1 #include <linux/prefetch.h>
4 * iommu_fill_pdir - Insert coalesced scatter/gather chunks into the I/O Pdir.
5 * @ioc: The I/O Controller.
6 * @startsg: The scatter/gather list of coalesced chunks.
7 * @nents: The number of entries in the scatter/gather list.
10 * This function inserts the coalesced scatter/gather list chunks into the
11 * I/O Controller's I/O Pdir.
13 static inline unsigned int
14 iommu_fill_pdir(struct ioc *ioc, struct scatterlist *startsg, int nents,
16 void (*iommu_io_pdir_entry)(u64 *, space_t, unsigned long,
19 struct scatterlist *dma_sg = startsg; /* pointer to current DMA */
20 unsigned int n_mappings = 0;
21 unsigned long dma_offset = 0, dma_len = 0;
24 /* Horrible hack. For efficiency's sake, dma_sg starts one
25 * entry below the true start (it is immediately incremented
33 DBG_RUN_SG(" %d : %08lx/%05x %p/%05x\n", nents,
34 (unsigned long)sg_dma_address(startsg), cnt,
35 sg_virt(startsg), startsg->length
40 ** Look for the start of a new DMA stream
43 if (sg_dma_address(startsg) & PIDE_FLAG) {
44 u32 pide = sg_dma_address(startsg) & ~PIDE_FLAG;
46 BUG_ON(pdirp && (dma_len != sg_dma_len(dma_sg)));
50 dma_len = sg_dma_len(startsg);
51 sg_dma_len(startsg) = 0;
52 dma_offset = (unsigned long) pide & ~IOVP_MASK;
54 #if defined(ZX1_SUPPORT)
55 /* Pluto IOMMU IO Virt Address is not zero based */
56 sg_dma_address(dma_sg) = pide | ioc->ibase;
58 /* SBA, ccio, and dino are zero based.
59 * Trying to save a few CPU cycles for most users.
61 sg_dma_address(dma_sg) = pide;
63 pdirp = &(ioc->pdir_base[pide >> IOVP_SHIFT]);
67 BUG_ON(pdirp == NULL);
69 vaddr = (unsigned long)sg_virt(startsg);
70 sg_dma_len(dma_sg) += startsg->length;
71 size = startsg->length + dma_offset;
73 #ifdef IOMMU_MAP_STATS
74 ioc->msg_pages += startsg->length >> IOVP_SHIFT;
77 iommu_io_pdir_entry(pdirp, KERNEL_SPACE,
82 } while(unlikely(size > 0));
90 ** First pass is to walk the SG list and determine where the breaks are
91 ** in the DMA stream. Allocates PDIR entries but does not fill them.
92 ** Returns the number of DMA chunks.
94 ** Doing the fill separate from the coalescing/allocation keeps the
95 ** code simpler. Future enhancement could make one pass through
96 ** the sglist do both.
99 static inline unsigned int
100 iommu_coalesce_chunks(struct ioc *ioc, struct device *dev,
101 struct scatterlist *startsg, int nents,
102 int (*iommu_alloc_range)(struct ioc *, struct device *, size_t))
104 struct scatterlist *contig_sg; /* contig chunk head */
105 unsigned long dma_offset, dma_len; /* start/len of DMA stream */
106 unsigned int n_mappings = 0;
107 unsigned int max_seg_size = min(dma_get_max_seg_size(dev),
108 (unsigned)DMA_CHUNK_SIZE);
109 unsigned int max_seg_boundary = dma_get_seg_boundary(dev) + 1;
110 if (max_seg_boundary) /* check if the addition above didn't overflow */
111 max_seg_size = min(max_seg_size, max_seg_boundary);
116 ** Prepare for first/next DMA stream
119 dma_len = startsg->length;
120 dma_offset = startsg->offset;
122 /* PARANOID: clear entries */
123 sg_dma_address(startsg) = 0;
124 sg_dma_len(startsg) = 0;
127 ** This loop terminates one iteration "early" since
128 ** it's always looking one "ahead".
131 unsigned long prev_end, sg_start;
133 prev_end = (unsigned long)sg_virt(startsg) +
137 sg_start = (unsigned long)sg_virt(startsg);
139 /* PARANOID: clear entries */
140 sg_dma_address(startsg) = 0;
141 sg_dma_len(startsg) = 0;
144 ** First make sure current dma stream won't
145 ** exceed max_seg_size if we coalesce the
148 if (unlikely(ALIGN(dma_len + dma_offset + startsg->length, IOVP_SIZE) >
153 * Next see if we can append the next chunk (i.e.
154 * it must end on one page and begin on another, or
155 * it must start on the same address as the previous
158 if (unlikely((prev_end != sg_start) ||
159 ((prev_end | sg_start) & ~PAGE_MASK)))
162 dma_len += startsg->length;
167 ** Terminate last VCONTIG block.
168 ** Allocate space for DMA stream.
170 sg_dma_len(contig_sg) = dma_len;
171 dma_len = ALIGN(dma_len + dma_offset, IOVP_SIZE);
172 sg_dma_address(contig_sg) =
174 | (iommu_alloc_range(ioc, dev, dma_len) << IOVP_SHIFT)