diff options
Diffstat (limited to 'arch/powerpc/mm/init64.c')
-rw-r--r-- | arch/powerpc/mm/init64.c | 385 |
1 files changed, 385 insertions, 0 deletions
diff --git a/arch/powerpc/mm/init64.c b/arch/powerpc/mm/init64.c new file mode 100644 index 000000000000..81f6745b31ef --- /dev/null +++ b/arch/powerpc/mm/init64.c | |||
@@ -0,0 +1,385 @@ | |||
1 | /* | ||
2 | * PowerPC version | ||
3 | * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) | ||
4 | * | ||
5 | * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au) | ||
6 | * and Cort Dougan (PReP) (cort@cs.nmt.edu) | ||
7 | * Copyright (C) 1996 Paul Mackerras | ||
8 | * Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk). | ||
9 | * | ||
10 | * Derived from "arch/i386/mm/init.c" | ||
11 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | ||
12 | * | ||
13 | * Dave Engebretsen <engebret@us.ibm.com> | ||
14 | * Rework for PPC64 port. | ||
15 | * | ||
16 | * This program is free software; you can redistribute it and/or | ||
17 | * modify it under the terms of the GNU General Public License | ||
18 | * as published by the Free Software Foundation; either version | ||
19 | * 2 of the License, or (at your option) any later version. | ||
20 | * | ||
21 | */ | ||
22 | |||
23 | #include <linux/config.h> | ||
24 | #include <linux/signal.h> | ||
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/mman.h> | ||
31 | #include <linux/mm.h> | ||
32 | #include <linux/swap.h> | ||
33 | #include <linux/stddef.h> | ||
34 | #include <linux/vmalloc.h> | ||
35 | #include <linux/init.h> | ||
36 | #include <linux/delay.h> | ||
37 | #include <linux/bootmem.h> | ||
38 | #include <linux/highmem.h> | ||
39 | #include <linux/idr.h> | ||
40 | #include <linux/nodemask.h> | ||
41 | #include <linux/module.h> | ||
42 | |||
43 | #include <asm/pgalloc.h> | ||
44 | #include <asm/page.h> | ||
45 | #include <asm/prom.h> | ||
46 | #include <asm/lmb.h> | ||
47 | #include <asm/rtas.h> | ||
48 | #include <asm/io.h> | ||
49 | #include <asm/mmu_context.h> | ||
50 | #include <asm/pgtable.h> | ||
51 | #include <asm/mmu.h> | ||
52 | #include <asm/uaccess.h> | ||
53 | #include <asm/smp.h> | ||
54 | #include <asm/machdep.h> | ||
55 | #include <asm/tlb.h> | ||
56 | #include <asm/eeh.h> | ||
57 | #include <asm/processor.h> | ||
58 | #include <asm/mmzone.h> | ||
59 | #include <asm/cputable.h> | ||
60 | #include <asm/ppcdebug.h> | ||
61 | #include <asm/sections.h> | ||
62 | #include <asm/system.h> | ||
63 | #include <asm/iommu.h> | ||
64 | #include <asm/abs_addr.h> | ||
65 | #include <asm/vdso.h> | ||
66 | #include <asm/imalloc.h> | ||
67 | |||
68 | #if PGTABLE_RANGE > USER_VSID_RANGE | ||
69 | #warning Limited user VSID range means pagetable space is wasted | ||
70 | #endif | ||
71 | |||
72 | #if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE) | ||
73 | #warning TASK_SIZE is smaller than it needs to be. | ||
74 | #endif | ||
75 | |||
76 | int mem_init_done; | ||
77 | unsigned long ioremap_bot = IMALLOC_BASE; | ||
78 | static unsigned long phbs_io_bot = PHBS_IO_BASE; | ||
79 | |||
80 | extern pgd_t swapper_pg_dir[]; | ||
81 | extern struct task_struct *current_set[NR_CPUS]; | ||
82 | |||
83 | unsigned long klimit = (unsigned long)_end; | ||
84 | |||
85 | unsigned long _SDR1=0; | ||
86 | unsigned long _ASR=0; | ||
87 | |||
88 | /* max amount of RAM to use */ | ||
89 | unsigned long __max_memory; | ||
90 | |||
91 | /* info on what we think the IO hole is */ | ||
92 | unsigned long io_hole_start; | ||
93 | unsigned long io_hole_size; | ||
94 | |||
95 | /* | ||
96 | * Do very early mm setup. | ||
97 | */ | ||
98 | void __init mm_init_ppc64(void) | ||
99 | { | ||
100 | #ifndef CONFIG_PPC_ISERIES | ||
101 | unsigned long i; | ||
102 | #endif | ||
103 | |||
104 | ppc64_boot_msg(0x100, "MM Init"); | ||
105 | |||
106 | /* This is the story of the IO hole... please, keep seated, | ||
107 | * unfortunately, we are out of oxygen masks at the moment. | ||
108 | * So we need some rough way to tell where your big IO hole | ||
109 | * is. On pmac, it's between 2G and 4G, on POWER3, it's around | ||
110 | * that area as well, on POWER4 we don't have one, etc... | ||
111 | * We need that as a "hint" when sizing the TCE table on POWER3 | ||
112 | * So far, the simplest way that seem work well enough for us it | ||
113 | * to just assume that the first discontinuity in our physical | ||
114 | * RAM layout is the IO hole. That may not be correct in the future | ||
115 | * (and isn't on iSeries but then we don't care ;) | ||
116 | */ | ||
117 | |||
118 | #ifndef CONFIG_PPC_ISERIES | ||
119 | for (i = 1; i < lmb.memory.cnt; i++) { | ||
120 | unsigned long base, prevbase, prevsize; | ||
121 | |||
122 | prevbase = lmb.memory.region[i-1].base; | ||
123 | prevsize = lmb.memory.region[i-1].size; | ||
124 | base = lmb.memory.region[i].base; | ||
125 | if (base > (prevbase + prevsize)) { | ||
126 | io_hole_start = prevbase + prevsize; | ||
127 | io_hole_size = base - (prevbase + prevsize); | ||
128 | break; | ||
129 | } | ||
130 | } | ||
131 | #endif /* CONFIG_PPC_ISERIES */ | ||
132 | if (io_hole_start) | ||
133 | printk("IO Hole assumed to be %lx -> %lx\n", | ||
134 | io_hole_start, io_hole_start + io_hole_size - 1); | ||
135 | |||
136 | ppc64_boot_msg(0x100, "MM Init Done"); | ||
137 | } | ||
138 | |||
139 | void free_initmem(void) | ||
140 | { | ||
141 | unsigned long addr; | ||
142 | |||
143 | addr = (unsigned long)__init_begin; | ||
144 | for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) { | ||
145 | memset((void *)addr, 0xcc, PAGE_SIZE); | ||
146 | ClearPageReserved(virt_to_page(addr)); | ||
147 | set_page_count(virt_to_page(addr), 1); | ||
148 | free_page(addr); | ||
149 | totalram_pages++; | ||
150 | } | ||
151 | printk ("Freeing unused kernel memory: %luk freed\n", | ||
152 | ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10); | ||
153 | } | ||
154 | |||
155 | #ifdef CONFIG_BLK_DEV_INITRD | ||
156 | void free_initrd_mem(unsigned long start, unsigned long end) | ||
157 | { | ||
158 | if (start < end) | ||
159 | printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10); | ||
160 | for (; start < end; start += PAGE_SIZE) { | ||
161 | ClearPageReserved(virt_to_page(start)); | ||
162 | set_page_count(virt_to_page(start), 1); | ||
163 | free_page(start); | ||
164 | totalram_pages++; | ||
165 | } | ||
166 | } | ||
167 | #endif | ||
168 | |||
169 | /* | ||
170 | * Initialize the bootmem system and give it all the memory we | ||
171 | * have available. | ||
172 | */ | ||
173 | #ifndef CONFIG_NEED_MULTIPLE_NODES | ||
174 | void __init do_init_bootmem(void) | ||
175 | { | ||
176 | unsigned long i; | ||
177 | unsigned long start, bootmap_pages; | ||
178 | unsigned long total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT; | ||
179 | int boot_mapsize; | ||
180 | |||
181 | /* | ||
182 | * Find an area to use for the bootmem bitmap. Calculate the size of | ||
183 | * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE. | ||
184 | * Add 1 additional page in case the address isn't page-aligned. | ||
185 | */ | ||
186 | bootmap_pages = bootmem_bootmap_pages(total_pages); | ||
187 | |||
188 | start = lmb_alloc(bootmap_pages<<PAGE_SHIFT, PAGE_SIZE); | ||
189 | BUG_ON(!start); | ||
190 | |||
191 | boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages); | ||
192 | |||
193 | max_pfn = max_low_pfn; | ||
194 | |||
195 | /* Add all physical memory to the bootmem map, mark each area | ||
196 | * present. | ||
197 | */ | ||
198 | for (i=0; i < lmb.memory.cnt; i++) | ||
199 | free_bootmem(lmb.memory.region[i].base, | ||
200 | lmb_size_bytes(&lmb.memory, i)); | ||
201 | |||
202 | /* reserve the sections we're already using */ | ||
203 | for (i=0; i < lmb.reserved.cnt; i++) | ||
204 | reserve_bootmem(lmb.reserved.region[i].base, | ||
205 | lmb_size_bytes(&lmb.reserved, i)); | ||
206 | |||
207 | for (i=0; i < lmb.memory.cnt; i++) | ||
208 | memory_present(0, lmb_start_pfn(&lmb.memory, i), | ||
209 | lmb_end_pfn(&lmb.memory, i)); | ||
210 | } | ||
211 | |||
212 | /* | ||
213 | * paging_init() sets up the page tables - in fact we've already done this. | ||
214 | */ | ||
215 | void __init paging_init(void) | ||
216 | { | ||
217 | unsigned long zones_size[MAX_NR_ZONES]; | ||
218 | unsigned long zholes_size[MAX_NR_ZONES]; | ||
219 | unsigned long total_ram = lmb_phys_mem_size(); | ||
220 | unsigned long top_of_ram = lmb_end_of_DRAM(); | ||
221 | |||
222 | printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n", | ||
223 | top_of_ram, total_ram); | ||
224 | printk(KERN_INFO "Memory hole size: %ldMB\n", | ||
225 | (top_of_ram - total_ram) >> 20); | ||
226 | /* | ||
227 | * All pages are DMA-able so we put them all in the DMA zone. | ||
228 | */ | ||
229 | memset(zones_size, 0, sizeof(zones_size)); | ||
230 | memset(zholes_size, 0, sizeof(zholes_size)); | ||
231 | |||
232 | zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT; | ||
233 | zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT; | ||
234 | |||
235 | free_area_init_node(0, NODE_DATA(0), zones_size, | ||
236 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size); | ||
237 | } | ||
238 | #endif /* ! CONFIG_NEED_MULTIPLE_NODES */ | ||
239 | |||
240 | static struct kcore_list kcore_vmem; | ||
241 | |||
242 | static int __init setup_kcore(void) | ||
243 | { | ||
244 | int i; | ||
245 | |||
246 | for (i=0; i < lmb.memory.cnt; i++) { | ||
247 | unsigned long base, size; | ||
248 | struct kcore_list *kcore_mem; | ||
249 | |||
250 | base = lmb.memory.region[i].base; | ||
251 | size = lmb.memory.region[i].size; | ||
252 | |||
253 | /* GFP_ATOMIC to avoid might_sleep warnings during boot */ | ||
254 | kcore_mem = kmalloc(sizeof(struct kcore_list), GFP_ATOMIC); | ||
255 | if (!kcore_mem) | ||
256 | panic("mem_init: kmalloc failed\n"); | ||
257 | |||
258 | kclist_add(kcore_mem, __va(base), size); | ||
259 | } | ||
260 | |||
261 | kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START); | ||
262 | |||
263 | return 0; | ||
264 | } | ||
265 | module_init(setup_kcore); | ||
266 | |||
267 | void __init mem_init(void) | ||
268 | { | ||
269 | #ifdef CONFIG_NEED_MULTIPLE_NODES | ||
270 | int nid; | ||
271 | #endif | ||
272 | pg_data_t *pgdat; | ||
273 | unsigned long i; | ||
274 | struct page *page; | ||
275 | unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize; | ||
276 | |||
277 | num_physpages = max_low_pfn; /* RAM is assumed contiguous */ | ||
278 | high_memory = (void *) __va(max_low_pfn * PAGE_SIZE); | ||
279 | |||
280 | #ifdef CONFIG_NEED_MULTIPLE_NODES | ||
281 | for_each_online_node(nid) { | ||
282 | if (NODE_DATA(nid)->node_spanned_pages != 0) { | ||
283 | printk("freeing bootmem node %x\n", nid); | ||
284 | totalram_pages += | ||
285 | free_all_bootmem_node(NODE_DATA(nid)); | ||
286 | } | ||
287 | } | ||
288 | #else | ||
289 | max_mapnr = num_physpages; | ||
290 | totalram_pages += free_all_bootmem(); | ||
291 | #endif | ||
292 | |||
293 | for_each_pgdat(pgdat) { | ||
294 | for (i = 0; i < pgdat->node_spanned_pages; i++) { | ||
295 | page = pgdat_page_nr(pgdat, i); | ||
296 | if (PageReserved(page)) | ||
297 | reservedpages++; | ||
298 | } | ||
299 | } | ||
300 | |||
301 | codesize = (unsigned long)&_etext - (unsigned long)&_stext; | ||
302 | initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin; | ||
303 | datasize = (unsigned long)&_edata - (unsigned long)&__init_end; | ||
304 | bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start; | ||
305 | |||
306 | printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, " | ||
307 | "%luk reserved, %luk data, %luk bss, %luk init)\n", | ||
308 | (unsigned long)nr_free_pages() << (PAGE_SHIFT-10), | ||
309 | num_physpages << (PAGE_SHIFT-10), | ||
310 | codesize >> 10, | ||
311 | reservedpages << (PAGE_SHIFT-10), | ||
312 | datasize >> 10, | ||
313 | bsssize >> 10, | ||
314 | initsize >> 10); | ||
315 | |||
316 | mem_init_done = 1; | ||
317 | |||
318 | /* Initialize the vDSO */ | ||
319 | vdso_init(); | ||
320 | } | ||
321 | |||
322 | void __iomem * reserve_phb_iospace(unsigned long size) | ||
323 | { | ||
324 | void __iomem *virt_addr; | ||
325 | |||
326 | if (phbs_io_bot >= IMALLOC_BASE) | ||
327 | panic("reserve_phb_iospace(): phb io space overflow\n"); | ||
328 | |||
329 | virt_addr = (void __iomem *) phbs_io_bot; | ||
330 | phbs_io_bot += size; | ||
331 | |||
332 | return virt_addr; | ||
333 | } | ||
334 | |||
335 | static void zero_ctor(void *addr, kmem_cache_t *cache, unsigned long flags) | ||
336 | { | ||
337 | memset(addr, 0, kmem_cache_size(cache)); | ||
338 | } | ||
339 | |||
340 | static const int pgtable_cache_size[2] = { | ||
341 | PTE_TABLE_SIZE, PMD_TABLE_SIZE | ||
342 | }; | ||
343 | static const char *pgtable_cache_name[ARRAY_SIZE(pgtable_cache_size)] = { | ||
344 | "pgd_pte_cache", "pud_pmd_cache", | ||
345 | }; | ||
346 | |||
347 | kmem_cache_t *pgtable_cache[ARRAY_SIZE(pgtable_cache_size)]; | ||
348 | |||
349 | void pgtable_cache_init(void) | ||
350 | { | ||
351 | int i; | ||
352 | |||
353 | BUILD_BUG_ON(PTE_TABLE_SIZE != pgtable_cache_size[PTE_CACHE_NUM]); | ||
354 | BUILD_BUG_ON(PMD_TABLE_SIZE != pgtable_cache_size[PMD_CACHE_NUM]); | ||
355 | BUILD_BUG_ON(PUD_TABLE_SIZE != pgtable_cache_size[PUD_CACHE_NUM]); | ||
356 | BUILD_BUG_ON(PGD_TABLE_SIZE != pgtable_cache_size[PGD_CACHE_NUM]); | ||
357 | |||
358 | for (i = 0; i < ARRAY_SIZE(pgtable_cache_size); i++) { | ||
359 | int size = pgtable_cache_size[i]; | ||
360 | const char *name = pgtable_cache_name[i]; | ||
361 | |||
362 | pgtable_cache[i] = kmem_cache_create(name, | ||
363 | size, size, | ||
364 | SLAB_HWCACHE_ALIGN | ||
365 | | SLAB_MUST_HWCACHE_ALIGN, | ||
366 | zero_ctor, | ||
367 | NULL); | ||
368 | if (! pgtable_cache[i]) | ||
369 | panic("pgtable_cache_init(): could not create %s!\n", | ||
370 | name); | ||
371 | } | ||
372 | } | ||
373 | |||
374 | pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr, | ||
375 | unsigned long size, pgprot_t vma_prot) | ||
376 | { | ||
377 | if (ppc_md.phys_mem_access_prot) | ||
378 | return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot); | ||
379 | |||
380 | if (!page_is_ram(addr >> PAGE_SHIFT)) | ||
381 | vma_prot = __pgprot(pgprot_val(vma_prot) | ||
382 | | _PAGE_GUARDED | _PAGE_NO_CACHE); | ||
383 | return vma_prot; | ||
384 | } | ||
385 | EXPORT_SYMBOL(phys_mem_access_prot); | ||