xref: /freebsd/sys/x86/iommu/intel_utils.c (revision 8ef24a0d4b28fe230e20637f56869cc4148cd2ca)
1 /*-
2  * Copyright (c) 2013 The FreeBSD Foundation
3  * All rights reserved.
4  *
5  * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
6  * under sponsorship from the FreeBSD Foundation.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/param.h>
34 #include <sys/bus.h>
35 #include <sys/kernel.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/memdesc.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/queue.h>
42 #include <sys/rman.h>
43 #include <sys/rwlock.h>
44 #include <sys/sched.h>
45 #include <sys/sf_buf.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 #include <sys/taskqueue.h>
49 #include <sys/tree.h>
50 #include <sys/vmem.h>
51 #include <dev/pci/pcivar.h>
52 #include <vm/vm.h>
53 #include <vm/vm_extern.h>
54 #include <vm/vm_kern.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_map.h>
58 #include <vm/vm_pageout.h>
59 #include <machine/bus.h>
60 #include <machine/cpu.h>
61 #include <machine/intr_machdep.h>
62 #include <x86/include/apicvar.h>
63 #include <x86/include/busdma_impl.h>
64 #include <x86/iommu/intel_reg.h>
65 #include <x86/iommu/busdma_dmar.h>
66 #include <x86/iommu/intel_dmar.h>
67 
68 u_int
69 dmar_nd2mask(u_int nd)
70 {
71 	static const u_int masks[] = {
72 		0x000f,	/* nd == 0 */
73 		0x002f,	/* nd == 1 */
74 		0x00ff,	/* nd == 2 */
75 		0x02ff,	/* nd == 3 */
76 		0x0fff,	/* nd == 4 */
77 		0x2fff,	/* nd == 5 */
78 		0xffff,	/* nd == 6 */
79 		0x0000,	/* nd == 7 reserved */
80 	};
81 
82 	KASSERT(nd <= 6, ("number of domains %d", nd));
83 	return (masks[nd]);
84 }
85 
86 static const struct sagaw_bits_tag {
87 	int agaw;
88 	int cap;
89 	int awlvl;
90 	int pglvl;
91 } sagaw_bits[] = {
92 	{.agaw = 30, .cap = DMAR_CAP_SAGAW_2LVL, .awlvl = DMAR_CTX2_AW_2LVL,
93 	    .pglvl = 2},
94 	{.agaw = 39, .cap = DMAR_CAP_SAGAW_3LVL, .awlvl = DMAR_CTX2_AW_3LVL,
95 	    .pglvl = 3},
96 	{.agaw = 48, .cap = DMAR_CAP_SAGAW_4LVL, .awlvl = DMAR_CTX2_AW_4LVL,
97 	    .pglvl = 4},
98 	{.agaw = 57, .cap = DMAR_CAP_SAGAW_5LVL, .awlvl = DMAR_CTX2_AW_5LVL,
99 	    .pglvl = 5},
100 	{.agaw = 64, .cap = DMAR_CAP_SAGAW_6LVL, .awlvl = DMAR_CTX2_AW_6LVL,
101 	    .pglvl = 6}
102 };
103 
104 bool
105 dmar_pglvl_supported(struct dmar_unit *unit, int pglvl)
106 {
107 	int i;
108 
109 	for (i = 0; i < nitems(sagaw_bits); i++) {
110 		if (sagaw_bits[i].pglvl != pglvl)
111 			continue;
112 		if ((DMAR_CAP_SAGAW(unit->hw_cap) & sagaw_bits[i].cap) != 0)
113 			return (true);
114 	}
115 	return (false);
116 }
117 
118 int
119 domain_set_agaw(struct dmar_domain *domain, int mgaw)
120 {
121 	int sagaw, i;
122 
123 	domain->mgaw = mgaw;
124 	sagaw = DMAR_CAP_SAGAW(domain->dmar->hw_cap);
125 	for (i = 0; i < nitems(sagaw_bits); i++) {
126 		if (sagaw_bits[i].agaw >= mgaw) {
127 			domain->agaw = sagaw_bits[i].agaw;
128 			domain->pglvl = sagaw_bits[i].pglvl;
129 			domain->awlvl = sagaw_bits[i].awlvl;
130 			return (0);
131 		}
132 	}
133 	device_printf(domain->dmar->dev,
134 	    "context request mgaw %d: no agaw found, sagaw %x\n",
135 	    mgaw, sagaw);
136 	return (EINVAL);
137 }
138 
139 /*
140  * Find a best fit mgaw for the given maxaddr:
141  *   - if allow_less is false, must find sagaw which maps all requested
142  *     addresses (used by identity mappings);
143  *   - if allow_less is true, and no supported sagaw can map all requested
144  *     address space, accept the biggest sagaw, whatever is it.
145  */
146 int
147 dmar_maxaddr2mgaw(struct dmar_unit *unit, dmar_gaddr_t maxaddr, bool allow_less)
148 {
149 	int i;
150 
151 	for (i = 0; i < nitems(sagaw_bits); i++) {
152 		if ((1ULL << sagaw_bits[i].agaw) >= maxaddr &&
153 		    (DMAR_CAP_SAGAW(unit->hw_cap) & sagaw_bits[i].cap) != 0)
154 			break;
155 	}
156 	if (allow_less && i == nitems(sagaw_bits)) {
157 		do {
158 			i--;
159 		} while ((DMAR_CAP_SAGAW(unit->hw_cap) & sagaw_bits[i].cap)
160 		    == 0);
161 	}
162 	if (i < nitems(sagaw_bits))
163 		return (sagaw_bits[i].agaw);
164 	KASSERT(0, ("no mgaw for maxaddr %jx allow_less %d",
165 	    (uintmax_t) maxaddr, allow_less));
166 	return (-1);
167 }
168 
169 /*
170  * Calculate the total amount of page table pages needed to map the
171  * whole bus address space on the context with the selected agaw.
172  */
173 vm_pindex_t
174 pglvl_max_pages(int pglvl)
175 {
176 	vm_pindex_t res;
177 	int i;
178 
179 	for (res = 0, i = pglvl; i > 0; i--) {
180 		res *= DMAR_NPTEPG;
181 		res++;
182 	}
183 	return (res);
184 }
185 
186 /*
187  * Return true if the page table level lvl supports the superpage for
188  * the context ctx.
189  */
190 int
191 domain_is_sp_lvl(struct dmar_domain *domain, int lvl)
192 {
193 	int alvl, cap_sps;
194 	static const int sagaw_sp[] = {
195 		DMAR_CAP_SPS_2M,
196 		DMAR_CAP_SPS_1G,
197 		DMAR_CAP_SPS_512G,
198 		DMAR_CAP_SPS_1T
199 	};
200 
201 	alvl = domain->pglvl - lvl - 1;
202 	cap_sps = DMAR_CAP_SPS(domain->dmar->hw_cap);
203 	return (alvl < nitems(sagaw_sp) && (sagaw_sp[alvl] & cap_sps) != 0);
204 }
205 
206 dmar_gaddr_t
207 pglvl_page_size(int total_pglvl, int lvl)
208 {
209 	int rlvl;
210 	static const dmar_gaddr_t pg_sz[] = {
211 		(dmar_gaddr_t)DMAR_PAGE_SIZE,
212 		(dmar_gaddr_t)DMAR_PAGE_SIZE << DMAR_NPTEPGSHIFT,
213 		(dmar_gaddr_t)DMAR_PAGE_SIZE << (2 * DMAR_NPTEPGSHIFT),
214 		(dmar_gaddr_t)DMAR_PAGE_SIZE << (3 * DMAR_NPTEPGSHIFT),
215 		(dmar_gaddr_t)DMAR_PAGE_SIZE << (4 * DMAR_NPTEPGSHIFT),
216 		(dmar_gaddr_t)DMAR_PAGE_SIZE << (5 * DMAR_NPTEPGSHIFT)
217 	};
218 
219 	KASSERT(lvl >= 0 && lvl < total_pglvl,
220 	    ("total %d lvl %d", total_pglvl, lvl));
221 	rlvl = total_pglvl - lvl - 1;
222 	KASSERT(rlvl < nitems(pg_sz), ("sizeof pg_sz lvl %d", lvl));
223 	return (pg_sz[rlvl]);
224 }
225 
226 dmar_gaddr_t
227 domain_page_size(struct dmar_domain *domain, int lvl)
228 {
229 
230 	return (pglvl_page_size(domain->pglvl, lvl));
231 }
232 
233 int
234 calc_am(struct dmar_unit *unit, dmar_gaddr_t base, dmar_gaddr_t size,
235     dmar_gaddr_t *isizep)
236 {
237 	dmar_gaddr_t isize;
238 	int am;
239 
240 	for (am = DMAR_CAP_MAMV(unit->hw_cap);; am--) {
241 		isize = 1ULL << (am + DMAR_PAGE_SHIFT);
242 		if ((base & (isize - 1)) == 0 && size >= isize)
243 			break;
244 		if (am == 0)
245 			break;
246 	}
247 	*isizep = isize;
248 	return (am);
249 }
250 
251 dmar_haddr_t dmar_high;
252 int haw;
253 int dmar_tbl_pagecnt;
254 
255 vm_page_t
256 dmar_pgalloc(vm_object_t obj, vm_pindex_t idx, int flags)
257 {
258 	vm_page_t m;
259 	int zeroed;
260 
261 	zeroed = (flags & DMAR_PGF_ZERO) != 0 ? VM_ALLOC_ZERO : 0;
262 	for (;;) {
263 		if ((flags & DMAR_PGF_OBJL) == 0)
264 			VM_OBJECT_WLOCK(obj);
265 		m = vm_page_lookup(obj, idx);
266 		if ((flags & DMAR_PGF_NOALLOC) != 0 || m != NULL) {
267 			if ((flags & DMAR_PGF_OBJL) == 0)
268 				VM_OBJECT_WUNLOCK(obj);
269 			break;
270 		}
271 		m = vm_page_alloc_contig(obj, idx, VM_ALLOC_NOBUSY |
272 		    VM_ALLOC_SYSTEM | VM_ALLOC_NODUMP | zeroed, 1, 0,
273 		    dmar_high, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
274 		if ((flags & DMAR_PGF_OBJL) == 0)
275 			VM_OBJECT_WUNLOCK(obj);
276 		if (m != NULL) {
277 			if (zeroed && (m->flags & PG_ZERO) == 0)
278 				pmap_zero_page(m);
279 			atomic_add_int(&dmar_tbl_pagecnt, 1);
280 			break;
281 		}
282 		if ((flags & DMAR_PGF_WAITOK) == 0)
283 			break;
284 		if ((flags & DMAR_PGF_OBJL) != 0)
285 			VM_OBJECT_WUNLOCK(obj);
286 		VM_WAIT;
287 		if ((flags & DMAR_PGF_OBJL) != 0)
288 			VM_OBJECT_WLOCK(obj);
289 	}
290 	return (m);
291 }
292 
293 void
294 dmar_pgfree(vm_object_t obj, vm_pindex_t idx, int flags)
295 {
296 	vm_page_t m;
297 
298 	if ((flags & DMAR_PGF_OBJL) == 0)
299 		VM_OBJECT_WLOCK(obj);
300 	m = vm_page_lookup(obj, idx);
301 	if (m != NULL) {
302 		vm_page_free(m);
303 		atomic_subtract_int(&dmar_tbl_pagecnt, 1);
304 	}
305 	if ((flags & DMAR_PGF_OBJL) == 0)
306 		VM_OBJECT_WUNLOCK(obj);
307 }
308 
309 void *
310 dmar_map_pgtbl(vm_object_t obj, vm_pindex_t idx, int flags,
311     struct sf_buf **sf)
312 {
313 	vm_page_t m;
314 	bool allocated;
315 
316 	if ((flags & DMAR_PGF_OBJL) == 0)
317 		VM_OBJECT_WLOCK(obj);
318 	m = vm_page_lookup(obj, idx);
319 	if (m == NULL && (flags & DMAR_PGF_ALLOC) != 0) {
320 		m = dmar_pgalloc(obj, idx, flags | DMAR_PGF_OBJL);
321 		allocated = true;
322 	} else
323 		allocated = false;
324 	if (m == NULL) {
325 		if ((flags & DMAR_PGF_OBJL) == 0)
326 			VM_OBJECT_WUNLOCK(obj);
327 		return (NULL);
328 	}
329 	/* Sleepable allocations cannot fail. */
330 	if ((flags & DMAR_PGF_WAITOK) != 0)
331 		VM_OBJECT_WUNLOCK(obj);
332 	sched_pin();
333 	*sf = sf_buf_alloc(m, SFB_CPUPRIVATE | ((flags & DMAR_PGF_WAITOK)
334 	    == 0 ? SFB_NOWAIT : 0));
335 	if (*sf == NULL) {
336 		sched_unpin();
337 		if (allocated) {
338 			VM_OBJECT_ASSERT_WLOCKED(obj);
339 			dmar_pgfree(obj, m->pindex, flags | DMAR_PGF_OBJL);
340 		}
341 		if ((flags & DMAR_PGF_OBJL) == 0)
342 			VM_OBJECT_WUNLOCK(obj);
343 		return (NULL);
344 	}
345 	if ((flags & (DMAR_PGF_WAITOK | DMAR_PGF_OBJL)) ==
346 	    (DMAR_PGF_WAITOK | DMAR_PGF_OBJL))
347 		VM_OBJECT_WLOCK(obj);
348 	else if ((flags & (DMAR_PGF_WAITOK | DMAR_PGF_OBJL)) == 0)
349 		VM_OBJECT_WUNLOCK(obj);
350 	return ((void *)sf_buf_kva(*sf));
351 }
352 
353 void
354 dmar_unmap_pgtbl(struct sf_buf *sf)
355 {
356 
357 	sf_buf_free(sf);
358 	sched_unpin();
359 }
360 
361 static void
362 dmar_flush_transl_to_ram(struct dmar_unit *unit, void *dst, size_t sz)
363 {
364 
365 	if (DMAR_IS_COHERENT(unit))
366 		return;
367 	/*
368 	 * If DMAR does not snoop paging structures accesses, flush
369 	 * CPU cache to memory.
370 	 */
371 	pmap_invalidate_cache_range((uintptr_t)dst, (uintptr_t)dst + sz,
372 	    TRUE);
373 }
374 
375 void
376 dmar_flush_pte_to_ram(struct dmar_unit *unit, dmar_pte_t *dst)
377 {
378 
379 	dmar_flush_transl_to_ram(unit, dst, sizeof(*dst));
380 }
381 
382 void
383 dmar_flush_ctx_to_ram(struct dmar_unit *unit, dmar_ctx_entry_t *dst)
384 {
385 
386 	dmar_flush_transl_to_ram(unit, dst, sizeof(*dst));
387 }
388 
389 void
390 dmar_flush_root_to_ram(struct dmar_unit *unit, dmar_root_entry_t *dst)
391 {
392 
393 	dmar_flush_transl_to_ram(unit, dst, sizeof(*dst));
394 }
395 
396 /*
397  * Load the root entry pointer into the hardware, busily waiting for
398  * the completion.
399  */
400 int
401 dmar_load_root_entry_ptr(struct dmar_unit *unit)
402 {
403 	vm_page_t root_entry;
404 
405 	/*
406 	 * Access to the GCMD register must be serialized while the
407 	 * command is submitted.
408 	 */
409 	DMAR_ASSERT_LOCKED(unit);
410 
411 	VM_OBJECT_RLOCK(unit->ctx_obj);
412 	root_entry = vm_page_lookup(unit->ctx_obj, 0);
413 	VM_OBJECT_RUNLOCK(unit->ctx_obj);
414 	dmar_write8(unit, DMAR_RTADDR_REG, VM_PAGE_TO_PHYS(root_entry));
415 	dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd | DMAR_GCMD_SRTP);
416 	/* XXXKIB should have a timeout */
417 	while ((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_RTPS) == 0)
418 		cpu_spinwait();
419 	return (0);
420 }
421 
422 /*
423  * Globally invalidate the context entries cache, busily waiting for
424  * the completion.
425  */
426 int
427 dmar_inv_ctx_glob(struct dmar_unit *unit)
428 {
429 
430 	/*
431 	 * Access to the CCMD register must be serialized while the
432 	 * command is submitted.
433 	 */
434 	DMAR_ASSERT_LOCKED(unit);
435 	KASSERT(!unit->qi_enabled, ("QI enabled"));
436 
437 	/*
438 	 * The DMAR_CCMD_ICC bit in the upper dword should be written
439 	 * after the low dword write is completed.  Amd64
440 	 * dmar_write8() does not have this issue, i386 dmar_write8()
441 	 * writes the upper dword last.
442 	 */
443 	dmar_write8(unit, DMAR_CCMD_REG, DMAR_CCMD_ICC | DMAR_CCMD_CIRG_GLOB);
444 	/* XXXKIB should have a timeout */
445 	while ((dmar_read4(unit, DMAR_CCMD_REG + 4) & DMAR_CCMD_ICC32) != 0)
446 		cpu_spinwait();
447 	return (0);
448 }
449 
450 /*
451  * Globally invalidate the IOTLB, busily waiting for the completion.
452  */
453 int
454 dmar_inv_iotlb_glob(struct dmar_unit *unit)
455 {
456 	int reg;
457 
458 	DMAR_ASSERT_LOCKED(unit);
459 	KASSERT(!unit->qi_enabled, ("QI enabled"));
460 
461 	reg = 16 * DMAR_ECAP_IRO(unit->hw_ecap);
462 	/* See a comment about DMAR_CCMD_ICC in dmar_inv_ctx_glob. */
463 	dmar_write8(unit, reg + DMAR_IOTLB_REG_OFF, DMAR_IOTLB_IVT |
464 	    DMAR_IOTLB_IIRG_GLB | DMAR_IOTLB_DR | DMAR_IOTLB_DW);
465 	/* XXXKIB should have a timeout */
466 	while ((dmar_read4(unit, reg + DMAR_IOTLB_REG_OFF + 4) &
467 	    DMAR_IOTLB_IVT32) != 0)
468 		cpu_spinwait();
469 	return (0);
470 }
471 
472 /*
473  * Flush the chipset write buffers.  See 11.1 "Write Buffer Flushing"
474  * in the architecture specification.
475  */
476 int
477 dmar_flush_write_bufs(struct dmar_unit *unit)
478 {
479 
480 	DMAR_ASSERT_LOCKED(unit);
481 
482 	/*
483 	 * DMAR_GCMD_WBF is only valid when CAP_RWBF is reported.
484 	 */
485 	KASSERT((unit->hw_cap & DMAR_CAP_RWBF) != 0,
486 	    ("dmar%d: no RWBF", unit->unit));
487 
488 	dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd | DMAR_GCMD_WBF);
489 	/* XXXKIB should have a timeout */
490 	while ((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_WBFS) == 0)
491 		cpu_spinwait();
492 	return (0);
493 }
494 
495 int
496 dmar_enable_translation(struct dmar_unit *unit)
497 {
498 
499 	DMAR_ASSERT_LOCKED(unit);
500 	unit->hw_gcmd |= DMAR_GCMD_TE;
501 	dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
502 	/* XXXKIB should have a timeout */
503 	while ((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_TES) == 0)
504 		cpu_spinwait();
505 	return (0);
506 }
507 
508 int
509 dmar_disable_translation(struct dmar_unit *unit)
510 {
511 
512 	DMAR_ASSERT_LOCKED(unit);
513 	unit->hw_gcmd &= ~DMAR_GCMD_TE;
514 	dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
515 	/* XXXKIB should have a timeout */
516 	while ((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_TES) != 0)
517 		cpu_spinwait();
518 	return (0);
519 }
520 
521 int
522 dmar_load_irt_ptr(struct dmar_unit *unit)
523 {
524 	uint64_t irta, s;
525 
526 	DMAR_ASSERT_LOCKED(unit);
527 	irta = unit->irt_phys;
528 	if (DMAR_X2APIC(unit))
529 		irta |= DMAR_IRTA_EIME;
530 	s = fls(unit->irte_cnt) - 2;
531 	KASSERT(unit->irte_cnt >= 2 && s <= DMAR_IRTA_S_MASK &&
532 	    powerof2(unit->irte_cnt),
533 	    ("IRTA_REG_S overflow %x", unit->irte_cnt));
534 	irta |= s;
535 	dmar_write8(unit, DMAR_IRTA_REG, irta);
536 	dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd | DMAR_GCMD_SIRTP);
537 	/* XXXKIB should have a timeout */
538 	while ((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_IRTPS) == 0)
539 		cpu_spinwait();
540 	return (0);
541 }
542 
543 int
544 dmar_enable_ir(struct dmar_unit *unit)
545 {
546 
547 	DMAR_ASSERT_LOCKED(unit);
548 	unit->hw_gcmd |= DMAR_GCMD_IRE;
549 	unit->hw_gcmd &= ~DMAR_GCMD_CFI;
550 	dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
551 	/* XXXKIB should have a timeout */
552 	while ((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_IRES) == 0)
553 		cpu_spinwait();
554 	return (0);
555 }
556 
557 int
558 dmar_disable_ir(struct dmar_unit *unit)
559 {
560 
561 	DMAR_ASSERT_LOCKED(unit);
562 	unit->hw_gcmd &= ~DMAR_GCMD_IRE;
563 	dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
564 	/* XXXKIB should have a timeout */
565 	while ((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_IRES) != 0)
566 		cpu_spinwait();
567 	return (0);
568 }
569 
570 #define BARRIER_F				\
571 	u_int f_done, f_inproc, f_wakeup;	\
572 						\
573 	f_done = 1 << (barrier_id * 3);		\
574 	f_inproc = 1 << (barrier_id * 3 + 1);	\
575 	f_wakeup = 1 << (barrier_id * 3 + 2)
576 
577 bool
578 dmar_barrier_enter(struct dmar_unit *dmar, u_int barrier_id)
579 {
580 	BARRIER_F;
581 
582 	DMAR_LOCK(dmar);
583 	if ((dmar->barrier_flags & f_done) != 0) {
584 		DMAR_UNLOCK(dmar);
585 		return (false);
586 	}
587 
588 	if ((dmar->barrier_flags & f_inproc) != 0) {
589 		while ((dmar->barrier_flags & f_inproc) != 0) {
590 			dmar->barrier_flags |= f_wakeup;
591 			msleep(&dmar->barrier_flags, &dmar->lock, 0,
592 			    "dmarb", 0);
593 		}
594 		KASSERT((dmar->barrier_flags & f_done) != 0,
595 		    ("dmar%d barrier %d missing done", dmar->unit, barrier_id));
596 		DMAR_UNLOCK(dmar);
597 		return (false);
598 	}
599 
600 	dmar->barrier_flags |= f_inproc;
601 	DMAR_UNLOCK(dmar);
602 	return (true);
603 }
604 
605 void
606 dmar_barrier_exit(struct dmar_unit *dmar, u_int barrier_id)
607 {
608 	BARRIER_F;
609 
610 	DMAR_ASSERT_LOCKED(dmar);
611 	KASSERT((dmar->barrier_flags & (f_done | f_inproc)) == f_inproc,
612 	    ("dmar%d barrier %d missed entry", dmar->unit, barrier_id));
613 	dmar->barrier_flags |= f_done;
614 	if ((dmar->barrier_flags & f_wakeup) != 0)
615 		wakeup(&dmar->barrier_flags);
616 	dmar->barrier_flags &= ~(f_inproc | f_wakeup);
617 	DMAR_UNLOCK(dmar);
618 }
619 
620 int dmar_match_verbose;
621 int dmar_batch_coalesce = 100;
622 
623 static SYSCTL_NODE(_hw, OID_AUTO, dmar, CTLFLAG_RD, NULL, "");
624 SYSCTL_INT(_hw_dmar, OID_AUTO, tbl_pagecnt, CTLFLAG_RD,
625     &dmar_tbl_pagecnt, 0,
626     "Count of pages used for DMAR pagetables");
627 SYSCTL_INT(_hw_dmar, OID_AUTO, match_verbose, CTLFLAG_RWTUN,
628     &dmar_match_verbose, 0,
629     "Verbose matching of the PCI devices to DMAR paths");
630 SYSCTL_INT(_hw_dmar, OID_AUTO, batch_coalesce, CTLFLAG_RWTUN,
631     &dmar_batch_coalesce, 0,
632     "Number of qi batches between interrupt");
633 #ifdef INVARIANTS
634 int dmar_check_free;
635 SYSCTL_INT(_hw_dmar, OID_AUTO, check_free, CTLFLAG_RWTUN,
636     &dmar_check_free, 0,
637     "Check the GPA RBtree for free_down and free_after validity");
638 #endif
639 
640