1/*-
2 * Copyright (c) 1997, 1998 Justin T. Gibbs.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions, and the following disclaimer,
10 *    without modification, immediately at the beginning of the file.
11 * 2. The name of the author may not be used to endorse or promote products
12 *    derived from this software without specific prior written permission.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
18 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27/*
28 * From amd64/busdma_machdep.c, r204214
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD$");
33
34#include <sys/param.h>
35#include <sys/systm.h>
36#include <sys/malloc.h>
37#include <sys/bus.h>
38#include <sys/interrupt.h>
39#include <sys/kernel.h>
40#include <sys/ktr.h>
41#include <sys/lock.h>
42#include <sys/proc.h>
43#include <sys/memdesc.h>
44#include <sys/mutex.h>
45#include <sys/sysctl.h>
46#include <sys/uio.h>
47
48#include <vm/vm.h>
49#include <vm/vm_extern.h>
50#include <vm/vm_kern.h>
51#include <vm/vm_page.h>
52#include <vm/vm_map.h>
53
54#include <machine/atomic.h>
55#include <machine/bus.h>
56#include <machine/cpufunc.h>
57#include <machine/md_var.h>
58
59#include "iommu_if.h"
60
61#define MAX_BPAGES MIN(8192, physmem/40)
62
63struct bounce_zone;
64
65struct bus_dma_tag {
66	bus_dma_tag_t	  parent;
67	bus_size_t	  alignment;
68	bus_addr_t	  boundary;
69	bus_addr_t	  lowaddr;
70	bus_addr_t	  highaddr;
71	bus_dma_filter_t *filter;
72	void		 *filterarg;
73	bus_size_t	  maxsize;
74	u_int		  nsegments;
75	bus_size_t	  maxsegsz;
76	int		  flags;
77	int		  ref_count;
78	int		  map_count;
79	bus_dma_lock_t	 *lockfunc;
80	void		 *lockfuncarg;
81	struct bounce_zone *bounce_zone;
82	device_t	  iommu;
83	void		 *iommu_cookie;
84};
85
86struct bounce_page {
87	vm_offset_t	vaddr;		/* kva of bounce buffer */
88	bus_addr_t	busaddr;	/* Physical address */
89	vm_offset_t	datavaddr;	/* kva of client data */
90	bus_addr_t	dataaddr;	/* client physical address */
91	bus_size_t	datacount;	/* client data count */
92	STAILQ_ENTRY(bounce_page) links;
93};
94
95int busdma_swi_pending;
96
97struct bounce_zone {
98	STAILQ_ENTRY(bounce_zone) links;
99	STAILQ_HEAD(bp_list, bounce_page) bounce_page_list;
100	int		total_bpages;
101	int		free_bpages;
102	int		reserved_bpages;
103	int		active_bpages;
104	int		total_bounced;
105	int		total_deferred;
106	int		map_count;
107	bus_size_t	alignment;
108	bus_addr_t	lowaddr;
109	char		zoneid[8];
110	char		lowaddrid[20];
111	struct sysctl_ctx_list sysctl_tree;
112	struct sysctl_oid *sysctl_tree_top;
113};
114
115static struct mtx bounce_lock;
116static int total_bpages;
117static int busdma_zonecount;
118static STAILQ_HEAD(, bounce_zone) bounce_zone_list;
119
120static SYSCTL_NODE(_hw, OID_AUTO, busdma, CTLFLAG_RD, 0, "Busdma parameters");
121SYSCTL_INT(_hw_busdma, OID_AUTO, total_bpages, CTLFLAG_RD, &total_bpages, 0,
122	   "Total bounce pages");
123
124struct bus_dmamap {
125	struct bp_list	       bpages;
126	int		       pagesneeded;
127	int		       pagesreserved;
128	bus_dma_tag_t	       dmat;
129	struct memdesc	       mem;
130	bus_dma_segment_t     *segments;
131	int		       nsegs;
132	bus_dmamap_callback_t *callback;
133	void		      *callback_arg;
134	STAILQ_ENTRY(bus_dmamap) links;
135	int		       contigalloc;
136};
137
138static STAILQ_HEAD(, bus_dmamap) bounce_map_waitinglist;
139static STAILQ_HEAD(, bus_dmamap) bounce_map_callbacklist;
140
141static void init_bounce_pages(void *dummy);
142static int alloc_bounce_zone(bus_dma_tag_t dmat);
143static int alloc_bounce_pages(bus_dma_tag_t dmat, u_int numpages);
144static int reserve_bounce_pages(bus_dma_tag_t dmat, bus_dmamap_t map,
145				int commit);
146static bus_addr_t add_bounce_page(bus_dma_tag_t dmat, bus_dmamap_t map,
147				  vm_offset_t vaddr, bus_addr_t addr,
148				  bus_size_t size);
149static void free_bounce_page(bus_dma_tag_t dmat, struct bounce_page *bpage);
150static __inline int run_filter(bus_dma_tag_t dmat, bus_addr_t paddr);
151
152/*
153 * Return true if a match is made.
154 *
155 * To find a match walk the chain of bus_dma_tag_t's looking for 'paddr'.
156 *
157 * If paddr is within the bounds of the dma tag then call the filter callback
158 * to check for a match, if there is no filter callback then assume a match.
159 */
160static __inline int
161run_filter(bus_dma_tag_t dmat, bus_addr_t paddr)
162{
163	int retval;
164
165	retval = 0;
166
167	do {
168		if (dmat->filter == NULL && dmat->iommu == NULL &&
169		    paddr > dmat->lowaddr && paddr <= dmat->highaddr)
170			retval = 1;
171		if (dmat->filter == NULL &&
172		    (paddr & (dmat->alignment - 1)) != 0)
173			retval = 1;
174		if (dmat->filter != NULL &&
175		    (*dmat->filter)(dmat->filterarg, paddr) != 0)
176			retval = 1;
177
178		dmat = dmat->parent;
179	} while (retval == 0 && dmat != NULL);
180	return (retval);
181}
182
183/*
184 * Convenience function for manipulating driver locks from busdma (during
185 * busdma_swi, for example).  Drivers that don't provide their own locks
186 * should specify &Giant to dmat->lockfuncarg.  Drivers that use their own
187 * non-mutex locking scheme don't have to use this at all.
188 */
189void
190busdma_lock_mutex(void *arg, bus_dma_lock_op_t op)
191{
192	struct mtx *dmtx;
193
194	dmtx = (struct mtx *)arg;
195	switch (op) {
196	case BUS_DMA_LOCK:
197		mtx_lock(dmtx);
198		break;
199	case BUS_DMA_UNLOCK:
200		mtx_unlock(dmtx);
201		break;
202	default:
203		panic("Unknown operation 0x%x for busdma_lock_mutex!", op);
204	}
205}
206
207/*
208 * dflt_lock should never get called.  It gets put into the dma tag when
209 * lockfunc == NULL, which is only valid if the maps that are associated
210 * with the tag are meant to never be defered.
211 * XXX Should have a way to identify which driver is responsible here.
212 */
213static void
214dflt_lock(void *arg, bus_dma_lock_op_t op)
215{
216	panic("driver error: busdma dflt_lock called");
217}
218
219#define BUS_DMA_COULD_BOUNCE	BUS_DMA_BUS3
220#define BUS_DMA_MIN_ALLOC_COMP	BUS_DMA_BUS4
221/*
222 * Allocate a device specific dma_tag.
223 */
224int
225bus_dma_tag_create(bus_dma_tag_t parent, bus_size_t alignment,
226		   bus_addr_t boundary, bus_addr_t lowaddr,
227		   bus_addr_t highaddr, bus_dma_filter_t *filter,
228		   void *filterarg, bus_size_t maxsize, int nsegments,
229		   bus_size_t maxsegsz, int flags, bus_dma_lock_t *lockfunc,
230		   void *lockfuncarg, bus_dma_tag_t *dmat)
231{
232	bus_dma_tag_t newtag;
233	int error = 0;
234
235	/* Basic sanity checking */
236	if (boundary != 0 && boundary < maxsegsz)
237		maxsegsz = boundary;
238
239	if (maxsegsz == 0) {
240		return (EINVAL);
241	}
242
243	/* Return a NULL tag on failure */
244	*dmat = NULL;
245
246	newtag = (bus_dma_tag_t)malloc(sizeof(*newtag), M_DEVBUF,
247	    M_ZERO | M_NOWAIT);
248	if (newtag == NULL) {
249		CTR4(KTR_BUSDMA, "%s returned tag %p tag flags 0x%x error %d",
250		    __func__, newtag, 0, error);
251		return (ENOMEM);
252	}
253
254	newtag->parent = parent;
255	newtag->alignment = alignment;
256	newtag->boundary = boundary;
257	newtag->lowaddr = trunc_page((vm_paddr_t)lowaddr) + (PAGE_SIZE - 1);
258	newtag->highaddr = trunc_page((vm_paddr_t)highaddr) + (PAGE_SIZE - 1);
259	newtag->filter = filter;
260	newtag->filterarg = filterarg;
261	newtag->maxsize = maxsize;
262	newtag->nsegments = nsegments;
263	newtag->maxsegsz = maxsegsz;
264	newtag->flags = flags;
265	newtag->ref_count = 1; /* Count ourself */
266	newtag->map_count = 0;
267	if (lockfunc != NULL) {
268		newtag->lockfunc = lockfunc;
269		newtag->lockfuncarg = lockfuncarg;
270	} else {
271		newtag->lockfunc = dflt_lock;
272		newtag->lockfuncarg = NULL;
273	}
274
275	/* Take into account any restrictions imposed by our parent tag */
276	if (parent != NULL) {
277		newtag->lowaddr = MIN(parent->lowaddr, newtag->lowaddr);
278		newtag->highaddr = MAX(parent->highaddr, newtag->highaddr);
279		if (newtag->boundary == 0)
280			newtag->boundary = parent->boundary;
281		else if (parent->boundary != 0)
282			newtag->boundary = MIN(parent->boundary,
283					       newtag->boundary);
284		if (newtag->filter == NULL) {
285			/*
286			 * Short circuit looking at our parent directly
287			 * since we have encapsulated all of its information
288			 */
289			newtag->filter = parent->filter;
290			newtag->filterarg = parent->filterarg;
291			newtag->parent = parent->parent;
292		}
293		if (newtag->parent != NULL)
294			atomic_add_int(&parent->ref_count, 1);
295		newtag->iommu = parent->iommu;
296		newtag->iommu_cookie = parent->iommu_cookie;
297	}
298
299	if (newtag->lowaddr < ptoa((vm_paddr_t)Maxmem) && newtag->iommu == NULL)
300		newtag->flags |= BUS_DMA_COULD_BOUNCE;
301
302	if (newtag->alignment > 1)
303		newtag->flags |= BUS_DMA_COULD_BOUNCE;
304
305	if (((newtag->flags & BUS_DMA_COULD_BOUNCE) != 0) &&
306	    (flags & BUS_DMA_ALLOCNOW) != 0) {
307		struct bounce_zone *bz;
308
309		/* Must bounce */
310
311		if ((error = alloc_bounce_zone(newtag)) != 0) {
312			free(newtag, M_DEVBUF);
313			return (error);
314		}
315		bz = newtag->bounce_zone;
316
317		if (ptoa(bz->total_bpages) < maxsize) {
318			int pages;
319
320			pages = atop(maxsize) - bz->total_bpages;
321
322			/* Add pages to our bounce pool */
323			if (alloc_bounce_pages(newtag, pages) < pages)
324				error = ENOMEM;
325		}
326		/* Performed initial allocation */
327		newtag->flags |= BUS_DMA_MIN_ALLOC_COMP;
328	}
329
330	if (error != 0) {
331		free(newtag, M_DEVBUF);
332	} else {
333		*dmat = newtag;
334	}
335	CTR4(KTR_BUSDMA, "%s returned tag %p tag flags 0x%x error %d",
336	    __func__, newtag, (newtag != NULL ? newtag->flags : 0), error);
337	return (error);
338}
339
340int
341bus_dma_tag_destroy(bus_dma_tag_t dmat)
342{
343	bus_dma_tag_t dmat_copy;
344	int error;
345
346	error = 0;
347	dmat_copy = dmat;
348
349	if (dmat != NULL) {
350
351		if (dmat->map_count != 0) {
352			error = EBUSY;
353			goto out;
354		}
355
356		while (dmat != NULL) {
357			bus_dma_tag_t parent;
358
359			parent = dmat->parent;
360			atomic_subtract_int(&dmat->ref_count, 1);
361			if (dmat->ref_count == 0) {
362				free(dmat, M_DEVBUF);
363				/*
364				 * Last reference count, so
365				 * release our reference
366				 * count on our parent.
367				 */
368				dmat = parent;
369			} else
370				dmat = NULL;
371		}
372	}
373out:
374	CTR3(KTR_BUSDMA, "%s tag %p error %d", __func__, dmat_copy, error);
375	return (error);
376}
377
378/*
379 * Allocate a handle for mapping from kva/uva/physical
380 * address space into bus device space.
381 */
382int
383bus_dmamap_create(bus_dma_tag_t dmat, int flags, bus_dmamap_t *mapp)
384{
385	int error;
386
387	error = 0;
388
389	*mapp = (bus_dmamap_t)malloc(sizeof(**mapp), M_DEVBUF,
390				     M_NOWAIT | M_ZERO);
391	if (*mapp == NULL) {
392		CTR3(KTR_BUSDMA, "%s: tag %p error %d",
393		    __func__, dmat, ENOMEM);
394		return (ENOMEM);
395	}
396
397
398	/*
399	 * Bouncing might be required if the driver asks for an active
400	 * exclusion region, a data alignment that is stricter than 1, and/or
401	 * an active address boundary.
402	 */
403	if (dmat->flags & BUS_DMA_COULD_BOUNCE) {
404
405		/* Must bounce */
406		struct bounce_zone *bz;
407		int maxpages;
408
409		if (dmat->bounce_zone == NULL) {
410			if ((error = alloc_bounce_zone(dmat)) != 0)
411				return (error);
412		}
413		bz = dmat->bounce_zone;
414
415		/* Initialize the new map */
416		STAILQ_INIT(&((*mapp)->bpages));
417
418		/*
419		 * Attempt to add pages to our pool on a per-instance
420		 * basis up to a sane limit.
421		 */
422		if (dmat->alignment > 1)
423			maxpages = MAX_BPAGES;
424		else
425			maxpages = MIN(MAX_BPAGES, Maxmem -atop(dmat->lowaddr));
426		if ((dmat->flags & BUS_DMA_MIN_ALLOC_COMP) == 0
427		 || (bz->map_count > 0 && bz->total_bpages < maxpages)) {
428			int pages;
429
430			pages = MAX(atop(dmat->maxsize), 1);
431			pages = MIN(maxpages - bz->total_bpages, pages);
432			pages = MAX(pages, 1);
433			if (alloc_bounce_pages(dmat, pages) < pages)
434				error = ENOMEM;
435
436			if ((dmat->flags & BUS_DMA_MIN_ALLOC_COMP) == 0) {
437				if (error == 0)
438					dmat->flags |= BUS_DMA_MIN_ALLOC_COMP;
439			} else {
440				error = 0;
441			}
442		}
443		bz->map_count++;
444	}
445
446	(*mapp)->nsegs = 0;
447	(*mapp)->segments = (bus_dma_segment_t *)malloc(
448	    sizeof(bus_dma_segment_t) * dmat->nsegments, M_DEVBUF,
449	    M_NOWAIT);
450	if ((*mapp)->segments == NULL) {
451		CTR3(KTR_BUSDMA, "%s: tag %p error %d",
452		    __func__, dmat, ENOMEM);
453		return (ENOMEM);
454	}
455
456	if (error == 0)
457		dmat->map_count++;
458	CTR4(KTR_BUSDMA, "%s: tag %p tag flags 0x%x error %d",
459	    __func__, dmat, dmat->flags, error);
460	return (error);
461}
462
463/*
464 * Destroy a handle for mapping from kva/uva/physical
465 * address space into bus device space.
466 */
467int
468bus_dmamap_destroy(bus_dma_tag_t dmat, bus_dmamap_t map)
469{
470	if (dmat->flags & BUS_DMA_COULD_BOUNCE) {
471		if (STAILQ_FIRST(&map->bpages) != NULL) {
472			CTR3(KTR_BUSDMA, "%s: tag %p error %d",
473			    __func__, dmat, EBUSY);
474			return (EBUSY);
475		}
476		if (dmat->bounce_zone)
477			dmat->bounce_zone->map_count--;
478	}
479	free(map->segments, M_DEVBUF);
480	free(map, M_DEVBUF);
481	dmat->map_count--;
482	CTR2(KTR_BUSDMA, "%s: tag %p error 0", __func__, dmat);
483	return (0);
484}
485
486
487/*
488 * Allocate a piece of memory that can be efficiently mapped into
489 * bus device space based on the constraints lited in the dma tag.
490 * A dmamap to for use with dmamap_load is also allocated.
491 */
492int
493bus_dmamem_alloc(bus_dma_tag_t dmat, void** vaddr, int flags,
494		 bus_dmamap_t *mapp)
495{
496	vm_memattr_t attr;
497	int mflags;
498
499	if (flags & BUS_DMA_NOWAIT)
500		mflags = M_NOWAIT;
501	else
502		mflags = M_WAITOK;
503
504	bus_dmamap_create(dmat, flags, mapp);
505
506	if (flags & BUS_DMA_ZERO)
507		mflags |= M_ZERO;
508#ifdef NOTYET
509	if (flags & BUS_DMA_NOCACHE)
510		attr = VM_MEMATTR_UNCACHEABLE;
511	else
512#endif
513		attr = VM_MEMATTR_DEFAULT;
514
515	/*
516	 * XXX:
517	 * (dmat->alignment <= dmat->maxsize) is just a quick hack; the exact
518	 * alignment guarantees of malloc need to be nailed down, and the
519	 * code below should be rewritten to take that into account.
520	 *
521	 * In the meantime, we'll warn the user if malloc gets it wrong.
522	 */
523	if ((dmat->maxsize <= PAGE_SIZE) &&
524	   (dmat->alignment <= dmat->maxsize) &&
525	    dmat->lowaddr >= ptoa((vm_paddr_t)Maxmem) &&
526	    attr == VM_MEMATTR_DEFAULT) {
527		*vaddr = malloc(dmat->maxsize, M_DEVBUF, mflags);
528	} else {
529		/*
530		 * XXX Use Contigmalloc until it is merged into this facility
531		 *     and handles multi-seg allocations.  Nobody is doing
532		 *     multi-seg allocations yet though.
533		 * XXX Certain AGP hardware does.
534		 */
535		*vaddr = (void *)kmem_alloc_contig(kmem_arena, dmat->maxsize,
536		    mflags, 0ul, dmat->lowaddr, dmat->alignment ?
537		    dmat->alignment : 1ul, dmat->boundary, attr);
538		(*mapp)->contigalloc = 1;
539	}
540	if (*vaddr == NULL) {
541		CTR4(KTR_BUSDMA, "%s: tag %p tag flags 0x%x error %d",
542		    __func__, dmat, dmat->flags, ENOMEM);
543		return (ENOMEM);
544	} else if (vtophys(*vaddr) & (dmat->alignment - 1)) {
545		printf("bus_dmamem_alloc failed to align memory properly.\n");
546	}
547	CTR4(KTR_BUSDMA, "%s: tag %p tag flags 0x%x error %d",
548	    __func__, dmat, dmat->flags, 0);
549	return (0);
550}
551
552/*
553 * Free a piece of memory and it's allociated dmamap, that was allocated
554 * via bus_dmamem_alloc.  Make the same choice for free/contigfree.
555 */
556void
557bus_dmamem_free(bus_dma_tag_t dmat, void *vaddr, bus_dmamap_t map)
558{
559
560	if (!map->contigalloc)
561		free(vaddr, M_DEVBUF);
562	else
563		kmem_free(kmem_arena, (vm_offset_t)vaddr, dmat->maxsize);
564	bus_dmamap_destroy(dmat, map);
565	CTR3(KTR_BUSDMA, "%s: tag %p flags 0x%x", __func__, dmat, dmat->flags);
566}
567
568static void
569_bus_dmamap_count_phys(bus_dma_tag_t dmat, bus_dmamap_t map, vm_paddr_t buf,
570    bus_size_t buflen, int flags)
571{
572	bus_addr_t curaddr;
573	bus_size_t sgsize;
574
575	if (map->pagesneeded == 0) {
576		CTR4(KTR_BUSDMA, "lowaddr= %d Maxmem= %d, boundary= %d, "
577		    "alignment= %d", dmat->lowaddr, ptoa((vm_paddr_t)Maxmem),
578		    dmat->boundary, dmat->alignment);
579		CTR2(KTR_BUSDMA, "map= %p, pagesneeded= %d", map, map->pagesneeded);
580		/*
581		 * Count the number of bounce pages
582		 * needed in order to complete this transfer
583		 */
584		curaddr = buf;
585		while (buflen != 0) {
586			sgsize = MIN(buflen, dmat->maxsegsz);
587			if (run_filter(dmat, curaddr) != 0) {
588				sgsize = MIN(sgsize, PAGE_SIZE);
589				map->pagesneeded++;
590			}
591			curaddr += sgsize;
592			buflen -= sgsize;
593		}
594		CTR1(KTR_BUSDMA, "pagesneeded= %d\n", map->pagesneeded);
595	}
596}
597
598static void
599_bus_dmamap_count_pages(bus_dma_tag_t dmat, bus_dmamap_t map, pmap_t pmap,
600    void *buf, bus_size_t buflen, int flags)
601{
602        vm_offset_t vaddr;
603        vm_offset_t vendaddr;
604        bus_addr_t paddr;
605
606	if (map->pagesneeded == 0) {
607		CTR4(KTR_BUSDMA, "lowaddr= %d Maxmem= %d, boundary= %d, "
608		    "alignment= %d", dmat->lowaddr, ptoa((vm_paddr_t)Maxmem),
609		    dmat->boundary, dmat->alignment);
610		CTR2(KTR_BUSDMA, "map= %p, pagesneeded= %d", map, map->pagesneeded);
611		/*
612		 * Count the number of bounce pages
613		 * needed in order to complete this transfer
614		 */
615		vaddr = (vm_offset_t)buf;
616		vendaddr = (vm_offset_t)buf + buflen;
617
618		while (vaddr < vendaddr) {
619			bus_size_t sg_len;
620
621			sg_len = PAGE_SIZE - ((vm_offset_t)vaddr & PAGE_MASK);
622			if (pmap == kernel_pmap)
623				paddr = pmap_kextract(vaddr);
624			else
625				paddr = pmap_extract(pmap, vaddr);
626			if (run_filter(dmat, paddr) != 0) {
627				sg_len = roundup2(sg_len, dmat->alignment);
628				map->pagesneeded++;
629			}
630			vaddr += sg_len;
631		}
632		CTR1(KTR_BUSDMA, "pagesneeded= %d\n", map->pagesneeded);
633	}
634}
635
636static int
637_bus_dmamap_reserve_pages(bus_dma_tag_t dmat, bus_dmamap_t map, int flags)
638{
639
640	/* Reserve Necessary Bounce Pages */
641	mtx_lock(&bounce_lock);
642	if (flags & BUS_DMA_NOWAIT) {
643		if (reserve_bounce_pages(dmat, map, 0) != 0) {
644			mtx_unlock(&bounce_lock);
645			return (ENOMEM);
646		}
647	} else {
648		if (reserve_bounce_pages(dmat, map, 1) != 0) {
649			/* Queue us for resources */
650			STAILQ_INSERT_TAIL(&bounce_map_waitinglist,
651			    map, links);
652			mtx_unlock(&bounce_lock);
653			return (EINPROGRESS);
654		}
655	}
656	mtx_unlock(&bounce_lock);
657
658	return (0);
659}
660
661/*
662 * Add a single contiguous physical range to the segment list.
663 */
664static int
665_bus_dmamap_addseg(bus_dma_tag_t dmat, bus_dmamap_t map, bus_addr_t curaddr,
666		   bus_size_t sgsize, bus_dma_segment_t *segs, int *segp)
667{
668	bus_addr_t baddr, bmask;
669	int seg;
670
671	/*
672	 * Make sure we don't cross any boundaries.
673	 */
674	bmask = ~(dmat->boundary - 1);
675	if (dmat->boundary > 0) {
676		baddr = (curaddr + dmat->boundary) & bmask;
677		if (sgsize > (baddr - curaddr))
678			sgsize = (baddr - curaddr);
679	}
680
681	/*
682	 * Insert chunk into a segment, coalescing with
683	 * previous segment if possible.
684	 */
685	seg = *segp;
686	if (seg == -1) {
687		seg = 0;
688		segs[seg].ds_addr = curaddr;
689		segs[seg].ds_len = sgsize;
690	} else {
691		if (curaddr == segs[seg].ds_addr + segs[seg].ds_len &&
692		    (segs[seg].ds_len + sgsize) <= dmat->maxsegsz &&
693		    (dmat->boundary == 0 ||
694		     (segs[seg].ds_addr & bmask) == (curaddr & bmask)))
695			segs[seg].ds_len += sgsize;
696		else {
697			if (++seg >= dmat->nsegments)
698				return (0);
699			segs[seg].ds_addr = curaddr;
700			segs[seg].ds_len = sgsize;
701		}
702	}
703	*segp = seg;
704	return (sgsize);
705}
706
707/*
708 * Utility function to load a physical buffer.  segp contains
709 * the starting segment on entrace, and the ending segment on exit.
710 */
711int
712_bus_dmamap_load_phys(bus_dma_tag_t dmat,
713		      bus_dmamap_t map,
714		      vm_paddr_t buf, bus_size_t buflen,
715		      int flags,
716		      bus_dma_segment_t *segs,
717		      int *segp)
718{
719	bus_addr_t curaddr;
720	bus_size_t sgsize;
721	int error;
722
723	if (segs == NULL)
724		segs = map->segments;
725
726	if ((dmat->flags & BUS_DMA_COULD_BOUNCE) != 0) {
727		_bus_dmamap_count_phys(dmat, map, buf, buflen, flags);
728		if (map->pagesneeded != 0) {
729			error = _bus_dmamap_reserve_pages(dmat, map, flags);
730			if (error)
731				return (error);
732		}
733	}
734
735	while (buflen > 0) {
736		curaddr = buf;
737		sgsize = MIN(buflen, dmat->maxsegsz);
738		if (map->pagesneeded != 0 && run_filter(dmat, curaddr)) {
739			sgsize = MIN(sgsize, PAGE_SIZE);
740			curaddr = add_bounce_page(dmat, map, 0, curaddr,
741			    sgsize);
742		}
743		sgsize = _bus_dmamap_addseg(dmat, map, curaddr, sgsize, segs,
744		    segp);
745		if (sgsize == 0)
746			break;
747		buf += sgsize;
748		buflen -= sgsize;
749	}
750
751	/*
752	 * Did we fit?
753	 */
754	return (buflen != 0 ? EFBIG : 0); /* XXX better return value here? */
755}
756
757int
758_bus_dmamap_load_ma(bus_dma_tag_t dmat, bus_dmamap_t map,
759    struct vm_page **ma, bus_size_t tlen, int ma_offs, int flags,
760    bus_dma_segment_t *segs, int *segp)
761{
762
763	return (bus_dmamap_load_ma_triv(dmat, map, ma, tlen, ma_offs, flags,
764	    segs, segp));
765}
766
767/*
768 * Utility function to load a linear buffer.  segp contains
769 * the starting segment on entrance, and the ending segment on exit.
770 */
771int
772_bus_dmamap_load_buffer(bus_dma_tag_t dmat,
773    			bus_dmamap_t map,
774			void *buf, bus_size_t buflen,
775			pmap_t pmap,
776			int flags,
777			bus_dma_segment_t *segs,
778			int *segp)
779{
780	bus_size_t sgsize;
781	bus_addr_t curaddr;
782	vm_offset_t vaddr;
783	int error;
784
785	if (segs == NULL)
786		segs = map->segments;
787
788	if ((dmat->flags & BUS_DMA_COULD_BOUNCE) != 0) {
789		_bus_dmamap_count_pages(dmat, map, pmap, buf, buflen, flags);
790		if (map->pagesneeded != 0) {
791			error = _bus_dmamap_reserve_pages(dmat, map, flags);
792			if (error)
793				return (error);
794		}
795	}
796
797	vaddr = (vm_offset_t)buf;
798
799	while (buflen > 0) {
800		bus_size_t max_sgsize;
801
802		/*
803		 * Get the physical address for this segment.
804		 */
805		if (pmap == kernel_pmap)
806			curaddr = pmap_kextract(vaddr);
807		else
808			curaddr = pmap_extract(pmap, vaddr);
809
810		/*
811		 * Compute the segment size, and adjust counts.
812		 */
813		max_sgsize = MIN(buflen, dmat->maxsegsz);
814		sgsize = PAGE_SIZE - ((vm_offset_t)curaddr & PAGE_MASK);
815		if (map->pagesneeded != 0 && run_filter(dmat, curaddr)) {
816			sgsize = roundup2(sgsize, dmat->alignment);
817			sgsize = MIN(sgsize, max_sgsize);
818			curaddr = add_bounce_page(dmat, map, vaddr, curaddr,
819			    sgsize);
820		} else {
821			sgsize = MIN(sgsize, max_sgsize);
822		}
823
824		sgsize = _bus_dmamap_addseg(dmat, map, curaddr, sgsize, segs,
825		    segp);
826		if (sgsize == 0)
827			break;
828		vaddr += sgsize;
829		buflen -= sgsize;
830	}
831
832	/*
833	 * Did we fit?
834	 */
835	return (buflen != 0 ? EFBIG : 0); /* XXX better return value here? */
836}
837
838void
839__bus_dmamap_waitok(bus_dma_tag_t dmat, bus_dmamap_t map,
840		    struct memdesc *mem, bus_dmamap_callback_t *callback,
841		    void *callback_arg)
842{
843
844	if (dmat->flags & BUS_DMA_COULD_BOUNCE) {
845		map->dmat = dmat;
846		map->mem = *mem;
847		map->callback = callback;
848		map->callback_arg = callback_arg;
849	}
850}
851
852bus_dma_segment_t *
853_bus_dmamap_complete(bus_dma_tag_t dmat, bus_dmamap_t map,
854		     bus_dma_segment_t *segs, int nsegs, int error)
855{
856
857	map->nsegs = nsegs;
858	if (segs != NULL)
859		memcpy(map->segments, segs, map->nsegs*sizeof(segs[0]));
860	if (dmat->iommu != NULL)
861		IOMMU_MAP(dmat->iommu, map->segments, &map->nsegs,
862		    dmat->lowaddr, dmat->highaddr, dmat->alignment,
863		    dmat->boundary, dmat->iommu_cookie);
864
865	if (segs != NULL)
866		memcpy(segs, map->segments, map->nsegs*sizeof(segs[0]));
867	else
868		segs = map->segments;
869
870	return (segs);
871}
872
873/*
874 * Release the mapping held by map.
875 */
876void
877_bus_dmamap_unload(bus_dma_tag_t dmat, bus_dmamap_t map)
878{
879	struct bounce_page *bpage;
880
881	if (dmat->iommu) {
882		IOMMU_UNMAP(dmat->iommu, map->segments, map->nsegs, dmat->iommu_cookie);
883		map->nsegs = 0;
884	}
885
886	while ((bpage = STAILQ_FIRST(&map->bpages)) != NULL) {
887		STAILQ_REMOVE_HEAD(&map->bpages, links);
888		free_bounce_page(dmat, bpage);
889	}
890}
891
892void
893_bus_dmamap_sync(bus_dma_tag_t dmat, bus_dmamap_t map, bus_dmasync_op_t op)
894{
895	struct bounce_page *bpage;
896
897	if ((bpage = STAILQ_FIRST(&map->bpages)) != NULL) {
898		/*
899		 * Handle data bouncing.  We might also
900		 * want to add support for invalidating
901		 * the caches on broken hardware
902		 */
903		CTR4(KTR_BUSDMA, "%s: tag %p tag flags 0x%x op 0x%x "
904		    "performing bounce", __func__, dmat, dmat->flags, op);
905
906		if (op & BUS_DMASYNC_PREWRITE) {
907			while (bpage != NULL) {
908				if (bpage->datavaddr != 0)
909					bcopy((void *)bpage->datavaddr,
910					      (void *)bpage->vaddr,
911					      bpage->datacount);
912				else
913					physcopyout(bpage->dataaddr,
914					    (void *)bpage->vaddr,
915					    bpage->datacount);
916				bpage = STAILQ_NEXT(bpage, links);
917			}
918			dmat->bounce_zone->total_bounced++;
919		}
920
921		if (op & BUS_DMASYNC_POSTREAD) {
922			while (bpage != NULL) {
923				if (bpage->datavaddr != 0)
924					bcopy((void *)bpage->vaddr,
925					      (void *)bpage->datavaddr,
926					      bpage->datacount);
927				else
928					physcopyin((void *)bpage->vaddr,
929					    bpage->dataaddr, bpage->datacount);
930				bpage = STAILQ_NEXT(bpage, links);
931			}
932			dmat->bounce_zone->total_bounced++;
933		}
934	}
935
936	powerpc_sync();
937}
938
939static void
940init_bounce_pages(void *dummy __unused)
941{
942
943	total_bpages = 0;
944	STAILQ_INIT(&bounce_zone_list);
945	STAILQ_INIT(&bounce_map_waitinglist);
946	STAILQ_INIT(&bounce_map_callbacklist);
947	mtx_init(&bounce_lock, "bounce pages lock", NULL, MTX_DEF);
948}
949SYSINIT(bpages, SI_SUB_LOCK, SI_ORDER_ANY, init_bounce_pages, NULL);
950
951static struct sysctl_ctx_list *
952busdma_sysctl_tree(struct bounce_zone *bz)
953{
954	return (&bz->sysctl_tree);
955}
956
957static struct sysctl_oid *
958busdma_sysctl_tree_top(struct bounce_zone *bz)
959{
960	return (bz->sysctl_tree_top);
961}
962
963static int
964alloc_bounce_zone(bus_dma_tag_t dmat)
965{
966	struct bounce_zone *bz;
967
968	/* Check to see if we already have a suitable zone */
969	STAILQ_FOREACH(bz, &bounce_zone_list, links) {
970		if ((dmat->alignment <= bz->alignment)
971		 && (dmat->lowaddr >= bz->lowaddr)) {
972			dmat->bounce_zone = bz;
973			return (0);
974		}
975	}
976
977	if ((bz = (struct bounce_zone *)malloc(sizeof(*bz), M_DEVBUF,
978	    M_NOWAIT | M_ZERO)) == NULL)
979		return (ENOMEM);
980
981	STAILQ_INIT(&bz->bounce_page_list);
982	bz->free_bpages = 0;
983	bz->reserved_bpages = 0;
984	bz->active_bpages = 0;
985	bz->lowaddr = dmat->lowaddr;
986	bz->alignment = MAX(dmat->alignment, PAGE_SIZE);
987	bz->map_count = 0;
988	snprintf(bz->zoneid, 8, "zone%d", busdma_zonecount);
989	busdma_zonecount++;
990	snprintf(bz->lowaddrid, 18, "%#jx", (uintmax_t)bz->lowaddr);
991	STAILQ_INSERT_TAIL(&bounce_zone_list, bz, links);
992	dmat->bounce_zone = bz;
993
994	sysctl_ctx_init(&bz->sysctl_tree);
995	bz->sysctl_tree_top = SYSCTL_ADD_NODE(&bz->sysctl_tree,
996	    SYSCTL_STATIC_CHILDREN(_hw_busdma), OID_AUTO, bz->zoneid,
997	    CTLFLAG_RD, 0, "");
998	if (bz->sysctl_tree_top == NULL) {
999		sysctl_ctx_free(&bz->sysctl_tree);
1000		return (0);	/* XXX error code? */
1001	}
1002
1003	SYSCTL_ADD_INT(busdma_sysctl_tree(bz),
1004	    SYSCTL_CHILDREN(busdma_sysctl_tree_top(bz)), OID_AUTO,
1005	    "total_bpages", CTLFLAG_RD, &bz->total_bpages, 0,
1006	    "Total bounce pages");
1007	SYSCTL_ADD_INT(busdma_sysctl_tree(bz),
1008	    SYSCTL_CHILDREN(busdma_sysctl_tree_top(bz)), OID_AUTO,
1009	    "free_bpages", CTLFLAG_RD, &bz->free_bpages, 0,
1010	    "Free bounce pages");
1011	SYSCTL_ADD_INT(busdma_sysctl_tree(bz),
1012	    SYSCTL_CHILDREN(busdma_sysctl_tree_top(bz)), OID_AUTO,
1013	    "reserved_bpages", CTLFLAG_RD, &bz->reserved_bpages, 0,
1014	    "Reserved bounce pages");
1015	SYSCTL_ADD_INT(busdma_sysctl_tree(bz),
1016	    SYSCTL_CHILDREN(busdma_sysctl_tree_top(bz)), OID_AUTO,
1017	    "active_bpages", CTLFLAG_RD, &bz->active_bpages, 0,
1018	    "Active bounce pages");
1019	SYSCTL_ADD_INT(busdma_sysctl_tree(bz),
1020	    SYSCTL_CHILDREN(busdma_sysctl_tree_top(bz)), OID_AUTO,
1021	    "total_bounced", CTLFLAG_RD, &bz->total_bounced, 0,
1022	    "Total bounce requests");
1023	SYSCTL_ADD_INT(busdma_sysctl_tree(bz),
1024	    SYSCTL_CHILDREN(busdma_sysctl_tree_top(bz)), OID_AUTO,
1025	    "total_deferred", CTLFLAG_RD, &bz->total_deferred, 0,
1026	    "Total bounce requests that were deferred");
1027	SYSCTL_ADD_STRING(busdma_sysctl_tree(bz),
1028	    SYSCTL_CHILDREN(busdma_sysctl_tree_top(bz)), OID_AUTO,
1029	    "lowaddr", CTLFLAG_RD, bz->lowaddrid, 0, "");
1030	SYSCTL_ADD_UAUTO(busdma_sysctl_tree(bz),
1031	    SYSCTL_CHILDREN(busdma_sysctl_tree_top(bz)), OID_AUTO,
1032	    "alignment", CTLFLAG_RD, &bz->alignment, "");
1033
1034	return (0);
1035}
1036
1037static int
1038alloc_bounce_pages(bus_dma_tag_t dmat, u_int numpages)
1039{
1040	struct bounce_zone *bz;
1041	int count;
1042
1043	bz = dmat->bounce_zone;
1044	count = 0;
1045	while (numpages > 0) {
1046		struct bounce_page *bpage;
1047
1048		bpage = (struct bounce_page *)malloc(sizeof(*bpage), M_DEVBUF,
1049						     M_NOWAIT | M_ZERO);
1050
1051		if (bpage == NULL)
1052			break;
1053		bpage->vaddr = (vm_offset_t)contigmalloc(PAGE_SIZE, M_DEVBUF,
1054							 M_NOWAIT, 0ul,
1055							 bz->lowaddr,
1056							 PAGE_SIZE,
1057							 0);
1058		if (bpage->vaddr == 0) {
1059			free(bpage, M_DEVBUF);
1060			break;
1061		}
1062		bpage->busaddr = pmap_kextract(bpage->vaddr);
1063		mtx_lock(&bounce_lock);
1064		STAILQ_INSERT_TAIL(&bz->bounce_page_list, bpage, links);
1065		total_bpages++;
1066		bz->total_bpages++;
1067		bz->free_bpages++;
1068		mtx_unlock(&bounce_lock);
1069		count++;
1070		numpages--;
1071	}
1072	return (count);
1073}
1074
1075static int
1076reserve_bounce_pages(bus_dma_tag_t dmat, bus_dmamap_t map, int commit)
1077{
1078	struct bounce_zone *bz;
1079	int pages;
1080
1081	mtx_assert(&bounce_lock, MA_OWNED);
1082	bz = dmat->bounce_zone;
1083	pages = MIN(bz->free_bpages, map->pagesneeded - map->pagesreserved);
1084	if (commit == 0 && map->pagesneeded > (map->pagesreserved + pages))
1085		return (map->pagesneeded - (map->pagesreserved + pages));
1086	bz->free_bpages -= pages;
1087	bz->reserved_bpages += pages;
1088	map->pagesreserved += pages;
1089	pages = map->pagesneeded - map->pagesreserved;
1090
1091	return (pages);
1092}
1093
1094static bus_addr_t
1095add_bounce_page(bus_dma_tag_t dmat, bus_dmamap_t map, vm_offset_t vaddr,
1096		bus_addr_t addr, bus_size_t size)
1097{
1098	struct bounce_zone *bz;
1099	struct bounce_page *bpage;
1100
1101	KASSERT(dmat->bounce_zone != NULL, ("no bounce zone in dma tag"));
1102
1103	bz = dmat->bounce_zone;
1104	if (map->pagesneeded == 0)
1105		panic("add_bounce_page: map doesn't need any pages");
1106	map->pagesneeded--;
1107
1108	if (map->pagesreserved == 0)
1109		panic("add_bounce_page: map doesn't need any pages");
1110	map->pagesreserved--;
1111
1112	mtx_lock(&bounce_lock);
1113	bpage = STAILQ_FIRST(&bz->bounce_page_list);
1114	if (bpage == NULL)
1115		panic("add_bounce_page: free page list is empty");
1116
1117	STAILQ_REMOVE_HEAD(&bz->bounce_page_list, links);
1118	bz->reserved_bpages--;
1119	bz->active_bpages++;
1120	mtx_unlock(&bounce_lock);
1121
1122	if (dmat->flags & BUS_DMA_KEEP_PG_OFFSET) {
1123		/* Page offset needs to be preserved. */
1124		bpage->vaddr |= addr & PAGE_MASK;
1125		bpage->busaddr |= addr & PAGE_MASK;
1126	}
1127	bpage->datavaddr = vaddr;
1128	bpage->dataaddr = addr;
1129	bpage->datacount = size;
1130	STAILQ_INSERT_TAIL(&(map->bpages), bpage, links);
1131	return (bpage->busaddr);
1132}
1133
1134static void
1135free_bounce_page(bus_dma_tag_t dmat, struct bounce_page *bpage)
1136{
1137	struct bus_dmamap *map;
1138	struct bounce_zone *bz;
1139
1140	bz = dmat->bounce_zone;
1141	bpage->datavaddr = 0;
1142	bpage->datacount = 0;
1143	if (dmat->flags & BUS_DMA_KEEP_PG_OFFSET) {
1144		/*
1145		 * Reset the bounce page to start at offset 0.  Other uses
1146		 * of this bounce page may need to store a full page of
1147		 * data and/or assume it starts on a page boundary.
1148		 */
1149		bpage->vaddr &= ~PAGE_MASK;
1150		bpage->busaddr &= ~PAGE_MASK;
1151	}
1152
1153	mtx_lock(&bounce_lock);
1154	STAILQ_INSERT_HEAD(&bz->bounce_page_list, bpage, links);
1155	bz->free_bpages++;
1156	bz->active_bpages--;
1157	if ((map = STAILQ_FIRST(&bounce_map_waitinglist)) != NULL) {
1158		if (reserve_bounce_pages(map->dmat, map, 1) == 0) {
1159			STAILQ_REMOVE_HEAD(&bounce_map_waitinglist, links);
1160			STAILQ_INSERT_TAIL(&bounce_map_callbacklist,
1161					   map, links);
1162			busdma_swi_pending = 1;
1163			bz->total_deferred++;
1164			swi_sched(vm_ih, 0);
1165		}
1166	}
1167	mtx_unlock(&bounce_lock);
1168}
1169
1170void
1171busdma_swi(void)
1172{
1173	bus_dma_tag_t dmat;
1174	struct bus_dmamap *map;
1175
1176	mtx_lock(&bounce_lock);
1177	while ((map = STAILQ_FIRST(&bounce_map_callbacklist)) != NULL) {
1178		STAILQ_REMOVE_HEAD(&bounce_map_callbacklist, links);
1179		mtx_unlock(&bounce_lock);
1180		dmat = map->dmat;
1181		(dmat->lockfunc)(dmat->lockfuncarg, BUS_DMA_LOCK);
1182		bus_dmamap_load_mem(map->dmat, map, &map->mem,
1183				    map->callback, map->callback_arg,
1184				    BUS_DMA_WAITOK);
1185		(dmat->lockfunc)(dmat->lockfuncarg, BUS_DMA_UNLOCK);
1186		mtx_lock(&bounce_lock);
1187	}
1188	mtx_unlock(&bounce_lock);
1189}
1190
1191int
1192bus_dma_tag_set_iommu(bus_dma_tag_t tag, struct device *iommu, void *cookie)
1193{
1194	tag->iommu = iommu;
1195	tag->iommu_cookie = cookie;
1196
1197	return (0);
1198}
1199
1200