xref: /openbmc/linux/mm/sparse-vmemmap.c (revision 6c33a6f4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Virtual Memory Map support
4  *
5  * (C) 2007 sgi. Christoph Lameter.
6  *
7  * Virtual memory maps allow VM primitives pfn_to_page, page_to_pfn,
8  * virt_to_page, page_address() to be implemented as a base offset
9  * calculation without memory access.
10  *
11  * However, virtual mappings need a page table and TLBs. Many Linux
12  * architectures already map their physical space using 1-1 mappings
13  * via TLBs. For those arches the virtual memory map is essentially
14  * for free if we use the same page size as the 1-1 mappings. In that
15  * case the overhead consists of a few additional pages that are
16  * allocated to create a view of memory for vmemmap.
17  *
18  * The architecture is expected to provide a vmemmap_populate() function
19  * to instantiate the mapping.
20  */
21 #include <linux/mm.h>
22 #include <linux/mmzone.h>
23 #include <linux/memblock.h>
24 #include <linux/memremap.h>
25 #include <linux/highmem.h>
26 #include <linux/slab.h>
27 #include <linux/spinlock.h>
28 #include <linux/vmalloc.h>
29 #include <linux/sched.h>
30 #include <asm/dma.h>
31 #include <asm/pgalloc.h>
32 #include <asm/pgtable.h>
33 
34 /*
35  * Allocate a block of memory to be used to back the virtual memory map
36  * or to back the page tables that are used to create the mapping.
37  * Uses the main allocators if they are available, else bootmem.
38  */
39 
40 static void * __ref __earlyonly_bootmem_alloc(int node,
41 				unsigned long size,
42 				unsigned long align,
43 				unsigned long goal)
44 {
45 	return memblock_alloc_try_nid_raw(size, align, goal,
46 					       MEMBLOCK_ALLOC_ACCESSIBLE, node);
47 }
48 
49 void * __meminit vmemmap_alloc_block(unsigned long size, int node)
50 {
51 	/* If the main allocator is up use that, fallback to bootmem. */
52 	if (slab_is_available()) {
53 		gfp_t gfp_mask = GFP_KERNEL|__GFP_RETRY_MAYFAIL|__GFP_NOWARN;
54 		int order = get_order(size);
55 		static bool warned;
56 		struct page *page;
57 
58 		page = alloc_pages_node(node, gfp_mask, order);
59 		if (page)
60 			return page_address(page);
61 
62 		if (!warned) {
63 			warn_alloc(gfp_mask & ~__GFP_NOWARN, NULL,
64 				   "vmemmap alloc failure: order:%u", order);
65 			warned = true;
66 		}
67 		return NULL;
68 	} else
69 		return __earlyonly_bootmem_alloc(node, size, size,
70 				__pa(MAX_DMA_ADDRESS));
71 }
72 
73 /* need to make sure size is all the same during early stage */
74 void * __meminit vmemmap_alloc_block_buf(unsigned long size, int node)
75 {
76 	void *ptr = sparse_buffer_alloc(size);
77 
78 	if (!ptr)
79 		ptr = vmemmap_alloc_block(size, node);
80 	return ptr;
81 }
82 
83 static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
84 {
85 	return altmap->base_pfn + altmap->reserve + altmap->alloc
86 		+ altmap->align;
87 }
88 
89 static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
90 {
91 	unsigned long allocated = altmap->alloc + altmap->align;
92 
93 	if (altmap->free > allocated)
94 		return altmap->free - allocated;
95 	return 0;
96 }
97 
98 /**
99  * altmap_alloc_block_buf - allocate pages from the device page map
100  * @altmap:	device page map
101  * @size:	size (in bytes) of the allocation
102  *
103  * Allocations are aligned to the size of the request.
104  */
105 void * __meminit altmap_alloc_block_buf(unsigned long size,
106 		struct vmem_altmap *altmap)
107 {
108 	unsigned long pfn, nr_pfns, nr_align;
109 
110 	if (size & ~PAGE_MASK) {
111 		pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
112 				__func__, size);
113 		return NULL;
114 	}
115 
116 	pfn = vmem_altmap_next_pfn(altmap);
117 	nr_pfns = size >> PAGE_SHIFT;
118 	nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
119 	nr_align = ALIGN(pfn, nr_align) - pfn;
120 	if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
121 		return NULL;
122 
123 	altmap->alloc += nr_pfns;
124 	altmap->align += nr_align;
125 	pfn += nr_align;
126 
127 	pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
128 			__func__, pfn, altmap->alloc, altmap->align, nr_pfns);
129 	return __va(__pfn_to_phys(pfn));
130 }
131 
132 void __meminit vmemmap_verify(pte_t *pte, int node,
133 				unsigned long start, unsigned long end)
134 {
135 	unsigned long pfn = pte_pfn(*pte);
136 	int actual_node = early_pfn_to_nid(pfn);
137 
138 	if (node_distance(actual_node, node) > LOCAL_DISTANCE)
139 		pr_warn("[%lx-%lx] potential offnode page_structs\n",
140 			start, end - 1);
141 }
142 
143 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
144 {
145 	pte_t *pte = pte_offset_kernel(pmd, addr);
146 	if (pte_none(*pte)) {
147 		pte_t entry;
148 		void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
149 		if (!p)
150 			return NULL;
151 		entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
152 		set_pte_at(&init_mm, addr, pte, entry);
153 	}
154 	return pte;
155 }
156 
157 static void * __meminit vmemmap_alloc_block_zero(unsigned long size, int node)
158 {
159 	void *p = vmemmap_alloc_block(size, node);
160 
161 	if (!p)
162 		return NULL;
163 	memset(p, 0, size);
164 
165 	return p;
166 }
167 
168 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
169 {
170 	pmd_t *pmd = pmd_offset(pud, addr);
171 	if (pmd_none(*pmd)) {
172 		void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
173 		if (!p)
174 			return NULL;
175 		pmd_populate_kernel(&init_mm, pmd, p);
176 	}
177 	return pmd;
178 }
179 
180 pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
181 {
182 	pud_t *pud = pud_offset(p4d, addr);
183 	if (pud_none(*pud)) {
184 		void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
185 		if (!p)
186 			return NULL;
187 		pud_populate(&init_mm, pud, p);
188 	}
189 	return pud;
190 }
191 
192 p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
193 {
194 	p4d_t *p4d = p4d_offset(pgd, addr);
195 	if (p4d_none(*p4d)) {
196 		void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
197 		if (!p)
198 			return NULL;
199 		p4d_populate(&init_mm, p4d, p);
200 	}
201 	return p4d;
202 }
203 
204 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
205 {
206 	pgd_t *pgd = pgd_offset_k(addr);
207 	if (pgd_none(*pgd)) {
208 		void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
209 		if (!p)
210 			return NULL;
211 		pgd_populate(&init_mm, pgd, p);
212 	}
213 	return pgd;
214 }
215 
216 int __meminit vmemmap_populate_basepages(unsigned long start,
217 					 unsigned long end, int node)
218 {
219 	unsigned long addr = start;
220 	pgd_t *pgd;
221 	p4d_t *p4d;
222 	pud_t *pud;
223 	pmd_t *pmd;
224 	pte_t *pte;
225 
226 	for (; addr < end; addr += PAGE_SIZE) {
227 		pgd = vmemmap_pgd_populate(addr, node);
228 		if (!pgd)
229 			return -ENOMEM;
230 		p4d = vmemmap_p4d_populate(pgd, addr, node);
231 		if (!p4d)
232 			return -ENOMEM;
233 		pud = vmemmap_pud_populate(p4d, addr, node);
234 		if (!pud)
235 			return -ENOMEM;
236 		pmd = vmemmap_pmd_populate(pud, addr, node);
237 		if (!pmd)
238 			return -ENOMEM;
239 		pte = vmemmap_pte_populate(pmd, addr, node);
240 		if (!pte)
241 			return -ENOMEM;
242 		vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
243 	}
244 
245 	return 0;
246 }
247 
248 struct page * __meminit __populate_section_memmap(unsigned long pfn,
249 		unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
250 {
251 	unsigned long start;
252 	unsigned long end;
253 
254 	/*
255 	 * The minimum granularity of memmap extensions is
256 	 * PAGES_PER_SUBSECTION as allocations are tracked in the
257 	 * 'subsection_map' bitmap of the section.
258 	 */
259 	end = ALIGN(pfn + nr_pages, PAGES_PER_SUBSECTION);
260 	pfn &= PAGE_SUBSECTION_MASK;
261 	nr_pages = end - pfn;
262 
263 	start = (unsigned long) pfn_to_page(pfn);
264 	end = start + nr_pages * sizeof(struct page);
265 
266 	if (vmemmap_populate(start, end, nid, altmap))
267 		return NULL;
268 
269 	return pfn_to_page(pfn);
270 }
271