1 /*
2  * OpenRISC Linux
3  *
4  * Linux architectural port borrowing liberally from similar works of
5  * others.  All original copyrights apply as per the original source
6  * declaration.
7  *
8  * Modifications for the OpenRISC architecture:
9  * Copyright (C) 2003 Matjaz Breskvar <phoenix@bsemi.com>
10  * Copyright (C) 2010-2011 Jonas Bonn <jonas@southpole.se>
11  *
12  *      This program is free software; you can redistribute it and/or
13  *      modify it under the terms of the GNU General Public License
14  *      as published by the Free Software Foundation; either version
15  *      2 of the License, or (at your option) any later version.
16  *
17  * DMA mapping callbacks...
18  * As alloc_coherent is the only DMA callback being used currently, that's
19  * the only thing implemented properly.  The rest need looking into...
20  */
21 
22 #include <linux/dma-mapping.h>
23 #include <linux/dma-debug.h>
24 
25 #include <asm/cpuinfo.h>
26 #include <asm/spr_defs.h>
27 #include <asm/tlbflush.h>
28 
page_set_nocache(pte_t * pte,unsigned long addr,unsigned long next,struct mm_walk * walk)29 static int page_set_nocache(pte_t *pte, unsigned long addr,
30 			    unsigned long next, struct mm_walk *walk)
31 {
32 	unsigned long cl;
33 
34 	pte_val(*pte) |= _PAGE_CI;
35 
36 	/*
37 	 * Flush the page out of the TLB so that the new page flags get
38 	 * picked up next time there's an access
39 	 */
40 	flush_tlb_page(NULL, addr);
41 
42 	/* Flush page out of dcache */
43 	for (cl = __pa(addr); cl < __pa(next); cl += cpuinfo.dcache_block_size)
44 		mtspr(SPR_DCBFR, cl);
45 
46 	return 0;
47 }
48 
page_clear_nocache(pte_t * pte,unsigned long addr,unsigned long next,struct mm_walk * walk)49 static int page_clear_nocache(pte_t *pte, unsigned long addr,
50 			      unsigned long next, struct mm_walk *walk)
51 {
52 	pte_val(*pte) &= ~_PAGE_CI;
53 
54 	/*
55 	 * Flush the page out of the TLB so that the new page flags get
56 	 * picked up next time there's an access
57 	 */
58 	flush_tlb_page(NULL, addr);
59 
60 	return 0;
61 }
62 
63 /*
64  * Alloc "coherent" memory, which for OpenRISC means simply uncached.
65  *
66  * This function effectively just calls __get_free_pages, sets the
67  * cache-inhibit bit on those pages, and makes sure that the pages are
68  * flushed out of the cache before they are used.
69  *
70  */
or1k_dma_alloc_coherent(struct device * dev,size_t size,dma_addr_t * dma_handle,gfp_t gfp)71 void *or1k_dma_alloc_coherent(struct device *dev, size_t size,
72 			      dma_addr_t *dma_handle, gfp_t gfp)
73 {
74 	unsigned long va;
75 	void *page;
76 	struct mm_walk walk = {
77 		.pte_entry = page_set_nocache,
78 		.mm = &init_mm
79 	};
80 
81 	page = alloc_pages_exact(size, gfp);
82 	if (!page)
83 		return NULL;
84 
85 	/* This gives us the real physical address of the first page. */
86 	*dma_handle = __pa(page);
87 
88 	va = (unsigned long)page;
89 
90 	/*
91 	 * We need to iterate through the pages, clearing the dcache for
92 	 * them and setting the cache-inhibit bit.
93 	 */
94 	if (walk_page_range(va, va + size, &walk)) {
95 		free_pages_exact(page, size);
96 		return NULL;
97 	}
98 
99 	return (void *)va;
100 }
101 
or1k_dma_free_coherent(struct device * dev,size_t size,void * vaddr,dma_addr_t dma_handle)102 void or1k_dma_free_coherent(struct device *dev, size_t size, void *vaddr,
103 			    dma_addr_t dma_handle)
104 {
105 	unsigned long va = (unsigned long)vaddr;
106 	struct mm_walk walk = {
107 		.pte_entry = page_clear_nocache,
108 		.mm = &init_mm
109 	};
110 
111 	/* walk_page_range shouldn't be able to fail here */
112 	WARN_ON(walk_page_range(va, va + size, &walk));
113 
114 	free_pages_exact(vaddr, size);
115 }
116 
or1k_map_page(struct device * dev,struct page * page,unsigned long offset,size_t size,enum dma_data_direction dir,struct dma_attrs * attrs)117 dma_addr_t or1k_map_page(struct device *dev, struct page *page,
118 			 unsigned long offset, size_t size,
119 			 enum dma_data_direction dir,
120 			 struct dma_attrs *attrs)
121 {
122 	unsigned long cl;
123 	dma_addr_t addr = page_to_phys(page) + offset;
124 
125 	switch (dir) {
126 	case DMA_TO_DEVICE:
127 		/* Flush the dcache for the requested range */
128 		for (cl = addr; cl < addr + size;
129 		     cl += cpuinfo.dcache_block_size)
130 			mtspr(SPR_DCBFR, cl);
131 		break;
132 	case DMA_FROM_DEVICE:
133 		/* Invalidate the dcache for the requested range */
134 		for (cl = addr; cl < addr + size;
135 		     cl += cpuinfo.dcache_block_size)
136 			mtspr(SPR_DCBIR, cl);
137 		break;
138 	default:
139 		/*
140 		 * NOTE: If dir == DMA_BIDIRECTIONAL then there's no need to
141 		 * flush nor invalidate the cache here as the area will need
142 		 * to be manually synced anyway.
143 		 */
144 		break;
145 	}
146 
147 	return addr;
148 }
149 
or1k_unmap_page(struct device * dev,dma_addr_t dma_handle,size_t size,enum dma_data_direction dir,struct dma_attrs * attrs)150 void or1k_unmap_page(struct device *dev, dma_addr_t dma_handle,
151 		     size_t size, enum dma_data_direction dir,
152 		     struct dma_attrs *attrs)
153 {
154 	/* Nothing special to do here... */
155 }
156 
or1k_map_sg(struct device * dev,struct scatterlist * sg,int nents,enum dma_data_direction dir,struct dma_attrs * attrs)157 int or1k_map_sg(struct device *dev, struct scatterlist *sg,
158 		int nents, enum dma_data_direction dir,
159 		struct dma_attrs *attrs)
160 {
161 	struct scatterlist *s;
162 	int i;
163 
164 	for_each_sg(sg, s, nents, i) {
165 		s->dma_address = or1k_map_page(dev, sg_page(s), s->offset,
166 					       s->length, dir, NULL);
167 	}
168 
169 	return nents;
170 }
171 
or1k_unmap_sg(struct device * dev,struct scatterlist * sg,int nents,enum dma_data_direction dir,struct dma_attrs * attrs)172 void or1k_unmap_sg(struct device *dev, struct scatterlist *sg,
173 		   int nents, enum dma_data_direction dir,
174 		   struct dma_attrs *attrs)
175 {
176 	struct scatterlist *s;
177 	int i;
178 
179 	for_each_sg(sg, s, nents, i) {
180 		or1k_unmap_page(dev, sg_dma_address(s), sg_dma_len(s), dir, NULL);
181 	}
182 }
183 
or1k_sync_single_for_cpu(struct device * dev,dma_addr_t dma_handle,size_t size,enum dma_data_direction dir)184 void or1k_sync_single_for_cpu(struct device *dev,
185 			      dma_addr_t dma_handle, size_t size,
186 			      enum dma_data_direction dir)
187 {
188 	unsigned long cl;
189 	dma_addr_t addr = dma_handle;
190 
191 	/* Invalidate the dcache for the requested range */
192 	for (cl = addr; cl < addr + size; cl += cpuinfo.dcache_block_size)
193 		mtspr(SPR_DCBIR, cl);
194 }
195 
or1k_sync_single_for_device(struct device * dev,dma_addr_t dma_handle,size_t size,enum dma_data_direction dir)196 void or1k_sync_single_for_device(struct device *dev,
197 			         dma_addr_t dma_handle, size_t size,
198 			         enum dma_data_direction dir)
199 {
200 	unsigned long cl;
201 	dma_addr_t addr = dma_handle;
202 
203 	/* Flush the dcache for the requested range */
204 	for (cl = addr; cl < addr + size; cl += cpuinfo.dcache_block_size)
205 		mtspr(SPR_DCBFR, cl);
206 }
207 
208 /* Number of entries preallocated for DMA-API debugging */
209 #define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
210 
dma_init(void)211 static int __init dma_init(void)
212 {
213 	dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES);
214 
215 	return 0;
216 }
217 fs_initcall(dma_init);
218