Commit 22d48269984fc93a71f65a54aa422aacf5fdb926
1 parent
bfc5501f6d
Exists in
master
and in
7 other branches
swiotlb: search and replace "int dir" with "enum dma_data_direction dir"
.. to catch anybody doing something funky. See "swiotlb: swiotlb: add swiotlb_tbl_map_single library function" for full description of patchset. [v2: swiotlb_sync_single_range_* no more - removed usage] [v3: enum dma_data_direction direction -> enum dma_data_direction dir] Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> Tested-by: Albert Herranz <albert_herranz@yahoo.es>
Showing 2 changed files with 15 additions and 12 deletions Inline Diff
include/linux/swiotlb.h
1 | #ifndef __LINUX_SWIOTLB_H | 1 | #ifndef __LINUX_SWIOTLB_H |
2 | #define __LINUX_SWIOTLB_H | 2 | #define __LINUX_SWIOTLB_H |
3 | 3 | ||
4 | #include <linux/types.h> | 4 | #include <linux/types.h> |
5 | 5 | ||
6 | struct device; | 6 | struct device; |
7 | struct dma_attrs; | 7 | struct dma_attrs; |
8 | struct scatterlist; | 8 | struct scatterlist; |
9 | 9 | ||
10 | extern int swiotlb_force; | 10 | extern int swiotlb_force; |
11 | 11 | ||
12 | /* | 12 | /* |
13 | * Maximum allowable number of contiguous slabs to map, | 13 | * Maximum allowable number of contiguous slabs to map, |
14 | * must be a power of 2. What is the appropriate value ? | 14 | * must be a power of 2. What is the appropriate value ? |
15 | * The complexity of {map,unmap}_single is linearly dependent on this value. | 15 | * The complexity of {map,unmap}_single is linearly dependent on this value. |
16 | */ | 16 | */ |
17 | #define IO_TLB_SEGSIZE 128 | 17 | #define IO_TLB_SEGSIZE 128 |
18 | 18 | ||
19 | /* | 19 | /* |
20 | * log of the size of each IO TLB slab. The number of slabs is command line | 20 | * log of the size of each IO TLB slab. The number of slabs is command line |
21 | * controllable. | 21 | * controllable. |
22 | */ | 22 | */ |
23 | #define IO_TLB_SHIFT 11 | 23 | #define IO_TLB_SHIFT 11 |
24 | 24 | ||
25 | extern void swiotlb_init(int verbose); | 25 | extern void swiotlb_init(int verbose); |
26 | extern void swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose); | 26 | extern void swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose); |
27 | 27 | ||
28 | extern void | 28 | extern void |
29 | *swiotlb_alloc_coherent(struct device *hwdev, size_t size, | 29 | *swiotlb_alloc_coherent(struct device *hwdev, size_t size, |
30 | dma_addr_t *dma_handle, gfp_t flags); | 30 | dma_addr_t *dma_handle, gfp_t flags); |
31 | 31 | ||
32 | extern void | 32 | extern void |
33 | swiotlb_free_coherent(struct device *hwdev, size_t size, | 33 | swiotlb_free_coherent(struct device *hwdev, size_t size, |
34 | void *vaddr, dma_addr_t dma_handle); | 34 | void *vaddr, dma_addr_t dma_handle); |
35 | 35 | ||
36 | extern dma_addr_t swiotlb_map_page(struct device *dev, struct page *page, | 36 | extern dma_addr_t swiotlb_map_page(struct device *dev, struct page *page, |
37 | unsigned long offset, size_t size, | 37 | unsigned long offset, size_t size, |
38 | enum dma_data_direction dir, | 38 | enum dma_data_direction dir, |
39 | struct dma_attrs *attrs); | 39 | struct dma_attrs *attrs); |
40 | extern void swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, | 40 | extern void swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, |
41 | size_t size, enum dma_data_direction dir, | 41 | size_t size, enum dma_data_direction dir, |
42 | struct dma_attrs *attrs); | 42 | struct dma_attrs *attrs); |
43 | 43 | ||
44 | extern int | 44 | extern int |
45 | swiotlb_map_sg(struct device *hwdev, struct scatterlist *sg, int nents, | 45 | swiotlb_map_sg(struct device *hwdev, struct scatterlist *sg, int nents, |
46 | int direction); | 46 | enum dma_data_direction dir); |
47 | 47 | ||
48 | extern void | 48 | extern void |
49 | swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sg, int nents, | 49 | swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sg, int nents, |
50 | int direction); | 50 | enum dma_data_direction dir); |
51 | 51 | ||
52 | extern int | 52 | extern int |
53 | swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, | 53 | swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, |
54 | enum dma_data_direction dir, struct dma_attrs *attrs); | 54 | enum dma_data_direction dir, struct dma_attrs *attrs); |
55 | 55 | ||
56 | extern void | 56 | extern void |
57 | swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | 57 | swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, |
58 | int nelems, enum dma_data_direction dir, | 58 | int nelems, enum dma_data_direction dir, |
59 | struct dma_attrs *attrs); | 59 | struct dma_attrs *attrs); |
60 | 60 | ||
61 | extern void | 61 | extern void |
62 | swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, | 62 | swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, |
63 | size_t size, enum dma_data_direction dir); | 63 | size_t size, enum dma_data_direction dir); |
64 | 64 | ||
65 | extern void | 65 | extern void |
66 | swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, | 66 | swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, |
67 | int nelems, enum dma_data_direction dir); | 67 | int nelems, enum dma_data_direction dir); |
68 | 68 | ||
69 | extern void | 69 | extern void |
70 | swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, | 70 | swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, |
71 | size_t size, enum dma_data_direction dir); | 71 | size_t size, enum dma_data_direction dir); |
72 | 72 | ||
73 | extern void | 73 | extern void |
74 | swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, | 74 | swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, |
75 | int nelems, enum dma_data_direction dir); | 75 | int nelems, enum dma_data_direction dir); |
76 | 76 | ||
77 | extern int | 77 | extern int |
78 | swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr); | 78 | swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr); |
79 | 79 | ||
80 | extern int | 80 | extern int |
81 | swiotlb_dma_supported(struct device *hwdev, u64 mask); | 81 | swiotlb_dma_supported(struct device *hwdev, u64 mask); |
82 | 82 | ||
83 | #ifdef CONFIG_SWIOTLB | 83 | #ifdef CONFIG_SWIOTLB |
84 | extern void __init swiotlb_free(void); | 84 | extern void __init swiotlb_free(void); |
85 | #else | 85 | #else |
86 | static inline void swiotlb_free(void) { } | 86 | static inline void swiotlb_free(void) { } |
87 | #endif | 87 | #endif |
88 | 88 | ||
89 | extern void swiotlb_print_info(void); | 89 | extern void swiotlb_print_info(void); |
90 | #endif /* __LINUX_SWIOTLB_H */ | 90 | #endif /* __LINUX_SWIOTLB_H */ |
91 | 91 |
lib/swiotlb.c
1 | /* | 1 | /* |
2 | * Dynamic DMA mapping support. | 2 | * Dynamic DMA mapping support. |
3 | * | 3 | * |
4 | * This implementation is a fallback for platforms that do not support | 4 | * This implementation is a fallback for platforms that do not support |
5 | * I/O TLBs (aka DMA address translation hardware). | 5 | * I/O TLBs (aka DMA address translation hardware). |
6 | * Copyright (C) 2000 Asit Mallick <Asit.K.Mallick@intel.com> | 6 | * Copyright (C) 2000 Asit Mallick <Asit.K.Mallick@intel.com> |
7 | * Copyright (C) 2000 Goutham Rao <goutham.rao@intel.com> | 7 | * Copyright (C) 2000 Goutham Rao <goutham.rao@intel.com> |
8 | * Copyright (C) 2000, 2003 Hewlett-Packard Co | 8 | * Copyright (C) 2000, 2003 Hewlett-Packard Co |
9 | * David Mosberger-Tang <davidm@hpl.hp.com> | 9 | * David Mosberger-Tang <davidm@hpl.hp.com> |
10 | * | 10 | * |
11 | * 03/05/07 davidm Switch from PCI-DMA to generic device DMA API. | 11 | * 03/05/07 davidm Switch from PCI-DMA to generic device DMA API. |
12 | * 00/12/13 davidm Rename to swiotlb.c and add mark_clean() to avoid | 12 | * 00/12/13 davidm Rename to swiotlb.c and add mark_clean() to avoid |
13 | * unnecessary i-cache flushing. | 13 | * unnecessary i-cache flushing. |
14 | * 04/07/.. ak Better overflow handling. Assorted fixes. | 14 | * 04/07/.. ak Better overflow handling. Assorted fixes. |
15 | * 05/09/10 linville Add support for syncing ranges, support syncing for | 15 | * 05/09/10 linville Add support for syncing ranges, support syncing for |
16 | * DMA_BIDIRECTIONAL mappings, miscellaneous cleanup. | 16 | * DMA_BIDIRECTIONAL mappings, miscellaneous cleanup. |
17 | * 08/12/11 beckyb Add highmem support | 17 | * 08/12/11 beckyb Add highmem support |
18 | */ | 18 | */ |
19 | 19 | ||
20 | #include <linux/cache.h> | 20 | #include <linux/cache.h> |
21 | #include <linux/dma-mapping.h> | 21 | #include <linux/dma-mapping.h> |
22 | #include <linux/mm.h> | 22 | #include <linux/mm.h> |
23 | #include <linux/module.h> | 23 | #include <linux/module.h> |
24 | #include <linux/spinlock.h> | 24 | #include <linux/spinlock.h> |
25 | #include <linux/string.h> | 25 | #include <linux/string.h> |
26 | #include <linux/swiotlb.h> | 26 | #include <linux/swiotlb.h> |
27 | #include <linux/pfn.h> | 27 | #include <linux/pfn.h> |
28 | #include <linux/types.h> | 28 | #include <linux/types.h> |
29 | #include <linux/ctype.h> | 29 | #include <linux/ctype.h> |
30 | #include <linux/highmem.h> | 30 | #include <linux/highmem.h> |
31 | #include <linux/gfp.h> | 31 | #include <linux/gfp.h> |
32 | 32 | ||
33 | #include <asm/io.h> | 33 | #include <asm/io.h> |
34 | #include <asm/dma.h> | 34 | #include <asm/dma.h> |
35 | #include <asm/scatterlist.h> | 35 | #include <asm/scatterlist.h> |
36 | 36 | ||
37 | #include <linux/init.h> | 37 | #include <linux/init.h> |
38 | #include <linux/bootmem.h> | 38 | #include <linux/bootmem.h> |
39 | #include <linux/iommu-helper.h> | 39 | #include <linux/iommu-helper.h> |
40 | 40 | ||
41 | #define OFFSET(val,align) ((unsigned long) \ | 41 | #define OFFSET(val,align) ((unsigned long) \ |
42 | ( (val) & ( (align) - 1))) | 42 | ( (val) & ( (align) - 1))) |
43 | 43 | ||
44 | #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT)) | 44 | #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT)) |
45 | 45 | ||
46 | /* | 46 | /* |
47 | * Minimum IO TLB size to bother booting with. Systems with mainly | 47 | * Minimum IO TLB size to bother booting with. Systems with mainly |
48 | * 64bit capable cards will only lightly use the swiotlb. If we can't | 48 | * 64bit capable cards will only lightly use the swiotlb. If we can't |
49 | * allocate a contiguous 1MB, we're probably in trouble anyway. | 49 | * allocate a contiguous 1MB, we're probably in trouble anyway. |
50 | */ | 50 | */ |
51 | #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT) | 51 | #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT) |
52 | 52 | ||
53 | /* | 53 | /* |
54 | * Enumeration for sync targets | 54 | * Enumeration for sync targets |
55 | */ | 55 | */ |
56 | enum dma_sync_target { | 56 | enum dma_sync_target { |
57 | SYNC_FOR_CPU = 0, | 57 | SYNC_FOR_CPU = 0, |
58 | SYNC_FOR_DEVICE = 1, | 58 | SYNC_FOR_DEVICE = 1, |
59 | }; | 59 | }; |
60 | 60 | ||
61 | int swiotlb_force; | 61 | int swiotlb_force; |
62 | 62 | ||
63 | /* | 63 | /* |
64 | * Used to do a quick range check in swiotlb_tbl_unmap_single and | 64 | * Used to do a quick range check in swiotlb_tbl_unmap_single and |
65 | * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this | 65 | * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this |
66 | * API. | 66 | * API. |
67 | */ | 67 | */ |
68 | static char *io_tlb_start, *io_tlb_end; | 68 | static char *io_tlb_start, *io_tlb_end; |
69 | 69 | ||
70 | /* | 70 | /* |
71 | * The number of IO TLB blocks (in groups of 64) betweeen io_tlb_start and | 71 | * The number of IO TLB blocks (in groups of 64) betweeen io_tlb_start and |
72 | * io_tlb_end. This is command line adjustable via setup_io_tlb_npages. | 72 | * io_tlb_end. This is command line adjustable via setup_io_tlb_npages. |
73 | */ | 73 | */ |
74 | static unsigned long io_tlb_nslabs; | 74 | static unsigned long io_tlb_nslabs; |
75 | 75 | ||
76 | /* | 76 | /* |
77 | * When the IOMMU overflows we return a fallback buffer. This sets the size. | 77 | * When the IOMMU overflows we return a fallback buffer. This sets the size. |
78 | */ | 78 | */ |
79 | static unsigned long io_tlb_overflow = 32*1024; | 79 | static unsigned long io_tlb_overflow = 32*1024; |
80 | 80 | ||
81 | void *io_tlb_overflow_buffer; | 81 | void *io_tlb_overflow_buffer; |
82 | 82 | ||
83 | /* | 83 | /* |
84 | * This is a free list describing the number of free entries available from | 84 | * This is a free list describing the number of free entries available from |
85 | * each index | 85 | * each index |
86 | */ | 86 | */ |
87 | static unsigned int *io_tlb_list; | 87 | static unsigned int *io_tlb_list; |
88 | static unsigned int io_tlb_index; | 88 | static unsigned int io_tlb_index; |
89 | 89 | ||
90 | /* | 90 | /* |
91 | * We need to save away the original address corresponding to a mapped entry | 91 | * We need to save away the original address corresponding to a mapped entry |
92 | * for the sync operations. | 92 | * for the sync operations. |
93 | */ | 93 | */ |
94 | static phys_addr_t *io_tlb_orig_addr; | 94 | static phys_addr_t *io_tlb_orig_addr; |
95 | 95 | ||
96 | /* | 96 | /* |
97 | * Protect the above data structures in the map and unmap calls | 97 | * Protect the above data structures in the map and unmap calls |
98 | */ | 98 | */ |
99 | static DEFINE_SPINLOCK(io_tlb_lock); | 99 | static DEFINE_SPINLOCK(io_tlb_lock); |
100 | 100 | ||
101 | static int late_alloc; | 101 | static int late_alloc; |
102 | 102 | ||
103 | static int __init | 103 | static int __init |
104 | setup_io_tlb_npages(char *str) | 104 | setup_io_tlb_npages(char *str) |
105 | { | 105 | { |
106 | if (isdigit(*str)) { | 106 | if (isdigit(*str)) { |
107 | io_tlb_nslabs = simple_strtoul(str, &str, 0); | 107 | io_tlb_nslabs = simple_strtoul(str, &str, 0); |
108 | /* avoid tail segment of size < IO_TLB_SEGSIZE */ | 108 | /* avoid tail segment of size < IO_TLB_SEGSIZE */ |
109 | io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE); | 109 | io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE); |
110 | } | 110 | } |
111 | if (*str == ',') | 111 | if (*str == ',') |
112 | ++str; | 112 | ++str; |
113 | if (!strcmp(str, "force")) | 113 | if (!strcmp(str, "force")) |
114 | swiotlb_force = 1; | 114 | swiotlb_force = 1; |
115 | 115 | ||
116 | return 1; | 116 | return 1; |
117 | } | 117 | } |
118 | __setup("swiotlb=", setup_io_tlb_npages); | 118 | __setup("swiotlb=", setup_io_tlb_npages); |
119 | /* make io_tlb_overflow tunable too? */ | 119 | /* make io_tlb_overflow tunable too? */ |
120 | 120 | ||
121 | /* Note that this doesn't work with highmem page */ | 121 | /* Note that this doesn't work with highmem page */ |
122 | static dma_addr_t swiotlb_virt_to_bus(struct device *hwdev, | 122 | static dma_addr_t swiotlb_virt_to_bus(struct device *hwdev, |
123 | volatile void *address) | 123 | volatile void *address) |
124 | { | 124 | { |
125 | return phys_to_dma(hwdev, virt_to_phys(address)); | 125 | return phys_to_dma(hwdev, virt_to_phys(address)); |
126 | } | 126 | } |
127 | 127 | ||
128 | void swiotlb_print_info(void) | 128 | void swiotlb_print_info(void) |
129 | { | 129 | { |
130 | unsigned long bytes = io_tlb_nslabs << IO_TLB_SHIFT; | 130 | unsigned long bytes = io_tlb_nslabs << IO_TLB_SHIFT; |
131 | phys_addr_t pstart, pend; | 131 | phys_addr_t pstart, pend; |
132 | 132 | ||
133 | pstart = virt_to_phys(io_tlb_start); | 133 | pstart = virt_to_phys(io_tlb_start); |
134 | pend = virt_to_phys(io_tlb_end); | 134 | pend = virt_to_phys(io_tlb_end); |
135 | 135 | ||
136 | printk(KERN_INFO "Placing %luMB software IO TLB between %p - %p\n", | 136 | printk(KERN_INFO "Placing %luMB software IO TLB between %p - %p\n", |
137 | bytes >> 20, io_tlb_start, io_tlb_end); | 137 | bytes >> 20, io_tlb_start, io_tlb_end); |
138 | printk(KERN_INFO "software IO TLB at phys %#llx - %#llx\n", | 138 | printk(KERN_INFO "software IO TLB at phys %#llx - %#llx\n", |
139 | (unsigned long long)pstart, | 139 | (unsigned long long)pstart, |
140 | (unsigned long long)pend); | 140 | (unsigned long long)pend); |
141 | } | 141 | } |
142 | 142 | ||
143 | void __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose) | 143 | void __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose) |
144 | { | 144 | { |
145 | unsigned long i, bytes; | 145 | unsigned long i, bytes; |
146 | 146 | ||
147 | bytes = nslabs << IO_TLB_SHIFT; | 147 | bytes = nslabs << IO_TLB_SHIFT; |
148 | 148 | ||
149 | io_tlb_nslabs = nslabs; | 149 | io_tlb_nslabs = nslabs; |
150 | io_tlb_start = tlb; | 150 | io_tlb_start = tlb; |
151 | io_tlb_end = io_tlb_start + bytes; | 151 | io_tlb_end = io_tlb_start + bytes; |
152 | 152 | ||
153 | /* | 153 | /* |
154 | * Allocate and initialize the free list array. This array is used | 154 | * Allocate and initialize the free list array. This array is used |
155 | * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE | 155 | * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE |
156 | * between io_tlb_start and io_tlb_end. | 156 | * between io_tlb_start and io_tlb_end. |
157 | */ | 157 | */ |
158 | io_tlb_list = alloc_bootmem(io_tlb_nslabs * sizeof(int)); | 158 | io_tlb_list = alloc_bootmem(io_tlb_nslabs * sizeof(int)); |
159 | for (i = 0; i < io_tlb_nslabs; i++) | 159 | for (i = 0; i < io_tlb_nslabs; i++) |
160 | io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE); | 160 | io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE); |
161 | io_tlb_index = 0; | 161 | io_tlb_index = 0; |
162 | io_tlb_orig_addr = alloc_bootmem(io_tlb_nslabs * sizeof(phys_addr_t)); | 162 | io_tlb_orig_addr = alloc_bootmem(io_tlb_nslabs * sizeof(phys_addr_t)); |
163 | 163 | ||
164 | /* | 164 | /* |
165 | * Get the overflow emergency buffer | 165 | * Get the overflow emergency buffer |
166 | */ | 166 | */ |
167 | io_tlb_overflow_buffer = alloc_bootmem_low(io_tlb_overflow); | 167 | io_tlb_overflow_buffer = alloc_bootmem_low(io_tlb_overflow); |
168 | if (!io_tlb_overflow_buffer) | 168 | if (!io_tlb_overflow_buffer) |
169 | panic("Cannot allocate SWIOTLB overflow buffer!\n"); | 169 | panic("Cannot allocate SWIOTLB overflow buffer!\n"); |
170 | if (verbose) | 170 | if (verbose) |
171 | swiotlb_print_info(); | 171 | swiotlb_print_info(); |
172 | } | 172 | } |
173 | 173 | ||
174 | /* | 174 | /* |
175 | * Statically reserve bounce buffer space and initialize bounce buffer data | 175 | * Statically reserve bounce buffer space and initialize bounce buffer data |
176 | * structures for the software IO TLB used to implement the DMA API. | 176 | * structures for the software IO TLB used to implement the DMA API. |
177 | */ | 177 | */ |
178 | void __init | 178 | void __init |
179 | swiotlb_init_with_default_size(size_t default_size, int verbose) | 179 | swiotlb_init_with_default_size(size_t default_size, int verbose) |
180 | { | 180 | { |
181 | unsigned long bytes; | 181 | unsigned long bytes; |
182 | 182 | ||
183 | if (!io_tlb_nslabs) { | 183 | if (!io_tlb_nslabs) { |
184 | io_tlb_nslabs = (default_size >> IO_TLB_SHIFT); | 184 | io_tlb_nslabs = (default_size >> IO_TLB_SHIFT); |
185 | io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE); | 185 | io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE); |
186 | } | 186 | } |
187 | 187 | ||
188 | bytes = io_tlb_nslabs << IO_TLB_SHIFT; | 188 | bytes = io_tlb_nslabs << IO_TLB_SHIFT; |
189 | 189 | ||
190 | /* | 190 | /* |
191 | * Get IO TLB memory from the low pages | 191 | * Get IO TLB memory from the low pages |
192 | */ | 192 | */ |
193 | io_tlb_start = alloc_bootmem_low_pages(bytes); | 193 | io_tlb_start = alloc_bootmem_low_pages(bytes); |
194 | if (!io_tlb_start) | 194 | if (!io_tlb_start) |
195 | panic("Cannot allocate SWIOTLB buffer"); | 195 | panic("Cannot allocate SWIOTLB buffer"); |
196 | 196 | ||
197 | swiotlb_init_with_tbl(io_tlb_start, io_tlb_nslabs, verbose); | 197 | swiotlb_init_with_tbl(io_tlb_start, io_tlb_nslabs, verbose); |
198 | } | 198 | } |
199 | 199 | ||
200 | void __init | 200 | void __init |
201 | swiotlb_init(int verbose) | 201 | swiotlb_init(int verbose) |
202 | { | 202 | { |
203 | swiotlb_init_with_default_size(64 * (1<<20), verbose); /* default to 64MB */ | 203 | swiotlb_init_with_default_size(64 * (1<<20), verbose); /* default to 64MB */ |
204 | } | 204 | } |
205 | 205 | ||
206 | /* | 206 | /* |
207 | * Systems with larger DMA zones (those that don't support ISA) can | 207 | * Systems with larger DMA zones (those that don't support ISA) can |
208 | * initialize the swiotlb later using the slab allocator if needed. | 208 | * initialize the swiotlb later using the slab allocator if needed. |
209 | * This should be just like above, but with some error catching. | 209 | * This should be just like above, but with some error catching. |
210 | */ | 210 | */ |
211 | int | 211 | int |
212 | swiotlb_late_init_with_default_size(size_t default_size) | 212 | swiotlb_late_init_with_default_size(size_t default_size) |
213 | { | 213 | { |
214 | unsigned long i, bytes, req_nslabs = io_tlb_nslabs; | 214 | unsigned long i, bytes, req_nslabs = io_tlb_nslabs; |
215 | unsigned int order; | 215 | unsigned int order; |
216 | 216 | ||
217 | if (!io_tlb_nslabs) { | 217 | if (!io_tlb_nslabs) { |
218 | io_tlb_nslabs = (default_size >> IO_TLB_SHIFT); | 218 | io_tlb_nslabs = (default_size >> IO_TLB_SHIFT); |
219 | io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE); | 219 | io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE); |
220 | } | 220 | } |
221 | 221 | ||
222 | /* | 222 | /* |
223 | * Get IO TLB memory from the low pages | 223 | * Get IO TLB memory from the low pages |
224 | */ | 224 | */ |
225 | order = get_order(io_tlb_nslabs << IO_TLB_SHIFT); | 225 | order = get_order(io_tlb_nslabs << IO_TLB_SHIFT); |
226 | io_tlb_nslabs = SLABS_PER_PAGE << order; | 226 | io_tlb_nslabs = SLABS_PER_PAGE << order; |
227 | bytes = io_tlb_nslabs << IO_TLB_SHIFT; | 227 | bytes = io_tlb_nslabs << IO_TLB_SHIFT; |
228 | 228 | ||
229 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { | 229 | while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { |
230 | io_tlb_start = (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN, | 230 | io_tlb_start = (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN, |
231 | order); | 231 | order); |
232 | if (io_tlb_start) | 232 | if (io_tlb_start) |
233 | break; | 233 | break; |
234 | order--; | 234 | order--; |
235 | } | 235 | } |
236 | 236 | ||
237 | if (!io_tlb_start) | 237 | if (!io_tlb_start) |
238 | goto cleanup1; | 238 | goto cleanup1; |
239 | 239 | ||
240 | if (order != get_order(bytes)) { | 240 | if (order != get_order(bytes)) { |
241 | printk(KERN_WARNING "Warning: only able to allocate %ld MB " | 241 | printk(KERN_WARNING "Warning: only able to allocate %ld MB " |
242 | "for software IO TLB\n", (PAGE_SIZE << order) >> 20); | 242 | "for software IO TLB\n", (PAGE_SIZE << order) >> 20); |
243 | io_tlb_nslabs = SLABS_PER_PAGE << order; | 243 | io_tlb_nslabs = SLABS_PER_PAGE << order; |
244 | bytes = io_tlb_nslabs << IO_TLB_SHIFT; | 244 | bytes = io_tlb_nslabs << IO_TLB_SHIFT; |
245 | } | 245 | } |
246 | io_tlb_end = io_tlb_start + bytes; | 246 | io_tlb_end = io_tlb_start + bytes; |
247 | memset(io_tlb_start, 0, bytes); | 247 | memset(io_tlb_start, 0, bytes); |
248 | 248 | ||
249 | /* | 249 | /* |
250 | * Allocate and initialize the free list array. This array is used | 250 | * Allocate and initialize the free list array. This array is used |
251 | * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE | 251 | * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE |
252 | * between io_tlb_start and io_tlb_end. | 252 | * between io_tlb_start and io_tlb_end. |
253 | */ | 253 | */ |
254 | io_tlb_list = (unsigned int *)__get_free_pages(GFP_KERNEL, | 254 | io_tlb_list = (unsigned int *)__get_free_pages(GFP_KERNEL, |
255 | get_order(io_tlb_nslabs * sizeof(int))); | 255 | get_order(io_tlb_nslabs * sizeof(int))); |
256 | if (!io_tlb_list) | 256 | if (!io_tlb_list) |
257 | goto cleanup2; | 257 | goto cleanup2; |
258 | 258 | ||
259 | for (i = 0; i < io_tlb_nslabs; i++) | 259 | for (i = 0; i < io_tlb_nslabs; i++) |
260 | io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE); | 260 | io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE); |
261 | io_tlb_index = 0; | 261 | io_tlb_index = 0; |
262 | 262 | ||
263 | io_tlb_orig_addr = (phys_addr_t *) | 263 | io_tlb_orig_addr = (phys_addr_t *) |
264 | __get_free_pages(GFP_KERNEL, | 264 | __get_free_pages(GFP_KERNEL, |
265 | get_order(io_tlb_nslabs * | 265 | get_order(io_tlb_nslabs * |
266 | sizeof(phys_addr_t))); | 266 | sizeof(phys_addr_t))); |
267 | if (!io_tlb_orig_addr) | 267 | if (!io_tlb_orig_addr) |
268 | goto cleanup3; | 268 | goto cleanup3; |
269 | 269 | ||
270 | memset(io_tlb_orig_addr, 0, io_tlb_nslabs * sizeof(phys_addr_t)); | 270 | memset(io_tlb_orig_addr, 0, io_tlb_nslabs * sizeof(phys_addr_t)); |
271 | 271 | ||
272 | /* | 272 | /* |
273 | * Get the overflow emergency buffer | 273 | * Get the overflow emergency buffer |
274 | */ | 274 | */ |
275 | io_tlb_overflow_buffer = (void *)__get_free_pages(GFP_DMA, | 275 | io_tlb_overflow_buffer = (void *)__get_free_pages(GFP_DMA, |
276 | get_order(io_tlb_overflow)); | 276 | get_order(io_tlb_overflow)); |
277 | if (!io_tlb_overflow_buffer) | 277 | if (!io_tlb_overflow_buffer) |
278 | goto cleanup4; | 278 | goto cleanup4; |
279 | 279 | ||
280 | swiotlb_print_info(); | 280 | swiotlb_print_info(); |
281 | 281 | ||
282 | late_alloc = 1; | 282 | late_alloc = 1; |
283 | 283 | ||
284 | return 0; | 284 | return 0; |
285 | 285 | ||
286 | cleanup4: | 286 | cleanup4: |
287 | free_pages((unsigned long)io_tlb_orig_addr, | 287 | free_pages((unsigned long)io_tlb_orig_addr, |
288 | get_order(io_tlb_nslabs * sizeof(phys_addr_t))); | 288 | get_order(io_tlb_nslabs * sizeof(phys_addr_t))); |
289 | io_tlb_orig_addr = NULL; | 289 | io_tlb_orig_addr = NULL; |
290 | cleanup3: | 290 | cleanup3: |
291 | free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * | 291 | free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * |
292 | sizeof(int))); | 292 | sizeof(int))); |
293 | io_tlb_list = NULL; | 293 | io_tlb_list = NULL; |
294 | cleanup2: | 294 | cleanup2: |
295 | io_tlb_end = NULL; | 295 | io_tlb_end = NULL; |
296 | free_pages((unsigned long)io_tlb_start, order); | 296 | free_pages((unsigned long)io_tlb_start, order); |
297 | io_tlb_start = NULL; | 297 | io_tlb_start = NULL; |
298 | cleanup1: | 298 | cleanup1: |
299 | io_tlb_nslabs = req_nslabs; | 299 | io_tlb_nslabs = req_nslabs; |
300 | return -ENOMEM; | 300 | return -ENOMEM; |
301 | } | 301 | } |
302 | 302 | ||
303 | void __init swiotlb_free(void) | 303 | void __init swiotlb_free(void) |
304 | { | 304 | { |
305 | if (!io_tlb_overflow_buffer) | 305 | if (!io_tlb_overflow_buffer) |
306 | return; | 306 | return; |
307 | 307 | ||
308 | if (late_alloc) { | 308 | if (late_alloc) { |
309 | free_pages((unsigned long)io_tlb_overflow_buffer, | 309 | free_pages((unsigned long)io_tlb_overflow_buffer, |
310 | get_order(io_tlb_overflow)); | 310 | get_order(io_tlb_overflow)); |
311 | free_pages((unsigned long)io_tlb_orig_addr, | 311 | free_pages((unsigned long)io_tlb_orig_addr, |
312 | get_order(io_tlb_nslabs * sizeof(phys_addr_t))); | 312 | get_order(io_tlb_nslabs * sizeof(phys_addr_t))); |
313 | free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * | 313 | free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * |
314 | sizeof(int))); | 314 | sizeof(int))); |
315 | free_pages((unsigned long)io_tlb_start, | 315 | free_pages((unsigned long)io_tlb_start, |
316 | get_order(io_tlb_nslabs << IO_TLB_SHIFT)); | 316 | get_order(io_tlb_nslabs << IO_TLB_SHIFT)); |
317 | } else { | 317 | } else { |
318 | free_bootmem_late(__pa(io_tlb_overflow_buffer), | 318 | free_bootmem_late(__pa(io_tlb_overflow_buffer), |
319 | io_tlb_overflow); | 319 | io_tlb_overflow); |
320 | free_bootmem_late(__pa(io_tlb_orig_addr), | 320 | free_bootmem_late(__pa(io_tlb_orig_addr), |
321 | io_tlb_nslabs * sizeof(phys_addr_t)); | 321 | io_tlb_nslabs * sizeof(phys_addr_t)); |
322 | free_bootmem_late(__pa(io_tlb_list), | 322 | free_bootmem_late(__pa(io_tlb_list), |
323 | io_tlb_nslabs * sizeof(int)); | 323 | io_tlb_nslabs * sizeof(int)); |
324 | free_bootmem_late(__pa(io_tlb_start), | 324 | free_bootmem_late(__pa(io_tlb_start), |
325 | io_tlb_nslabs << IO_TLB_SHIFT); | 325 | io_tlb_nslabs << IO_TLB_SHIFT); |
326 | } | 326 | } |
327 | } | 327 | } |
328 | 328 | ||
329 | static int is_swiotlb_buffer(phys_addr_t paddr) | 329 | static int is_swiotlb_buffer(phys_addr_t paddr) |
330 | { | 330 | { |
331 | return paddr >= virt_to_phys(io_tlb_start) && | 331 | return paddr >= virt_to_phys(io_tlb_start) && |
332 | paddr < virt_to_phys(io_tlb_end); | 332 | paddr < virt_to_phys(io_tlb_end); |
333 | } | 333 | } |
334 | 334 | ||
335 | /* | 335 | /* |
336 | * Bounce: copy the swiotlb buffer back to the original dma location | 336 | * Bounce: copy the swiotlb buffer back to the original dma location |
337 | */ | 337 | */ |
338 | static void swiotlb_bounce(phys_addr_t phys, char *dma_addr, size_t size, | 338 | static void swiotlb_bounce(phys_addr_t phys, char *dma_addr, size_t size, |
339 | enum dma_data_direction dir) | 339 | enum dma_data_direction dir) |
340 | { | 340 | { |
341 | unsigned long pfn = PFN_DOWN(phys); | 341 | unsigned long pfn = PFN_DOWN(phys); |
342 | 342 | ||
343 | if (PageHighMem(pfn_to_page(pfn))) { | 343 | if (PageHighMem(pfn_to_page(pfn))) { |
344 | /* The buffer does not have a mapping. Map it in and copy */ | 344 | /* The buffer does not have a mapping. Map it in and copy */ |
345 | unsigned int offset = phys & ~PAGE_MASK; | 345 | unsigned int offset = phys & ~PAGE_MASK; |
346 | char *buffer; | 346 | char *buffer; |
347 | unsigned int sz = 0; | 347 | unsigned int sz = 0; |
348 | unsigned long flags; | 348 | unsigned long flags; |
349 | 349 | ||
350 | while (size) { | 350 | while (size) { |
351 | sz = min_t(size_t, PAGE_SIZE - offset, size); | 351 | sz = min_t(size_t, PAGE_SIZE - offset, size); |
352 | 352 | ||
353 | local_irq_save(flags); | 353 | local_irq_save(flags); |
354 | buffer = kmap_atomic(pfn_to_page(pfn), | 354 | buffer = kmap_atomic(pfn_to_page(pfn), |
355 | KM_BOUNCE_READ); | 355 | KM_BOUNCE_READ); |
356 | if (dir == DMA_TO_DEVICE) | 356 | if (dir == DMA_TO_DEVICE) |
357 | memcpy(dma_addr, buffer + offset, sz); | 357 | memcpy(dma_addr, buffer + offset, sz); |
358 | else | 358 | else |
359 | memcpy(buffer + offset, dma_addr, sz); | 359 | memcpy(buffer + offset, dma_addr, sz); |
360 | kunmap_atomic(buffer, KM_BOUNCE_READ); | 360 | kunmap_atomic(buffer, KM_BOUNCE_READ); |
361 | local_irq_restore(flags); | 361 | local_irq_restore(flags); |
362 | 362 | ||
363 | size -= sz; | 363 | size -= sz; |
364 | pfn++; | 364 | pfn++; |
365 | dma_addr += sz; | 365 | dma_addr += sz; |
366 | offset = 0; | 366 | offset = 0; |
367 | } | 367 | } |
368 | } else { | 368 | } else { |
369 | if (dir == DMA_TO_DEVICE) | 369 | if (dir == DMA_TO_DEVICE) |
370 | memcpy(dma_addr, phys_to_virt(phys), size); | 370 | memcpy(dma_addr, phys_to_virt(phys), size); |
371 | else | 371 | else |
372 | memcpy(phys_to_virt(phys), dma_addr, size); | 372 | memcpy(phys_to_virt(phys), dma_addr, size); |
373 | } | 373 | } |
374 | } | 374 | } |
375 | 375 | ||
376 | void *swiotlb_tbl_map_single(struct device *hwdev, dma_addr_t tbl_dma_addr, | 376 | void *swiotlb_tbl_map_single(struct device *hwdev, dma_addr_t tbl_dma_addr, |
377 | phys_addr_t phys, size_t size, int dir) | 377 | phys_addr_t phys, size_t size, |
378 | enum dma_data_direction dir) | ||
378 | { | 379 | { |
379 | unsigned long flags; | 380 | unsigned long flags; |
380 | char *dma_addr; | 381 | char *dma_addr; |
381 | unsigned int nslots, stride, index, wrap; | 382 | unsigned int nslots, stride, index, wrap; |
382 | int i; | 383 | int i; |
383 | unsigned long mask; | 384 | unsigned long mask; |
384 | unsigned long offset_slots; | 385 | unsigned long offset_slots; |
385 | unsigned long max_slots; | 386 | unsigned long max_slots; |
386 | 387 | ||
387 | mask = dma_get_seg_boundary(hwdev); | 388 | mask = dma_get_seg_boundary(hwdev); |
388 | 389 | ||
389 | tbl_dma_addr &= mask; | 390 | tbl_dma_addr &= mask; |
390 | 391 | ||
391 | offset_slots = ALIGN(tbl_dma_addr, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; | 392 | offset_slots = ALIGN(tbl_dma_addr, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; |
392 | 393 | ||
393 | /* | 394 | /* |
394 | * Carefully handle integer overflow which can occur when mask == ~0UL. | 395 | * Carefully handle integer overflow which can occur when mask == ~0UL. |
395 | */ | 396 | */ |
396 | max_slots = mask + 1 | 397 | max_slots = mask + 1 |
397 | ? ALIGN(mask + 1, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT | 398 | ? ALIGN(mask + 1, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT |
398 | : 1UL << (BITS_PER_LONG - IO_TLB_SHIFT); | 399 | : 1UL << (BITS_PER_LONG - IO_TLB_SHIFT); |
399 | 400 | ||
400 | /* | 401 | /* |
401 | * For mappings greater than a page, we limit the stride (and | 402 | * For mappings greater than a page, we limit the stride (and |
402 | * hence alignment) to a page size. | 403 | * hence alignment) to a page size. |
403 | */ | 404 | */ |
404 | nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; | 405 | nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; |
405 | if (size > PAGE_SIZE) | 406 | if (size > PAGE_SIZE) |
406 | stride = (1 << (PAGE_SHIFT - IO_TLB_SHIFT)); | 407 | stride = (1 << (PAGE_SHIFT - IO_TLB_SHIFT)); |
407 | else | 408 | else |
408 | stride = 1; | 409 | stride = 1; |
409 | 410 | ||
410 | BUG_ON(!nslots); | 411 | BUG_ON(!nslots); |
411 | 412 | ||
412 | /* | 413 | /* |
413 | * Find suitable number of IO TLB entries size that will fit this | 414 | * Find suitable number of IO TLB entries size that will fit this |
414 | * request and allocate a buffer from that IO TLB pool. | 415 | * request and allocate a buffer from that IO TLB pool. |
415 | */ | 416 | */ |
416 | spin_lock_irqsave(&io_tlb_lock, flags); | 417 | spin_lock_irqsave(&io_tlb_lock, flags); |
417 | index = ALIGN(io_tlb_index, stride); | 418 | index = ALIGN(io_tlb_index, stride); |
418 | if (index >= io_tlb_nslabs) | 419 | if (index >= io_tlb_nslabs) |
419 | index = 0; | 420 | index = 0; |
420 | wrap = index; | 421 | wrap = index; |
421 | 422 | ||
422 | do { | 423 | do { |
423 | while (iommu_is_span_boundary(index, nslots, offset_slots, | 424 | while (iommu_is_span_boundary(index, nslots, offset_slots, |
424 | max_slots)) { | 425 | max_slots)) { |
425 | index += stride; | 426 | index += stride; |
426 | if (index >= io_tlb_nslabs) | 427 | if (index >= io_tlb_nslabs) |
427 | index = 0; | 428 | index = 0; |
428 | if (index == wrap) | 429 | if (index == wrap) |
429 | goto not_found; | 430 | goto not_found; |
430 | } | 431 | } |
431 | 432 | ||
432 | /* | 433 | /* |
433 | * If we find a slot that indicates we have 'nslots' number of | 434 | * If we find a slot that indicates we have 'nslots' number of |
434 | * contiguous buffers, we allocate the buffers from that slot | 435 | * contiguous buffers, we allocate the buffers from that slot |
435 | * and mark the entries as '0' indicating unavailable. | 436 | * and mark the entries as '0' indicating unavailable. |
436 | */ | 437 | */ |
437 | if (io_tlb_list[index] >= nslots) { | 438 | if (io_tlb_list[index] >= nslots) { |
438 | int count = 0; | 439 | int count = 0; |
439 | 440 | ||
440 | for (i = index; i < (int) (index + nslots); i++) | 441 | for (i = index; i < (int) (index + nslots); i++) |
441 | io_tlb_list[i] = 0; | 442 | io_tlb_list[i] = 0; |
442 | for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE - 1) && io_tlb_list[i]; i--) | 443 | for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE - 1) && io_tlb_list[i]; i--) |
443 | io_tlb_list[i] = ++count; | 444 | io_tlb_list[i] = ++count; |
444 | dma_addr = io_tlb_start + (index << IO_TLB_SHIFT); | 445 | dma_addr = io_tlb_start + (index << IO_TLB_SHIFT); |
445 | 446 | ||
446 | /* | 447 | /* |
447 | * Update the indices to avoid searching in the next | 448 | * Update the indices to avoid searching in the next |
448 | * round. | 449 | * round. |
449 | */ | 450 | */ |
450 | io_tlb_index = ((index + nslots) < io_tlb_nslabs | 451 | io_tlb_index = ((index + nslots) < io_tlb_nslabs |
451 | ? (index + nslots) : 0); | 452 | ? (index + nslots) : 0); |
452 | 453 | ||
453 | goto found; | 454 | goto found; |
454 | } | 455 | } |
455 | index += stride; | 456 | index += stride; |
456 | if (index >= io_tlb_nslabs) | 457 | if (index >= io_tlb_nslabs) |
457 | index = 0; | 458 | index = 0; |
458 | } while (index != wrap); | 459 | } while (index != wrap); |
459 | 460 | ||
460 | not_found: | 461 | not_found: |
461 | spin_unlock_irqrestore(&io_tlb_lock, flags); | 462 | spin_unlock_irqrestore(&io_tlb_lock, flags); |
462 | return NULL; | 463 | return NULL; |
463 | found: | 464 | found: |
464 | spin_unlock_irqrestore(&io_tlb_lock, flags); | 465 | spin_unlock_irqrestore(&io_tlb_lock, flags); |
465 | 466 | ||
466 | /* | 467 | /* |
467 | * Save away the mapping from the original address to the DMA address. | 468 | * Save away the mapping from the original address to the DMA address. |
468 | * This is needed when we sync the memory. Then we sync the buffer if | 469 | * This is needed when we sync the memory. Then we sync the buffer if |
469 | * needed. | 470 | * needed. |
470 | */ | 471 | */ |
471 | for (i = 0; i < nslots; i++) | 472 | for (i = 0; i < nslots; i++) |
472 | io_tlb_orig_addr[index+i] = phys + (i << IO_TLB_SHIFT); | 473 | io_tlb_orig_addr[index+i] = phys + (i << IO_TLB_SHIFT); |
473 | if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) | 474 | if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) |
474 | swiotlb_bounce(phys, dma_addr, size, DMA_TO_DEVICE); | 475 | swiotlb_bounce(phys, dma_addr, size, DMA_TO_DEVICE); |
475 | 476 | ||
476 | return dma_addr; | 477 | return dma_addr; |
477 | } | 478 | } |
478 | 479 | ||
479 | /* | 480 | /* |
480 | * Allocates bounce buffer and returns its kernel virtual address. | 481 | * Allocates bounce buffer and returns its kernel virtual address. |
481 | */ | 482 | */ |
482 | 483 | ||
483 | static void * | 484 | static void * |
484 | map_single(struct device *hwdev, phys_addr_t phys, size_t size, int dir) | 485 | map_single(struct device *hwdev, phys_addr_t phys, size_t size, |
486 | enum dma_data_direction dir) | ||
485 | { | 487 | { |
486 | dma_addr_t start_dma_addr = swiotlb_virt_to_bus(hwdev, io_tlb_start); | 488 | dma_addr_t start_dma_addr = swiotlb_virt_to_bus(hwdev, io_tlb_start); |
487 | 489 | ||
488 | return swiotlb_tbl_map_single(hwdev, start_dma_addr, phys, size, dir); | 490 | return swiotlb_tbl_map_single(hwdev, start_dma_addr, phys, size, dir); |
489 | } | 491 | } |
490 | 492 | ||
491 | /* | 493 | /* |
492 | * dma_addr is the kernel virtual address of the bounce buffer to unmap. | 494 | * dma_addr is the kernel virtual address of the bounce buffer to unmap. |
493 | */ | 495 | */ |
494 | static void | 496 | static void |
495 | swiotlb_tbl_unmap_single(struct device *hwdev, char *dma_addr, size_t size, | 497 | swiotlb_tbl_unmap_single(struct device *hwdev, char *dma_addr, size_t size, |
496 | int dir) | 498 | enum dma_data_direction dir) |
497 | { | 499 | { |
498 | unsigned long flags; | 500 | unsigned long flags; |
499 | int i, count, nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; | 501 | int i, count, nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; |
500 | int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; | 502 | int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; |
501 | phys_addr_t phys = io_tlb_orig_addr[index]; | 503 | phys_addr_t phys = io_tlb_orig_addr[index]; |
502 | 504 | ||
503 | /* | 505 | /* |
504 | * First, sync the memory before unmapping the entry | 506 | * First, sync the memory before unmapping the entry |
505 | */ | 507 | */ |
506 | if (phys && ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL))) | 508 | if (phys && ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL))) |
507 | swiotlb_bounce(phys, dma_addr, size, DMA_FROM_DEVICE); | 509 | swiotlb_bounce(phys, dma_addr, size, DMA_FROM_DEVICE); |
508 | 510 | ||
509 | /* | 511 | /* |
510 | * Return the buffer to the free list by setting the corresponding | 512 | * Return the buffer to the free list by setting the corresponding |
511 | * entries to indicate the number of contiguous entries available. | 513 | * entries to indicate the number of contiguous entries available. |
512 | * While returning the entries to the free list, we merge the entries | 514 | * While returning the entries to the free list, we merge the entries |
513 | * with slots below and above the pool being returned. | 515 | * with slots below and above the pool being returned. |
514 | */ | 516 | */ |
515 | spin_lock_irqsave(&io_tlb_lock, flags); | 517 | spin_lock_irqsave(&io_tlb_lock, flags); |
516 | { | 518 | { |
517 | count = ((index + nslots) < ALIGN(index + 1, IO_TLB_SEGSIZE) ? | 519 | count = ((index + nslots) < ALIGN(index + 1, IO_TLB_SEGSIZE) ? |
518 | io_tlb_list[index + nslots] : 0); | 520 | io_tlb_list[index + nslots] : 0); |
519 | /* | 521 | /* |
520 | * Step 1: return the slots to the free list, merging the | 522 | * Step 1: return the slots to the free list, merging the |
521 | * slots with superceeding slots | 523 | * slots with superceeding slots |
522 | */ | 524 | */ |
523 | for (i = index + nslots - 1; i >= index; i--) | 525 | for (i = index + nslots - 1; i >= index; i--) |
524 | io_tlb_list[i] = ++count; | 526 | io_tlb_list[i] = ++count; |
525 | /* | 527 | /* |
526 | * Step 2: merge the returned slots with the preceding slots, | 528 | * Step 2: merge the returned slots with the preceding slots, |
527 | * if available (non zero) | 529 | * if available (non zero) |
528 | */ | 530 | */ |
529 | for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE -1) && io_tlb_list[i]; i--) | 531 | for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE -1) && io_tlb_list[i]; i--) |
530 | io_tlb_list[i] = ++count; | 532 | io_tlb_list[i] = ++count; |
531 | } | 533 | } |
532 | spin_unlock_irqrestore(&io_tlb_lock, flags); | 534 | spin_unlock_irqrestore(&io_tlb_lock, flags); |
533 | } | 535 | } |
534 | 536 | ||
535 | static void | 537 | static void |
536 | swiotlb_tbl_sync_single(struct device *hwdev, char *dma_addr, size_t size, | 538 | swiotlb_tbl_sync_single(struct device *hwdev, char *dma_addr, size_t size, |
537 | int dir, int target) | 539 | enum dma_data_direction dir, int target) |
538 | { | 540 | { |
539 | int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; | 541 | int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; |
540 | phys_addr_t phys = io_tlb_orig_addr[index]; | 542 | phys_addr_t phys = io_tlb_orig_addr[index]; |
541 | 543 | ||
542 | phys += ((unsigned long)dma_addr & ((1 << IO_TLB_SHIFT) - 1)); | 544 | phys += ((unsigned long)dma_addr & ((1 << IO_TLB_SHIFT) - 1)); |
543 | 545 | ||
544 | switch (target) { | 546 | switch (target) { |
545 | case SYNC_FOR_CPU: | 547 | case SYNC_FOR_CPU: |
546 | if (likely(dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)) | 548 | if (likely(dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)) |
547 | swiotlb_bounce(phys, dma_addr, size, DMA_FROM_DEVICE); | 549 | swiotlb_bounce(phys, dma_addr, size, DMA_FROM_DEVICE); |
548 | else | 550 | else |
549 | BUG_ON(dir != DMA_TO_DEVICE); | 551 | BUG_ON(dir != DMA_TO_DEVICE); |
550 | break; | 552 | break; |
551 | case SYNC_FOR_DEVICE: | 553 | case SYNC_FOR_DEVICE: |
552 | if (likely(dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)) | 554 | if (likely(dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)) |
553 | swiotlb_bounce(phys, dma_addr, size, DMA_TO_DEVICE); | 555 | swiotlb_bounce(phys, dma_addr, size, DMA_TO_DEVICE); |
554 | else | 556 | else |
555 | BUG_ON(dir != DMA_FROM_DEVICE); | 557 | BUG_ON(dir != DMA_FROM_DEVICE); |
556 | break; | 558 | break; |
557 | default: | 559 | default: |
558 | BUG(); | 560 | BUG(); |
559 | } | 561 | } |
560 | } | 562 | } |
561 | 563 | ||
562 | void * | 564 | void * |
563 | swiotlb_alloc_coherent(struct device *hwdev, size_t size, | 565 | swiotlb_alloc_coherent(struct device *hwdev, size_t size, |
564 | dma_addr_t *dma_handle, gfp_t flags) | 566 | dma_addr_t *dma_handle, gfp_t flags) |
565 | { | 567 | { |
566 | dma_addr_t dev_addr; | 568 | dma_addr_t dev_addr; |
567 | void *ret; | 569 | void *ret; |
568 | int order = get_order(size); | 570 | int order = get_order(size); |
569 | u64 dma_mask = DMA_BIT_MASK(32); | 571 | u64 dma_mask = DMA_BIT_MASK(32); |
570 | 572 | ||
571 | if (hwdev && hwdev->coherent_dma_mask) | 573 | if (hwdev && hwdev->coherent_dma_mask) |
572 | dma_mask = hwdev->coherent_dma_mask; | 574 | dma_mask = hwdev->coherent_dma_mask; |
573 | 575 | ||
574 | ret = (void *)__get_free_pages(flags, order); | 576 | ret = (void *)__get_free_pages(flags, order); |
575 | if (ret && swiotlb_virt_to_bus(hwdev, ret) + size - 1 > dma_mask) { | 577 | if (ret && swiotlb_virt_to_bus(hwdev, ret) + size - 1 > dma_mask) { |
576 | /* | 578 | /* |
577 | * The allocated memory isn't reachable by the device. | 579 | * The allocated memory isn't reachable by the device. |
578 | */ | 580 | */ |
579 | free_pages((unsigned long) ret, order); | 581 | free_pages((unsigned long) ret, order); |
580 | ret = NULL; | 582 | ret = NULL; |
581 | } | 583 | } |
582 | if (!ret) { | 584 | if (!ret) { |
583 | /* | 585 | /* |
584 | * We are either out of memory or the device can't DMA to | 586 | * We are either out of memory or the device can't DMA to |
585 | * GFP_DMA memory; fall back on map_single(), which | 587 | * GFP_DMA memory; fall back on map_single(), which |
586 | * will grab memory from the lowest available address range. | 588 | * will grab memory from the lowest available address range. |
587 | */ | 589 | */ |
588 | ret = map_single(hwdev, 0, size, DMA_FROM_DEVICE); | 590 | ret = map_single(hwdev, 0, size, DMA_FROM_DEVICE); |
589 | if (!ret) | 591 | if (!ret) |
590 | return NULL; | 592 | return NULL; |
591 | } | 593 | } |
592 | 594 | ||
593 | memset(ret, 0, size); | 595 | memset(ret, 0, size); |
594 | dev_addr = swiotlb_virt_to_bus(hwdev, ret); | 596 | dev_addr = swiotlb_virt_to_bus(hwdev, ret); |
595 | 597 | ||
596 | /* Confirm address can be DMA'd by device */ | 598 | /* Confirm address can be DMA'd by device */ |
597 | if (dev_addr + size - 1 > dma_mask) { | 599 | if (dev_addr + size - 1 > dma_mask) { |
598 | printk("hwdev DMA mask = 0x%016Lx, dev_addr = 0x%016Lx\n", | 600 | printk("hwdev DMA mask = 0x%016Lx, dev_addr = 0x%016Lx\n", |
599 | (unsigned long long)dma_mask, | 601 | (unsigned long long)dma_mask, |
600 | (unsigned long long)dev_addr); | 602 | (unsigned long long)dev_addr); |
601 | 603 | ||
602 | /* DMA_TO_DEVICE to avoid memcpy in unmap_single */ | 604 | /* DMA_TO_DEVICE to avoid memcpy in unmap_single */ |
603 | swiotlb_tbl_unmap_single(hwdev, ret, size, DMA_TO_DEVICE); | 605 | swiotlb_tbl_unmap_single(hwdev, ret, size, DMA_TO_DEVICE); |
604 | return NULL; | 606 | return NULL; |
605 | } | 607 | } |
606 | *dma_handle = dev_addr; | 608 | *dma_handle = dev_addr; |
607 | return ret; | 609 | return ret; |
608 | } | 610 | } |
609 | EXPORT_SYMBOL(swiotlb_alloc_coherent); | 611 | EXPORT_SYMBOL(swiotlb_alloc_coherent); |
610 | 612 | ||
611 | void | 613 | void |
612 | swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, | 614 | swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, |
613 | dma_addr_t dev_addr) | 615 | dma_addr_t dev_addr) |
614 | { | 616 | { |
615 | phys_addr_t paddr = dma_to_phys(hwdev, dev_addr); | 617 | phys_addr_t paddr = dma_to_phys(hwdev, dev_addr); |
616 | 618 | ||
617 | WARN_ON(irqs_disabled()); | 619 | WARN_ON(irqs_disabled()); |
618 | if (!is_swiotlb_buffer(paddr)) | 620 | if (!is_swiotlb_buffer(paddr)) |
619 | free_pages((unsigned long)vaddr, get_order(size)); | 621 | free_pages((unsigned long)vaddr, get_order(size)); |
620 | else | 622 | else |
621 | /* DMA_TO_DEVICE to avoid memcpy in swiotlb_tbl_unmap_single */ | 623 | /* DMA_TO_DEVICE to avoid memcpy in swiotlb_tbl_unmap_single */ |
622 | swiotlb_tbl_unmap_single(hwdev, vaddr, size, DMA_TO_DEVICE); | 624 | swiotlb_tbl_unmap_single(hwdev, vaddr, size, DMA_TO_DEVICE); |
623 | } | 625 | } |
624 | EXPORT_SYMBOL(swiotlb_free_coherent); | 626 | EXPORT_SYMBOL(swiotlb_free_coherent); |
625 | 627 | ||
626 | static void | 628 | static void |
627 | swiotlb_full(struct device *dev, size_t size, int dir, int do_panic) | 629 | swiotlb_full(struct device *dev, size_t size, enum dma_data_direction dir, |
630 | int do_panic) | ||
628 | { | 631 | { |
629 | /* | 632 | /* |
630 | * Ran out of IOMMU space for this operation. This is very bad. | 633 | * Ran out of IOMMU space for this operation. This is very bad. |
631 | * Unfortunately the drivers cannot handle this operation properly. | 634 | * Unfortunately the drivers cannot handle this operation properly. |
632 | * unless they check for dma_mapping_error (most don't) | 635 | * unless they check for dma_mapping_error (most don't) |
633 | * When the mapping is small enough return a static buffer to limit | 636 | * When the mapping is small enough return a static buffer to limit |
634 | * the damage, or panic when the transfer is too big. | 637 | * the damage, or panic when the transfer is too big. |
635 | */ | 638 | */ |
636 | printk(KERN_ERR "DMA: Out of SW-IOMMU space for %zu bytes at " | 639 | printk(KERN_ERR "DMA: Out of SW-IOMMU space for %zu bytes at " |
637 | "device %s\n", size, dev ? dev_name(dev) : "?"); | 640 | "device %s\n", size, dev ? dev_name(dev) : "?"); |
638 | 641 | ||
639 | if (size <= io_tlb_overflow || !do_panic) | 642 | if (size <= io_tlb_overflow || !do_panic) |
640 | return; | 643 | return; |
641 | 644 | ||
642 | if (dir == DMA_BIDIRECTIONAL) | 645 | if (dir == DMA_BIDIRECTIONAL) |
643 | panic("DMA: Random memory could be DMA accessed\n"); | 646 | panic("DMA: Random memory could be DMA accessed\n"); |
644 | if (dir == DMA_FROM_DEVICE) | 647 | if (dir == DMA_FROM_DEVICE) |
645 | panic("DMA: Random memory could be DMA written\n"); | 648 | panic("DMA: Random memory could be DMA written\n"); |
646 | if (dir == DMA_TO_DEVICE) | 649 | if (dir == DMA_TO_DEVICE) |
647 | panic("DMA: Random memory could be DMA read\n"); | 650 | panic("DMA: Random memory could be DMA read\n"); |
648 | } | 651 | } |
649 | 652 | ||
650 | /* | 653 | /* |
651 | * Map a single buffer of the indicated size for DMA in streaming mode. The | 654 | * Map a single buffer of the indicated size for DMA in streaming mode. The |
652 | * physical address to use is returned. | 655 | * physical address to use is returned. |
653 | * | 656 | * |
654 | * Once the device is given the dma address, the device owns this memory until | 657 | * Once the device is given the dma address, the device owns this memory until |
655 | * either swiotlb_unmap_page or swiotlb_dma_sync_single is performed. | 658 | * either swiotlb_unmap_page or swiotlb_dma_sync_single is performed. |
656 | */ | 659 | */ |
657 | dma_addr_t swiotlb_map_page(struct device *dev, struct page *page, | 660 | dma_addr_t swiotlb_map_page(struct device *dev, struct page *page, |
658 | unsigned long offset, size_t size, | 661 | unsigned long offset, size_t size, |
659 | enum dma_data_direction dir, | 662 | enum dma_data_direction dir, |
660 | struct dma_attrs *attrs) | 663 | struct dma_attrs *attrs) |
661 | { | 664 | { |
662 | phys_addr_t phys = page_to_phys(page) + offset; | 665 | phys_addr_t phys = page_to_phys(page) + offset; |
663 | dma_addr_t dev_addr = phys_to_dma(dev, phys); | 666 | dma_addr_t dev_addr = phys_to_dma(dev, phys); |
664 | void *map; | 667 | void *map; |
665 | 668 | ||
666 | BUG_ON(dir == DMA_NONE); | 669 | BUG_ON(dir == DMA_NONE); |
667 | /* | 670 | /* |
668 | * If the address happens to be in the device's DMA window, | 671 | * If the address happens to be in the device's DMA window, |
669 | * we can safely return the device addr and not worry about bounce | 672 | * we can safely return the device addr and not worry about bounce |
670 | * buffering it. | 673 | * buffering it. |
671 | */ | 674 | */ |
672 | if (dma_capable(dev, dev_addr, size) && !swiotlb_force) | 675 | if (dma_capable(dev, dev_addr, size) && !swiotlb_force) |
673 | return dev_addr; | 676 | return dev_addr; |
674 | 677 | ||
675 | /* | 678 | /* |
676 | * Oh well, have to allocate and map a bounce buffer. | 679 | * Oh well, have to allocate and map a bounce buffer. |
677 | */ | 680 | */ |
678 | map = map_single(dev, phys, size, dir); | 681 | map = map_single(dev, phys, size, dir); |
679 | if (!map) { | 682 | if (!map) { |
680 | swiotlb_full(dev, size, dir, 1); | 683 | swiotlb_full(dev, size, dir, 1); |
681 | map = io_tlb_overflow_buffer; | 684 | map = io_tlb_overflow_buffer; |
682 | } | 685 | } |
683 | 686 | ||
684 | dev_addr = swiotlb_virt_to_bus(dev, map); | 687 | dev_addr = swiotlb_virt_to_bus(dev, map); |
685 | 688 | ||
686 | /* | 689 | /* |
687 | * Ensure that the address returned is DMA'ble | 690 | * Ensure that the address returned is DMA'ble |
688 | */ | 691 | */ |
689 | if (!dma_capable(dev, dev_addr, size)) | 692 | if (!dma_capable(dev, dev_addr, size)) |
690 | panic("map_single: bounce buffer is not DMA'ble"); | 693 | panic("map_single: bounce buffer is not DMA'ble"); |
691 | 694 | ||
692 | return dev_addr; | 695 | return dev_addr; |
693 | } | 696 | } |
694 | EXPORT_SYMBOL_GPL(swiotlb_map_page); | 697 | EXPORT_SYMBOL_GPL(swiotlb_map_page); |
695 | 698 | ||
696 | /* | 699 | /* |
697 | * Unmap a single streaming mode DMA translation. The dma_addr and size must | 700 | * Unmap a single streaming mode DMA translation. The dma_addr and size must |
698 | * match what was provided for in a previous swiotlb_map_page call. All | 701 | * match what was provided for in a previous swiotlb_map_page call. All |
699 | * other usages are undefined. | 702 | * other usages are undefined. |
700 | * | 703 | * |
701 | * After this call, reads by the cpu to the buffer are guaranteed to see | 704 | * After this call, reads by the cpu to the buffer are guaranteed to see |
702 | * whatever the device wrote there. | 705 | * whatever the device wrote there. |
703 | */ | 706 | */ |
704 | static void unmap_single(struct device *hwdev, dma_addr_t dev_addr, | 707 | static void unmap_single(struct device *hwdev, dma_addr_t dev_addr, |
705 | size_t size, int dir) | 708 | size_t size, enum dma_data_direction dir) |
706 | { | 709 | { |
707 | phys_addr_t paddr = dma_to_phys(hwdev, dev_addr); | 710 | phys_addr_t paddr = dma_to_phys(hwdev, dev_addr); |
708 | 711 | ||
709 | BUG_ON(dir == DMA_NONE); | 712 | BUG_ON(dir == DMA_NONE); |
710 | 713 | ||
711 | if (is_swiotlb_buffer(paddr)) { | 714 | if (is_swiotlb_buffer(paddr)) { |
712 | swiotlb_tbl_unmap_single(hwdev, phys_to_virt(paddr), size, dir); | 715 | swiotlb_tbl_unmap_single(hwdev, phys_to_virt(paddr), size, dir); |
713 | return; | 716 | return; |
714 | } | 717 | } |
715 | 718 | ||
716 | if (dir != DMA_FROM_DEVICE) | 719 | if (dir != DMA_FROM_DEVICE) |
717 | return; | 720 | return; |
718 | 721 | ||
719 | /* | 722 | /* |
720 | * phys_to_virt doesn't work with hihgmem page but we could | 723 | * phys_to_virt doesn't work with hihgmem page but we could |
721 | * call dma_mark_clean() with hihgmem page here. However, we | 724 | * call dma_mark_clean() with hihgmem page here. However, we |
722 | * are fine since dma_mark_clean() is null on POWERPC. We can | 725 | * are fine since dma_mark_clean() is null on POWERPC. We can |
723 | * make dma_mark_clean() take a physical address if necessary. | 726 | * make dma_mark_clean() take a physical address if necessary. |
724 | */ | 727 | */ |
725 | dma_mark_clean(phys_to_virt(paddr), size); | 728 | dma_mark_clean(phys_to_virt(paddr), size); |
726 | } | 729 | } |
727 | 730 | ||
728 | void swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, | 731 | void swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr, |
729 | size_t size, enum dma_data_direction dir, | 732 | size_t size, enum dma_data_direction dir, |
730 | struct dma_attrs *attrs) | 733 | struct dma_attrs *attrs) |
731 | { | 734 | { |
732 | unmap_single(hwdev, dev_addr, size, dir); | 735 | unmap_single(hwdev, dev_addr, size, dir); |
733 | } | 736 | } |
734 | EXPORT_SYMBOL_GPL(swiotlb_unmap_page); | 737 | EXPORT_SYMBOL_GPL(swiotlb_unmap_page); |
735 | 738 | ||
736 | /* | 739 | /* |
737 | * Make physical memory consistent for a single streaming mode DMA translation | 740 | * Make physical memory consistent for a single streaming mode DMA translation |
738 | * after a transfer. | 741 | * after a transfer. |
739 | * | 742 | * |
740 | * If you perform a swiotlb_map_page() but wish to interrogate the buffer | 743 | * If you perform a swiotlb_map_page() but wish to interrogate the buffer |
741 | * using the cpu, yet do not wish to teardown the dma mapping, you must | 744 | * using the cpu, yet do not wish to teardown the dma mapping, you must |
742 | * call this function before doing so. At the next point you give the dma | 745 | * call this function before doing so. At the next point you give the dma |
743 | * address back to the card, you must first perform a | 746 | * address back to the card, you must first perform a |
744 | * swiotlb_dma_sync_for_device, and then the device again owns the buffer | 747 | * swiotlb_dma_sync_for_device, and then the device again owns the buffer |
745 | */ | 748 | */ |
746 | static void | 749 | static void |
747 | swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr, | 750 | swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr, |
748 | size_t size, int dir, int target) | 751 | size_t size, enum dma_data_direction dir, int target) |
749 | { | 752 | { |
750 | phys_addr_t paddr = dma_to_phys(hwdev, dev_addr); | 753 | phys_addr_t paddr = dma_to_phys(hwdev, dev_addr); |
751 | 754 | ||
752 | BUG_ON(dir == DMA_NONE); | 755 | BUG_ON(dir == DMA_NONE); |
753 | 756 | ||
754 | if (is_swiotlb_buffer(paddr)) { | 757 | if (is_swiotlb_buffer(paddr)) { |
755 | swiotlb_tbl_sync_single(hwdev, phys_to_virt(paddr), size, dir, | 758 | swiotlb_tbl_sync_single(hwdev, phys_to_virt(paddr), size, dir, |
756 | target); | 759 | target); |
757 | return; | 760 | return; |
758 | } | 761 | } |
759 | 762 | ||
760 | if (dir != DMA_FROM_DEVICE) | 763 | if (dir != DMA_FROM_DEVICE) |
761 | return; | 764 | return; |
762 | 765 | ||
763 | dma_mark_clean(phys_to_virt(paddr), size); | 766 | dma_mark_clean(phys_to_virt(paddr), size); |
764 | } | 767 | } |
765 | 768 | ||
766 | void | 769 | void |
767 | swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, | 770 | swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, |
768 | size_t size, enum dma_data_direction dir) | 771 | size_t size, enum dma_data_direction dir) |
769 | { | 772 | { |
770 | swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU); | 773 | swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU); |
771 | } | 774 | } |
772 | EXPORT_SYMBOL(swiotlb_sync_single_for_cpu); | 775 | EXPORT_SYMBOL(swiotlb_sync_single_for_cpu); |
773 | 776 | ||
774 | void | 777 | void |
775 | swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, | 778 | swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, |
776 | size_t size, enum dma_data_direction dir) | 779 | size_t size, enum dma_data_direction dir) |
777 | { | 780 | { |
778 | swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE); | 781 | swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE); |
779 | } | 782 | } |
780 | EXPORT_SYMBOL(swiotlb_sync_single_for_device); | 783 | EXPORT_SYMBOL(swiotlb_sync_single_for_device); |
781 | 784 | ||
782 | /* | 785 | /* |
783 | * Map a set of buffers described by scatterlist in streaming mode for DMA. | 786 | * Map a set of buffers described by scatterlist in streaming mode for DMA. |
784 | * This is the scatter-gather version of the above swiotlb_map_page | 787 | * This is the scatter-gather version of the above swiotlb_map_page |
785 | * interface. Here the scatter gather list elements are each tagged with the | 788 | * interface. Here the scatter gather list elements are each tagged with the |
786 | * appropriate dma address and length. They are obtained via | 789 | * appropriate dma address and length. They are obtained via |
787 | * sg_dma_{address,length}(SG). | 790 | * sg_dma_{address,length}(SG). |
788 | * | 791 | * |
789 | * NOTE: An implementation may be able to use a smaller number of | 792 | * NOTE: An implementation may be able to use a smaller number of |
790 | * DMA address/length pairs than there are SG table elements. | 793 | * DMA address/length pairs than there are SG table elements. |
791 | * (for example via virtual mapping capabilities) | 794 | * (for example via virtual mapping capabilities) |
792 | * The routine returns the number of addr/length pairs actually | 795 | * The routine returns the number of addr/length pairs actually |
793 | * used, at most nents. | 796 | * used, at most nents. |
794 | * | 797 | * |
795 | * Device ownership issues as mentioned above for swiotlb_map_page are the | 798 | * Device ownership issues as mentioned above for swiotlb_map_page are the |
796 | * same here. | 799 | * same here. |
797 | */ | 800 | */ |
798 | int | 801 | int |
799 | swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, | 802 | swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, |
800 | enum dma_data_direction dir, struct dma_attrs *attrs) | 803 | enum dma_data_direction dir, struct dma_attrs *attrs) |
801 | { | 804 | { |
802 | struct scatterlist *sg; | 805 | struct scatterlist *sg; |
803 | int i; | 806 | int i; |
804 | 807 | ||
805 | BUG_ON(dir == DMA_NONE); | 808 | BUG_ON(dir == DMA_NONE); |
806 | 809 | ||
807 | for_each_sg(sgl, sg, nelems, i) { | 810 | for_each_sg(sgl, sg, nelems, i) { |
808 | phys_addr_t paddr = sg_phys(sg); | 811 | phys_addr_t paddr = sg_phys(sg); |
809 | dma_addr_t dev_addr = phys_to_dma(hwdev, paddr); | 812 | dma_addr_t dev_addr = phys_to_dma(hwdev, paddr); |
810 | 813 | ||
811 | if (swiotlb_force || | 814 | if (swiotlb_force || |
812 | !dma_capable(hwdev, dev_addr, sg->length)) { | 815 | !dma_capable(hwdev, dev_addr, sg->length)) { |
813 | void *map = map_single(hwdev, sg_phys(sg), | 816 | void *map = map_single(hwdev, sg_phys(sg), |
814 | sg->length, dir); | 817 | sg->length, dir); |
815 | if (!map) { | 818 | if (!map) { |
816 | /* Don't panic here, we expect map_sg users | 819 | /* Don't panic here, we expect map_sg users |
817 | to do proper error handling. */ | 820 | to do proper error handling. */ |
818 | swiotlb_full(hwdev, sg->length, dir, 0); | 821 | swiotlb_full(hwdev, sg->length, dir, 0); |
819 | swiotlb_unmap_sg_attrs(hwdev, sgl, i, dir, | 822 | swiotlb_unmap_sg_attrs(hwdev, sgl, i, dir, |
820 | attrs); | 823 | attrs); |
821 | sgl[0].dma_length = 0; | 824 | sgl[0].dma_length = 0; |
822 | return 0; | 825 | return 0; |
823 | } | 826 | } |
824 | sg->dma_address = swiotlb_virt_to_bus(hwdev, map); | 827 | sg->dma_address = swiotlb_virt_to_bus(hwdev, map); |
825 | } else | 828 | } else |
826 | sg->dma_address = dev_addr; | 829 | sg->dma_address = dev_addr; |
827 | sg->dma_length = sg->length; | 830 | sg->dma_length = sg->length; |
828 | } | 831 | } |
829 | return nelems; | 832 | return nelems; |
830 | } | 833 | } |
831 | EXPORT_SYMBOL(swiotlb_map_sg_attrs); | 834 | EXPORT_SYMBOL(swiotlb_map_sg_attrs); |
832 | 835 | ||
833 | int | 836 | int |
834 | swiotlb_map_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, | 837 | swiotlb_map_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, |
835 | int dir) | 838 | enum dma_data_direction dir) |
836 | { | 839 | { |
837 | return swiotlb_map_sg_attrs(hwdev, sgl, nelems, dir, NULL); | 840 | return swiotlb_map_sg_attrs(hwdev, sgl, nelems, dir, NULL); |
838 | } | 841 | } |
839 | EXPORT_SYMBOL(swiotlb_map_sg); | 842 | EXPORT_SYMBOL(swiotlb_map_sg); |
840 | 843 | ||
841 | /* | 844 | /* |
842 | * Unmap a set of streaming mode DMA translations. Again, cpu read rules | 845 | * Unmap a set of streaming mode DMA translations. Again, cpu read rules |
843 | * concerning calls here are the same as for swiotlb_unmap_page() above. | 846 | * concerning calls here are the same as for swiotlb_unmap_page() above. |
844 | */ | 847 | */ |
845 | void | 848 | void |
846 | swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, | 849 | swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, |
847 | int nelems, enum dma_data_direction dir, struct dma_attrs *attrs) | 850 | int nelems, enum dma_data_direction dir, struct dma_attrs *attrs) |
848 | { | 851 | { |
849 | struct scatterlist *sg; | 852 | struct scatterlist *sg; |
850 | int i; | 853 | int i; |
851 | 854 | ||
852 | BUG_ON(dir == DMA_NONE); | 855 | BUG_ON(dir == DMA_NONE); |
853 | 856 | ||
854 | for_each_sg(sgl, sg, nelems, i) | 857 | for_each_sg(sgl, sg, nelems, i) |
855 | unmap_single(hwdev, sg->dma_address, sg->dma_length, dir); | 858 | unmap_single(hwdev, sg->dma_address, sg->dma_length, dir); |
856 | 859 | ||
857 | } | 860 | } |
858 | EXPORT_SYMBOL(swiotlb_unmap_sg_attrs); | 861 | EXPORT_SYMBOL(swiotlb_unmap_sg_attrs); |
859 | 862 | ||
860 | void | 863 | void |
861 | swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, | 864 | swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, |
862 | int dir) | 865 | enum dma_data_direction dir) |
863 | { | 866 | { |
864 | return swiotlb_unmap_sg_attrs(hwdev, sgl, nelems, dir, NULL); | 867 | return swiotlb_unmap_sg_attrs(hwdev, sgl, nelems, dir, NULL); |
865 | } | 868 | } |
866 | EXPORT_SYMBOL(swiotlb_unmap_sg); | 869 | EXPORT_SYMBOL(swiotlb_unmap_sg); |
867 | 870 | ||
868 | /* | 871 | /* |
869 | * Make physical memory consistent for a set of streaming mode DMA translations | 872 | * Make physical memory consistent for a set of streaming mode DMA translations |
870 | * after a transfer. | 873 | * after a transfer. |
871 | * | 874 | * |
872 | * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules | 875 | * The same as swiotlb_sync_single_* but for a scatter-gather list, same rules |
873 | * and usage. | 876 | * and usage. |
874 | */ | 877 | */ |
875 | static void | 878 | static void |
876 | swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, | 879 | swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, |
877 | int nelems, int dir, int target) | 880 | int nelems, enum dma_data_direction dir, int target) |
878 | { | 881 | { |
879 | struct scatterlist *sg; | 882 | struct scatterlist *sg; |
880 | int i; | 883 | int i; |
881 | 884 | ||
882 | for_each_sg(sgl, sg, nelems, i) | 885 | for_each_sg(sgl, sg, nelems, i) |
883 | swiotlb_sync_single(hwdev, sg->dma_address, | 886 | swiotlb_sync_single(hwdev, sg->dma_address, |
884 | sg->dma_length, dir, target); | 887 | sg->dma_length, dir, target); |
885 | } | 888 | } |
886 | 889 | ||
887 | void | 890 | void |
888 | swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, | 891 | swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, |
889 | int nelems, enum dma_data_direction dir) | 892 | int nelems, enum dma_data_direction dir) |
890 | { | 893 | { |
891 | swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU); | 894 | swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU); |
892 | } | 895 | } |
893 | EXPORT_SYMBOL(swiotlb_sync_sg_for_cpu); | 896 | EXPORT_SYMBOL(swiotlb_sync_sg_for_cpu); |
894 | 897 | ||
895 | void | 898 | void |
896 | swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, | 899 | swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, |
897 | int nelems, enum dma_data_direction dir) | 900 | int nelems, enum dma_data_direction dir) |
898 | { | 901 | { |
899 | swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE); | 902 | swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE); |
900 | } | 903 | } |
901 | EXPORT_SYMBOL(swiotlb_sync_sg_for_device); | 904 | EXPORT_SYMBOL(swiotlb_sync_sg_for_device); |
902 | 905 | ||
903 | int | 906 | int |
904 | swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) | 907 | swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) |
905 | { | 908 | { |
906 | return (dma_addr == swiotlb_virt_to_bus(hwdev, io_tlb_overflow_buffer)); | 909 | return (dma_addr == swiotlb_virt_to_bus(hwdev, io_tlb_overflow_buffer)); |
907 | } | 910 | } |
908 | EXPORT_SYMBOL(swiotlb_dma_mapping_error); | 911 | EXPORT_SYMBOL(swiotlb_dma_mapping_error); |
909 | 912 | ||
910 | /* | 913 | /* |
911 | * Return whether the given device DMA address mask can be supported | 914 | * Return whether the given device DMA address mask can be supported |
912 | * properly. For example, if your device can only drive the low 24-bits | 915 | * properly. For example, if your device can only drive the low 24-bits |
913 | * during bus mastering, then you would pass 0x00ffffff as the mask to | 916 | * during bus mastering, then you would pass 0x00ffffff as the mask to |
914 | * this function. | 917 | * this function. |
915 | */ | 918 | */ |
916 | int | 919 | int |
917 | swiotlb_dma_supported(struct device *hwdev, u64 mask) | 920 | swiotlb_dma_supported(struct device *hwdev, u64 mask) |
918 | { | 921 | { |
919 | return swiotlb_virt_to_bus(hwdev, io_tlb_end - 1) <= mask; | 922 | return swiotlb_virt_to_bus(hwdev, io_tlb_end - 1) <= mask; |
920 | } | 923 | } |
921 | EXPORT_SYMBOL(swiotlb_dma_supported); | 924 | EXPORT_SYMBOL(swiotlb_dma_supported); |
922 | 925 |