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
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#include <linux/hardirq.h>
40#include <linux/gfp.h>
41
42#include <asm/pgalloc.h>
43#include <asm/io.h>
44#include <asm/mmu_context.h>
45#include <asm/pgtable.h>
46#include <asm/mmu.h>
47#include <asm/uaccess.h>
48#include <asm/smp.h>
49
50static int map_page(unsigned long va, unsigned long pa, pgprot_t prot)
51{
52	pgd_t *pge;
53	pud_t *pue;
54	pmd_t *pme;
55	pte_t *pte;
56	int err = -ENOMEM;
57
58	/* Use upper 10 bits of VA to index the first level map */
59	pge = pgd_offset_k(va);
60	pue = pud_offset(pge, va);
61	pme = pmd_offset(pue, va);
62
63	/* Use middle 10 bits of VA to index the second-level map */
64	pte = pte_alloc_kernel(pme, va);
65	if (pte != 0) {
66		err = 0;
67		set_pte(pte, mk_pte_phys(pa & PAGE_MASK, prot));
68	}
69
70	return err;
71}
72
73/*
74 * This function will allocate the requested contiguous pages and
75 * map them into the kernel's vmalloc() space.  This is done so we
76 * get unique mapping for these pages, outside of the kernel's 1:1
77 * virtual:physical mapping.  This is necessary so we can cover large
78 * portions of the kernel with single large page TLB entries, and
79 * still get unique uncached pages for consistent DMA.
80 */
81void *consistent_alloc(gfp_t gfp, size_t size, dma_addr_t *dma_handle)
82{
83	struct vm_struct *area;
84	unsigned long page, va, pa;
85	void *ret;
86	int order, err, i;
87
88	if (in_interrupt())
89		BUG();
90
91	/* only allocate page size areas */
92	size = PAGE_ALIGN(size);
93	order = get_order(size);
94
95	page = __get_free_pages(gfp, order);
96	if (!page) {
97		BUG();
98		return NULL;
99	}
100
101	/* allocate some common virtual space to map the new pages */
102	area = get_vm_area(size, VM_ALLOC);
103	if (area == 0) {
104		free_pages(page, order);
105		return NULL;
106	}
107	va = VMALLOC_VMADDR(area->addr);
108	ret = (void *) va;
109
110	/* this gives us the real physical address of the first page */
111	*dma_handle = pa = virt_to_bus((void *) page);
112
113	/* set refcount=1 on all pages in an order>0 allocation so that vfree() will actually free
114	 * all pages that were allocated.
115	 */
116	if (order > 0) {
117		struct page *rpage = virt_to_page(page);
118		split_page(rpage, order);
119	}
120
121	err = 0;
122	for (i = 0; i < size && err == 0; i += PAGE_SIZE)
123		err = map_page(va + i, pa + i, PAGE_KERNEL_NOCACHE);
124
125	if (err) {
126		vfree((void *) va);
127		return NULL;
128	}
129
130	/* we need to ensure that there are no cachelines in use, or worse dirty in this area
131	 * - can't do until after virtual address mappings are created
132	 */
133	frv_cache_invalidate(va, va + size);
134
135	return ret;
136}
137
138/*
139 * free page(s) as defined by the above mapping.
140 */
141void consistent_free(void *vaddr)
142{
143	if (in_interrupt())
144		BUG();
145	vfree(vaddr);
146}
147
148/*
149 * make an area consistent.
150 */
151void consistent_sync(void *vaddr, size_t size, int direction)
152{
153	unsigned long start = (unsigned long) vaddr;
154	unsigned long end   = start + size;
155
156	switch (direction) {
157	case PCI_DMA_NONE:
158		BUG();
159	case PCI_DMA_FROMDEVICE:	/* invalidate only */
160		frv_cache_invalidate(start, end);
161		break;
162	case PCI_DMA_TODEVICE:		/* writeback only */
163		frv_dcache_writeback(start, end);
164		break;
165	case PCI_DMA_BIDIRECTIONAL:	/* writeback and invalidate */
166		frv_dcache_writeback(start, end);
167		break;
168	}
169}
170
171/*
172 * consistent_sync_page make a page are consistent. identical
173 * to consistent_sync, but takes a struct page instead of a virtual address
174 */
175
176void consistent_sync_page(struct page *page, unsigned long offset,
177			  size_t size, int direction)
178{
179	void *start;
180
181	start = page_address(page) + offset;
182	consistent_sync(start, size, direction);
183}
184