xref: /linux/mm/sparse-vmemmap.c (revision a3b072cd180c12e8fe0ece9487b9065808327640)
18f6aac41SChristoph Lameter /*
28f6aac41SChristoph Lameter  * Virtual Memory Map support
38f6aac41SChristoph Lameter  *
4cde53535SChristoph Lameter  * (C) 2007 sgi. Christoph Lameter.
58f6aac41SChristoph Lameter  *
68f6aac41SChristoph Lameter  * Virtual memory maps allow VM primitives pfn_to_page, page_to_pfn,
78f6aac41SChristoph Lameter  * virt_to_page, page_address() to be implemented as a base offset
88f6aac41SChristoph Lameter  * calculation without memory access.
98f6aac41SChristoph Lameter  *
108f6aac41SChristoph Lameter  * However, virtual mappings need a page table and TLBs. Many Linux
118f6aac41SChristoph Lameter  * architectures already map their physical space using 1-1 mappings
12b595076aSUwe Kleine-König  * via TLBs. For those arches the virtual memory map is essentially
138f6aac41SChristoph Lameter  * for free if we use the same page size as the 1-1 mappings. In that
148f6aac41SChristoph Lameter  * case the overhead consists of a few additional pages that are
158f6aac41SChristoph Lameter  * allocated to create a view of memory for vmemmap.
168f6aac41SChristoph Lameter  *
1729c71111SAndy Whitcroft  * The architecture is expected to provide a vmemmap_populate() function
1829c71111SAndy Whitcroft  * to instantiate the mapping.
198f6aac41SChristoph Lameter  */
208f6aac41SChristoph Lameter #include <linux/mm.h>
218f6aac41SChristoph Lameter #include <linux/mmzone.h>
228f6aac41SChristoph Lameter #include <linux/bootmem.h>
238f6aac41SChristoph Lameter #include <linux/highmem.h>
245a0e3ad6STejun Heo #include <linux/slab.h>
258f6aac41SChristoph Lameter #include <linux/spinlock.h>
268f6aac41SChristoph Lameter #include <linux/vmalloc.h>
278bca44bbSGlauber de Oliveira Costa #include <linux/sched.h>
288f6aac41SChristoph Lameter #include <asm/dma.h>
298f6aac41SChristoph Lameter #include <asm/pgalloc.h>
308f6aac41SChristoph Lameter #include <asm/pgtable.h>
318f6aac41SChristoph Lameter 
328f6aac41SChristoph Lameter /*
338f6aac41SChristoph Lameter  * Allocate a block of memory to be used to back the virtual memory map
348f6aac41SChristoph Lameter  * or to back the page tables that are used to create the mapping.
358f6aac41SChristoph Lameter  * Uses the main allocators if they are available, else bootmem.
368f6aac41SChristoph Lameter  */
37e0dc3a53SKAMEZAWA Hiroyuki 
38e0dc3a53SKAMEZAWA Hiroyuki static void * __init_refok __earlyonly_bootmem_alloc(int node,
39e0dc3a53SKAMEZAWA Hiroyuki 				unsigned long size,
40e0dc3a53SKAMEZAWA Hiroyuki 				unsigned long align,
41e0dc3a53SKAMEZAWA Hiroyuki 				unsigned long goal)
42e0dc3a53SKAMEZAWA Hiroyuki {
43*bb016b84SSantosh Shilimkar 	return memblock_virt_alloc_try_nid(size, align, goal,
44*bb016b84SSantosh Shilimkar 					    BOOTMEM_ALLOC_ACCESSIBLE, node);
45e0dc3a53SKAMEZAWA Hiroyuki }
46e0dc3a53SKAMEZAWA Hiroyuki 
479bdac914SYinghai Lu static void *vmemmap_buf;
489bdac914SYinghai Lu static void *vmemmap_buf_end;
49e0dc3a53SKAMEZAWA Hiroyuki 
508f6aac41SChristoph Lameter void * __meminit vmemmap_alloc_block(unsigned long size, int node)
518f6aac41SChristoph Lameter {
528f6aac41SChristoph Lameter 	/* If the main allocator is up use that, fallback to bootmem. */
538f6aac41SChristoph Lameter 	if (slab_is_available()) {
54f52407ceSShaohua Li 		struct page *page;
55f52407ceSShaohua Li 
56f52407ceSShaohua Li 		if (node_state(node, N_HIGH_MEMORY))
57055e4fd9SBen Hutchings 			page = alloc_pages_node(
58055e4fd9SBen Hutchings 				node, GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
59055e4fd9SBen Hutchings 				get_order(size));
60f52407ceSShaohua Li 		else
61055e4fd9SBen Hutchings 			page = alloc_pages(
62055e4fd9SBen Hutchings 				GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
63f52407ceSShaohua Li 				get_order(size));
648f6aac41SChristoph Lameter 		if (page)
658f6aac41SChristoph Lameter 			return page_address(page);
668f6aac41SChristoph Lameter 		return NULL;
678f6aac41SChristoph Lameter 	} else
68e0dc3a53SKAMEZAWA Hiroyuki 		return __earlyonly_bootmem_alloc(node, size, size,
698f6aac41SChristoph Lameter 				__pa(MAX_DMA_ADDRESS));
708f6aac41SChristoph Lameter }
718f6aac41SChristoph Lameter 
729bdac914SYinghai Lu /* need to make sure size is all the same during early stage */
739bdac914SYinghai Lu void * __meminit vmemmap_alloc_block_buf(unsigned long size, int node)
749bdac914SYinghai Lu {
759bdac914SYinghai Lu 	void *ptr;
769bdac914SYinghai Lu 
779bdac914SYinghai Lu 	if (!vmemmap_buf)
789bdac914SYinghai Lu 		return vmemmap_alloc_block(size, node);
799bdac914SYinghai Lu 
809bdac914SYinghai Lu 	/* take the from buf */
819bdac914SYinghai Lu 	ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
829bdac914SYinghai Lu 	if (ptr + size > vmemmap_buf_end)
839bdac914SYinghai Lu 		return vmemmap_alloc_block(size, node);
849bdac914SYinghai Lu 
859bdac914SYinghai Lu 	vmemmap_buf = ptr + size;
869bdac914SYinghai Lu 
879bdac914SYinghai Lu 	return ptr;
889bdac914SYinghai Lu }
899bdac914SYinghai Lu 
908f6aac41SChristoph Lameter void __meminit vmemmap_verify(pte_t *pte, int node,
918f6aac41SChristoph Lameter 				unsigned long start, unsigned long end)
928f6aac41SChristoph Lameter {
938f6aac41SChristoph Lameter 	unsigned long pfn = pte_pfn(*pte);
948f6aac41SChristoph Lameter 	int actual_node = early_pfn_to_nid(pfn);
958f6aac41SChristoph Lameter 
96b41ad14cSDavid Rientjes 	if (node_distance(actual_node, node) > LOCAL_DISTANCE)
978f6aac41SChristoph Lameter 		printk(KERN_WARNING "[%lx-%lx] potential offnode "
988f6aac41SChristoph Lameter 			"page_structs\n", start, end - 1);
998f6aac41SChristoph Lameter }
1008f6aac41SChristoph Lameter 
10129c71111SAndy Whitcroft pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
1028f6aac41SChristoph Lameter {
10329c71111SAndy Whitcroft 	pte_t *pte = pte_offset_kernel(pmd, addr);
1048f6aac41SChristoph Lameter 	if (pte_none(*pte)) {
1058f6aac41SChristoph Lameter 		pte_t entry;
1069bdac914SYinghai Lu 		void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
1078f6aac41SChristoph Lameter 		if (!p)
1089dce07f1SAl Viro 			return NULL;
10929c71111SAndy Whitcroft 		entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
11029c71111SAndy Whitcroft 		set_pte_at(&init_mm, addr, pte, entry);
11129c71111SAndy Whitcroft 	}
11229c71111SAndy Whitcroft 	return pte;
1138f6aac41SChristoph Lameter }
1148f6aac41SChristoph Lameter 
11529c71111SAndy Whitcroft pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
1168f6aac41SChristoph Lameter {
11729c71111SAndy Whitcroft 	pmd_t *pmd = pmd_offset(pud, addr);
1188f6aac41SChristoph Lameter 	if (pmd_none(*pmd)) {
1198f6aac41SChristoph Lameter 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
1208f6aac41SChristoph Lameter 		if (!p)
1219dce07f1SAl Viro 			return NULL;
1228f6aac41SChristoph Lameter 		pmd_populate_kernel(&init_mm, pmd, p);
1238f6aac41SChristoph Lameter 	}
12429c71111SAndy Whitcroft 	return pmd;
1258f6aac41SChristoph Lameter }
1268f6aac41SChristoph Lameter 
12729c71111SAndy Whitcroft pud_t * __meminit vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node)
1288f6aac41SChristoph Lameter {
12929c71111SAndy Whitcroft 	pud_t *pud = pud_offset(pgd, addr);
1308f6aac41SChristoph Lameter 	if (pud_none(*pud)) {
1318f6aac41SChristoph Lameter 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
1328f6aac41SChristoph Lameter 		if (!p)
1339dce07f1SAl Viro 			return NULL;
1348f6aac41SChristoph Lameter 		pud_populate(&init_mm, pud, p);
1358f6aac41SChristoph Lameter 	}
13629c71111SAndy Whitcroft 	return pud;
1378f6aac41SChristoph Lameter }
1388f6aac41SChristoph Lameter 
13929c71111SAndy Whitcroft pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
1408f6aac41SChristoph Lameter {
14129c71111SAndy Whitcroft 	pgd_t *pgd = pgd_offset_k(addr);
1428f6aac41SChristoph Lameter 	if (pgd_none(*pgd)) {
1438f6aac41SChristoph Lameter 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
1448f6aac41SChristoph Lameter 		if (!p)
1459dce07f1SAl Viro 			return NULL;
1468f6aac41SChristoph Lameter 		pgd_populate(&init_mm, pgd, p);
1478f6aac41SChristoph Lameter 	}
14829c71111SAndy Whitcroft 	return pgd;
1498f6aac41SChristoph Lameter }
15029c71111SAndy Whitcroft 
1510aad818bSJohannes Weiner int __meminit vmemmap_populate_basepages(unsigned long start,
1520aad818bSJohannes Weiner 					 unsigned long end, int node)
15329c71111SAndy Whitcroft {
1540aad818bSJohannes Weiner 	unsigned long addr = start;
15529c71111SAndy Whitcroft 	pgd_t *pgd;
15629c71111SAndy Whitcroft 	pud_t *pud;
15729c71111SAndy Whitcroft 	pmd_t *pmd;
15829c71111SAndy Whitcroft 	pte_t *pte;
15929c71111SAndy Whitcroft 
16029c71111SAndy Whitcroft 	for (; addr < end; addr += PAGE_SIZE) {
16129c71111SAndy Whitcroft 		pgd = vmemmap_pgd_populate(addr, node);
16229c71111SAndy Whitcroft 		if (!pgd)
16329c71111SAndy Whitcroft 			return -ENOMEM;
16429c71111SAndy Whitcroft 		pud = vmemmap_pud_populate(pgd, addr, node);
16529c71111SAndy Whitcroft 		if (!pud)
16629c71111SAndy Whitcroft 			return -ENOMEM;
16729c71111SAndy Whitcroft 		pmd = vmemmap_pmd_populate(pud, addr, node);
16829c71111SAndy Whitcroft 		if (!pmd)
16929c71111SAndy Whitcroft 			return -ENOMEM;
17029c71111SAndy Whitcroft 		pte = vmemmap_pte_populate(pmd, addr, node);
17129c71111SAndy Whitcroft 		if (!pte)
17229c71111SAndy Whitcroft 			return -ENOMEM;
17329c71111SAndy Whitcroft 		vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
1748f6aac41SChristoph Lameter 	}
17529c71111SAndy Whitcroft 
17629c71111SAndy Whitcroft 	return 0;
17729c71111SAndy Whitcroft }
1788f6aac41SChristoph Lameter 
17998f3cfc1SYasunori Goto struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
1808f6aac41SChristoph Lameter {
1810aad818bSJohannes Weiner 	unsigned long start;
1820aad818bSJohannes Weiner 	unsigned long end;
1830aad818bSJohannes Weiner 	struct page *map;
1840aad818bSJohannes Weiner 
1850aad818bSJohannes Weiner 	map = pfn_to_page(pnum * PAGES_PER_SECTION);
1860aad818bSJohannes Weiner 	start = (unsigned long)map;
1870aad818bSJohannes Weiner 	end = (unsigned long)(map + PAGES_PER_SECTION);
1880aad818bSJohannes Weiner 
1890aad818bSJohannes Weiner 	if (vmemmap_populate(start, end, nid))
1908f6aac41SChristoph Lameter 		return NULL;
1918f6aac41SChristoph Lameter 
1928f6aac41SChristoph Lameter 	return map;
1938f6aac41SChristoph Lameter }
1949bdac914SYinghai Lu 
1959bdac914SYinghai Lu void __init sparse_mem_maps_populate_node(struct page **map_map,
1969bdac914SYinghai Lu 					  unsigned long pnum_begin,
1979bdac914SYinghai Lu 					  unsigned long pnum_end,
1989bdac914SYinghai Lu 					  unsigned long map_count, int nodeid)
1999bdac914SYinghai Lu {
2009bdac914SYinghai Lu 	unsigned long pnum;
2019bdac914SYinghai Lu 	unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
2029bdac914SYinghai Lu 	void *vmemmap_buf_start;
2039bdac914SYinghai Lu 
2049bdac914SYinghai Lu 	size = ALIGN(size, PMD_SIZE);
2059bdac914SYinghai Lu 	vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
2069bdac914SYinghai Lu 			 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
2079bdac914SYinghai Lu 
2089bdac914SYinghai Lu 	if (vmemmap_buf_start) {
2099bdac914SYinghai Lu 		vmemmap_buf = vmemmap_buf_start;
2109bdac914SYinghai Lu 		vmemmap_buf_end = vmemmap_buf_start + size * map_count;
2119bdac914SYinghai Lu 	}
2129bdac914SYinghai Lu 
2139bdac914SYinghai Lu 	for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
2149bdac914SYinghai Lu 		struct mem_section *ms;
2159bdac914SYinghai Lu 
2169bdac914SYinghai Lu 		if (!present_section_nr(pnum))
2179bdac914SYinghai Lu 			continue;
2189bdac914SYinghai Lu 
2199bdac914SYinghai Lu 		map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
2209bdac914SYinghai Lu 		if (map_map[pnum])
2219bdac914SYinghai Lu 			continue;
2229bdac914SYinghai Lu 		ms = __nr_to_section(pnum);
2239bdac914SYinghai Lu 		printk(KERN_ERR "%s: sparsemem memory map backing failed "
2249bdac914SYinghai Lu 			"some memory will not be available.\n", __func__);
2259bdac914SYinghai Lu 		ms->section_mem_map = 0;
2269bdac914SYinghai Lu 	}
2279bdac914SYinghai Lu 
2289bdac914SYinghai Lu 	if (vmemmap_buf_start) {
2299bdac914SYinghai Lu 		/* need to free left buf */
230*bb016b84SSantosh Shilimkar 		memblock_free_early(__pa(vmemmap_buf),
231*bb016b84SSantosh Shilimkar 				    vmemmap_buf_end - vmemmap_buf);
2329bdac914SYinghai Lu 		vmemmap_buf = NULL;
2339bdac914SYinghai Lu 		vmemmap_buf_end = NULL;
2349bdac914SYinghai Lu 	}
2359bdac914SYinghai Lu }
236