI_IVRS_DEVICE4 *d4;
703 
704 			d8a = (ACPI_IVRS_DEVICE8A *)db;
705 			d4 = (ACPI_IVRS_DEVICE4 *)(d8a + 1);
706 			if (d4->Header.Type != ACPI_IVRS_TYPE_END) {
707 				printf("IVRS alias start not followed by END "
708 				    "(%#x)\n", d4->Header.Type);
709 				return (false);
710 			}
711 			if (d8a->Header.Id <= ifu->rid &&
712 			    ifu->rid <= d4->Header.Id) {
713 				ifu->dte = d8a->Header.DataSetting;
714 				ifu->rid_real = d8a->UsedId;
715 				return (true);
716 			}
717 			len = sizeof(*d8a) + sizeof(*d4);
718 		} else if (d->Type == ACPI_IVRS_TYPE_EXT_SELECT) {
719 			ACPI_IVRS_DEVICE8B *d8b;
720 
721 			d8b = (ACPI_IVRS_DEVICE8B *)db;
722 			if (d8b->Header.Id == ifu->rid) {
723 				ifu->dte = d8b->Header.DataSetting;
724 				ifu->rid_real = ifu->rid;
725 				ifu->edte = d8b->ExtendedData;
726 				return (true);
727 			}
728 			len = sizeof(*d8b);
729 		} else if (d->Type == ACPI_IVRS_TYPE_EXT_START) {
730 			ACPI_IVRS_DEVICE8B *d8b;
731 			ACPI_IVRS_DEVICE4 *d4;
732 
733 			d8b = (ACPI_IVRS_DEVICE8B *)db;
734 			d4 = (ACPI_IVRS_DEVICE4 *)(db + sizeof(*d8b));
735 			if (d4->Header.Type != ACPI_IVRS_TYPE_END) {
736 				printf("IVRS ext start not followed by END "
737 				    "(%#x)\n", d4->Header.Type);
738 				return (false);
739 			}
740 			if (d8b->Header.Id >= ifu->rid &&
741 			    ifu->rid <= d4->Header.Id) {
742 				ifu->dte = d8b->Header.DataSetting;
743 				ifu->rid_real = ifu->rid;
744 				ifu->edte = d8b->ExtendedData;
745 				return (true);
746 			}
747 			len = sizeof(*d8b) + sizeof(*d4);
748 		} else if (d->Type == ACPI_IVRS_TYPE_SPECIAL) {
749 			ACPI_IVRS_DEVICE8C *d8c;
750 
751 			d8c = (ACPI_IVRS_DEVICE8C *)db;
752 			if (((ifu->type == IFU_DEV_IOAPIC &&
753 			    d8c->Variety == ACPI_IVHD_IOAPIC) ||
754 			    (ifu->type == IFU_DEV_HPET &&
755 			    d8c->Variety == ACPI_IVHD_HPET)) &&
756 			    ifu->devno == d8c->Handle) {
757 				ifu->dte = d8c->Header.DataSetting;
758 				ifu->rid_real = d8c->UsedId;
759 				return (true);
760 			}
761 			len = sizeof(*d8c);
762 		} else if (d->Type == ACPI_IVRS_TYPE_HID) {
763 			ACPI_IVRS_DEVICE_HID *dh;
764 
765 			dh = (ACPI_IVRS_DEVICE_HID *)db;
766 			len = sizeof(*dh) + dh->UidLength;
767 			/* XXXKIB */
768 		} else {
769 #if 0
770 			printf("amdiommu: unknown IVRS device entry type %#x\n",
771 			    d->Type);
772 #endif
773 			if (d->Type <= 63)
774 				len = sizeof(ACPI_IVRS_DEVICE4);
775 			else if (d->Type <= 127)
776 				len = sizeof(ACPI_IVRS_DEVICE8A);
777 			else {
778 				printf("amdiommu: abort, cannot "
779 				    "advance iterator, item type %#x\n",
780 				    d->Type);
781 				return (false);
782 			}
783 		}
784 	}
785 	return (false);
786 }
787 
788 static bool
amdiommu_find_unit_scan_0x11(ACPI_IVRS_HARDWARE2 * ivrs,void * arg)789 amdiommu_find_unit_scan_0x11(ACPI_IVRS_HARDWARE2 *ivrs, void *arg)
790 {
791 	struct ivhd_find_unit *ifu = arg;
792 	ACPI_IVRS_DE_HEADER *d;
793 	bool res;
794 
795 	KASSERT(ivrs->Header.Type == ACPI_IVRS_TYPE_HARDWARE2 ||
796 	    ivrs->Header.Type == ACPI_IVRS_TYPE_HARDWARE3,
797 	    ("Misparsed IVHD h2, ivrs type %#x", ivrs->Header.Type));
798 
799 	if (ifu->domain != ivrs->PciSegmentGroup)
800 		return (false);
801 	d = (ACPI_IVRS_DE_HEADER *)(ivrs + 1);
802 	res = amdiommu_find_unit_scan_ivrs(d, ivrs->Header.Length, ifu);
803 	if (res)
804 		ifu->device_id = ivrs->Header.DeviceId;
805 	return (res);
806 }
807 
808 static bool
amdiommu_find_unit_scan_0x10(ACPI_IVRS_HARDWARE1 * ivrs,void * arg)809 amdiommu_find_unit_scan_0x10(ACPI_IVRS_HARDWARE1 *ivrs, void *arg)
810 {
811 	struct ivhd_find_unit *ifu = arg;
812 	ACPI_IVRS_DE_HEADER *d;
813 	bool res;
814 
815 	KASSERT(ivrs->Header.Type == ACPI_IVRS_TYPE_HARDWARE1,
816 	    ("Misparsed IVHD h1, ivrs type %#x", ivrs->Header.Type));
817 
818 	if (ifu->domain != ivrs->PciSegmentGroup)
819 		return (false);
820 	d = (ACPI_IVRS_DE_HEADER *)(ivrs + 1);
821 	res = amdiommu_find_unit_scan_ivrs(d, ivrs->Header.Length, ifu);
822 	if (res)
823 		ifu->device_id = ivrs->Header.DeviceId;
824 	return (res);
825 }
826 
827 static void
amdiommu_dev_prop_dtr(device_t dev,const char * name,void * val,void * dtr_ctx)828 amdiommu_dev_prop_dtr(device_t dev, const char *name, void *val, void *dtr_ctx)
829 {
830 	free(val, M_DEVBUF);
831 }
832 
833 static int *
amdiommu_dev_fetch_flagsp(struct amdiommu_unit * unit,device_t dev)834 amdiommu_dev_fetch_flagsp(struct amdiommu_unit *unit, device_t dev)
835 {
836 	int *flagsp, error;
837 
838 	bus_topo_assert();
839 	error = device_get_prop(dev, device_get_nameunit(unit->iommu.dev),
840 	    (void **)&flagsp);
841 	if (error == ENOENT) {
842 		flagsp = malloc(sizeof(int), M_DEVBUF, M_WAITOK | M_ZERO);
843 		device_set_prop(dev, device_get_nameunit(unit->iommu.dev),
844 		    flagsp, amdiommu_dev_prop_dtr, unit);
845 	}
846 	return (flagsp);
847 }
848 
849 static int
amdiommu_get_dev_prop_flags(struct amdiommu_unit * unit,device_t dev)850 amdiommu_get_dev_prop_flags(struct amdiommu_unit *unit, device_t dev)
851 {
852 	int *flagsp, flags;
853 
854 	bus_topo_lock();
855 	flagsp = amdiommu_dev_fetch_flagsp(unit, dev);
856 	flags = *flagsp;
857 	bus_topo_unlock();
858 	return (flags);
859 }
860 
861 static void
amdiommu_set_dev_prop_flags(struct amdiommu_unit * unit,device_t dev,int flag)862 amdiommu_set_dev_prop_flags(struct amdiommu_unit *unit, device_t dev,
863     int flag)
864 {
865 	int *flagsp;
866 
867 	bus_topo_lock();
868 	flagsp = amdiommu_dev_fetch_flagsp(unit, dev);
869 	*flagsp |= flag;
870 	bus_topo_unlock();
871 }
872 
873 int
amdiommu_find_unit(device_t dev,struct amdiommu_unit ** unitp,uint16_t * ridp,uint8_t * dtep,uint32_t * edtep,bool verbose)874 amdiommu_find_unit(device_t dev, struct amdiommu_unit **unitp, uint16_t *ridp,
875     uint8_t *dtep, uint32_t *edtep, bool verbose)
876 {
877 	struct ivhd_find_unit ifu;
878 	struct amdiommu_unit *unit;
879 	int error, flags;
880 	bool res;
881 
882 	if (!amdiommu_enable)
883 		return (ENXIO);
884 
885 	if (device_get_devclass(device_get_parent(dev)) !=
886 	    devclass_find("pci"))
887 		return (ENXIO);
888 
889 	bzero(&ifu, sizeof(ifu));
890 	ifu.type = IFU_DEV_PCI;
891 
892 	error = pci_get_id(dev, PCI_ID_RID, &ifu.rid);
893 	if (error != 0) {
894 		if (verbose)
895 			device_printf(dev,
896 			    "amdiommu cannot get rid, error %d\n", error);
897 		return (ENXIO);
898 	}
899 
900 	ifu.domain = pci_get_domain(dev);
901 	res = amdiommu_ivrs_iterate_tbl(amdiommu_find_unit_scan_0x11,
902 	    amdiommu_find_unit_scan_0x11, amdiommu_find_unit_scan_0x10, &ifu);
903 	if (!res) {
904 		if (verbose)
905 			device_printf(dev,
906 			    "(%#06x:%#06x) amdiommu cannot match rid in IVHD\n",
907 			    ifu.domain, (unsigned)ifu.rid);
908 		return (ENXIO);
909 	}
910 
911 	unit = amdiommu_unit_by_device_id(ifu.domain, ifu.device_id);
912 	if (unit == NULL) {
913 		if (verbose)
914 			device_printf(dev,
915 			    "(%#06x:%#06x) amdiommu cannot find unit\n",
916 			    ifu.domain, (unsigned)ifu.rid);
917 		return (ENXIO);
918 	}
919 	*unitp = unit;
920 	iommu_device_set_iommu_prop(dev, unit->iommu.dev);
921 	if (ridp != NULL)
922 		*ridp = ifu.rid_real;
923 	if (dtep != NULL)
924 		*dtep = ifu.dte;
925 	if (edtep != NULL)
926 		*edtep = ifu.edte;
927 	if (verbose) {
928 		flags = amdiommu_get_dev_prop_flags(unit, dev);
929 		if ((flags & AMDIOMMU_DEV_REPORTED) == 0) {
930 			amdiommu_set_dev_prop_flags(unit, dev,
931 			    AMDIOMMU_DEV_REPORTED);
932 			device_printf(dev, "amdiommu%d "
933 			    "initiator rid %#06x dte %#x edte %#x\n",
934 			    unit->iommu.unit, ifu.rid_real, ifu.dte, ifu.edte);
935 		}
936 	}
937 	return (0);
938 }
939 
940 int
amdiommu_find_unit_for_ioapic(int apic_id,struct amdiommu_unit ** unitp,uint16_t * ridp,uint8_t * dtep,uint32_t * edtep,bool verbose)941 amdiommu_find_unit_for_ioapic(int apic_id, struct amdiommu_unit **unitp,
942     uint16_t *ridp, uint8_t *dtep, uint32_t *edtep, bool verbose)
943 {
944 	struct ivhd_find_unit ifu;
945 	struct amdiommu_unit *unit;
946 	device_t apic_dev;
947 	bool res;
948 
949 	if (!amdiommu_enable)
950 		return (ENXIO);
951 
952 	bzero(&ifu, sizeof(ifu));
953 	ifu.type = IFU_DEV_IOAPIC;
954 	ifu.devno = apic_id;
955 	ifu.rid = -1;
956 
957 	res = amdiommu_ivrs_iterate_tbl(amdiommu_find_unit_scan_0x11,
958 	    amdiommu_find_unit_scan_0x11, amdiommu_find_unit_scan_0x10, &ifu);
959 	if (!res) {
960 		if (verbose)
961 			printf("amdiommu cannot match ioapic no %d in IVHD\n",
962 			    apic_id);
963 		return (ENXIO);
964 	}
965 
966 	unit = amdiommu_unit_by_device_id(0, ifu.device_id);
967 	apic_dev = ioapic_get_dev(apic_id);
968 	if (apic_dev != NULL)
969 		iommu_device_set_iommu_prop(apic_dev, unit->iommu.dev);
970 	if (unit == NULL) {
971 		if (verbose)
972 			printf("amdiommu cannot find unit by dev id %#x\n",
973 			    ifu.device_id);
974 		return (ENXIO);
975 	}
976 	*unitp = unit;
977 	if (ridp != NULL)
978 		*ridp = ifu.rid_real;
979 	if (dtep != NULL)
980 		*dtep = ifu.dte;
981 	if (edtep != NULL)
982 		*edtep = ifu.edte;
983 	if (verbose) {
984 		printf("amdiommu%d IOAPIC %d "
985 		    "initiator rid %#06x dte %#x edte %#x\n",
986 		    unit->iommu.unit, apic_id, ifu.rid_real, ifu.dte,
987 		    ifu.edte);
988 	}
989 	return (0);
990 }
991 
992 int
amdiommu_find_unit_for_hpet(device_t hpet,struct amdiommu_unit ** unitp,uint16_t * ridp,uint8_t * dtep,uint32_t * edtep,bool verbose)993 amdiommu_find_unit_for_hpet(device_t hpet, struct amdiommu_unit **unitp,
994     uint16_t *ridp, uint8_t *dtep, uint32_t *edtep, bool verbose)
995 {
996 	struct ivhd_find_unit ifu;
997 	struct amdiommu_unit *unit;
998 	int hpet_no;
999 	bool res;
1000 
1001 	if (!amdiommu_enable)
1002 		return (ENXIO);
1003 
1004 	hpet_no = hpet_get_uid(hpet);
1005 	bzero(&ifu, sizeof(ifu));
1006 	ifu.type = IFU_DEV_HPET;
1007 	ifu.devno = hpet_no;
1008 	ifu.rid = -1;
1009 
1010 	res = amdiommu_ivrs_iterate_tbl(amdiommu_find_unit_scan_0x11,
1011 	    amdiommu_find_unit_scan_0x11, amdiommu_find_unit_scan_0x10, &ifu);
1012 	if (!res) {
1013 		if (verbose)
1014 			printf("amdiommu cannot match hpet no %d in IVHD\n",
1015 			    hpet_no);
1016 		return (ENXIO);
1017 	}
1018 
1019 	unit = amdiommu_unit_by_device_id(0, ifu.device_id);
1020 	if (unit == NULL) {
1021 		if (verbose)
1022 			printf("amdiommu cannot find unit id %d\n",
1023 			    hpet_no);
1024 		return (ENXIO);
1025 	}
1026 	*unitp = unit;
1027 	iommu_device_set_iommu_prop(hpet, unit->iommu.dev);
1028 	if (ridp != NULL)
1029 		*ridp = ifu.rid_real;
1030 	if (dtep != NULL)
1031 		*dtep = ifu.dte;
1032 	if (edtep != NULL)
1033 		*edtep = ifu.edte;
1034 	if (verbose) {
1035 		printf("amdiommu%d HPET no %d "
1036 		    "initiator rid %#06x dte %#x edte %#x\n",
1037 		    unit->iommu.unit, hpet_no, ifu.rid_real, ifu.dte,
1038 		    ifu.edte);
1039 	}
1040 	return (0);
1041 }
1042 
1043 static struct iommu_unit *
amdiommu_find_method(device_t dev,bool verbose)1044 amdiommu_find_method(device_t dev, bool verbose)
1045 {
1046 	struct amdiommu_unit *unit;
1047 	int error;
1048 	uint32_t edte;
1049 	uint16_t rid;
1050 	uint8_t dte;
1051 
1052 	error = amdiommu_find_unit(dev, &unit, &rid, &dte, &edte, verbose);
1053 	if (error != 0) {
1054 		if (verbose && amdiommu_enable)
1055 			device_printf(dev,
1056 			    "cannot find amdiommu unit, error %d\n",
1057 			    error);
1058 		return (NULL);
1059 	}
1060 	return (&unit->iommu);
1061 }
1062 
1063 static struct x86_unit_common *
amdiommu_get_x86_common(struct iommu_unit * unit)1064 amdiommu_get_x86_common(struct iommu_unit *unit)
1065 {
1066 	struct amdiommu_unit *iommu;
1067 
1068 	iommu = IOMMU2AMD(unit);
1069 	return (&iommu->x86c);
1070 }
1071 
1072 static void
amdiommu_unit_pre_instantiate_ctx(struct iommu_unit * unit)1073 amdiommu_unit_pre_instantiate_ctx(struct iommu_unit *unit)
1074 {
1075 }
1076 
1077 static struct x86_iommu amd_x86_iommu = {
1078 	.get_x86_common = amdiommu_get_x86_common,
1079 	.unit_pre_instantiate_ctx = amdiommu_unit_pre_instantiate_ctx,
1080 	.find = amdiommu_find_method,
1081 	.domain_unload_entry = amdiommu_domain_unload_entry,
1082 	.domain_unload = amdiommu_domain_unload,
1083 	.get_ctx = amdiommu_get_ctx,
1084 	.free_ctx_locked = amdiommu_free_ctx_locked_method,
1085 	.alloc_msi_intr = amdiommu_alloc_msi_intr,
1086 	.map_msi_intr = amdiommu_map_msi_intr,
1087 	.unmap_msi_intr = amdiommu_unmap_msi_intr,
1088 	.map_ioapic_intr = amdiommu_map_ioapic_intr,
1089 	.unmap_ioapic_intr = amdiommu_unmap_ioapic_intr,
1090 };
1091 
1092 static void
x86_iommu_set_amd(void * arg __unused)1093 x86_iommu_set_amd(void *arg __unused)
1094 {
1095 	if (cpu_vendor_id == CPU_VENDOR_AMD)
1096 		set_x86_iommu(&amd_x86_iommu);
1097 }
1098 
1099 SYSINIT(x86_iommu, SI_SUB_TUNABLES, SI_ORDER_ANY, x86_iommu_set_amd, NULL);
1100 
1101 #ifdef DDB
1102 #include <ddb/ddb.h>
1103 #include <ddb/db_lex.h>
1104 
1105 static void
amdiommu_print_domain(struct amdiommu_domain * domain,bool show_mappings)1106 amdiommu_print_domain(struct amdiommu_domain *domain, bool show_mappings)
1107 {
1108 	struct iommu_domain *iodom;
1109 
1110 	iodom = DOM2IODOM(domain);
1111 
1112 	db_printf(
1113 	    "  @%p dom %d pglvl %d end %jx refs %d\n"
1114 	    "   ctx_cnt %d flags %x pgobj %p map_ents %u\n",
1115 	    domain, domain->domain, domain->pglvl,
1116 	    (uintmax_t)domain->iodom.end, domain->refs, domain->ctx_cnt,
1117 	    domain->iodom.flags, domain->pgtbl_obj, domain->iodom.entries_cnt);
1118 
1119 	iommu_db_domain_print_contexts(iodom);
1120 
1121 	if (show_mappings)
1122 		iommu_db_domain_print_mappings(iodom);
1123 }
1124 
1125 static void
amdiommu_print_one(struct amdiommu_unit * unit,bool show_domains,bool show_mappings,bool show_cmdq)1126 amdiommu_print_one(struct amdiommu_unit *unit, bool show_domains,
1127     bool show_mappings, bool show_cmdq)
1128 {
1129 	struct amdiommu_domain *domain;
1130 	struct amdiommu_cmd_generic *cp;
1131 	u_int cmd_head, cmd_tail, ci;
1132 
1133 	cmd_head = amdiommu_read4(unit, AMDIOMMU_CMDBUF_HEAD);
1134 	cmd_tail = amdiommu_read4(unit, AMDIOMMU_CMDBUF_TAIL);
1135 	db_printf("amdiommu%d at %p, mmio at %#jx/sz %#jx\n",
1136 	    unit->iommu.unit, unit, (uintmax_t)unit->mmio_base,
1137 	    (uintmax_t)unit->mmio_sz);
1138 	db_printf("  hw ctrl %#018jx cmdevst %#018jx\n",
1139 	    (uintmax_t)amdiommu_read8(unit, AMDIOMMU_CTRL),
1140 	    (uintmax_t)amdiommu_read8(unit, AMDIOMMU_CMDEV_STATUS));
1141 	db_printf("  devtbl at %p\n", unit->dev_tbl);
1142 	db_printf("  hwseq at %p phys %#jx val %#jx\n",
1143 	    &unit->x86c.inv_waitd_seq_hw,
1144 	    pmap_kextract((vm_offset_t)&unit->x86c.inv_waitd_seq_hw),
1145 	    unit->x86c.inv_waitd_seq_hw);
1146 	db_printf("  invq at %p base %#jx hw head/tail %#x/%#x\n",
1147 	    unit->x86c.inv_queue,
1148 	    (uintmax_t)amdiommu_read8(unit, AMDIOMMU_CMDBUF_BASE),
1149 	    cmd_head, cmd_tail);
1150 
1151 	if (show_cmdq) {
1152 		db_printf("  cmd q:\n");
1153 		for (ci = cmd_head; ci != cmd_tail;) {
1154 			cp = (struct amdiommu_cmd_generic *)(unit->
1155 			    x86c.inv_queue + ci);
1156 			db_printf(
1157 		    "    idx %#x op %#x %#010x %#010x %#010x %#010x\n",
1158 			    ci >> AMDIOMMU_CMD_SZ_SHIFT, cp->op,
1159 		    	    cp->w0, cp->ww1, cp->w2, cp->w3);
1160 
1161 			ci += AMDIOMMU_CMD_SZ;
1162 			if (ci == unit->x86c.inv_queue_size)
1163 				ci = 0;
1164 		}
1165 	}
1166 
1167 	if (show_domains) {
1168 		db_printf("  domains:\n");
1169 		LIST_FOREACH(domain, &unit->domains, link) {
1170 			amdiommu_print_domain(domain, show_mappings);
1171 			if (db_pager_quit)
1172 				break;
1173 		}
1174 	}
1175 }
1176 
DB_SHOW_COMMAND(amdiommu,db_amdiommu_print)1177 DB_SHOW_COMMAND(amdiommu, db_amdiommu_print)
1178 {
1179 	struct amdiommu_unit *unit;
1180 	bool show_domains, show_mappings, show_cmdq;
1181 
1182 	show_domains = strchr(modif, 'd') != NULL;
1183 	show_mappings = strchr(modif, 'm') != NULL;
1184 	show_cmdq = strchr(modif, 'q') != NULL;
1185 	if (!have_addr) {
1186 		db_printf("usage: show amdiommu [/d] [/m] [/q] index\n");
1187 		return;
1188 	}
1189 	if ((vm_offset_t)addr < 0x10000)
1190 		unit = amdiommu_unit_by_device_id(0, (u_int)addr);
1191 	else
1192 		unit = (struct amdiommu_unit *)addr;
1193 	amdiommu_print_one(unit, show_domains, show_mappings, show_cmdq);
1194 }
1195 
DB_SHOW_ALL_COMMAND(amdiommus,db_show_all_amdiommus)1196 DB_SHOW_ALL_COMMAND(amdiommus, db_show_all_amdiommus)
1197 {
1198 	struct amdiommu_unit *unit;
1199 	bool show_domains, show_mappings, show_cmdq;
1200 
1201 	show_domains = strchr(modif, 'd') != NULL;
1202 	show_mappings = strchr(modif, 'm') != NULL;
1203 	show_cmdq = strchr(modif, 'q') != NULL;
1204 
1205 	TAILQ_FOREACH(unit, &amdiommu_units, unit_next) {
1206 		amdiommu_print_one(unit, show_domains, show_mappings,
1207 		    show_cmdq);
1208 		if (db_pager_quit)
1209 			break;
1210 	}
1211 }
1212 #endif
1213