Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* dma-alloc.c: consistent DMA memory allocation |
| 2 | * |
| 3 | * Derived from arch/ppc/mm/cachemap.c |
| 4 | * |
| 5 | * PowerPC version derived from arch/arm/mm/consistent.c |
| 6 | * Copyright (C) 2001 Dan Malek (dmalek@jlc.net) |
| 7 | * |
| 8 | * linux/arch/arm/mm/consistent.c |
| 9 | * |
| 10 | * Copyright (C) 2000 Russell King |
| 11 | * |
| 12 | * Consistent memory allocators. Used for DMA devices that want to |
| 13 | * share uncached memory with the processor core. The function return |
| 14 | * is the virtual address and 'dma_handle' is the physical address. |
| 15 | * Mostly stolen from the ARM port, with some changes for PowerPC. |
| 16 | * -- Dan |
| 17 | * Modified for 36-bit support. -Matt |
| 18 | * |
| 19 | * This program is free software; you can redistribute it and/or modify |
| 20 | * it under the terms of the GNU General Public License version 2 as |
| 21 | * published by the Free Software Foundation. |
| 22 | */ |
| 23 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/module.h> |
| 25 | #include <linux/signal.h> |
| 26 | #include <linux/sched.h> |
| 27 | #include <linux/kernel.h> |
| 28 | #include <linux/errno.h> |
| 29 | #include <linux/string.h> |
| 30 | #include <linux/types.h> |
| 31 | #include <linux/ptrace.h> |
| 32 | #include <linux/mman.h> |
| 33 | #include <linux/mm.h> |
| 34 | #include <linux/swap.h> |
| 35 | #include <linux/stddef.h> |
| 36 | #include <linux/vmalloc.h> |
| 37 | #include <linux/init.h> |
| 38 | #include <linux/pci.h> |
| 39 | |
| 40 | #include <asm/pgalloc.h> |
| 41 | #include <asm/io.h> |
| 42 | #include <asm/hardirq.h> |
| 43 | #include <asm/mmu_context.h> |
| 44 | #include <asm/pgtable.h> |
| 45 | #include <asm/mmu.h> |
| 46 | #include <asm/uaccess.h> |
| 47 | #include <asm/smp.h> |
| 48 | |
| 49 | static int map_page(unsigned long va, unsigned long pa, pgprot_t prot) |
| 50 | { |
| 51 | pgd_t *pge; |
| 52 | pud_t *pue; |
| 53 | pmd_t *pme; |
| 54 | pte_t *pte; |
| 55 | int err = -ENOMEM; |
| 56 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | /* Use upper 10 bits of VA to index the first level map */ |
| 58 | pge = pgd_offset_k(va); |
| 59 | pue = pud_offset(pge, va); |
| 60 | pme = pmd_offset(pue, va); |
| 61 | |
| 62 | /* Use middle 10 bits of VA to index the second-level map */ |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 63 | pte = pte_alloc_kernel(pme, va); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | if (pte != 0) { |
| 65 | err = 0; |
| 66 | set_pte(pte, mk_pte_phys(pa & PAGE_MASK, prot)); |
| 67 | } |
| 68 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | return err; |
| 70 | } |
| 71 | |
| 72 | /* |
| 73 | * This function will allocate the requested contiguous pages and |
| 74 | * map them into the kernel's vmalloc() space. This is done so we |
| 75 | * get unique mapping for these pages, outside of the kernel's 1:1 |
| 76 | * virtual:physical mapping. This is necessary so we can cover large |
| 77 | * portions of the kernel with single large page TLB entries, and |
| 78 | * still get unique uncached pages for consistent DMA. |
| 79 | */ |
Al Viro | a5da7d3 | 2005-10-21 03:21:18 -0400 | [diff] [blame] | 80 | void *consistent_alloc(gfp_t gfp, size_t size, dma_addr_t *dma_handle) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | { |
| 82 | struct vm_struct *area; |
| 83 | unsigned long page, va, pa; |
| 84 | void *ret; |
| 85 | int order, err, i; |
| 86 | |
| 87 | if (in_interrupt()) |
| 88 | BUG(); |
| 89 | |
| 90 | /* only allocate page size areas */ |
| 91 | size = PAGE_ALIGN(size); |
| 92 | order = get_order(size); |
| 93 | |
| 94 | page = __get_free_pages(gfp, order); |
| 95 | if (!page) { |
| 96 | BUG(); |
| 97 | return NULL; |
| 98 | } |
| 99 | |
| 100 | /* allocate some common virtual space to map the new pages */ |
| 101 | area = get_vm_area(size, VM_ALLOC); |
| 102 | if (area == 0) { |
| 103 | free_pages(page, order); |
| 104 | return NULL; |
| 105 | } |
| 106 | va = VMALLOC_VMADDR(area->addr); |
| 107 | ret = (void *) va; |
| 108 | |
| 109 | /* this gives us the real physical address of the first page */ |
| 110 | *dma_handle = pa = virt_to_bus((void *) page); |
| 111 | |
| 112 | /* set refcount=1 on all pages in an order>0 allocation so that vfree() will actually free |
| 113 | * all pages that were allocated. |
| 114 | */ |
| 115 | if (order > 0) { |
| 116 | struct page *rpage = virt_to_page(page); |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 117 | split_page(rpage, order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | } |
| 119 | |
| 120 | err = 0; |
| 121 | for (i = 0; i < size && err == 0; i += PAGE_SIZE) |
| 122 | err = map_page(va + i, pa + i, PAGE_KERNEL_NOCACHE); |
| 123 | |
| 124 | if (err) { |
| 125 | vfree((void *) va); |
| 126 | return NULL; |
| 127 | } |
| 128 | |
| 129 | /* we need to ensure that there are no cachelines in use, or worse dirty in this area |
| 130 | * - can't do until after virtual address mappings are created |
| 131 | */ |
| 132 | frv_cache_invalidate(va, va + size); |
| 133 | |
| 134 | return ret; |
| 135 | } |
| 136 | |
| 137 | /* |
| 138 | * free page(s) as defined by the above mapping. |
| 139 | */ |
| 140 | void consistent_free(void *vaddr) |
| 141 | { |
| 142 | if (in_interrupt()) |
| 143 | BUG(); |
| 144 | vfree(vaddr); |
| 145 | } |
| 146 | |
| 147 | /* |
| 148 | * make an area consistent. |
| 149 | */ |
| 150 | void consistent_sync(void *vaddr, size_t size, int direction) |
| 151 | { |
| 152 | unsigned long start = (unsigned long) vaddr; |
| 153 | unsigned long end = start + size; |
| 154 | |
| 155 | switch (direction) { |
| 156 | case PCI_DMA_NONE: |
| 157 | BUG(); |
| 158 | case PCI_DMA_FROMDEVICE: /* invalidate only */ |
| 159 | frv_cache_invalidate(start, end); |
| 160 | break; |
| 161 | case PCI_DMA_TODEVICE: /* writeback only */ |
| 162 | frv_dcache_writeback(start, end); |
| 163 | break; |
| 164 | case PCI_DMA_BIDIRECTIONAL: /* writeback and invalidate */ |
| 165 | frv_dcache_writeback(start, end); |
| 166 | break; |
| 167 | } |
| 168 | } |
| 169 | |
| 170 | /* |
| 171 | * consistent_sync_page make a page are consistent. identical |
| 172 | * to consistent_sync, but takes a struct page instead of a virtual address |
| 173 | */ |
| 174 | |
| 175 | void consistent_sync_page(struct page *page, unsigned long offset, |
| 176 | size_t size, int direction) |
| 177 | { |
| 178 | void *start; |
| 179 | |
| 180 | start = page_address(page) + offset; |
| 181 | consistent_sync(start, size, direction); |
| 182 | } |