summaryrefslogtreecommitdiff
path: root/arch/powerpc/platforms/cell/iommu.c
blob: eeacb3a52ca13c55edac28556e5ba4e8bd61c332 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
/*
 * IOMMU implementation for Cell Broadband Processor Architecture
 *
 * (C) Copyright IBM Corporation 2006-2008
 *
 * Author: Jeremy Kerr <jk@ozlabs.org>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2, or (at your option)
 * any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

#undef DEBUG

#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/notifier.h>
#include <linux/of.h>
#include <linux/of_platform.h>
#include <linux/lmb.h>

#include <asm/prom.h>
#include <asm/iommu.h>
#include <asm/machdep.h>
#include <asm/pci-bridge.h>
#include <asm/udbg.h>
#include <asm/firmware.h>
#include <asm/cell-regs.h>

#include "interrupt.h"

/* Define CELL_IOMMU_REAL_UNMAP to actually unmap non-used pages
 * instead of leaving them mapped to some dummy page. This can be
 * enabled once the appropriate workarounds for spider bugs have
 * been enabled
 */
#define CELL_IOMMU_REAL_UNMAP

/* Define CELL_IOMMU_STRICT_PROTECTION to enforce protection of
 * IO PTEs based on the transfer direction. That can be enabled
 * once spider-net has been fixed to pass the correct direction
 * to the DMA mapping functions
 */
#define CELL_IOMMU_STRICT_PROTECTION


#define NR_IOMMUS			2

/* IOC mmap registers */
#define IOC_Reg_Size			0x2000

#define IOC_IOPT_CacheInvd		0x908
#define IOC_IOPT_CacheInvd_NE_Mask	0xffe0000000000000ul
#define IOC_IOPT_CacheInvd_IOPTE_Mask	0x000003fffffffff8ul
#define IOC_IOPT_CacheInvd_Busy		0x0000000000000001ul

#define IOC_IOST_Origin			0x918
#define IOC_IOST_Origin_E		0x8000000000000000ul
#define IOC_IOST_Origin_HW		0x0000000000000800ul
#define IOC_IOST_Origin_HL		0x0000000000000400ul

#define IOC_IO_ExcpStat			0x920
#define IOC_IO_ExcpStat_V		0x8000000000000000ul
#define IOC_IO_ExcpStat_SPF_Mask	0x6000000000000000ul
#define IOC_IO_ExcpStat_SPF_S		0x6000000000000000ul
#define IOC_IO_ExcpStat_SPF_P		0x4000000000000000ul
#define IOC_IO_ExcpStat_ADDR_Mask	0x00000007fffff000ul
#define IOC_IO_ExcpStat_RW_Mask		0x0000000000000800ul
#define IOC_IO_ExcpStat_IOID_Mask	0x00000000000007fful

#define IOC_IO_ExcpMask			0x928
#define IOC_IO_ExcpMask_SFE		0x4000000000000000ul
#define IOC_IO_ExcpMask_PFE		0x2000000000000000ul

#define IOC_IOCmd_Offset		0x1000

#define IOC_IOCmd_Cfg			0xc00
#define IOC_IOCmd_Cfg_TE		0x0000800000000000ul


/* Segment table entries */
#define IOSTE_V			0x8000000000000000ul /* valid */
#define IOSTE_H			0x4000000000000000ul /* cache hint */
#define IOSTE_PT_Base_RPN_Mask  0x3ffffffffffff000ul /* base RPN of IOPT */
#define IOSTE_NPPT_Mask		0x0000000000000fe0ul /* no. pages in IOPT */
#define IOSTE_PS_Mask		0x0000000000000007ul /* page size */
#define IOSTE_PS_4K		0x0000000000000001ul /*   - 4kB  */
#define IOSTE_PS_64K		0x0000000000000003ul /*   - 64kB */
#define IOSTE_PS_1M		0x0000000000000005ul /*   - 1MB  */
#define IOSTE_PS_16M		0x0000000000000007ul /*   - 16MB */

/* Page table entries */
#define IOPTE_PP_W		0x8000000000000000ul /* protection: write */
#define IOPTE_PP_R		0x4000000000000000ul /* protection: read */
#define IOPTE_M			0x2000000000000000ul /* coherency required */
#define IOPTE_SO_R		0x1000000000000000ul /* ordering: writes */
#define IOPTE_SO_RW             0x1800000000000000ul /* ordering: r & w */
#define IOPTE_RPN_Mask		0x07fffffffffff000ul /* RPN */
#define IOPTE_H			0x0000000000000800ul /* cache hint */
#define IOPTE_IOID_Mask		0x00000000000007fful /* ioid */


/* IOMMU sizing */
#define IO_SEGMENT_SHIFT	28
#define IO_PAGENO_BITS(shift)	(IO_SEGMENT_SHIFT - (shift))

/* The high bit needs to be set on every DMA address */
#define SPIDER_DMA_OFFSET	0x80000000ul

struct iommu_window {
	struct list_head list;
	struct cbe_iommu *iommu;
	unsigned long offset;
	unsigned long size;
	unsigned int ioid;
	struct iommu_table table;
};

#define NAMESIZE 8
struct cbe_iommu {
	int nid;
	char name[NAMESIZE];
	void __iomem *xlate_regs;
	void __iomem *cmd_regs;
	unsigned long *stab;
	unsigned long *ptab;
	void *pad_page;
	struct list_head windows;
};

/* Static array of iommus, one per node
 *   each contains a list of windows, keyed from dma_window property
 *   - on bus setup, look for a matching window, or create one
 *   - on dev setup, assign iommu_table ptr
 */
static struct cbe_iommu iommus[NR_IOMMUS];
static int cbe_nr_iommus;

static void invalidate_tce_cache(struct cbe_iommu *iommu, unsigned long *pte,
		long n_ptes)
{
	unsigned long __iomem *reg;
	unsigned long val;
	long n;

	reg = iommu->xlate_regs + IOC_IOPT_CacheInvd;

	while (n_ptes > 0) {
		/* we can invalidate up to 1 << 11 PTEs at once */
		n = min(n_ptes, 1l << 11);
		val = (((n /*- 1*/) << 53) & IOC_IOPT_CacheInvd_NE_Mask)
			| (__pa(pte) & IOC_IOPT_CacheInvd_IOPTE_Mask)
		        | IOC_IOPT_CacheInvd_Busy;

		out_be64(reg, val);
		while (in_be64(reg) & IOC_IOPT_CacheInvd_Busy)
			;

		n_ptes -= n;
		pte += n;
	}
}

static void tce_build_cell(struct iommu_table *tbl, long index, long npages,
		unsigned long uaddr, enum dma_data_direction direction)
{
	int i;
	unsigned long *io_pte, base_pte;
	struct iommu_window *window =
		container_of(tbl, struct iommu_window, table);

	/* implementing proper protection causes problems with the spidernet
	 * driver - check mapping directions later, but allow read & write by
	 * default for now.*/
#ifdef CELL_IOMMU_STRICT_PROTECTION
	/* to avoid referencing a global, we use a trick here to setup the
	 * protection bit. "prot" is setup to be 3 fields of 4 bits apprended
	 * together for each of the 3 supported direction values. It is then
	 * shifted left so that the fields matching the desired direction
	 * lands on the appropriate bits, and other bits are masked out.
	 */
	const unsigned long prot = 0xc48;
	base_pte =
		((prot << (52 + 4 * direction)) & (IOPTE_PP_W | IOPTE_PP_R))
		| IOPTE_M | IOPTE_SO_RW | (window->ioid & IOPTE_IOID_Mask);
#else
	base_pte = IOPTE_PP_W | IOPTE_PP_R | IOPTE_M | IOPTE_SO_RW |
		(window->ioid & IOPTE_IOID_Mask);
#endif

	io_pte = (unsigned long *)tbl->it_base + (index - tbl->it_offset);

	for (i = 0; i < npages; i++, uaddr += IOMMU_PAGE_SIZE)
		io_pte[i] = base_pte | (__pa(uaddr) & IOPTE_RPN_Mask);

	mb();

	invalidate_tce_cache(window->iommu, io_pte, npages);

	pr_debug("tce_build_cell(index=%lx,n=%lx,dir=%d,base_pte=%lx)\n",
		 index, npages, direction, base_pte);
}

static void tce_free_cell(struct iommu_table *tbl, long index, long npages)
{

	int i;
	unsigned long *io_pte, pte;
	struct iommu_window *window =
		container_of(tbl, struct iommu_window, table);

	pr_debug("tce_free_cell(index=%lx,n=%lx)\n", index, npages);

#ifdef CELL_IOMMU_REAL_UNMAP
	pte = 0;
#else
	/* spider bridge does PCI reads after freeing - insert a mapping
	 * to a scratch page instead of an invalid entry */
	pte = IOPTE_PP_R | IOPTE_M | IOPTE_SO_RW | __pa(window->iommu->pad_page)
		| (window->ioid & IOPTE_IOID_Mask);
#endif

	io_pte = (unsigned long *)tbl->it_base + (index - tbl->it_offset);

	for (i = 0; i < npages; i++)
		io_pte[i] = pte;

	mb();

	invalidate_tce_cache(window->iommu, io_pte, npages);
}

static irqreturn_t ioc_interrupt(int irq, void *data)
{
	unsigned long stat;
	struct cbe_iommu *iommu = data;

	stat = in_be64(iommu->xlate_regs + IOC_IO_ExcpStat);

	/* Might want to rate limit it */
	printk(KERN_ERR "iommu: DMA exception 0x%016lx\n", stat);
	printk(KERN_ERR "  V=%d, SPF=[%c%c], RW=%s, IOID=0x%04x\n",
	       !!(stat & IOC_IO_ExcpStat_V),
	       (stat & IOC_IO_ExcpStat_SPF_S) ? 'S' : ' ',
	       (stat & IOC_IO_ExcpStat_SPF_P) ? 'P' : ' ',
	       (stat & IOC_IO_ExcpStat_RW_Mask) ? "Read" : "Write",
	       (unsigned int)(stat & IOC_IO_ExcpStat_IOID_Mask));
	printk(KERN_ERR "  page=0x%016lx\n",
	       stat & IOC_IO_ExcpStat_ADDR_Mask);

	/* clear interrupt */
	stat &= ~IOC_IO_ExcpStat_V;
	out_be64(iommu->xlate_regs + IOC_IO_ExcpStat, stat);

	return IRQ_HANDLED;
}

static int cell_iommu_find_ioc(int nid, unsigned long *base)
{
	struct device_node *np;
	struct resource r;

	*base = 0;

	/* First look for new style /be nodes */
	for_each_node_by_name(np, "ioc") {
		if (of_node_to_nid(np) != nid)
			continue;
		if (of_address_to_resource(np, 0, &r)) {
			printk(KERN_ERR "iommu: can't get address for %s\n",
			       np->full_name);
			continue;
		}
		*base = r.start;
		of_node_put(np);
		return 0;
	}

	/* Ok, let's try the old way */
	for_each_node_by_type(np, "cpu") {
		const unsigned int *nidp;
		const unsigned long *tmp;

		nidp = of_get_property(np, "node-id", NULL);
		if (nidp && *nidp == nid) {
			tmp = of_get_property(np, "ioc-translation", NULL);
			if (tmp) {
				*base = *tmp;
				of_node_put(np);
				return 0;
			}
		}
	}

	return -ENODEV;
}

static void cell_iommu_setup_stab(struct cbe_iommu *iommu,
				unsigned long dbase, unsigned long dsize,
				unsigned long fbase, unsigned long fsize)
{
	struct page *page;
	unsigned long segments, stab_size;

	segments = max(dbase + dsize, fbase + fsize) >> IO_SEGMENT_SHIFT;

	pr_debug("%s: iommu[%d]: segments: %lu\n",
			__func__, iommu->nid, segments);

	/* set up the segment table */
	stab_size = segments * sizeof(unsigned long);
	page = alloc_pages_node(iommu->nid, GFP_KERNEL, get_order(stab_size));
	BUG_ON(!page);
	iommu->stab = page_address(page);
	memset(iommu->stab, 0, stab_size);
}

static unsigned long *cell_iommu_alloc_ptab(struct cbe_iommu *iommu,
		unsigned long base, unsigned long size, unsigned long gap_base,
		unsigned long gap_size, unsigned long page_shift)
{
	struct page *page;
	int i;
	unsigned long reg, segments, pages_per_segment, ptab_size,
		      n_pte_pages, start_seg, *ptab;

	start_seg = base >> IO_SEGMENT_SHIFT;
	segments  = size >> IO_SEGMENT_SHIFT;
	pages_per_segment = 1ull << IO_PAGENO_BITS(page_shift);
	/* PTEs for each segment must start on a 4K bounday */
	pages_per_segment = max(pages_per_segment,
				(1 << 12) / sizeof(unsigned long));

	ptab_size = segments * pages_per_segment * sizeof(unsigned long);
	pr_debug("%s: iommu[%d]: ptab_size: %lu, order: %d\n", __func__,
			iommu->nid, ptab_size, get_order(ptab_size));
	page = alloc_pages_node(iommu->nid, GFP_KERNEL, get_order(ptab_size));
	BUG_ON(!page);

	ptab = page_address(page);
	memset(ptab, 0, ptab_size);

	/* number of 4K pages needed for a page table */
	n_pte_pages = (pages_per_segment * sizeof(unsigned long)) >> 12;

	pr_debug("%s: iommu[%d]: stab at %p, ptab at %p, n_pte_pages: %lu\n",
			__func__, iommu->nid, iommu->stab, ptab,
			n_pte_pages);

	/* initialise the STEs */
	reg = IOSTE_V | ((n_pte_pages - 1) << 5);

	switch (page_shift) {
	case 12: reg |= IOSTE_PS_4K;  break;
	case 16: reg |= IOSTE_PS_64K; break;
	case 20: reg |= IOSTE_PS_1M;  break;
	case 24: reg |= IOSTE_PS_16M; break;
	default: BUG();
	}

	gap_base = gap_base >> IO_SEGMENT_SHIFT;
	gap_size = gap_size >> IO_SEGMENT_SHIFT;

	pr_debug("Setting up IOMMU stab:\n");
	for (i = start_seg; i < (start_seg + segments); i++) {
		if (i >= gap_base && i < (gap_base + gap_size)) {
			pr_debug("\toverlap at %d, skipping\n", i);
			continue;
		}
		iommu->stab[i] = reg | (__pa(ptab) + (n_pte_pages << 12) *
					(i - start_seg));
		pr_debug("\t[%d] 0x%016lx\n", i, iommu->stab[i]);
	}

	return ptab;
}

static void cell_iommu_enable_hardware(struct cbe_iommu *iommu)
{
	int ret;
	unsigned long reg, xlate_base;
	unsigned int virq;

	if (cell_iommu_find_ioc(iommu->nid, &xlate_base))
		panic("%s: missing IOC register mappings for node %d\n",
		      __func__, iommu->nid);

	iommu->xlate_regs = ioremap(xlate_base, IOC_Reg_Size);
	iommu->cmd_regs = iommu->xlate_regs + IOC_IOCmd_Offset;

	/* ensure that the STEs have updated */
	mb();

	/* setup interrupts for the iommu. */
	reg = in_be64(iommu->xlate_regs + IOC_IO_ExcpStat);
	out_be64(iommu->xlate_regs + IOC_IO_ExcpStat,
			reg & ~IOC_IO_ExcpStat_V);
	out_be64(iommu->xlate_regs + IOC_IO_ExcpMask,
			IOC_IO_ExcpMask_PFE | IOC_IO_ExcpMask_SFE);

	virq = irq_create_mapping(NULL,
			IIC_IRQ_IOEX_ATI | (iommu->nid << IIC_IRQ_NODE_SHIFT));
	BUG_ON(virq == NO_IRQ);

	ret = request_irq(virq, ioc_interrupt, IRQF_DISABLED,
			iommu->name, iommu);
	BUG_ON(ret);

	/* set the IOC segment table origin register (and turn on the iommu) */
	reg = IOC_IOST_Origin_E | __pa(iommu->stab) | IOC_IOST_Origin_HW;
	out_be64(iommu->xlate_regs + IOC_IOST_Origin, reg);
	in_be64(iommu->xlate_regs + IOC_IOST_Origin);

	/* turn on IO translation */
	reg = in_be64(iommu->cmd_regs + IOC_IOCmd_Cfg) | IOC_IOCmd_Cfg_TE;
	out_be64(iommu->cmd_regs + IOC_IOCmd_Cfg, reg);
}

static void cell_iommu_setup_hardware(struct cbe_iommu *iommu,
	unsigned long base, unsigned long size)
{
	cell_iommu_setup_stab(iommu, base, size, 0, 0);
	iommu->ptab = cell_iommu_alloc_ptab(iommu, base, size, 0, 0,
					    IOMMU_PAGE_SHIFT);
	cell_iommu_enable_hardware(iommu);
}

#if 0/* Unused for now */
static struct iommu_window *find_window(struct cbe_iommu *iommu,
		unsigned long offset, unsigned long size)
{
	struct iommu_window *window;

	/* todo: check for overlapping (but not equal) windows) */

	list_for_each_entry(window, &(iommu->windows), list) {
		if (window->offset == offset && window->size == size)
			return window;
	}

	return NULL;
}
#endif

static inline u32 cell_iommu_get_ioid(struct device_node *np)
{
	const u32 *ioid;

	ioid = of_get_property(np, "ioid", NULL);
	if (ioid == NULL) {
		printk(KERN_WARNING "iommu: missing ioid for %s using 0\n",
		       np->full_name);
		return 0;
	}

	return *ioid;
}

static struct iommu_window * __init
cell_iommu_setup_window(struct cbe_iommu *iommu, struct device_node *np,
			unsigned long offset, unsigned long size,
			unsigned long pte_offset)
{
	struct iommu_window *window;
	struct page *page;
	u32 ioid;

	ioid = cell_iommu_get_ioid(np);

	window = kmalloc_node(sizeof(*window), GFP_KERNEL, iommu->nid);
	BUG_ON(window == NULL);

	window->offset = offset;
	window->size = size;
	window->ioid = ioid;
	window->iommu = iommu;

	window->table.it_blocksize = 16;
	window->table.it_base = (unsigned long)iommu->ptab;
	window->table.it_index = iommu->nid;
	window->table.it_offset = (offset >> IOMMU_PAGE_SHIFT) + pte_offset;
	window->table.it_size = size >> IOMMU_PAGE_SHIFT;

	iommu_init_table(&window->table, iommu->nid);

	pr_debug("\tioid      %d\n", window->ioid);
	pr_debug("\tblocksize %ld\n", window->table.it_blocksize);
	pr_debug("\tbase      0x%016lx\n", window->table.it_base);
	pr_debug("\toffset    0x%lx\n", window->table.it_offset);
	pr_debug("\tsize      %ld\n", window->table.it_size);

	list_add(&window->list, &iommu->windows);

	if (offset != 0)
		return window;

	/* We need to map and reserve the first IOMMU page since it's used
	 * by the spider workaround. In theory, we only need to do that when
	 * running on spider but it doesn't really matter.
	 *
	 * This code also assumes that we have a window that starts at 0,
	 * which is the case on all spider based blades.
	 */
	page = alloc_pages_node(iommu->nid, GFP_KERNEL, 0);
	BUG_ON(!page);
	iommu->pad_page = page_address(page);
	clear_page(iommu->pad_page);

	__set_bit(0, window->table.it_map);
	tce_build_cell(&window->table, window->table.it_offset, 1,
		       (unsigned long)iommu->pad_page, DMA_TO_DEVICE);
	window->table.it_hint = window->table.it_blocksize;

	return window;
}

static struct cbe_iommu *cell_iommu_for_node(int nid)
{
	int i;

	for (i = 0; i < cbe_nr_iommus; i++)
		if (iommus[i].nid == nid)
			return &iommus[i];
	return NULL;
}

static unsigned long cell_dma_direct_offset;

static unsigned long dma_iommu_fixed_base;
struct dma_mapping_ops dma_iommu_fixed_ops;

static struct iommu_table *cell_get_iommu_table(struct device *dev)
{
	struct iommu_window *window;
	struct cbe_iommu *iommu;
	struct dev_archdata *archdata = &dev->archdata;

	/* Current implementation uses the first window available in that
	 * node's iommu. We -might- do something smarter later though it may
	 * never be necessary
	 */
	iommu = cell_iommu_for_node(archdata->numa_node);
	if (iommu == NULL || list_empty(&iommu->windows)) {
		printk(KERN_ERR "iommu: missing iommu for %s (node %d)\n",
		       archdata->of_node ? archdata->of_node->full_name : "?",
		       archdata->numa_node);
		return NULL;
	}
	window = list_entry(iommu->windows.next, struct iommu_window, list);

	return &window->table;
}

static void cell_dma_dev_setup_fixed(struct device *dev);

static void cell_dma_dev_setup(struct device *dev)
{
	struct dev_archdata *archdata = &dev->archdata;

	/* Order is important here, these are not mutually exclusive */
	if (get_dma_ops(dev) == &dma_iommu_fixed_ops)
		cell_dma_dev_setup_fixed(dev);
	else if (get_pci_dma_ops() == &dma_iommu_ops)
		archdata->dma_data = cell_get_iommu_table(dev);
	else if (get_pci_dma_ops() == &dma_direct_ops)
		archdata->dma_data = (void *)cell_dma_direct_offset;
	else
		BUG();
}

static void cell_pci_dma_dev_setup(struct pci_dev *dev)
{
	cell_dma_dev_setup(&dev->dev);
}

static int cell_of_bus_notify(struct notifier_block *nb, unsigned long action,
			      void *data)
{
	struct device *dev = data;

	/* We are only intereted in device addition */
	if (action != BUS_NOTIFY_ADD_DEVICE)
		return 0;

	/* We use the PCI DMA ops */
	dev->archdata.dma_ops = get_pci_dma_ops();

	cell_dma_dev_setup(dev);

	return 0;
}

static struct notifier_block cell_of_bus_notifier = {
	.notifier_call = cell_of_bus_notify
};

static int __init cell_iommu_get_window(struct device_node *np,
					 unsigned long *base,
					 unsigned long *size)
{
	const void *dma_window;
	unsigned long index;

	/* Use ibm,dma-window if available, else, hard code ! */
	dma_window = of_get_property(np, "ibm,dma-window", NULL);
	if (dma_window == NULL) {
		*base = 0;
		*size = 0x80000000u;
		return -ENODEV;
	}

	of_parse_dma_window(np, dma_window, &index, base, size);
	return 0;
}

static struct cbe_iommu * __init cell_iommu_alloc(struct device_node *np)
{
	struct cbe_iommu *iommu;
	int nid, i;

	/* Get node ID */
	nid = of_node_to_nid(np);
	if (nid < 0) {
		printk(KERN_ERR "iommu: failed to get node for %s\n",
		       np->full_name);
		return NULL;
	}
	pr_debug("iommu: setting up iommu for node %d (%s)\n",
		 nid, np->full_name);

	/* XXX todo: If we can have multiple windows on the same IOMMU, which
	 * isn't the case today, we probably want here to check wether the
	 * iommu for that node is already setup.
	 * However, there might be issue with getting the size right so let's
	 * ignore that for now. We might want to completely get rid of the
	 * multiple window support since the cell iommu supports per-page ioids
	 */

	if (cbe_nr_iommus >= NR_IOMMUS) {
		printk(KERN_ERR "iommu: too many IOMMUs detected ! (%s)\n",
		       np->full_name);
		return NULL;
	}

	/* Init base fields */
	i = cbe_nr_iommus++;
	iommu = &iommus[i];
	iommu->stab = NULL;
	iommu->nid = nid;
	snprintf(iommu->name, sizeof(iommu->name), "iommu%d", i);
	INIT_LIST_HEAD(&iommu->windows);

	return iommu;
}

static void __init cell_iommu_init_one(struct device_node *np,
				       unsigned long offset)
{
	struct cbe_iommu *iommu;
	unsigned long base, size;

	iommu = cell_iommu_alloc(np);
	if (!iommu)
		return;

	/* Obtain a window for it */
	cell_iommu_get_window(np, &base, &size);

	pr_debug("\ttranslating window 0x%lx...0x%lx\n",
		 base, base + size - 1);

	/* Initialize the hardware */
	cell_iommu_setup_hardware(iommu, base, size);

	/* Setup the iommu_table */
	cell_iommu_setup_window(iommu, np, base, size,
				offset >> IOMMU_PAGE_SHIFT);
}

static void __init cell_disable_iommus(void)
{
	int node;
	unsigned long base, val;
	void __iomem *xregs, *cregs;

	/* Make sure IOC translation is disabled on all nodes */
	for_each_online_node(node) {
		if (cell_iommu_find_ioc(node, &base))
			continue;
		xregs = ioremap(base, IOC_Reg_Size);
		if (xregs == NULL)
			continue;
		cregs = xregs + IOC_IOCmd_Offset;

		pr_debug("iommu: cleaning up iommu on node %d\n", node);

		out_be64(xregs + IOC_IOST_Origin, 0);
		(void)in_be64(xregs + IOC_IOST_Origin);
		val = in_be64(cregs + IOC_IOCmd_Cfg);
		val &= ~IOC_IOCmd_Cfg_TE;
		out_be64(cregs + IOC_IOCmd_Cfg, val);
		(void)in_be64(cregs + IOC_IOCmd_Cfg);

		iounmap(xregs);
	}
}

static int __init cell_iommu_init_disabled(void)
{
	struct device_node *np = NULL;
	unsigned long base = 0, size;

	/* When no iommu is present, we use direct DMA ops */
	set_pci_dma_ops(&dma_direct_ops);

	/* First make sure all IOC translation is turned off */
	cell_disable_iommus();

	/* If we have no Axon, we set up the spider DMA magic offset */
	if (of_find_node_by_name(NULL, "axon") == NULL)
		cell_dma_direct_offset = SPIDER_DMA_OFFSET;

	/* Now we need to check to see where the memory is mapped
	 * in PCI space. We assume that all busses use the same dma
	 * window which is always the case so far on Cell, thus we
	 * pick up the first pci-internal node we can find and check
	 * the DMA window from there.
	 */
	for_each_node_by_name(np, "axon") {
		if (np->parent == NULL || np->parent->parent != NULL)
			continue;
		if (cell_iommu_get_window(np, &base, &size) == 0)
			break;
	}
	if (np == NULL) {
		for_each_node_by_name(np, "pci-internal") {
			if (np->parent == NULL || np->parent->parent != NULL)
				continue;
			if (cell_iommu_get_window(np, &base, &size) == 0)
				break;
		}
	}
	of_node_put(np);

	/* If we found a DMA window, we check if it's big enough to enclose
	 * all of physical memory. If not, we force enable IOMMU
	 */
	if (np && size < lmb_end_of_DRAM()) {
		printk(KERN_WARNING "iommu: force-enabled, dma window"
		       " (%ldMB) smaller than total memory (%ldMB)\n",
		       size >> 20, lmb_end_of_DRAM() >> 20);
		return -ENODEV;
	}

	cell_dma_direct_offset += base;

	if (cell_dma_direct_offset != 0)
		ppc_md.pci_dma_dev_setup = cell_pci_dma_dev_setup;

	printk("iommu: disabled, direct DMA offset is 0x%lx\n",
	       cell_dma_direct_offset);

	return 0;
}

/*
 *  Fixed IOMMU mapping support
 *
 *  This code adds support for setting up a fixed IOMMU mapping on certain
 *  cell machines. For 64-bit devices this avoids the performance overhead of
 *  mapping and unmapping pages at runtime. 32-bit devices are unable to use
 *  the fixed mapping.
 *
 *  The fixed mapping is established at boot, and maps all of physical memory
 *  1:1 into device space at some offset. On machines with < 30 GB of memory
 *  we setup the fixed mapping immediately above the normal IOMMU window.
 *
 *  For example a machine with 4GB of memory would end up with the normal
 *  IOMMU window from 0-2GB and the fixed mapping window from 2GB to 6GB. In
 *  this case a 64-bit device wishing to DMA to 1GB would be told to DMA to
 *  3GB, plus any offset required by firmware. The firmware offset is encoded
 *  in the "dma-ranges" property.
 *
 *  On machines with 30GB or more of memory, we are unable to place the fixed
 *  mapping above the normal IOMMU window as we would run out of address space.
 *  Instead we move the normal IOMMU window to coincide with the hash page
 *  table, this region does not need to be part of the fixed mapping as no
 *  device should ever be DMA'ing to it. We then setup the fixed mapping
 *  from 0 to 32GB.
 */

static u64 cell_iommu_get_fixed_address(struct device *dev)
{
	u64 cpu_addr, size, best_size, dev_addr = OF_BAD_ADDR;
	struct device_node *np;
	const u32 *ranges = NULL;
	int i, len, best, naddr, nsize, pna, range_size;

	np = of_node_get(dev->archdata.of_node);
	while (1) {
		naddr = of_n_addr_cells(np);
		nsize = of_n_size_cells(np);
		np = of_get_next_parent(np);
		if (!np)
			break;

		ranges = of_get_property(np, "dma-ranges", &len);

		/* Ignore empty ranges, they imply no translation required */
		if (ranges && len > 0)
			break;
	}

	if (!ranges) {
		dev_dbg(dev, "iommu: no dma-ranges found\n");
		goto out;
	}

	len /= sizeof(u32);

	pna = of_n_addr_cells(np);
	range_size = naddr + nsize + pna;

	/* dma-ranges format:
	 * child addr	: naddr cells
	 * parent addr	: pna cells
	 * size		: nsize cells
	 */
	for (i = 0, best = -1, best_size = 0; i < len; i += range_size) {
		cpu_addr = of_translate_dma_address(np, ranges + i + naddr);
		size = of_read_number(ranges + i + naddr + pna, nsize);

		if (cpu_addr == 0 && size > best_size) {
			best = i;
			best_size = size;
		}
	}

	if (best >= 0) {
		dev_addr = of_read_number(ranges + best, naddr);
	} else
		dev_dbg(dev, "iommu: no suitable range found!\n");

out:
	of_node_put(np);

	return dev_addr;
}

static int dma_set_mask_and_switch(struct device *dev, u64 dma_mask)
{
	if (!dev->dma_mask || !dma_supported(dev, dma_mask))
		return -EIO;

	if (dma_mask == DMA_BIT_MASK(64) &&
		cell_iommu_get_fixed_address(dev) != OF_BAD_ADDR)
	{
		dev_dbg(dev, "iommu: 64-bit OK, using fixed ops\n");
		set_dma_ops(dev, &dma_iommu_fixed_ops);
	} else {
		dev_dbg(dev, "iommu: not 64-bit, using default ops\n");
		set_dma_ops(dev, get_pci_dma_ops());
	}

	cell_dma_dev_setup(dev);

	*dev->dma_mask = dma_mask;

	return 0;
}

static void cell_dma_dev_setup_fixed(struct device *dev)
{
	struct dev_archdata *archdata = &dev->archdata;
	u64 addr;

	addr = cell_iommu_get_fixed_address(dev) + dma_iommu_fixed_base;
	archdata->dma_data = (void *)addr;

	dev_dbg(dev, "iommu: fixed addr = %lx\n", addr);
}

static void insert_16M_pte(unsigned long addr, unsigned long *ptab,
			   unsigned long base_pte)
{
	unsigned long segment, offset;

	segment = addr >> IO_SEGMENT_SHIFT;
	offset = (addr >> 24) - (segment << IO_PAGENO_BITS(24));
	ptab = ptab + (segment * (1 << 12) / sizeof(unsigned long));

	pr_debug("iommu: addr %lx ptab %p segment %lx offset %lx\n",
		  addr, ptab, segment, offset);

	ptab[offset] = base_pte | (__pa(addr) & IOPTE_RPN_Mask);
}

static void cell_iommu_setup_fixed_ptab(struct cbe_iommu *iommu,
	struct device_node *np, unsigned long dbase, unsigned long dsize,
	unsigned long fbase, unsigned long fsize)
{
	unsigned long base_pte, uaddr, ioaddr, *ptab;

	ptab = cell_iommu_alloc_ptab(iommu, fbase, fsize, dbase, dsize, 24);

	dma_iommu_fixed_base = fbase;

	pr_debug("iommu: mapping 0x%lx pages from 0x%lx\n", fsize, fbase);

	base_pte = IOPTE_PP_W | IOPTE_PP_R | IOPTE_M | IOPTE_SO_RW
		    | (cell_iommu_get_ioid(np) & IOPTE_IOID_Mask);

	for (uaddr = 0; uaddr < fsize; uaddr += (1 << 24)) {
		/* Don't touch the dynamic region */
		ioaddr = uaddr + fbase;
		if (ioaddr >= dbase && ioaddr < (dbase + dsize)) {
			pr_debug("iommu: fixed/dynamic overlap, skipping\n");
			continue;
		}

		insert_16M_pte(uaddr, ptab, base_pte);
	}

	mb();
}

static int __init cell_iommu_fixed_mapping_init(void)
{
	unsigned long dbase, dsize, fbase, fsize, hbase, hend;
	struct cbe_iommu *iommu;
	struct device_node *np;

	/* The fixed mapping is only supported on axon machines */
	np = of_find_node_by_name(NULL, "axon");
	if (!np) {
		pr_debug("iommu: fixed mapping disabled, no axons found\n");
		return -1;
	}

	/* We must have dma-ranges properties for fixed mapping to work */
	for (np = NULL; (np = of_find_all_nodes(np));) {
		if (of_find_property(np, "dma-ranges", NULL))
			break;
	}
	of_node_put(np);

	if (!np) {
		pr_debug("iommu: no dma-ranges found, no fixed mapping\n");
		return -1;
	}

	/* The default setup is to have the fixed mapping sit after the
	 * dynamic region, so find the top of the largest IOMMU window
	 * on any axon, then add the size of RAM and that's our max value.
	 * If that is > 32GB we have to do other shennanigans.
	 */
	fbase = 0;
	for_each_node_by_name(np, "axon") {
		cell_iommu_get_window(np, &dbase, &dsize);
		fbase = max(fbase, dbase + dsize);
	}

	fbase = _ALIGN_UP(fbase, 1 << IO_SEGMENT_SHIFT);
	fsize = lmb_phys_mem_size();

	if ((fbase + fsize) <= 0x800000000)
		hbase = 0; /* use the device tree window */
	else {
		/* If we're over 32 GB we need to cheat. We can't map all of
		 * RAM with the fixed mapping, and also fit the dynamic
		 * region. So try to place the dynamic region where the hash
		 * table sits, drivers never need to DMA to it, we don't
		 * need a fixed mapping for that area.
		 */
		if (!htab_address) {
			pr_debug("iommu: htab is NULL, on LPAR? Huh?\n");
			return -1;
		}
		hbase = __pa(htab_address);
		hend  = hbase + htab_size_bytes;

		/* The window must start and end on a segment boundary */
		if ((hbase != _ALIGN_UP(hbase, 1 << IO_SEGMENT_SHIFT)) ||
		    (hend != _ALIGN_UP(hend, 1 << IO_SEGMENT_SHIFT))) {
			pr_debug("iommu: hash window not segment aligned\n");
			return -1;
		}

		/* Check the hash window fits inside the real DMA window */
		for_each_node_by_name(np, "axon") {
			cell_iommu_get_window(np, &dbase, &dsize);

			if (hbase < dbase || (hend > (dbase + dsize))) {
				pr_debug("iommu: hash window doesn't fit in"
					 "real DMA window\n");
				return -1;
			}
		}

		fbase = 0;
	}

	/* Setup the dynamic regions */
	for_each_node_by_name(np, "axon") {
		iommu = cell_iommu_alloc(np);
		BUG_ON(!iommu);

		if (hbase == 0)
			cell_iommu_get_window(np, &dbase, &dsize);
		else {
			dbase = hbase;
			dsize = htab_size_bytes;
		}

		printk(KERN_DEBUG "iommu: node %d, dynamic window 0x%lx-0x%lx "
			"fixed window 0x%lx-0x%lx\n", iommu->nid, dbase,
			 dbase + dsize, fbase, fbase + fsize);

		cell_iommu_setup_stab(iommu, dbase, dsize, fbase, fsize);
		iommu->ptab = cell_iommu_alloc_ptab(iommu, dbase, dsize, 0, 0,
						    IOMMU_PAGE_SHIFT);
		cell_iommu_setup_fixed_ptab(iommu, np, dbase, dsize,
					     fbase, fsize);
		cell_iommu_enable_hardware(iommu);
		cell_iommu_setup_window(iommu, np, dbase, dsize, 0);
	}

	dma_iommu_fixed_ops = dma_direct_ops;
	dma_iommu_fixed_ops.set_dma_mask = dma_set_mask_and_switch;

	dma_iommu_ops.set_dma_mask = dma_set_mask_and_switch;
	set_pci_dma_ops(&dma_iommu_ops);

	return 0;
}

static int iommu_fixed_disabled;

static int __init setup_iommu_fixed(char *str)
{
	if (strcmp(str, "off") == 0)
		iommu_fixed_disabled = 1;

	return 1;
}
__setup("iommu_fixed=", setup_iommu_fixed);

static int __init cell_iommu_init(void)
{
	struct device_node *np;

	/* If IOMMU is disabled or we have little enough RAM to not need
	 * to enable it, we setup a direct mapping.
	 *
	 * Note: should we make sure we have the IOMMU actually disabled ?
	 */
	if (iommu_is_off ||
	    (!iommu_force_on && lmb_end_of_DRAM() <= 0x80000000ull))
		if (cell_iommu_init_disabled() == 0)
			goto bail;

	/* Setup various ppc_md. callbacks */
	ppc_md.pci_dma_dev_setup = cell_pci_dma_dev_setup;
	ppc_md.tce_build = tce_build_cell;
	ppc_md.tce_free = tce_free_cell;

	if (!iommu_fixed_disabled && cell_iommu_fixed_mapping_init() == 0)
		goto bail;

	/* Create an iommu for each /axon node.  */
	for_each_node_by_name(np, "axon") {
		if (np->parent == NULL || np->parent->parent != NULL)
			continue;
		cell_iommu_init_one(np, 0);
	}

	/* Create an iommu for each toplevel /pci-internal node for
	 * old hardware/firmware
	 */
	for_each_node_by_name(np, "pci-internal") {
		if (np->parent == NULL || np->parent->parent != NULL)
			continue;
		cell_iommu_init_one(np, SPIDER_DMA_OFFSET);
	}

	/* Setup default PCI iommu ops */
	set_pci_dma_ops(&dma_iommu_ops);

 bail:
	/* Register callbacks on OF platform device addition/removal
	 * to handle linking them to the right DMA operations
	 */
	bus_register_notifier(&of_platform_bus_type, &cell_of_bus_notifier);

	return 0;
}
machine_arch_initcall(cell, cell_iommu_init);
machine_arch_initcall(celleb_native, cell_iommu_init);