diff options
author | David S. Miller <davem@huronp11.davemloft.net> | 2008-02-08 21:05:46 -0500 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2008-02-09 06:15:04 -0500 |
commit | d284142cbad66832d5072a0aebeca7bd9ca841b7 (patch) | |
tree | e5c5ad6271b3a61e28f1767b744e0696af0cd1a4 /arch/sparc64/kernel/iommu.c | |
parent | 19814ea24e9d80583504e336340ab4590841b0b1 (diff) |
[SPARC64]: IOMMU allocations using iommu-helper layer.
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc64/kernel/iommu.c')
-rw-r--r-- | arch/sparc64/kernel/iommu.c | 125 |
1 files changed, 81 insertions, 44 deletions
diff --git a/arch/sparc64/kernel/iommu.c b/arch/sparc64/kernel/iommu.c index 5623a4d59dff..90a5907080a1 100644 --- a/arch/sparc64/kernel/iommu.c +++ b/arch/sparc64/kernel/iommu.c | |||
@@ -1,6 +1,6 @@ | |||
1 | /* iommu.c: Generic sparc64 IOMMU support. | 1 | /* iommu.c: Generic sparc64 IOMMU support. |
2 | * | 2 | * |
3 | * Copyright (C) 1999, 2007 David S. Miller (davem@davemloft.net) | 3 | * Copyright (C) 1999, 2007, 2008 David S. Miller (davem@davemloft.net) |
4 | * Copyright (C) 1999, 2000 Jakub Jelinek (jakub@redhat.com) | 4 | * Copyright (C) 1999, 2000 Jakub Jelinek (jakub@redhat.com) |
5 | */ | 5 | */ |
6 | 6 | ||
@@ -10,6 +10,7 @@ | |||
10 | #include <linux/device.h> | 10 | #include <linux/device.h> |
11 | #include <linux/dma-mapping.h> | 11 | #include <linux/dma-mapping.h> |
12 | #include <linux/errno.h> | 12 | #include <linux/errno.h> |
13 | #include <linux/iommu-helper.h> | ||
13 | 14 | ||
14 | #ifdef CONFIG_PCI | 15 | #ifdef CONFIG_PCI |
15 | #include <linux/pci.h> | 16 | #include <linux/pci.h> |
@@ -41,7 +42,7 @@ | |||
41 | "i" (ASI_PHYS_BYPASS_EC_E)) | 42 | "i" (ASI_PHYS_BYPASS_EC_E)) |
42 | 43 | ||
43 | /* Must be invoked under the IOMMU lock. */ | 44 | /* Must be invoked under the IOMMU lock. */ |
44 | static void __iommu_flushall(struct iommu *iommu) | 45 | static void iommu_flushall(struct iommu *iommu) |
45 | { | 46 | { |
46 | if (iommu->iommu_flushinv) { | 47 | if (iommu->iommu_flushinv) { |
47 | iommu_write(iommu->iommu_flushinv, ~(u64)0); | 48 | iommu_write(iommu->iommu_flushinv, ~(u64)0); |
@@ -83,54 +84,91 @@ static inline void iopte_make_dummy(struct iommu *iommu, iopte_t *iopte) | |||
83 | iopte_val(*iopte) = val; | 84 | iopte_val(*iopte) = val; |
84 | } | 85 | } |
85 | 86 | ||
86 | /* Based largely upon the ppc64 iommu allocator. */ | 87 | /* Based almost entirely upon the ppc64 iommu allocator. If you use the 'handle' |
87 | static long arena_alloc(struct iommu *iommu, unsigned long npages) | 88 | * facility it must all be done in one pass while under the iommu lock. |
89 | * | ||
90 | * On sun4u platforms, we only flush the IOMMU once every time we've passed | ||
91 | * over the entire page table doing allocations. Therefore we only ever advance | ||
92 | * the hint and cannot backtrack it. | ||
93 | */ | ||
94 | unsigned long iommu_range_alloc(struct device *dev, | ||
95 | struct iommu *iommu, | ||
96 | unsigned long npages, | ||
97 | unsigned long *handle) | ||
88 | { | 98 | { |
99 | unsigned long n, end, start, limit, boundary_size; | ||
89 | struct iommu_arena *arena = &iommu->arena; | 100 | struct iommu_arena *arena = &iommu->arena; |
90 | unsigned long n, i, start, end, limit; | 101 | int pass = 0; |
91 | int pass; | 102 | |
103 | /* This allocator was derived from x86_64's bit string search */ | ||
104 | |||
105 | /* Sanity check */ | ||
106 | if (unlikely(npages == 0)) { | ||
107 | if (printk_ratelimit()) | ||
108 | WARN_ON(1); | ||
109 | return DMA_ERROR_CODE; | ||
110 | } | ||
111 | |||
112 | if (handle && *handle) | ||
113 | start = *handle; | ||
114 | else | ||
115 | start = arena->hint; | ||
92 | 116 | ||
93 | limit = arena->limit; | 117 | limit = arena->limit; |
94 | start = arena->hint; | ||
95 | pass = 0; | ||
96 | 118 | ||
97 | again: | 119 | /* The case below can happen if we have a small segment appended |
98 | n = find_next_zero_bit(arena->map, limit, start); | 120 | * to a large, or when the previous alloc was at the very end of |
99 | end = n + npages; | 121 | * the available space. If so, go back to the beginning and flush. |
100 | if (unlikely(end >= limit)) { | 122 | */ |
123 | if (start >= limit) { | ||
124 | start = 0; | ||
125 | if (iommu->flush_all) | ||
126 | iommu->flush_all(iommu); | ||
127 | } | ||
128 | |||
129 | again: | ||
130 | |||
131 | if (dev) | ||
132 | boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1, | ||
133 | 1 << IO_PAGE_SHIFT); | ||
134 | else | ||
135 | boundary_size = ALIGN(1UL << 32, 1 << IO_PAGE_SHIFT); | ||
136 | |||
137 | n = iommu_area_alloc(arena->map, limit, start, npages, 0, | ||
138 | boundary_size >> IO_PAGE_SHIFT, 0); | ||
139 | if (n == -1) { | ||
101 | if (likely(pass < 1)) { | 140 | if (likely(pass < 1)) { |
102 | limit = start; | 141 | /* First failure, rescan from the beginning. */ |
103 | start = 0; | 142 | start = 0; |
104 | __iommu_flushall(iommu); | 143 | if (iommu->flush_all) |
144 | iommu->flush_all(iommu); | ||
105 | pass++; | 145 | pass++; |
106 | goto again; | 146 | goto again; |
107 | } else { | 147 | } else { |
108 | /* Scanned the whole thing, give up. */ | 148 | /* Second failure, give up */ |
109 | return -1; | 149 | return DMA_ERROR_CODE; |
110 | } | 150 | } |
111 | } | 151 | } |
112 | 152 | ||
113 | for (i = n; i < end; i++) { | 153 | end = n + npages; |
114 | if (test_bit(i, arena->map)) { | ||
115 | start = i + 1; | ||
116 | goto again; | ||
117 | } | ||
118 | } | ||
119 | |||
120 | for (i = n; i < end; i++) | ||
121 | __set_bit(i, arena->map); | ||
122 | 154 | ||
123 | arena->hint = end; | 155 | arena->hint = end; |
124 | 156 | ||
157 | /* Update handle for SG allocations */ | ||
158 | if (handle) | ||
159 | *handle = end; | ||
160 | |||
125 | return n; | 161 | return n; |
126 | } | 162 | } |
127 | 163 | ||
128 | static void arena_free(struct iommu_arena *arena, unsigned long base, unsigned long npages) | 164 | void iommu_range_free(struct iommu *iommu, dma_addr_t dma_addr, unsigned long npages) |
129 | { | 165 | { |
130 | unsigned long i; | 166 | struct iommu_arena *arena = &iommu->arena; |
167 | unsigned long entry; | ||
131 | 168 | ||
132 | for (i = base; i < (base + npages); i++) | 169 | entry = (dma_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT; |
133 | __clear_bit(i, arena->map); | 170 | |
171 | iommu_area_free(arena->map, entry, npages); | ||
134 | } | 172 | } |
135 | 173 | ||
136 | int iommu_table_init(struct iommu *iommu, int tsbsize, | 174 | int iommu_table_init(struct iommu *iommu, int tsbsize, |
@@ -156,6 +194,9 @@ int iommu_table_init(struct iommu *iommu, int tsbsize, | |||
156 | } | 194 | } |
157 | iommu->arena.limit = num_tsb_entries; | 195 | iommu->arena.limit = num_tsb_entries; |
158 | 196 | ||
197 | if (tlb_type != hypervisor) | ||
198 | iommu->flush_all = iommu_flushall; | ||
199 | |||
159 | /* Allocate and initialize the dummy page which we | 200 | /* Allocate and initialize the dummy page which we |
160 | * set inactive IO PTEs to point to. | 201 | * set inactive IO PTEs to point to. |
161 | */ | 202 | */ |
@@ -192,22 +233,18 @@ out_free_map: | |||
192 | return -ENOMEM; | 233 | return -ENOMEM; |
193 | } | 234 | } |
194 | 235 | ||
195 | static inline iopte_t *alloc_npages(struct iommu *iommu, unsigned long npages) | 236 | static inline iopte_t *alloc_npages(struct device *dev, struct iommu *iommu, |
237 | unsigned long npages) | ||
196 | { | 238 | { |
197 | long entry; | 239 | unsigned long entry; |
198 | 240 | ||
199 | entry = arena_alloc(iommu, npages); | 241 | entry = iommu_range_alloc(dev, iommu, npages, NULL); |
200 | if (unlikely(entry < 0)) | 242 | if (unlikely(entry == DMA_ERROR_CODE)) |
201 | return NULL; | 243 | return NULL; |
202 | 244 | ||
203 | return iommu->page_table + entry; | 245 | return iommu->page_table + entry; |
204 | } | 246 | } |
205 | 247 | ||
206 | static inline void free_npages(struct iommu *iommu, dma_addr_t base, unsigned long npages) | ||
207 | { | ||
208 | arena_free(&iommu->arena, base >> IO_PAGE_SHIFT, npages); | ||
209 | } | ||
210 | |||
211 | static int iommu_alloc_ctx(struct iommu *iommu) | 248 | static int iommu_alloc_ctx(struct iommu *iommu) |
212 | { | 249 | { |
213 | int lowest = iommu->ctx_lowest_free; | 250 | int lowest = iommu->ctx_lowest_free; |
@@ -258,7 +295,7 @@ static void *dma_4u_alloc_coherent(struct device *dev, size_t size, | |||
258 | iommu = dev->archdata.iommu; | 295 | iommu = dev->archdata.iommu; |
259 | 296 | ||
260 | spin_lock_irqsave(&iommu->lock, flags); | 297 | spin_lock_irqsave(&iommu->lock, flags); |
261 | iopte = alloc_npages(iommu, size >> IO_PAGE_SHIFT); | 298 | iopte = alloc_npages(dev, iommu, size >> IO_PAGE_SHIFT); |
262 | spin_unlock_irqrestore(&iommu->lock, flags); | 299 | spin_unlock_irqrestore(&iommu->lock, flags); |
263 | 300 | ||
264 | if (unlikely(iopte == NULL)) { | 301 | if (unlikely(iopte == NULL)) { |
@@ -296,7 +333,7 @@ static void dma_4u_free_coherent(struct device *dev, size_t size, | |||
296 | 333 | ||
297 | spin_lock_irqsave(&iommu->lock, flags); | 334 | spin_lock_irqsave(&iommu->lock, flags); |
298 | 335 | ||
299 | free_npages(iommu, dvma - iommu->page_table_map_base, npages); | 336 | iommu_range_free(iommu, dvma, npages); |
300 | 337 | ||
301 | spin_unlock_irqrestore(&iommu->lock, flags); | 338 | spin_unlock_irqrestore(&iommu->lock, flags); |
302 | 339 | ||
@@ -327,7 +364,7 @@ static dma_addr_t dma_4u_map_single(struct device *dev, void *ptr, size_t sz, | |||
327 | npages >>= IO_PAGE_SHIFT; | 364 | npages >>= IO_PAGE_SHIFT; |
328 | 365 | ||
329 | spin_lock_irqsave(&iommu->lock, flags); | 366 | spin_lock_irqsave(&iommu->lock, flags); |
330 | base = alloc_npages(iommu, npages); | 367 | base = alloc_npages(dev, iommu, npages); |
331 | ctx = 0; | 368 | ctx = 0; |
332 | if (iommu->iommu_ctxflush) | 369 | if (iommu->iommu_ctxflush) |
333 | ctx = iommu_alloc_ctx(iommu); | 370 | ctx = iommu_alloc_ctx(iommu); |
@@ -465,7 +502,7 @@ static void dma_4u_unmap_single(struct device *dev, dma_addr_t bus_addr, | |||
465 | for (i = 0; i < npages; i++) | 502 | for (i = 0; i < npages; i++) |
466 | iopte_make_dummy(iommu, base + i); | 503 | iopte_make_dummy(iommu, base + i); |
467 | 504 | ||
468 | free_npages(iommu, bus_addr - iommu->page_table_map_base, npages); | 505 | iommu_range_free(iommu, bus_addr, npages); |
469 | 506 | ||
470 | iommu_free_ctx(iommu, ctx); | 507 | iommu_free_ctx(iommu, ctx); |
471 | 508 | ||
@@ -503,7 +540,7 @@ static int dma_4u_map_sg(struct device *dev, struct scatterlist *sglist, | |||
503 | 540 | ||
504 | spin_lock_irqsave(&iommu->lock, flags); | 541 | spin_lock_irqsave(&iommu->lock, flags); |
505 | 542 | ||
506 | base = alloc_npages(iommu, npages); | 543 | base = alloc_npages(dev, iommu, npages); |
507 | ctx = 0; | 544 | ctx = 0; |
508 | if (iommu->iommu_ctxflush) | 545 | if (iommu->iommu_ctxflush) |
509 | ctx = iommu_alloc_ctx(iommu); | 546 | ctx = iommu_alloc_ctx(iommu); |
@@ -592,7 +629,7 @@ static void dma_4u_unmap_sg(struct device *dev, struct scatterlist *sglist, | |||
592 | for (i = 0; i < npages; i++) | 629 | for (i = 0; i < npages; i++) |
593 | iopte_make_dummy(iommu, base + i); | 630 | iopte_make_dummy(iommu, base + i); |
594 | 631 | ||
595 | free_npages(iommu, bus_addr - iommu->page_table_map_base, npages); | 632 | iommu_range_free(iommu, bus_addr, npages); |
596 | 633 | ||
597 | iommu_free_ctx(iommu, ctx); | 634 | iommu_free_ctx(iommu, ctx); |
598 | 635 | ||