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>
45 #include <linux/of_fdt.h>
46 #include <linux/libfdt.h>
47 #include <linux/memremap.h>
49 #include <asm/pgalloc.h>
54 #include <asm/mmu_context.h>
55 #include <asm/pgtable.h>
57 #include <linux/uaccess.h>
59 #include <asm/machdep.h>
62 #include <asm/processor.h>
63 #include <asm/mmzone.h>
64 #include <asm/cputable.h>
65 #include <asm/sections.h>
66 #include <asm/iommu.h>
69 #include <mm/mmu_decl.h>
71 phys_addr_t memstart_addr
= ~0;
72 EXPORT_SYMBOL_GPL(memstart_addr
);
73 phys_addr_t kernstart_addr
;
74 EXPORT_SYMBOL_GPL(kernstart_addr
);
76 #ifdef CONFIG_SPARSEMEM_VMEMMAP
78 * Given an address within the vmemmap, determine the pfn of the page that
79 * represents the start of the section it is within. Note that we have to
80 * do this by hand as the proffered address may not be correctly aligned.
81 * Subtraction of non-aligned pointers produces undefined results.
83 static unsigned long __meminit
vmemmap_section_start(unsigned long page
)
85 unsigned long offset
= page
- ((unsigned long)(vmemmap
));
87 /* Return the pfn of the start of the section. */
88 return (offset
/ sizeof(struct page
)) & PAGE_SECTION_MASK
;
92 * Check if this vmemmap page is already initialised. If any section
93 * which overlaps this vmemmap page is initialised then this page is
94 * initialised already.
96 static int __meminit
vmemmap_populated(unsigned long start
, int page_size
)
98 unsigned long end
= start
+ page_size
;
99 start
= (unsigned long)(pfn_to_page(vmemmap_section_start(start
)));
101 for (; start
< end
; start
+= (PAGES_PER_SECTION
* sizeof(struct page
)))
102 if (pfn_valid(page_to_pfn((struct page
*)start
)))
109 * vmemmap virtual address space management does not have a traditonal page
110 * table to track which virtual struct pages are backed by physical mapping.
111 * The virtual to physical mappings are tracked in a simple linked list
112 * format. 'vmemmap_list' maintains the entire vmemmap physical mapping at
113 * all times where as the 'next' list maintains the available
114 * vmemmap_backing structures which have been deleted from the
115 * 'vmemmap_global' list during system runtime (memory hotplug remove
116 * operation). The freed 'vmemmap_backing' structures are reused later when
117 * new requests come in without allocating fresh memory. This pointer also
118 * tracks the allocated 'vmemmap_backing' structures as we allocate one
119 * full page memory at a time when we dont have any.
121 struct vmemmap_backing
*vmemmap_list
;
122 static struct vmemmap_backing
*next
;
125 * The same pointer 'next' tracks individual chunks inside the allocated
126 * full page during the boot time and again tracks the freeed nodes during
127 * runtime. It is racy but it does not happen as they are separated by the
128 * boot process. Will create problem if some how we have memory hotplug
129 * operation during boot !!
132 static int num_freed
;
134 static __meminit
struct vmemmap_backing
* vmemmap_list_alloc(int node
)
136 struct vmemmap_backing
*vmem_back
;
137 /* get from freed entries first */
146 /* allocate a page when required and hand out chunks */
148 next
= vmemmap_alloc_block(PAGE_SIZE
, node
);
149 if (unlikely(!next
)) {
153 num_left
= PAGE_SIZE
/ sizeof(struct vmemmap_backing
);
161 static __meminit
void vmemmap_list_populate(unsigned long phys
,
165 struct vmemmap_backing
*vmem_back
;
167 vmem_back
= vmemmap_list_alloc(node
);
168 if (unlikely(!vmem_back
)) {
173 vmem_back
->phys
= phys
;
174 vmem_back
->virt_addr
= start
;
175 vmem_back
->list
= vmemmap_list
;
177 vmemmap_list
= vmem_back
;
180 int __meminit
vmemmap_populate(unsigned long start
, unsigned long end
, int node
,
181 struct vmem_altmap
*altmap
)
183 unsigned long page_size
= 1 << mmu_psize_defs
[mmu_vmemmap_psize
].shift
;
185 /* Align to the page size of the linear mapping. */
186 start
= _ALIGN_DOWN(start
, page_size
);
188 pr_debug("vmemmap_populate %lx..%lx, node %d\n", start
, end
, node
);
190 for (; start
< end
; start
+= page_size
) {
194 if (vmemmap_populated(start
, page_size
))
198 * Allocate from the altmap first if we have one. This may
199 * fail due to alignment issues when using 16MB hugepages, so
200 * fall back to system memory if the altmap allocation fail.
203 p
= altmap_alloc_block_buf(page_size
, altmap
);
205 p
= vmemmap_alloc_block_buf(page_size
, node
);
209 vmemmap_list_populate(__pa(p
), start
, node
);
211 pr_debug(" * %016lx..%016lx allocated at %p\n",
212 start
, start
+ page_size
, p
);
214 rc
= vmemmap_create_mapping(start
, page_size
, __pa(p
));
216 pr_warn("%s: Unable to create vmemmap mapping: %d\n",
225 #ifdef CONFIG_MEMORY_HOTPLUG
226 static unsigned long vmemmap_list_free(unsigned long start
)
228 struct vmemmap_backing
*vmem_back
, *vmem_back_prev
;
230 vmem_back_prev
= vmem_back
= vmemmap_list
;
232 /* look for it with prev pointer recorded */
233 for (; vmem_back
; vmem_back
= vmem_back
->list
) {
234 if (vmem_back
->virt_addr
== start
)
236 vmem_back_prev
= vmem_back
;
239 if (unlikely(!vmem_back
)) {
244 /* remove it from vmemmap_list */
245 if (vmem_back
== vmemmap_list
) /* remove head */
246 vmemmap_list
= vmem_back
->list
;
248 vmem_back_prev
->list
= vmem_back
->list
;
250 /* next point to this freed entry */
251 vmem_back
->list
= next
;
255 return vmem_back
->phys
;
258 void __ref
vmemmap_free(unsigned long start
, unsigned long end
,
259 struct vmem_altmap
*altmap
)
261 unsigned long page_size
= 1 << mmu_psize_defs
[mmu_vmemmap_psize
].shift
;
262 unsigned long page_order
= get_order(page_size
);
263 unsigned long alt_start
= ~0, alt_end
= ~0;
264 unsigned long base_pfn
;
266 start
= _ALIGN_DOWN(start
, page_size
);
268 alt_start
= altmap
->base_pfn
;
269 alt_end
= altmap
->base_pfn
+ altmap
->reserve
+
270 altmap
->free
+ altmap
->alloc
+ altmap
->align
;
273 pr_debug("vmemmap_free %lx...%lx\n", start
, end
);
275 for (; start
< end
; start
+= page_size
) {
276 unsigned long nr_pages
, addr
;
280 * the section has already be marked as invalid, so
281 * vmemmap_populated() true means some other sections still
282 * in this page, so skip it.
284 if (vmemmap_populated(start
, page_size
))
287 addr
= vmemmap_list_free(start
);
291 page
= pfn_to_page(addr
>> PAGE_SHIFT
);
292 nr_pages
= 1 << page_order
;
293 base_pfn
= PHYS_PFN(addr
);
295 if (base_pfn
>= alt_start
&& base_pfn
< alt_end
) {
296 vmem_altmap_free(altmap
, nr_pages
);
297 } else if (PageReserved(page
)) {
298 /* allocated from bootmem */
299 if (page_size
< PAGE_SIZE
) {
301 * this shouldn't happen, but if it is
302 * the case, leave the memory there
307 free_reserved_page(page
++);
310 free_pages((unsigned long)(__va(addr
)), page_order
);
313 vmemmap_remove_mapping(start
, page_size
);
317 void register_page_bootmem_memmap(unsigned long section_nr
,
318 struct page
*start_page
, unsigned long size
)
322 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
324 #ifdef CONFIG_PPC_BOOK3S_64
325 static bool disable_radix
= !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT
);
327 static int __init
parse_disable_radix(char *p
)
333 else if (kstrtobool(p
, &val
))
340 early_param("disable_radix", parse_disable_radix
);
343 * If we're running under a hypervisor, we need to check the contents of
344 * /chosen/ibm,architecture-vec-5 to see if the hypervisor is willing to do
345 * radix. If not, we clear the radix feature bit so we fall back to hash.
347 static void __init
early_check_vec5(void)
349 unsigned long root
, chosen
;
354 root
= of_get_flat_dt_root();
355 chosen
= of_get_flat_dt_subnode_by_name(root
, "chosen");
356 if (chosen
== -FDT_ERR_NOTFOUND
) {
357 cur_cpu_spec
->mmu_features
&= ~MMU_FTR_TYPE_RADIX
;
360 vec5
= of_get_flat_dt_prop(chosen
, "ibm,architecture-vec-5", &size
);
362 cur_cpu_spec
->mmu_features
&= ~MMU_FTR_TYPE_RADIX
;
365 if (size
<= OV5_INDX(OV5_MMU_SUPPORT
)) {
366 cur_cpu_spec
->mmu_features
&= ~MMU_FTR_TYPE_RADIX
;
370 /* Check for supported configuration */
371 mmu_supported
= vec5
[OV5_INDX(OV5_MMU_SUPPORT
)] &
372 OV5_FEAT(OV5_MMU_SUPPORT
);
373 if (mmu_supported
== OV5_FEAT(OV5_MMU_RADIX
)) {
374 /* Hypervisor only supports radix - check enabled && GTSE */
375 if (!early_radix_enabled()) {
376 pr_warn("WARNING: Ignoring cmdline option disable_radix\n");
378 if (!(vec5
[OV5_INDX(OV5_RADIX_GTSE
)] &
379 OV5_FEAT(OV5_RADIX_GTSE
))) {
380 pr_warn("WARNING: Hypervisor doesn't support RADIX with GTSE\n");
382 /* Do radix anyway - the hypervisor said we had to */
383 cur_cpu_spec
->mmu_features
|= MMU_FTR_TYPE_RADIX
;
384 } else if (mmu_supported
== OV5_FEAT(OV5_MMU_HASH
)) {
385 /* Hypervisor only supports hash - disable radix */
386 cur_cpu_spec
->mmu_features
&= ~MMU_FTR_TYPE_RADIX
;
390 void __init
mmu_early_init_devtree(void)
392 /* Disable radix mode based on kernel command line. */
394 cur_cpu_spec
->mmu_features
&= ~MMU_FTR_TYPE_RADIX
;
397 * Check /chosen/ibm,architecture-vec-5 if running as a guest.
398 * When running bare-metal, we can use radix if we like
399 * even though the ibm,architecture-vec-5 property created by
400 * skiboot doesn't have the necessary bits set.
402 if (!(mfmsr() & MSR_HV
))
405 if (early_radix_enabled())
406 radix__early_init_devtree();
408 hash__early_init_devtree();
410 #endif /* CONFIG_PPC_BOOK3S_64 */