iommu sg merging: x86: make pci-gart iommu respect the segment size limits
[linux-2.6-block.git] / arch / powerpc / kernel / iommu.c
CommitLineData
1da177e4 1/*
1da177e4
LT
2 * Copyright (C) 2001 Mike Corrigan & Dave Engebretsen, IBM Corporation
3 *
4 * Rewrite, cleanup, new allocation schemes, virtual merging:
5 * Copyright (C) 2004 Olof Johansson, IBM Corporation
6 * and Ben. Herrenschmidt, IBM Corporation
7 *
8 * Dynamic DMA mapping support, bus-independent parts.
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 */
24
25
1da177e4
LT
26#include <linux/init.h>
27#include <linux/types.h>
28#include <linux/slab.h>
29#include <linux/mm.h>
30#include <linux/spinlock.h>
31#include <linux/string.h>
32#include <linux/dma-mapping.h>
1da177e4
LT
33#include <linux/bitops.h>
34#include <asm/io.h>
35#include <asm/prom.h>
36#include <asm/iommu.h>
37#include <asm/pci-bridge.h>
38#include <asm/machdep.h>
5f50867b 39#include <asm/kdump.h>
1da177e4
LT
40
41#define DBG(...)
42
43#ifdef CONFIG_IOMMU_VMERGE
44static int novmerge = 0;
45#else
46static int novmerge = 1;
47#endif
48
56997559
JM
49static int protect4gb = 1;
50
5d2efba6
LV
51static inline unsigned long iommu_num_pages(unsigned long vaddr,
52 unsigned long slen)
53{
54 unsigned long npages;
55
56 npages = IOMMU_PAGE_ALIGN(vaddr + slen) - (vaddr & IOMMU_PAGE_MASK);
57 npages >>= IOMMU_PAGE_SHIFT;
58
59 return npages;
60}
61
56997559
JM
62static int __init setup_protect4gb(char *str)
63{
64 if (strcmp(str, "on") == 0)
65 protect4gb = 1;
66 else if (strcmp(str, "off") == 0)
67 protect4gb = 0;
68
69 return 1;
70}
71
1da177e4
LT
72static int __init setup_iommu(char *str)
73{
74 if (!strcmp(str, "novmerge"))
75 novmerge = 1;
76 else if (!strcmp(str, "vmerge"))
77 novmerge = 0;
78 return 1;
79}
80
56997559 81__setup("protect4gb=", setup_protect4gb);
1da177e4
LT
82__setup("iommu=", setup_iommu);
83
84static unsigned long iommu_range_alloc(struct iommu_table *tbl,
85 unsigned long npages,
86 unsigned long *handle,
7daa411b 87 unsigned long mask,
1da177e4
LT
88 unsigned int align_order)
89{
90 unsigned long n, end, i, start;
91 unsigned long limit;
92 int largealloc = npages > 15;
93 int pass = 0;
94 unsigned long align_mask;
95
96 align_mask = 0xffffffffffffffffl >> (64 - align_order);
97
98 /* This allocator was derived from x86_64's bit string search */
99
100 /* Sanity check */
13a2eea1 101 if (unlikely(npages == 0)) {
1da177e4
LT
102 if (printk_ratelimit())
103 WARN_ON(1);
104 return DMA_ERROR_CODE;
105 }
106
107 if (handle && *handle)
108 start = *handle;
109 else
110 start = largealloc ? tbl->it_largehint : tbl->it_hint;
111
112 /* Use only half of the table for small allocs (15 pages or less) */
113 limit = largealloc ? tbl->it_size : tbl->it_halfpoint;
114
115 if (largealloc && start < tbl->it_halfpoint)
116 start = tbl->it_halfpoint;
117
118 /* The case below can happen if we have a small segment appended
119 * to a large, or when the previous alloc was at the very end of
120 * the available space. If so, go back to the initial start.
121 */
122 if (start >= limit)
123 start = largealloc ? tbl->it_largehint : tbl->it_hint;
7daa411b 124
1da177e4
LT
125 again:
126
7daa411b
OJ
127 if (limit + tbl->it_offset > mask) {
128 limit = mask - tbl->it_offset + 1;
129 /* If we're constrained on address range, first try
130 * at the masked hint to avoid O(n) search complexity,
131 * but on second pass, start at 0.
132 */
133 if ((start & mask) >= limit || pass > 0)
134 start = 0;
135 else
136 start &= mask;
137 }
138
1da177e4
LT
139 n = find_next_zero_bit(tbl->it_map, limit, start);
140
141 /* Align allocation */
142 n = (n + align_mask) & ~align_mask;
143
144 end = n + npages;
145
146 if (unlikely(end >= limit)) {
147 if (likely(pass < 2)) {
148 /* First failure, just rescan the half of the table.
149 * Second failure, rescan the other half of the table.
150 */
151 start = (largealloc ^ pass) ? tbl->it_halfpoint : 0;
152 limit = pass ? tbl->it_size : limit;
153 pass++;
154 goto again;
155 } else {
156 /* Third failure, give up */
157 return DMA_ERROR_CODE;
158 }
159 }
160
161 for (i = n; i < end; i++)
162 if (test_bit(i, tbl->it_map)) {
163 start = i+1;
164 goto again;
165 }
166
167 for (i = n; i < end; i++)
168 __set_bit(i, tbl->it_map);
169
170 /* Bump the hint to a new block for small allocs. */
171 if (largealloc) {
172 /* Don't bump to new block to avoid fragmentation */
173 tbl->it_largehint = end;
174 } else {
175 /* Overflow will be taken care of at the next allocation */
176 tbl->it_hint = (end + tbl->it_blocksize - 1) &
177 ~(tbl->it_blocksize - 1);
178 }
179
180 /* Update handle for SG allocations */
181 if (handle)
182 *handle = end;
183
184 return n;
185}
186
187static dma_addr_t iommu_alloc(struct iommu_table *tbl, void *page,
188 unsigned int npages, enum dma_data_direction direction,
7daa411b 189 unsigned long mask, unsigned int align_order)
1da177e4
LT
190{
191 unsigned long entry, flags;
192 dma_addr_t ret = DMA_ERROR_CODE;
7daa411b 193
1da177e4
LT
194 spin_lock_irqsave(&(tbl->it_lock), flags);
195
7daa411b 196 entry = iommu_range_alloc(tbl, npages, NULL, mask, align_order);
1da177e4
LT
197
198 if (unlikely(entry == DMA_ERROR_CODE)) {
199 spin_unlock_irqrestore(&(tbl->it_lock), flags);
200 return DMA_ERROR_CODE;
201 }
202
203 entry += tbl->it_offset; /* Offset into real TCE table */
5d2efba6 204 ret = entry << IOMMU_PAGE_SHIFT; /* Set the return dma address */
1da177e4
LT
205
206 /* Put the TCEs in the HW table */
5d2efba6 207 ppc_md.tce_build(tbl, entry, npages, (unsigned long)page & IOMMU_PAGE_MASK,
1da177e4
LT
208 direction);
209
210
211 /* Flush/invalidate TLB caches if necessary */
212 if (ppc_md.tce_flush)
213 ppc_md.tce_flush(tbl);
214
215 spin_unlock_irqrestore(&(tbl->it_lock), flags);
216
217 /* Make sure updates are seen by hardware */
218 mb();
219
220 return ret;
221}
222
223static void __iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
224 unsigned int npages)
225{
226 unsigned long entry, free_entry;
227 unsigned long i;
228
5d2efba6 229 entry = dma_addr >> IOMMU_PAGE_SHIFT;
1da177e4
LT
230 free_entry = entry - tbl->it_offset;
231
232 if (((free_entry + npages) > tbl->it_size) ||
233 (entry < tbl->it_offset)) {
234 if (printk_ratelimit()) {
235 printk(KERN_INFO "iommu_free: invalid entry\n");
236 printk(KERN_INFO "\tentry = 0x%lx\n", entry);
237 printk(KERN_INFO "\tdma_addr = 0x%lx\n", (u64)dma_addr);
238 printk(KERN_INFO "\tTable = 0x%lx\n", (u64)tbl);
239 printk(KERN_INFO "\tbus# = 0x%lx\n", (u64)tbl->it_busno);
240 printk(KERN_INFO "\tsize = 0x%lx\n", (u64)tbl->it_size);
241 printk(KERN_INFO "\tstartOff = 0x%lx\n", (u64)tbl->it_offset);
242 printk(KERN_INFO "\tindex = 0x%lx\n", (u64)tbl->it_index);
243 WARN_ON(1);
244 }
245 return;
246 }
247
248 ppc_md.tce_free(tbl, entry, npages);
249
250 for (i = 0; i < npages; i++)
251 __clear_bit(free_entry+i, tbl->it_map);
252}
253
254static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
255 unsigned int npages)
256{
257 unsigned long flags;
258
259 spin_lock_irqsave(&(tbl->it_lock), flags);
260
261 __iommu_free(tbl, dma_addr, npages);
262
263 /* Make sure TLB cache is flushed if the HW needs it. We do
264 * not do an mb() here on purpose, it is not needed on any of
265 * the current platforms.
266 */
267 if (ppc_md.tce_flush)
268 ppc_md.tce_flush(tbl);
269
270 spin_unlock_irqrestore(&(tbl->it_lock), flags);
271}
272
12d04eef
BH
273int iommu_map_sg(struct iommu_table *tbl, struct scatterlist *sglist,
274 int nelems, unsigned long mask,
275 enum dma_data_direction direction)
1da177e4
LT
276{
277 dma_addr_t dma_next = 0, dma_addr;
278 unsigned long flags;
279 struct scatterlist *s, *outs, *segstart;
78bdc310 280 int outcount, incount, i;
d262c32a 281 unsigned int align;
1da177e4
LT
282 unsigned long handle;
283
284 BUG_ON(direction == DMA_NONE);
285
286 if ((nelems == 0) || !tbl)
287 return 0;
288
289 outs = s = segstart = &sglist[0];
290 outcount = 1;
ac9af7cb 291 incount = nelems;
1da177e4
LT
292 handle = 0;
293
294 /* Init first segment length for backout at failure */
295 outs->dma_length = 0;
296
5d2efba6 297 DBG("sg mapping %d elements:\n", nelems);
1da177e4
LT
298
299 spin_lock_irqsave(&(tbl->it_lock), flags);
300
78bdc310 301 for_each_sg(sglist, s, nelems, i) {
1da177e4
LT
302 unsigned long vaddr, npages, entry, slen;
303
304 slen = s->length;
305 /* Sanity check */
306 if (slen == 0) {
307 dma_next = 0;
308 continue;
309 }
310 /* Allocate iommu entries for that segment */
58b053e4 311 vaddr = (unsigned long) sg_virt(s);
5d2efba6 312 npages = iommu_num_pages(vaddr, slen);
d262c32a
BH
313 align = 0;
314 if (IOMMU_PAGE_SHIFT < PAGE_SHIFT && slen >= PAGE_SIZE &&
315 (vaddr & ~PAGE_MASK) == 0)
316 align = PAGE_SHIFT - IOMMU_PAGE_SHIFT;
317 entry = iommu_range_alloc(tbl, npages, &handle,
318 mask >> IOMMU_PAGE_SHIFT, align);
1da177e4
LT
319
320 DBG(" - vaddr: %lx, size: %lx\n", vaddr, slen);
321
322 /* Handle failure */
323 if (unlikely(entry == DMA_ERROR_CODE)) {
324 if (printk_ratelimit())
325 printk(KERN_INFO "iommu_alloc failed, tbl %p vaddr %lx"
326 " npages %lx\n", tbl, vaddr, npages);
327 goto failure;
328 }
329
330 /* Convert entry to a dma_addr_t */
331 entry += tbl->it_offset;
5d2efba6
LV
332 dma_addr = entry << IOMMU_PAGE_SHIFT;
333 dma_addr |= (s->offset & ~IOMMU_PAGE_MASK);
1da177e4 334
5d2efba6 335 DBG(" - %lu pages, entry: %lx, dma_addr: %lx\n",
1da177e4
LT
336 npages, entry, dma_addr);
337
338 /* Insert into HW table */
5d2efba6 339 ppc_md.tce_build(tbl, entry, npages, vaddr & IOMMU_PAGE_MASK, direction);
1da177e4
LT
340
341 /* If we are in an open segment, try merging */
342 if (segstart != s) {
343 DBG(" - trying merge...\n");
344 /* We cannot merge if:
345 * - allocated dma_addr isn't contiguous to previous allocation
346 */
347 if (novmerge || (dma_addr != dma_next)) {
348 /* Can't merge: create a new segment */
349 segstart = s;
78bdc310
JA
350 outcount++;
351 outs = sg_next(outs);
1da177e4
LT
352 DBG(" can't merge, new segment.\n");
353 } else {
354 outs->dma_length += s->length;
5d2efba6 355 DBG(" merged, new len: %ux\n", outs->dma_length);
1da177e4
LT
356 }
357 }
358
359 if (segstart == s) {
360 /* This is a new segment, fill entries */
361 DBG(" - filling new segment.\n");
362 outs->dma_address = dma_addr;
363 outs->dma_length = slen;
364 }
365
366 /* Calculate next page pointer for contiguous check */
367 dma_next = dma_addr + slen;
368
369 DBG(" - dma next is: %lx\n", dma_next);
370 }
371
372 /* Flush/invalidate TLB caches if necessary */
373 if (ppc_md.tce_flush)
374 ppc_md.tce_flush(tbl);
375
376 spin_unlock_irqrestore(&(tbl->it_lock), flags);
377
1da177e4
LT
378 DBG("mapped %d elements:\n", outcount);
379
ac9af7cb 380 /* For the sake of iommu_unmap_sg, we clear out the length in the
1da177e4
LT
381 * next entry of the sglist if we didn't fill the list completely
382 */
ac9af7cb 383 if (outcount < incount) {
78bdc310 384 outs = sg_next(outs);
1da177e4
LT
385 outs->dma_address = DMA_ERROR_CODE;
386 outs->dma_length = 0;
387 }
a958a264
JM
388
389 /* Make sure updates are seen by hardware */
390 mb();
391
1da177e4
LT
392 return outcount;
393
394 failure:
78bdc310 395 for_each_sg(sglist, s, nelems, i) {
1da177e4
LT
396 if (s->dma_length != 0) {
397 unsigned long vaddr, npages;
398
5d2efba6
LV
399 vaddr = s->dma_address & IOMMU_PAGE_MASK;
400 npages = iommu_num_pages(s->dma_address, s->dma_length);
1da177e4 401 __iommu_free(tbl, vaddr, npages);
a958a264
JM
402 s->dma_address = DMA_ERROR_CODE;
403 s->dma_length = 0;
1da177e4 404 }
78bdc310
JA
405 if (s == outs)
406 break;
1da177e4
LT
407 }
408 spin_unlock_irqrestore(&(tbl->it_lock), flags);
409 return 0;
410}
411
412
413void iommu_unmap_sg(struct iommu_table *tbl, struct scatterlist *sglist,
414 int nelems, enum dma_data_direction direction)
415{
78bdc310 416 struct scatterlist *sg;
1da177e4
LT
417 unsigned long flags;
418
419 BUG_ON(direction == DMA_NONE);
420
421 if (!tbl)
422 return;
423
424 spin_lock_irqsave(&(tbl->it_lock), flags);
425
78bdc310 426 sg = sglist;
1da177e4
LT
427 while (nelems--) {
428 unsigned int npages;
78bdc310 429 dma_addr_t dma_handle = sg->dma_address;
1da177e4 430
78bdc310 431 if (sg->dma_length == 0)
1da177e4 432 break;
78bdc310 433 npages = iommu_num_pages(dma_handle, sg->dma_length);
1da177e4 434 __iommu_free(tbl, dma_handle, npages);
78bdc310 435 sg = sg_next(sg);
1da177e4
LT
436 }
437
438 /* Flush/invalidate TLBs if necessary. As for iommu_free(), we
439 * do not do an mb() here, the affected platforms do not need it
440 * when freeing.
441 */
442 if (ppc_md.tce_flush)
443 ppc_md.tce_flush(tbl);
444
445 spin_unlock_irqrestore(&(tbl->it_lock), flags);
446}
447
448/*
449 * Build a iommu_table structure. This contains a bit map which
450 * is used to manage allocation of the tce space.
451 */
ca1588e7 452struct iommu_table *iommu_init_table(struct iommu_table *tbl, int nid)
1da177e4
LT
453{
454 unsigned long sz;
56997559
JM
455 unsigned long start_index, end_index;
456 unsigned long entries_per_4g;
457 unsigned long index;
1da177e4 458 static int welcomed = 0;
ca1588e7 459 struct page *page;
1da177e4
LT
460
461 /* Set aside 1/4 of the table for large allocations. */
462 tbl->it_halfpoint = tbl->it_size * 3 / 4;
463
464 /* number of bytes needed for the bitmap */
465 sz = (tbl->it_size + 7) >> 3;
466
ca1588e7
AB
467 page = alloc_pages_node(nid, GFP_ATOMIC, get_order(sz));
468 if (!page)
1da177e4 469 panic("iommu_init_table: Can't allocate %ld bytes\n", sz);
ca1588e7 470 tbl->it_map = page_address(page);
1da177e4
LT
471 memset(tbl->it_map, 0, sz);
472
473 tbl->it_hint = 0;
474 tbl->it_largehint = tbl->it_halfpoint;
475 spin_lock_init(&tbl->it_lock);
476
5f50867b
HM
477#ifdef CONFIG_CRASH_DUMP
478 if (ppc_md.tce_get) {
56997559 479 unsigned long tceval;
5f50867b
HM
480 unsigned long tcecount = 0;
481
482 /*
483 * Reserve the existing mappings left by the first kernel.
484 */
485 for (index = 0; index < tbl->it_size; index++) {
486 tceval = ppc_md.tce_get(tbl, index + tbl->it_offset);
487 /*
488 * Freed TCE entry contains 0x7fffffffffffffff on JS20
489 */
490 if (tceval && (tceval != 0x7fffffffffffffffUL)) {
491 __set_bit(index, tbl->it_map);
492 tcecount++;
493 }
494 }
495 if ((tbl->it_size - tcecount) < KDUMP_MIN_TCE_ENTRIES) {
496 printk(KERN_WARNING "TCE table is full; ");
497 printk(KERN_WARNING "freeing %d entries for the kdump boot\n",
498 KDUMP_MIN_TCE_ENTRIES);
499 for (index = tbl->it_size - KDUMP_MIN_TCE_ENTRIES;
500 index < tbl->it_size; index++)
501 __clear_bit(index, tbl->it_map);
502 }
503 }
504#else
d3588ba9
JR
505 /* Clear the hardware table in case firmware left allocations in it */
506 ppc_md.tce_free(tbl, tbl->it_offset, tbl->it_size);
5f50867b 507#endif
d3588ba9 508
56997559
JM
509 /*
510 * DMA cannot cross 4 GB boundary. Mark last entry of each 4
511 * GB chunk as reserved.
512 */
513 if (protect4gb) {
514 entries_per_4g = 0x100000000l >> IOMMU_PAGE_SHIFT;
515
516 /* Mark the last bit before a 4GB boundary as used */
517 start_index = tbl->it_offset | (entries_per_4g - 1);
518 start_index -= tbl->it_offset;
519
520 end_index = tbl->it_size;
521
522 for (index = start_index; index < end_index - 1; index += entries_per_4g)
523 __set_bit(index, tbl->it_map);
524 }
525
1da177e4
LT
526 if (!welcomed) {
527 printk(KERN_INFO "IOMMU table initialized, virtual merging %s\n",
528 novmerge ? "disabled" : "enabled");
529 welcomed = 1;
530 }
531
532 return tbl;
533}
534
68d315f5 535void iommu_free_table(struct iommu_table *tbl, const char *node_name)
1da177e4 536{
1da177e4
LT
537 unsigned long bitmap_sz, i;
538 unsigned int order;
539
540 if (!tbl || !tbl->it_map) {
541 printk(KERN_ERR "%s: expected TCE map for %s\n", __FUNCTION__,
68d315f5 542 node_name);
1da177e4
LT
543 return;
544 }
545
546 /* verify that table contains no entries */
547 /* it_size is in entries, and we're examining 64 at a time */
548 for (i = 0; i < (tbl->it_size/64); i++) {
549 if (tbl->it_map[i] != 0) {
550 printk(KERN_WARNING "%s: Unexpected TCEs for %s\n",
68d315f5 551 __FUNCTION__, node_name);
1da177e4
LT
552 break;
553 }
554 }
555
556 /* calculate bitmap size in bytes */
557 bitmap_sz = (tbl->it_size + 7) / 8;
558
559 /* free bitmap */
560 order = get_order(bitmap_sz);
561 free_pages((unsigned long) tbl->it_map, order);
562
563 /* free table */
564 kfree(tbl);
565}
566
567/* Creates TCEs for a user provided buffer. The user buffer must be
568 * contiguous real kernel storage (not vmalloc). The address of the buffer
569 * passed here is the kernel (virtual) address of the buffer. The buffer
570 * need not be page aligned, the dma_addr_t returned will point to the same
571 * byte within the page as vaddr.
572 */
573dma_addr_t iommu_map_single(struct iommu_table *tbl, void *vaddr,
7daa411b
OJ
574 size_t size, unsigned long mask,
575 enum dma_data_direction direction)
1da177e4
LT
576{
577 dma_addr_t dma_handle = DMA_ERROR_CODE;
578 unsigned long uaddr;
d262c32a 579 unsigned int npages, align;
1da177e4
LT
580
581 BUG_ON(direction == DMA_NONE);
582
583 uaddr = (unsigned long)vaddr;
5d2efba6 584 npages = iommu_num_pages(uaddr, size);
1da177e4
LT
585
586 if (tbl) {
d262c32a
BH
587 align = 0;
588 if (IOMMU_PAGE_SHIFT < PAGE_SHIFT && size >= PAGE_SIZE &&
589 ((unsigned long)vaddr & ~PAGE_MASK) == 0)
590 align = PAGE_SHIFT - IOMMU_PAGE_SHIFT;
591
7daa411b 592 dma_handle = iommu_alloc(tbl, vaddr, npages, direction,
d262c32a 593 mask >> IOMMU_PAGE_SHIFT, align);
1da177e4
LT
594 if (dma_handle == DMA_ERROR_CODE) {
595 if (printk_ratelimit()) {
596 printk(KERN_INFO "iommu_alloc failed, "
597 "tbl %p vaddr %p npages %d\n",
598 tbl, vaddr, npages);
599 }
600 } else
5d2efba6 601 dma_handle |= (uaddr & ~IOMMU_PAGE_MASK);
1da177e4
LT
602 }
603
604 return dma_handle;
605}
606
607void iommu_unmap_single(struct iommu_table *tbl, dma_addr_t dma_handle,
608 size_t size, enum dma_data_direction direction)
609{
5d2efba6
LV
610 unsigned int npages;
611
1da177e4
LT
612 BUG_ON(direction == DMA_NONE);
613
5d2efba6
LV
614 if (tbl) {
615 npages = iommu_num_pages(dma_handle, size);
616 iommu_free(tbl, dma_handle, npages);
617 }
1da177e4
LT
618}
619
620/* Allocates a contiguous real buffer and creates mappings over it.
621 * Returns the virtual address of the buffer and sets dma_handle
622 * to the dma address (mapping) of the first page.
623 */
624void *iommu_alloc_coherent(struct iommu_table *tbl, size_t size,
8eb6c6e3 625 dma_addr_t *dma_handle, unsigned long mask, gfp_t flag, int node)
1da177e4
LT
626{
627 void *ret = NULL;
628 dma_addr_t mapping;
5d2efba6
LV
629 unsigned int order;
630 unsigned int nio_pages, io_order;
8eb6c6e3 631 struct page *page;
1da177e4
LT
632
633 size = PAGE_ALIGN(size);
1da177e4
LT
634 order = get_order(size);
635
636 /*
637 * Client asked for way too much space. This is checked later
638 * anyway. It is easier to debug here for the drivers than in
639 * the tce tables.
640 */
641 if (order >= IOMAP_MAX_ORDER) {
642 printk("iommu_alloc_consistent size too large: 0x%lx\n", size);
643 return NULL;
644 }
645
646 if (!tbl)
647 return NULL;
648
649 /* Alloc enough pages (and possibly more) */
05061354 650 page = alloc_pages_node(node, flag, order);
8eb6c6e3 651 if (!page)
1da177e4 652 return NULL;
8eb6c6e3 653 ret = page_address(page);
1da177e4
LT
654 memset(ret, 0, size);
655
656 /* Set up tces to cover the allocated range */
5d2efba6
LV
657 nio_pages = size >> IOMMU_PAGE_SHIFT;
658 io_order = get_iommu_order(size);
659 mapping = iommu_alloc(tbl, ret, nio_pages, DMA_BIDIRECTIONAL,
660 mask >> IOMMU_PAGE_SHIFT, io_order);
1da177e4
LT
661 if (mapping == DMA_ERROR_CODE) {
662 free_pages((unsigned long)ret, order);
8eb6c6e3
CH
663 return NULL;
664 }
665 *dma_handle = mapping;
1da177e4
LT
666 return ret;
667}
668
669void iommu_free_coherent(struct iommu_table *tbl, size_t size,
670 void *vaddr, dma_addr_t dma_handle)
671{
1da177e4 672 if (tbl) {
5d2efba6
LV
673 unsigned int nio_pages;
674
675 size = PAGE_ALIGN(size);
676 nio_pages = size >> IOMMU_PAGE_SHIFT;
677 iommu_free(tbl, dma_handle, nio_pages);
1da177e4 678 size = PAGE_ALIGN(size);
1da177e4
LT
679 free_pages((unsigned long)vaddr, get_order(size));
680 }
681}