3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
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
9 * Derived from "arch/i386/mm/init.c"
10 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
12 * Dave Engebretsen <engebret@us.ibm.com>
13 * Rework for PPC64 port.
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
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>
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/highmem.h>
38 #include <linux/idr.h>
39 #include <linux/nodemask.h>
40 #include <linux/module.h>
41 #include <linux/poison.h>
42 #include <linux/memblock.h>
43 #include <linux/hugetlb.h>
44 #include <linux/slab.h>
46 #include <asm/pgalloc.h>
51 #include <asm/mmu_context.h>
52 #include <asm/pgtable.h>
54 #include <linux/uaccess.h>
56 #include <asm/machdep.h>
59 #include <asm/processor.h>
60 #include <asm/mmzone.h>
61 #include <asm/cputable.h>
62 #include <asm/sections.h>
63 #include <asm/iommu.h>
68 #ifdef CONFIG_PPC_STD_MMU_64
69 #if H_PGTABLE_RANGE > USER_VSID_RANGE
70 #warning Limited user VSID range means pagetable space is wasted
73 #if (TASK_SIZE_USER64 < H_PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
74 #warning TASK_SIZE is smaller than it needs to be.
76 #endif /* CONFIG_PPC_STD_MMU_64 */
78 phys_addr_t memstart_addr
= ~0;
79 EXPORT_SYMBOL_GPL(memstart_addr
);
80 phys_addr_t kernstart_addr
;
81 EXPORT_SYMBOL_GPL(kernstart_addr
);
83 #ifdef CONFIG_SPARSEMEM_VMEMMAP
85 * Given an address within the vmemmap, determine the pfn of the page that
86 * represents the start of the section it is within. Note that we have to
87 * do this by hand as the proffered address may not be correctly aligned.
88 * Subtraction of non-aligned pointers produces undefined results.
90 static unsigned long __meminit
vmemmap_section_start(unsigned long page
)
92 unsigned long offset
= page
- ((unsigned long)(vmemmap
));
94 /* Return the pfn of the start of the section. */
95 return (offset
/ sizeof(struct page
)) & PAGE_SECTION_MASK
;
99 * Check if this vmemmap page is already initialised. If any section
100 * which overlaps this vmemmap page is initialised then this page is
101 * initialised already.
103 static int __meminit
vmemmap_populated(unsigned long start
, int page_size
)
105 unsigned long end
= start
+ page_size
;
106 start
= (unsigned long)(pfn_to_page(vmemmap_section_start(start
)));
108 for (; start
< end
; start
+= (PAGES_PER_SECTION
* sizeof(struct page
)))
109 if (pfn_valid(page_to_pfn((struct page
*)start
)))
115 struct vmemmap_backing
*vmemmap_list
;
116 static struct vmemmap_backing
*next
;
118 static int num_freed
;
120 static __meminit
struct vmemmap_backing
* vmemmap_list_alloc(int node
)
122 struct vmemmap_backing
*vmem_back
;
123 /* get from freed entries first */
132 /* allocate a page when required and hand out chunks */
134 next
= vmemmap_alloc_block(PAGE_SIZE
, node
);
135 if (unlikely(!next
)) {
139 num_left
= PAGE_SIZE
/ sizeof(struct vmemmap_backing
);
147 static __meminit
void vmemmap_list_populate(unsigned long phys
,
151 struct vmemmap_backing
*vmem_back
;
153 vmem_back
= vmemmap_list_alloc(node
);
154 if (unlikely(!vmem_back
)) {
159 vmem_back
->phys
= phys
;
160 vmem_back
->virt_addr
= start
;
161 vmem_back
->list
= vmemmap_list
;
163 vmemmap_list
= vmem_back
;
166 int __meminit
vmemmap_populate(unsigned long start
, unsigned long end
, int node
)
168 unsigned long page_size
= 1 << mmu_psize_defs
[mmu_vmemmap_psize
].shift
;
170 /* Align to the page size of the linear mapping. */
171 start
= _ALIGN_DOWN(start
, page_size
);
173 pr_debug("vmemmap_populate %lx..%lx, node %d\n", start
, end
, node
);
175 for (; start
< end
; start
+= page_size
) {
179 if (vmemmap_populated(start
, page_size
))
182 p
= vmemmap_alloc_block(page_size
, node
);
186 vmemmap_list_populate(__pa(p
), start
, node
);
188 pr_debug(" * %016lx..%016lx allocated at %p\n",
189 start
, start
+ page_size
, p
);
191 rc
= vmemmap_create_mapping(start
, page_size
, __pa(p
));
194 "vmemmap_populate: Unable to create vmemmap mapping: %d\n",
203 #ifdef CONFIG_MEMORY_HOTPLUG
204 static unsigned long vmemmap_list_free(unsigned long start
)
206 struct vmemmap_backing
*vmem_back
, *vmem_back_prev
;
208 vmem_back_prev
= vmem_back
= vmemmap_list
;
210 /* look for it with prev pointer recorded */
211 for (; vmem_back
; vmem_back
= vmem_back
->list
) {
212 if (vmem_back
->virt_addr
== start
)
214 vmem_back_prev
= vmem_back
;
217 if (unlikely(!vmem_back
)) {
222 /* remove it from vmemmap_list */
223 if (vmem_back
== vmemmap_list
) /* remove head */
224 vmemmap_list
= vmem_back
->list
;
226 vmem_back_prev
->list
= vmem_back
->list
;
228 /* next point to this freed entry */
229 vmem_back
->list
= next
;
233 return vmem_back
->phys
;
236 void __ref
vmemmap_free(unsigned long start
, unsigned long end
)
238 unsigned long page_size
= 1 << mmu_psize_defs
[mmu_vmemmap_psize
].shift
;
240 start
= _ALIGN_DOWN(start
, page_size
);
242 pr_debug("vmemmap_free %lx...%lx\n", start
, end
);
244 for (; start
< end
; start
+= page_size
) {
248 * the section has already be marked as invalid, so
249 * vmemmap_populated() true means some other sections still
250 * in this page, so skip it.
252 if (vmemmap_populated(start
, page_size
))
255 addr
= vmemmap_list_free(start
);
257 struct page
*page
= pfn_to_page(addr
>> PAGE_SHIFT
);
259 if (PageReserved(page
)) {
260 /* allocated from bootmem */
261 if (page_size
< PAGE_SIZE
) {
263 * this shouldn't happen, but if it is
264 * the case, leave the memory there
268 unsigned int nr_pages
=
269 1 << get_order(page_size
);
271 free_reserved_page(page
++);
274 free_pages((unsigned long)(__va(addr
)),
275 get_order(page_size
));
277 vmemmap_remove_mapping(start
, page_size
);
282 void register_page_bootmem_memmap(unsigned long section_nr
,
283 struct page
*start_page
, unsigned long size
)
288 * We do not have access to the sparsemem vmemmap, so we fallback to
289 * walking the list of sparsemem blocks which we already maintain for
290 * the sake of crashdump. In the long run, we might want to maintain
291 * a tree if performance of that linear walk becomes a problem.
293 * realmode_pfn_to_page functions can fail due to:
294 * 1) As real sparsemem blocks do not lay in RAM continously (they
295 * are in virtual address space which is not available in the real mode),
296 * the requested page struct can be split between blocks so get_page/put_page
298 * 2) When huge pages are used, the get_page/put_page API will fail
299 * in real mode as the linked addresses in the page struct are virtual
302 struct page
*realmode_pfn_to_page(unsigned long pfn
)
304 struct vmemmap_backing
*vmem_back
;
306 unsigned long page_size
= 1 << mmu_psize_defs
[mmu_vmemmap_psize
].shift
;
307 unsigned long pg_va
= (unsigned long) pfn_to_page(pfn
);
309 for (vmem_back
= vmemmap_list
; vmem_back
; vmem_back
= vmem_back
->list
) {
310 if (pg_va
< vmem_back
->virt_addr
)
313 /* After vmemmap_list entry free is possible, need check all */
314 if ((pg_va
+ sizeof(struct page
)) <=
315 (vmem_back
->virt_addr
+ page_size
)) {
316 page
= (struct page
*) (vmem_back
->phys
+ pg_va
-
317 vmem_back
->virt_addr
);
322 /* Probably that page struct is split between real pages */
325 EXPORT_SYMBOL_GPL(realmode_pfn_to_page
);
327 #elif defined(CONFIG_FLATMEM)
329 struct page
*realmode_pfn_to_page(unsigned long pfn
)
331 struct page
*page
= pfn_to_page(pfn
);
334 EXPORT_SYMBOL_GPL(realmode_pfn_to_page
);
336 #endif /* CONFIG_SPARSEMEM_VMEMMAP/CONFIG_FLATMEM */
338 #ifdef CONFIG_PPC_STD_MMU_64
339 static bool disable_radix
;
340 static int __init
parse_disable_radix(char *p
)
342 disable_radix
= true;
345 early_param("disable_radix", parse_disable_radix
);
347 void __init
mmu_early_init_devtree(void)
349 /* Disable radix mode based on kernel command line. */
351 cur_cpu_spec
->mmu_features
&= ~MMU_FTR_TYPE_RADIX
;
353 if (early_radix_enabled())
354 radix__early_init_devtree();
356 hash__early_init_devtree();
358 #endif /* CONFIG_PPC_STD_MMU_64 */