diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2009-05-27 13:58:23 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2009-05-27 13:58:23 -0400 |
commit | b14f3bd90d2202a83f36eac85bcb3db0fba7d6a6 (patch) | |
tree | bad35a2937e40ca211a70dbfbb5432b864eab724 /arch | |
parent | 911e690e70540f009125bacd16c017eb1a7b1916 (diff) | |
parent | 8b31e49d1d75729c1da9009664ba52abd1adc628 (diff) |
Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc
* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc:
powerpc: Fix up dma_alloc_coherent() on platforms without cache coherency.
powerpc: Minor cleanups of kernel virt address space definitions
powerpc: Move dma-noncoherent.c from arch/powerpc/lib to arch/powerpc/mm
Revert "powerpc: Rework dma-noncoherent to use generic vmalloc layer"
Diffstat (limited to 'arch')
-rw-r--r-- | arch/powerpc/Kconfig | 12 | ||||
-rw-r--r-- | arch/powerpc/include/asm/dma-mapping.h | 6 | ||||
-rw-r--r-- | arch/powerpc/include/asm/fixmap.h | 4 | ||||
-rw-r--r-- | arch/powerpc/include/asm/pgtable-ppc32.h | 26 | ||||
-rw-r--r-- | arch/powerpc/kernel/dma.c | 2 | ||||
-rw-r--r-- | arch/powerpc/lib/Makefile | 1 | ||||
-rw-r--r-- | arch/powerpc/lib/dma-noncoherent.c | 237 | ||||
-rw-r--r-- | arch/powerpc/mm/Makefile | 1 | ||||
-rw-r--r-- | arch/powerpc/mm/dma-noncoherent.c | 400 | ||||
-rw-r--r-- | arch/powerpc/mm/init_32.c | 8 | ||||
-rw-r--r-- | arch/powerpc/mm/mem.c | 17 | ||||
-rw-r--r-- | arch/powerpc/mm/pgtable_32.c | 2 |
12 files changed, 463 insertions, 253 deletions
diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig index a0d1146a0578..cdc9a6ff4be8 100644 --- a/arch/powerpc/Kconfig +++ b/arch/powerpc/Kconfig | |||
@@ -868,6 +868,18 @@ config TASK_SIZE | |||
868 | default "0x80000000" if PPC_PREP || PPC_8xx | 868 | default "0x80000000" if PPC_PREP || PPC_8xx |
869 | default "0xc0000000" | 869 | default "0xc0000000" |
870 | 870 | ||
871 | config CONSISTENT_SIZE_BOOL | ||
872 | bool "Set custom consistent memory pool size" | ||
873 | depends on ADVANCED_OPTIONS && NOT_COHERENT_CACHE | ||
874 | help | ||
875 | This option allows you to set the size of the | ||
876 | consistent memory pool. This pool of virtual memory | ||
877 | is used to make consistent memory allocations. | ||
878 | |||
879 | config CONSISTENT_SIZE | ||
880 | hex "Size of consistent memory pool" if CONSISTENT_SIZE_BOOL | ||
881 | default "0x00200000" if NOT_COHERENT_CACHE | ||
882 | |||
871 | config PIN_TLB | 883 | config PIN_TLB |
872 | bool "Pinned Kernel TLBs (860 ONLY)" | 884 | bool "Pinned Kernel TLBs (860 ONLY)" |
873 | depends on ADVANCED_OPTIONS && 8xx | 885 | depends on ADVANCED_OPTIONS && 8xx |
diff --git a/arch/powerpc/include/asm/dma-mapping.h b/arch/powerpc/include/asm/dma-mapping.h index c69f2b5f0cc4..cb448d68452c 100644 --- a/arch/powerpc/include/asm/dma-mapping.h +++ b/arch/powerpc/include/asm/dma-mapping.h | |||
@@ -26,7 +26,9 @@ | |||
26 | * allocate the space "normally" and use the cache management functions | 26 | * allocate the space "normally" and use the cache management functions |
27 | * to ensure it is consistent. | 27 | * to ensure it is consistent. |
28 | */ | 28 | */ |
29 | extern void *__dma_alloc_coherent(size_t size, dma_addr_t *handle, gfp_t gfp); | 29 | struct device; |
30 | extern void *__dma_alloc_coherent(struct device *dev, size_t size, | ||
31 | dma_addr_t *handle, gfp_t gfp); | ||
30 | extern void __dma_free_coherent(size_t size, void *vaddr); | 32 | extern void __dma_free_coherent(size_t size, void *vaddr); |
31 | extern void __dma_sync(void *vaddr, size_t size, int direction); | 33 | extern void __dma_sync(void *vaddr, size_t size, int direction); |
32 | extern void __dma_sync_page(struct page *page, unsigned long offset, | 34 | extern void __dma_sync_page(struct page *page, unsigned long offset, |
@@ -37,7 +39,7 @@ extern void __dma_sync_page(struct page *page, unsigned long offset, | |||
37 | * Cache coherent cores. | 39 | * Cache coherent cores. |
38 | */ | 40 | */ |
39 | 41 | ||
40 | #define __dma_alloc_coherent(gfp, size, handle) NULL | 42 | #define __dma_alloc_coherent(dev, gfp, size, handle) NULL |
41 | #define __dma_free_coherent(size, addr) ((void)0) | 43 | #define __dma_free_coherent(size, addr) ((void)0) |
42 | #define __dma_sync(addr, size, rw) ((void)0) | 44 | #define __dma_sync(addr, size, rw) ((void)0) |
43 | #define __dma_sync_page(pg, off, sz, rw) ((void)0) | 45 | #define __dma_sync_page(pg, off, sz, rw) ((void)0) |
diff --git a/arch/powerpc/include/asm/fixmap.h b/arch/powerpc/include/asm/fixmap.h index d60fd18f428c..f1f4e23a84e9 100644 --- a/arch/powerpc/include/asm/fixmap.h +++ b/arch/powerpc/include/asm/fixmap.h | |||
@@ -14,8 +14,6 @@ | |||
14 | #ifndef _ASM_FIXMAP_H | 14 | #ifndef _ASM_FIXMAP_H |
15 | #define _ASM_FIXMAP_H | 15 | #define _ASM_FIXMAP_H |
16 | 16 | ||
17 | extern unsigned long FIXADDR_TOP; | ||
18 | |||
19 | #ifndef __ASSEMBLY__ | 17 | #ifndef __ASSEMBLY__ |
20 | #include <linux/kernel.h> | 18 | #include <linux/kernel.h> |
21 | #include <asm/page.h> | 19 | #include <asm/page.h> |
@@ -24,6 +22,8 @@ extern unsigned long FIXADDR_TOP; | |||
24 | #include <asm/kmap_types.h> | 22 | #include <asm/kmap_types.h> |
25 | #endif | 23 | #endif |
26 | 24 | ||
25 | #define FIXADDR_TOP ((unsigned long)(-PAGE_SIZE)) | ||
26 | |||
27 | /* | 27 | /* |
28 | * Here we define all the compile-time 'special' virtual | 28 | * Here we define all the compile-time 'special' virtual |
29 | * addresses. The point is to have a constant address at | 29 | * addresses. The point is to have a constant address at |
diff --git a/arch/powerpc/include/asm/pgtable-ppc32.h b/arch/powerpc/include/asm/pgtable-ppc32.h index ba45c997830f..c9ff9d75990e 100644 --- a/arch/powerpc/include/asm/pgtable-ppc32.h +++ b/arch/powerpc/include/asm/pgtable-ppc32.h | |||
@@ -10,7 +10,7 @@ | |||
10 | 10 | ||
11 | extern unsigned long va_to_phys(unsigned long address); | 11 | extern unsigned long va_to_phys(unsigned long address); |
12 | extern pte_t *va_to_pte(unsigned long address); | 12 | extern pte_t *va_to_pte(unsigned long address); |
13 | extern unsigned long ioremap_bot, ioremap_base; | 13 | extern unsigned long ioremap_bot; |
14 | 14 | ||
15 | #ifdef CONFIG_44x | 15 | #ifdef CONFIG_44x |
16 | extern int icache_44x_need_flush; | 16 | extern int icache_44x_need_flush; |
@@ -56,8 +56,30 @@ extern int icache_44x_need_flush; | |||
56 | printk("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e)) | 56 | printk("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e)) |
57 | 57 | ||
58 | /* | 58 | /* |
59 | * This is the bottom of the PKMAP area with HIGHMEM or an arbitrary | ||
60 | * value (for now) on others, from where we can start layout kernel | ||
61 | * virtual space that goes below PKMAP and FIXMAP | ||
62 | */ | ||
63 | #ifdef CONFIG_HIGHMEM | ||
64 | #define KVIRT_TOP PKMAP_BASE | ||
65 | #else | ||
66 | #define KVIRT_TOP (0xfe000000UL) /* for now, could be FIXMAP_BASE ? */ | ||
67 | #endif | ||
68 | |||
69 | /* | ||
70 | * ioremap_bot starts at that address. Early ioremaps move down from there, | ||
71 | * until mem_init() at which point this becomes the top of the vmalloc | ||
72 | * and ioremap space | ||
73 | */ | ||
74 | #ifdef CONFIG_NOT_COHERENT_CACHE | ||
75 | #define IOREMAP_TOP ((KVIRT_TOP - CONFIG_CONSISTENT_SIZE) & PAGE_MASK) | ||
76 | #else | ||
77 | #define IOREMAP_TOP KVIRT_TOP | ||
78 | #endif | ||
79 | |||
80 | /* | ||
59 | * Just any arbitrary offset to the start of the vmalloc VM area: the | 81 | * Just any arbitrary offset to the start of the vmalloc VM area: the |
60 | * current 64MB value just means that there will be a 64MB "hole" after the | 82 | * current 16MB value just means that there will be a 64MB "hole" after the |
61 | * physical memory until the kernel virtual memory starts. That means that | 83 | * physical memory until the kernel virtual memory starts. That means that |
62 | * any out-of-bounds memory accesses will hopefully be caught. | 84 | * any out-of-bounds memory accesses will hopefully be caught. |
63 | * The vmalloc() routines leaves a hole of 4kB between each vmalloced | 85 | * The vmalloc() routines leaves a hole of 4kB between each vmalloced |
diff --git a/arch/powerpc/kernel/dma.c b/arch/powerpc/kernel/dma.c index 53c7788cba78..6b02793dc75b 100644 --- a/arch/powerpc/kernel/dma.c +++ b/arch/powerpc/kernel/dma.c | |||
@@ -32,7 +32,7 @@ void *dma_direct_alloc_coherent(struct device *dev, size_t size, | |||
32 | { | 32 | { |
33 | void *ret; | 33 | void *ret; |
34 | #ifdef CONFIG_NOT_COHERENT_CACHE | 34 | #ifdef CONFIG_NOT_COHERENT_CACHE |
35 | ret = __dma_alloc_coherent(size, dma_handle, flag); | 35 | ret = __dma_alloc_coherent(dev, size, dma_handle, flag); |
36 | if (ret == NULL) | 36 | if (ret == NULL) |
37 | return NULL; | 37 | return NULL; |
38 | *dma_handle += get_dma_direct_offset(dev); | 38 | *dma_handle += get_dma_direct_offset(dev); |
diff --git a/arch/powerpc/lib/Makefile b/arch/powerpc/lib/Makefile index 8db35278a4b4..29b742b90f1f 100644 --- a/arch/powerpc/lib/Makefile +++ b/arch/powerpc/lib/Makefile | |||
@@ -18,7 +18,6 @@ obj-$(CONFIG_PPC64) += copypage_64.o copyuser_64.o \ | |||
18 | memcpy_64.o usercopy_64.o mem_64.o string.o | 18 | memcpy_64.o usercopy_64.o mem_64.o string.o |
19 | obj-$(CONFIG_XMON) += sstep.o | 19 | obj-$(CONFIG_XMON) += sstep.o |
20 | obj-$(CONFIG_KPROBES) += sstep.o | 20 | obj-$(CONFIG_KPROBES) += sstep.o |
21 | obj-$(CONFIG_NOT_COHERENT_CACHE) += dma-noncoherent.o | ||
22 | 21 | ||
23 | ifeq ($(CONFIG_PPC64),y) | 22 | ifeq ($(CONFIG_PPC64),y) |
24 | obj-$(CONFIG_SMP) += locks.o | 23 | obj-$(CONFIG_SMP) += locks.o |
diff --git a/arch/powerpc/lib/dma-noncoherent.c b/arch/powerpc/lib/dma-noncoherent.c deleted file mode 100644 index 005a28d380af..000000000000 --- a/arch/powerpc/lib/dma-noncoherent.c +++ /dev/null | |||
@@ -1,237 +0,0 @@ | |||
1 | /* | ||
2 | * PowerPC version derived from arch/arm/mm/consistent.c | ||
3 | * Copyright (C) 2001 Dan Malek (dmalek@jlc.net) | ||
4 | * | ||
5 | * Copyright (C) 2000 Russell King | ||
6 | * | ||
7 | * Consistent memory allocators. Used for DMA devices that want to | ||
8 | * share uncached memory with the processor core. The function return | ||
9 | * is the virtual address and 'dma_handle' is the physical address. | ||
10 | * Mostly stolen from the ARM port, with some changes for PowerPC. | ||
11 | * -- Dan | ||
12 | * | ||
13 | * Reorganized to get rid of the arch-specific consistent_* functions | ||
14 | * and provide non-coherent implementations for the DMA API. -Matt | ||
15 | * | ||
16 | * Added in_interrupt() safe dma_alloc_coherent()/dma_free_coherent() | ||
17 | * implementation. This is pulled straight from ARM and barely | ||
18 | * modified. -Matt | ||
19 | * | ||
20 | * This program is free software; you can redistribute it and/or modify | ||
21 | * it under the terms of the GNU General Public License version 2 as | ||
22 | * published by the Free Software Foundation. | ||
23 | */ | ||
24 | |||
25 | #include <linux/sched.h> | ||
26 | #include <linux/kernel.h> | ||
27 | #include <linux/errno.h> | ||
28 | #include <linux/string.h> | ||
29 | #include <linux/types.h> | ||
30 | #include <linux/highmem.h> | ||
31 | #include <linux/dma-mapping.h> | ||
32 | #include <linux/vmalloc.h> | ||
33 | |||
34 | #include <asm/tlbflush.h> | ||
35 | |||
36 | /* | ||
37 | * Allocate DMA-coherent memory space and return both the kernel remapped | ||
38 | * virtual and bus address for that space. | ||
39 | */ | ||
40 | void * | ||
41 | __dma_alloc_coherent(size_t size, dma_addr_t *handle, gfp_t gfp) | ||
42 | { | ||
43 | struct page *page; | ||
44 | unsigned long order; | ||
45 | int i; | ||
46 | unsigned int nr_pages = PAGE_ALIGN(size)>>PAGE_SHIFT; | ||
47 | unsigned int array_size = nr_pages * sizeof(struct page *); | ||
48 | struct page **pages; | ||
49 | struct page *end; | ||
50 | u64 mask = 0x00ffffff, limit; /* ISA default */ | ||
51 | struct vm_struct *area; | ||
52 | |||
53 | BUG_ON(!mem_init_done); | ||
54 | size = PAGE_ALIGN(size); | ||
55 | limit = (mask + 1) & ~mask; | ||
56 | if (limit && size >= limit) { | ||
57 | printk(KERN_WARNING "coherent allocation too big (requested " | ||
58 | "%#x mask %#Lx)\n", size, mask); | ||
59 | return NULL; | ||
60 | } | ||
61 | |||
62 | order = get_order(size); | ||
63 | |||
64 | if (mask != 0xffffffff) | ||
65 | gfp |= GFP_DMA; | ||
66 | |||
67 | page = alloc_pages(gfp, order); | ||
68 | if (!page) | ||
69 | goto no_page; | ||
70 | |||
71 | end = page + (1 << order); | ||
72 | |||
73 | /* | ||
74 | * Invalidate any data that might be lurking in the | ||
75 | * kernel direct-mapped region for device DMA. | ||
76 | */ | ||
77 | { | ||
78 | unsigned long kaddr = (unsigned long)page_address(page); | ||
79 | memset(page_address(page), 0, size); | ||
80 | flush_dcache_range(kaddr, kaddr + size); | ||
81 | } | ||
82 | |||
83 | split_page(page, order); | ||
84 | |||
85 | /* | ||
86 | * Set the "dma handle" | ||
87 | */ | ||
88 | *handle = page_to_phys(page); | ||
89 | |||
90 | area = get_vm_area_caller(size, VM_IOREMAP, | ||
91 | __builtin_return_address(1)); | ||
92 | if (!area) | ||
93 | goto out_free_pages; | ||
94 | |||
95 | if (array_size > PAGE_SIZE) { | ||
96 | pages = vmalloc(array_size); | ||
97 | area->flags |= VM_VPAGES; | ||
98 | } else { | ||
99 | pages = kmalloc(array_size, GFP_KERNEL); | ||
100 | } | ||
101 | if (!pages) | ||
102 | goto out_free_area; | ||
103 | |||
104 | area->pages = pages; | ||
105 | area->nr_pages = nr_pages; | ||
106 | |||
107 | for (i = 0; i < nr_pages; i++) | ||
108 | pages[i] = page + i; | ||
109 | |||
110 | if (map_vm_area(area, pgprot_noncached(PAGE_KERNEL), &pages)) | ||
111 | goto out_unmap; | ||
112 | |||
113 | /* | ||
114 | * Free the otherwise unused pages. | ||
115 | */ | ||
116 | page += nr_pages; | ||
117 | while (page < end) { | ||
118 | __free_page(page); | ||
119 | page++; | ||
120 | } | ||
121 | |||
122 | return area->addr; | ||
123 | out_unmap: | ||
124 | vunmap(area->addr); | ||
125 | if (array_size > PAGE_SIZE) | ||
126 | vfree(pages); | ||
127 | else | ||
128 | kfree(pages); | ||
129 | goto out_free_pages; | ||
130 | out_free_area: | ||
131 | free_vm_area(area); | ||
132 | out_free_pages: | ||
133 | if (page) | ||
134 | __free_pages(page, order); | ||
135 | no_page: | ||
136 | return NULL; | ||
137 | } | ||
138 | EXPORT_SYMBOL(__dma_alloc_coherent); | ||
139 | |||
140 | /* | ||
141 | * free a page as defined by the above mapping. | ||
142 | */ | ||
143 | void __dma_free_coherent(size_t size, void *vaddr) | ||
144 | { | ||
145 | vfree(vaddr); | ||
146 | |||
147 | } | ||
148 | EXPORT_SYMBOL(__dma_free_coherent); | ||
149 | |||
150 | /* | ||
151 | * make an area consistent. | ||
152 | */ | ||
153 | void __dma_sync(void *vaddr, size_t size, int direction) | ||
154 | { | ||
155 | unsigned long start = (unsigned long)vaddr; | ||
156 | unsigned long end = start + size; | ||
157 | |||
158 | switch (direction) { | ||
159 | case DMA_NONE: | ||
160 | BUG(); | ||
161 | case DMA_FROM_DEVICE: | ||
162 | /* | ||
163 | * invalidate only when cache-line aligned otherwise there is | ||
164 | * the potential for discarding uncommitted data from the cache | ||
165 | */ | ||
166 | if ((start & (L1_CACHE_BYTES - 1)) || (size & (L1_CACHE_BYTES - 1))) | ||
167 | flush_dcache_range(start, end); | ||
168 | else | ||
169 | invalidate_dcache_range(start, end); | ||
170 | break; | ||
171 | case DMA_TO_DEVICE: /* writeback only */ | ||
172 | clean_dcache_range(start, end); | ||
173 | break; | ||
174 | case DMA_BIDIRECTIONAL: /* writeback and invalidate */ | ||
175 | flush_dcache_range(start, end); | ||
176 | break; | ||
177 | } | ||
178 | } | ||
179 | EXPORT_SYMBOL(__dma_sync); | ||
180 | |||
181 | #ifdef CONFIG_HIGHMEM | ||
182 | /* | ||
183 | * __dma_sync_page() implementation for systems using highmem. | ||
184 | * In this case, each page of a buffer must be kmapped/kunmapped | ||
185 | * in order to have a virtual address for __dma_sync(). This must | ||
186 | * not sleep so kmap_atomic()/kunmap_atomic() are used. | ||
187 | * | ||
188 | * Note: yes, it is possible and correct to have a buffer extend | ||
189 | * beyond the first page. | ||
190 | */ | ||
191 | static inline void __dma_sync_page_highmem(struct page *page, | ||
192 | unsigned long offset, size_t size, int direction) | ||
193 | { | ||
194 | size_t seg_size = min((size_t)(PAGE_SIZE - offset), size); | ||
195 | size_t cur_size = seg_size; | ||
196 | unsigned long flags, start, seg_offset = offset; | ||
197 | int nr_segs = 1 + ((size - seg_size) + PAGE_SIZE - 1)/PAGE_SIZE; | ||
198 | int seg_nr = 0; | ||
199 | |||
200 | local_irq_save(flags); | ||
201 | |||
202 | do { | ||
203 | start = (unsigned long)kmap_atomic(page + seg_nr, | ||
204 | KM_PPC_SYNC_PAGE) + seg_offset; | ||
205 | |||
206 | /* Sync this buffer segment */ | ||
207 | __dma_sync((void *)start, seg_size, direction); | ||
208 | kunmap_atomic((void *)start, KM_PPC_SYNC_PAGE); | ||
209 | seg_nr++; | ||
210 | |||
211 | /* Calculate next buffer segment size */ | ||
212 | seg_size = min((size_t)PAGE_SIZE, size - cur_size); | ||
213 | |||
214 | /* Add the segment size to our running total */ | ||
215 | cur_size += seg_size; | ||
216 | seg_offset = 0; | ||
217 | } while (seg_nr < nr_segs); | ||
218 | |||
219 | local_irq_restore(flags); | ||
220 | } | ||
221 | #endif /* CONFIG_HIGHMEM */ | ||
222 | |||
223 | /* | ||
224 | * __dma_sync_page makes memory consistent. identical to __dma_sync, but | ||
225 | * takes a struct page instead of a virtual address | ||
226 | */ | ||
227 | void __dma_sync_page(struct page *page, unsigned long offset, | ||
228 | size_t size, int direction) | ||
229 | { | ||
230 | #ifdef CONFIG_HIGHMEM | ||
231 | __dma_sync_page_highmem(page, offset, size, direction); | ||
232 | #else | ||
233 | unsigned long start = (unsigned long)page_address(page) + offset; | ||
234 | __dma_sync((void *)start, size, direction); | ||
235 | #endif | ||
236 | } | ||
237 | EXPORT_SYMBOL(__dma_sync_page); | ||
diff --git a/arch/powerpc/mm/Makefile b/arch/powerpc/mm/Makefile index 17290bcedc5e..b746f4ca4209 100644 --- a/arch/powerpc/mm/Makefile +++ b/arch/powerpc/mm/Makefile | |||
@@ -26,3 +26,4 @@ obj-$(CONFIG_NEED_MULTIPLE_NODES) += numa.o | |||
26 | obj-$(CONFIG_PPC_MM_SLICES) += slice.o | 26 | obj-$(CONFIG_PPC_MM_SLICES) += slice.o |
27 | obj-$(CONFIG_HUGETLB_PAGE) += hugetlbpage.o | 27 | obj-$(CONFIG_HUGETLB_PAGE) += hugetlbpage.o |
28 | obj-$(CONFIG_PPC_SUBPAGE_PROT) += subpage-prot.o | 28 | obj-$(CONFIG_PPC_SUBPAGE_PROT) += subpage-prot.o |
29 | obj-$(CONFIG_NOT_COHERENT_CACHE) += dma-noncoherent.o | ||
diff --git a/arch/powerpc/mm/dma-noncoherent.c b/arch/powerpc/mm/dma-noncoherent.c new file mode 100644 index 000000000000..36692f5c9a76 --- /dev/null +++ b/arch/powerpc/mm/dma-noncoherent.c | |||
@@ -0,0 +1,400 @@ | |||
1 | /* | ||
2 | * PowerPC version derived from arch/arm/mm/consistent.c | ||
3 | * Copyright (C) 2001 Dan Malek (dmalek@jlc.net) | ||
4 | * | ||
5 | * Copyright (C) 2000 Russell King | ||
6 | * | ||
7 | * Consistent memory allocators. Used for DMA devices that want to | ||
8 | * share uncached memory with the processor core. The function return | ||
9 | * is the virtual address and 'dma_handle' is the physical address. | ||
10 | * Mostly stolen from the ARM port, with some changes for PowerPC. | ||
11 | * -- Dan | ||
12 | * | ||
13 | * Reorganized to get rid of the arch-specific consistent_* functions | ||
14 | * and provide non-coherent implementations for the DMA API. -Matt | ||
15 | * | ||
16 | * Added in_interrupt() safe dma_alloc_coherent()/dma_free_coherent() | ||
17 | * implementation. This is pulled straight from ARM and barely | ||
18 | * modified. -Matt | ||
19 | * | ||
20 | * This program is free software; you can redistribute it and/or modify | ||
21 | * it under the terms of the GNU General Public License version 2 as | ||
22 | * published by the Free Software Foundation. | ||
23 | */ | ||
24 | |||
25 | #include <linux/sched.h> | ||
26 | #include <linux/kernel.h> | ||
27 | #include <linux/errno.h> | ||
28 | #include <linux/string.h> | ||
29 | #include <linux/types.h> | ||
30 | #include <linux/highmem.h> | ||
31 | #include <linux/dma-mapping.h> | ||
32 | |||
33 | #include <asm/tlbflush.h> | ||
34 | |||
35 | #include "mmu_decl.h" | ||
36 | |||
37 | /* | ||
38 | * This address range defaults to a value that is safe for all | ||
39 | * platforms which currently set CONFIG_NOT_COHERENT_CACHE. It | ||
40 | * can be further configured for specific applications under | ||
41 | * the "Advanced Setup" menu. -Matt | ||
42 | */ | ||
43 | #define CONSISTENT_BASE (IOREMAP_TOP) | ||
44 | #define CONSISTENT_END (CONSISTENT_BASE + CONFIG_CONSISTENT_SIZE) | ||
45 | #define CONSISTENT_OFFSET(x) (((unsigned long)(x) - CONSISTENT_BASE) >> PAGE_SHIFT) | ||
46 | |||
47 | /* | ||
48 | * This is the page table (2MB) covering uncached, DMA consistent allocations | ||
49 | */ | ||
50 | static DEFINE_SPINLOCK(consistent_lock); | ||
51 | |||
52 | /* | ||
53 | * VM region handling support. | ||
54 | * | ||
55 | * This should become something generic, handling VM region allocations for | ||
56 | * vmalloc and similar (ioremap, module space, etc). | ||
57 | * | ||
58 | * I envisage vmalloc()'s supporting vm_struct becoming: | ||
59 | * | ||
60 | * struct vm_struct { | ||
61 | * struct vm_region region; | ||
62 | * unsigned long flags; | ||
63 | * struct page **pages; | ||
64 | * unsigned int nr_pages; | ||
65 | * unsigned long phys_addr; | ||
66 | * }; | ||
67 | * | ||
68 | * get_vm_area() would then call vm_region_alloc with an appropriate | ||
69 | * struct vm_region head (eg): | ||
70 | * | ||
71 | * struct vm_region vmalloc_head = { | ||
72 | * .vm_list = LIST_HEAD_INIT(vmalloc_head.vm_list), | ||
73 | * .vm_start = VMALLOC_START, | ||
74 | * .vm_end = VMALLOC_END, | ||
75 | * }; | ||
76 | * | ||
77 | * However, vmalloc_head.vm_start is variable (typically, it is dependent on | ||
78 | * the amount of RAM found at boot time.) I would imagine that get_vm_area() | ||
79 | * would have to initialise this each time prior to calling vm_region_alloc(). | ||
80 | */ | ||
81 | struct ppc_vm_region { | ||
82 | struct list_head vm_list; | ||
83 | unsigned long vm_start; | ||
84 | unsigned long vm_end; | ||
85 | }; | ||
86 | |||
87 | static struct ppc_vm_region consistent_head = { | ||
88 | .vm_list = LIST_HEAD_INIT(consistent_head.vm_list), | ||
89 | .vm_start = CONSISTENT_BASE, | ||
90 | .vm_end = CONSISTENT_END, | ||
91 | }; | ||
92 | |||
93 | static struct ppc_vm_region * | ||
94 | ppc_vm_region_alloc(struct ppc_vm_region *head, size_t size, gfp_t gfp) | ||
95 | { | ||
96 | unsigned long addr = head->vm_start, end = head->vm_end - size; | ||
97 | unsigned long flags; | ||
98 | struct ppc_vm_region *c, *new; | ||
99 | |||
100 | new = kmalloc(sizeof(struct ppc_vm_region), gfp); | ||
101 | if (!new) | ||
102 | goto out; | ||
103 | |||
104 | spin_lock_irqsave(&consistent_lock, flags); | ||
105 | |||
106 | list_for_each_entry(c, &head->vm_list, vm_list) { | ||
107 | if ((addr + size) < addr) | ||
108 | goto nospc; | ||
109 | if ((addr + size) <= c->vm_start) | ||
110 | goto found; | ||
111 | addr = c->vm_end; | ||
112 | if (addr > end) | ||
113 | goto nospc; | ||
114 | } | ||
115 | |||
116 | found: | ||
117 | /* | ||
118 | * Insert this entry _before_ the one we found. | ||
119 | */ | ||
120 | list_add_tail(&new->vm_list, &c->vm_list); | ||
121 | new->vm_start = addr; | ||
122 | new->vm_end = addr + size; | ||
123 | |||
124 | spin_unlock_irqrestore(&consistent_lock, flags); | ||
125 | return new; | ||
126 | |||
127 | nospc: | ||
128 | spin_unlock_irqrestore(&consistent_lock, flags); | ||
129 | kfree(new); | ||
130 | out: | ||
131 | return NULL; | ||
132 | } | ||
133 | |||
134 | static struct ppc_vm_region *ppc_vm_region_find(struct ppc_vm_region *head, unsigned long addr) | ||
135 | { | ||
136 | struct ppc_vm_region *c; | ||
137 | |||
138 | list_for_each_entry(c, &head->vm_list, vm_list) { | ||
139 | if (c->vm_start == addr) | ||
140 | goto out; | ||
141 | } | ||
142 | c = NULL; | ||
143 | out: | ||
144 | return c; | ||
145 | } | ||
146 | |||
147 | /* | ||
148 | * Allocate DMA-coherent memory space and return both the kernel remapped | ||
149 | * virtual and bus address for that space. | ||
150 | */ | ||
151 | void * | ||
152 | __dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *handle, gfp_t gfp) | ||
153 | { | ||
154 | struct page *page; | ||
155 | struct ppc_vm_region *c; | ||
156 | unsigned long order; | ||
157 | u64 mask = ISA_DMA_THRESHOLD, limit; | ||
158 | |||
159 | if (dev) { | ||
160 | mask = dev->coherent_dma_mask; | ||
161 | |||
162 | /* | ||
163 | * Sanity check the DMA mask - it must be non-zero, and | ||
164 | * must be able to be satisfied by a DMA allocation. | ||
165 | */ | ||
166 | if (mask == 0) { | ||
167 | dev_warn(dev, "coherent DMA mask is unset\n"); | ||
168 | goto no_page; | ||
169 | } | ||
170 | |||
171 | if ((~mask) & ISA_DMA_THRESHOLD) { | ||
172 | dev_warn(dev, "coherent DMA mask %#llx is smaller " | ||
173 | "than system GFP_DMA mask %#llx\n", | ||
174 | mask, (unsigned long long)ISA_DMA_THRESHOLD); | ||
175 | goto no_page; | ||
176 | } | ||
177 | } | ||
178 | |||
179 | |||
180 | size = PAGE_ALIGN(size); | ||
181 | limit = (mask + 1) & ~mask; | ||
182 | if ((limit && size >= limit) || | ||
183 | size >= (CONSISTENT_END - CONSISTENT_BASE)) { | ||
184 | printk(KERN_WARNING "coherent allocation too big (requested %#x mask %#Lx)\n", | ||
185 | size, mask); | ||
186 | return NULL; | ||
187 | } | ||
188 | |||
189 | order = get_order(size); | ||
190 | |||
191 | /* Might be useful if we ever have a real legacy DMA zone... */ | ||
192 | if (mask != 0xffffffff) | ||
193 | gfp |= GFP_DMA; | ||
194 | |||
195 | page = alloc_pages(gfp, order); | ||
196 | if (!page) | ||
197 | goto no_page; | ||
198 | |||
199 | /* | ||
200 | * Invalidate any data that might be lurking in the | ||
201 | * kernel direct-mapped region for device DMA. | ||
202 | */ | ||
203 | { | ||
204 | unsigned long kaddr = (unsigned long)page_address(page); | ||
205 | memset(page_address(page), 0, size); | ||
206 | flush_dcache_range(kaddr, kaddr + size); | ||
207 | } | ||
208 | |||
209 | /* | ||
210 | * Allocate a virtual address in the consistent mapping region. | ||
211 | */ | ||
212 | c = ppc_vm_region_alloc(&consistent_head, size, | ||
213 | gfp & ~(__GFP_DMA | __GFP_HIGHMEM)); | ||
214 | if (c) { | ||
215 | unsigned long vaddr = c->vm_start; | ||
216 | struct page *end = page + (1 << order); | ||
217 | |||
218 | split_page(page, order); | ||
219 | |||
220 | /* | ||
221 | * Set the "dma handle" | ||
222 | */ | ||
223 | *handle = page_to_phys(page); | ||
224 | |||
225 | do { | ||
226 | SetPageReserved(page); | ||
227 | map_page(vaddr, page_to_phys(page), | ||
228 | pgprot_noncached(PAGE_KERNEL)); | ||
229 | page++; | ||
230 | vaddr += PAGE_SIZE; | ||
231 | } while (size -= PAGE_SIZE); | ||
232 | |||
233 | /* | ||
234 | * Free the otherwise unused pages. | ||
235 | */ | ||
236 | while (page < end) { | ||
237 | __free_page(page); | ||
238 | page++; | ||
239 | } | ||
240 | |||
241 | return (void *)c->vm_start; | ||
242 | } | ||
243 | |||
244 | if (page) | ||
245 | __free_pages(page, order); | ||
246 | no_page: | ||
247 | return NULL; | ||
248 | } | ||
249 | EXPORT_SYMBOL(__dma_alloc_coherent); | ||
250 | |||
251 | /* | ||
252 | * free a page as defined by the above mapping. | ||
253 | */ | ||
254 | void __dma_free_coherent(size_t size, void *vaddr) | ||
255 | { | ||
256 | struct ppc_vm_region *c; | ||
257 | unsigned long flags, addr; | ||
258 | |||
259 | size = PAGE_ALIGN(size); | ||
260 | |||
261 | spin_lock_irqsave(&consistent_lock, flags); | ||
262 | |||
263 | c = ppc_vm_region_find(&consistent_head, (unsigned long)vaddr); | ||
264 | if (!c) | ||
265 | goto no_area; | ||
266 | |||
267 | if ((c->vm_end - c->vm_start) != size) { | ||
268 | printk(KERN_ERR "%s: freeing wrong coherent size (%ld != %d)\n", | ||
269 | __func__, c->vm_end - c->vm_start, size); | ||
270 | dump_stack(); | ||
271 | size = c->vm_end - c->vm_start; | ||
272 | } | ||
273 | |||
274 | addr = c->vm_start; | ||
275 | do { | ||
276 | pte_t *ptep; | ||
277 | unsigned long pfn; | ||
278 | |||
279 | ptep = pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(addr), | ||
280 | addr), | ||
281 | addr), | ||
282 | addr); | ||
283 | if (!pte_none(*ptep) && pte_present(*ptep)) { | ||
284 | pfn = pte_pfn(*ptep); | ||
285 | pte_clear(&init_mm, addr, ptep); | ||
286 | if (pfn_valid(pfn)) { | ||
287 | struct page *page = pfn_to_page(pfn); | ||
288 | |||
289 | ClearPageReserved(page); | ||
290 | __free_page(page); | ||
291 | } | ||
292 | } | ||
293 | addr += PAGE_SIZE; | ||
294 | } while (size -= PAGE_SIZE); | ||
295 | |||
296 | flush_tlb_kernel_range(c->vm_start, c->vm_end); | ||
297 | |||
298 | list_del(&c->vm_list); | ||
299 | |||
300 | spin_unlock_irqrestore(&consistent_lock, flags); | ||
301 | |||
302 | kfree(c); | ||
303 | return; | ||
304 | |||
305 | no_area: | ||
306 | spin_unlock_irqrestore(&consistent_lock, flags); | ||
307 | printk(KERN_ERR "%s: trying to free invalid coherent area: %p\n", | ||
308 | __func__, vaddr); | ||
309 | dump_stack(); | ||
310 | } | ||
311 | EXPORT_SYMBOL(__dma_free_coherent); | ||
312 | |||
313 | /* | ||
314 | * make an area consistent. | ||
315 | */ | ||
316 | void __dma_sync(void *vaddr, size_t size, int direction) | ||
317 | { | ||
318 | unsigned long start = (unsigned long)vaddr; | ||
319 | unsigned long end = start + size; | ||
320 | |||
321 | switch (direction) { | ||
322 | case DMA_NONE: | ||
323 | BUG(); | ||
324 | case DMA_FROM_DEVICE: | ||
325 | /* | ||
326 | * invalidate only when cache-line aligned otherwise there is | ||
327 | * the potential for discarding uncommitted data from the cache | ||
328 | */ | ||
329 | if ((start & (L1_CACHE_BYTES - 1)) || (size & (L1_CACHE_BYTES - 1))) | ||
330 | flush_dcache_range(start, end); | ||
331 | else | ||
332 | invalidate_dcache_range(start, end); | ||
333 | break; | ||
334 | case DMA_TO_DEVICE: /* writeback only */ | ||
335 | clean_dcache_range(start, end); | ||
336 | break; | ||
337 | case DMA_BIDIRECTIONAL: /* writeback and invalidate */ | ||
338 | flush_dcache_range(start, end); | ||
339 | break; | ||
340 | } | ||
341 | } | ||
342 | EXPORT_SYMBOL(__dma_sync); | ||
343 | |||
344 | #ifdef CONFIG_HIGHMEM | ||
345 | /* | ||
346 | * __dma_sync_page() implementation for systems using highmem. | ||
347 | * In this case, each page of a buffer must be kmapped/kunmapped | ||
348 | * in order to have a virtual address for __dma_sync(). This must | ||
349 | * not sleep so kmap_atomic()/kunmap_atomic() are used. | ||
350 | * | ||
351 | * Note: yes, it is possible and correct to have a buffer extend | ||
352 | * beyond the first page. | ||
353 | */ | ||
354 | static inline void __dma_sync_page_highmem(struct page *page, | ||
355 | unsigned long offset, size_t size, int direction) | ||
356 | { | ||
357 | size_t seg_size = min((size_t)(PAGE_SIZE - offset), size); | ||
358 | size_t cur_size = seg_size; | ||
359 | unsigned long flags, start, seg_offset = offset; | ||
360 | int nr_segs = 1 + ((size - seg_size) + PAGE_SIZE - 1)/PAGE_SIZE; | ||
361 | int seg_nr = 0; | ||
362 | |||
363 | local_irq_save(flags); | ||
364 | |||
365 | do { | ||
366 | start = (unsigned long)kmap_atomic(page + seg_nr, | ||
367 | KM_PPC_SYNC_PAGE) + seg_offset; | ||
368 | |||
369 | /* Sync this buffer segment */ | ||
370 | __dma_sync((void *)start, seg_size, direction); | ||
371 | kunmap_atomic((void *)start, KM_PPC_SYNC_PAGE); | ||
372 | seg_nr++; | ||
373 | |||
374 | /* Calculate next buffer segment size */ | ||
375 | seg_size = min((size_t)PAGE_SIZE, size - cur_size); | ||
376 | |||
377 | /* Add the segment size to our running total */ | ||
378 | cur_size += seg_size; | ||
379 | seg_offset = 0; | ||
380 | } while (seg_nr < nr_segs); | ||
381 | |||
382 | local_irq_restore(flags); | ||
383 | } | ||
384 | #endif /* CONFIG_HIGHMEM */ | ||
385 | |||
386 | /* | ||
387 | * __dma_sync_page makes memory consistent. identical to __dma_sync, but | ||
388 | * takes a struct page instead of a virtual address | ||
389 | */ | ||
390 | void __dma_sync_page(struct page *page, unsigned long offset, | ||
391 | size_t size, int direction) | ||
392 | { | ||
393 | #ifdef CONFIG_HIGHMEM | ||
394 | __dma_sync_page_highmem(page, offset, size, direction); | ||
395 | #else | ||
396 | unsigned long start = (unsigned long)page_address(page) + offset; | ||
397 | __dma_sync((void *)start, size, direction); | ||
398 | #endif | ||
399 | } | ||
400 | EXPORT_SYMBOL(__dma_sync_page); | ||
diff --git a/arch/powerpc/mm/init_32.c b/arch/powerpc/mm/init_32.c index 666a5e8a5be1..3de6a0d93824 100644 --- a/arch/powerpc/mm/init_32.c +++ b/arch/powerpc/mm/init_32.c | |||
@@ -168,12 +168,8 @@ void __init MMU_init(void) | |||
168 | ppc_md.progress("MMU:mapin", 0x301); | 168 | ppc_md.progress("MMU:mapin", 0x301); |
169 | mapin_ram(); | 169 | mapin_ram(); |
170 | 170 | ||
171 | #ifdef CONFIG_HIGHMEM | 171 | /* Initialize early top-down ioremap allocator */ |
172 | ioremap_base = PKMAP_BASE; | 172 | ioremap_bot = IOREMAP_TOP; |
173 | #else | ||
174 | ioremap_base = 0xfe000000UL; /* for now, could be 0xfffff000 */ | ||
175 | #endif /* CONFIG_HIGHMEM */ | ||
176 | ioremap_bot = ioremap_base; | ||
177 | 173 | ||
178 | /* Map in I/O resources */ | 174 | /* Map in I/O resources */ |
179 | if (ppc_md.progress) | 175 | if (ppc_md.progress) |
diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c index d0602a76bf7f..579382c163a9 100644 --- a/arch/powerpc/mm/mem.c +++ b/arch/powerpc/mm/mem.c | |||
@@ -380,6 +380,23 @@ void __init mem_init(void) | |||
380 | bsssize >> 10, | 380 | bsssize >> 10, |
381 | initsize >> 10); | 381 | initsize >> 10); |
382 | 382 | ||
383 | #ifdef CONFIG_PPC32 | ||
384 | pr_info("Kernel virtual memory layout:\n"); | ||
385 | pr_info(" * 0x%08lx..0x%08lx : fixmap\n", FIXADDR_START, FIXADDR_TOP); | ||
386 | #ifdef CONFIG_HIGHMEM | ||
387 | pr_info(" * 0x%08lx..0x%08lx : highmem PTEs\n", | ||
388 | PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP)); | ||
389 | #endif /* CONFIG_HIGHMEM */ | ||
390 | #ifdef CONFIG_NOT_COHERENT_CACHE | ||
391 | pr_info(" * 0x%08lx..0x%08lx : consistent mem\n", | ||
392 | IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE); | ||
393 | #endif /* CONFIG_NOT_COHERENT_CACHE */ | ||
394 | pr_info(" * 0x%08lx..0x%08lx : early ioremap\n", | ||
395 | ioremap_bot, IOREMAP_TOP); | ||
396 | pr_info(" * 0x%08lx..0x%08lx : vmalloc & ioremap\n", | ||
397 | VMALLOC_START, VMALLOC_END); | ||
398 | #endif /* CONFIG_PPC32 */ | ||
399 | |||
383 | mem_init_done = 1; | 400 | mem_init_done = 1; |
384 | } | 401 | } |
385 | 402 | ||
diff --git a/arch/powerpc/mm/pgtable_32.c b/arch/powerpc/mm/pgtable_32.c index 430d0908fa50..5422169626ba 100644 --- a/arch/powerpc/mm/pgtable_32.c +++ b/arch/powerpc/mm/pgtable_32.c | |||
@@ -399,8 +399,6 @@ void kernel_map_pages(struct page *page, int numpages, int enable) | |||
399 | #endif /* CONFIG_DEBUG_PAGEALLOC */ | 399 | #endif /* CONFIG_DEBUG_PAGEALLOC */ |
400 | 400 | ||
401 | static int fixmaps; | 401 | static int fixmaps; |
402 | unsigned long FIXADDR_TOP = (-PAGE_SIZE); | ||
403 | EXPORT_SYMBOL(FIXADDR_TOP); | ||
404 | 402 | ||
405 | void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags) | 403 | void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags) |
406 | { | 404 | { |