12874c5fdSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 295f72d1eSYinghai Lu /* 395f72d1eSYinghai Lu * Procedures for maintaining information about logical memory blocks. 495f72d1eSYinghai Lu * 595f72d1eSYinghai Lu * Peter Bergner, IBM Corp. June 2001. 695f72d1eSYinghai Lu * Copyright (C) 2001 Peter Bergner. 795f72d1eSYinghai Lu */ 895f72d1eSYinghai Lu 995f72d1eSYinghai Lu #include <linux/kernel.h> 10142b45a7SBenjamin Herrenschmidt #include <linux/slab.h> 1195f72d1eSYinghai Lu #include <linux/init.h> 1295f72d1eSYinghai Lu #include <linux/bitops.h> 13449e8df3SBenjamin Herrenschmidt #include <linux/poison.h> 14c196f76fSBenjamin Herrenschmidt #include <linux/pfn.h> 156d03b885SBenjamin Herrenschmidt #include <linux/debugfs.h> 16514c6032SRandy Dunlap #include <linux/kmemleak.h> 176d03b885SBenjamin Herrenschmidt #include <linux/seq_file.h> 1895f72d1eSYinghai Lu #include <linux/memblock.h> 1995f72d1eSYinghai Lu 20c4c5ad6bSChristoph Hellwig #include <asm/sections.h> 2126f09e9bSSantosh Shilimkar #include <linux/io.h> 2226f09e9bSSantosh Shilimkar 2326f09e9bSSantosh Shilimkar #include "internal.h" 2479442ed1STang Chen 258a5b403dSArd Biesheuvel #define INIT_MEMBLOCK_REGIONS 128 268a5b403dSArd Biesheuvel #define INIT_PHYSMEM_REGIONS 4 278a5b403dSArd Biesheuvel 288a5b403dSArd Biesheuvel #ifndef INIT_MEMBLOCK_RESERVED_REGIONS 298a5b403dSArd Biesheuvel # define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS 308a5b403dSArd Biesheuvel #endif 318a5b403dSArd Biesheuvel 32*450d0e74SZhou Guanghui #ifndef INIT_MEMBLOCK_MEMORY_REGIONS 33*450d0e74SZhou Guanghui #define INIT_MEMBLOCK_MEMORY_REGIONS INIT_MEMBLOCK_REGIONS 34*450d0e74SZhou Guanghui #endif 35*450d0e74SZhou Guanghui 363e039c5cSMike Rapoport /** 373e039c5cSMike Rapoport * DOC: memblock overview 383e039c5cSMike Rapoport * 393e039c5cSMike Rapoport * Memblock is a method of managing memory regions during the early 403e039c5cSMike Rapoport * boot period when the usual kernel memory allocators are not up and 413e039c5cSMike Rapoport * running. 423e039c5cSMike Rapoport * 433e039c5cSMike Rapoport * Memblock views the system memory as collections of contiguous 443e039c5cSMike Rapoport * regions. There are several types of these collections: 453e039c5cSMike Rapoport * 463e039c5cSMike Rapoport * * ``memory`` - describes the physical memory available to the 473e039c5cSMike Rapoport * kernel; this may differ from the actual physical memory installed 483e039c5cSMike Rapoport * in the system, for instance when the memory is restricted with 493e039c5cSMike Rapoport * ``mem=`` command line parameter 503e039c5cSMike Rapoport * * ``reserved`` - describes the regions that were allocated 5177649905SDavid Hildenbrand * * ``physmem`` - describes the actual physical memory available during 5277649905SDavid Hildenbrand * boot regardless of the possible restrictions and memory hot(un)plug; 5377649905SDavid Hildenbrand * the ``physmem`` type is only available on some architectures. 543e039c5cSMike Rapoport * 559303c9d5SMauro Carvalho Chehab * Each region is represented by struct memblock_region that 563e039c5cSMike Rapoport * defines the region extents, its attributes and NUMA node id on NUMA 571bf162e4SMauro Carvalho Chehab * systems. Every memory type is described by the struct memblock_type 581bf162e4SMauro Carvalho Chehab * which contains an array of memory regions along with 5977649905SDavid Hildenbrand * the allocator metadata. The "memory" and "reserved" types are nicely 609303c9d5SMauro Carvalho Chehab * wrapped with struct memblock. This structure is statically 6177649905SDavid Hildenbrand * initialized at build time. The region arrays are initially sized to 62*450d0e74SZhou Guanghui * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and 63*450d0e74SZhou Guanghui * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array 64*450d0e74SZhou Guanghui * for "physmem" is initially sized to %INIT_PHYSMEM_REGIONS. 656e5af9a8SCao jin * The memblock_allow_resize() enables automatic resizing of the region 666e5af9a8SCao jin * arrays during addition of new regions. This feature should be used 676e5af9a8SCao jin * with care so that memory allocated for the region array will not 686e5af9a8SCao jin * overlap with areas that should be reserved, for example initrd. 693e039c5cSMike Rapoport * 703e039c5cSMike Rapoport * The early architecture setup should tell memblock what the physical 716e5af9a8SCao jin * memory layout is by using memblock_add() or memblock_add_node() 726e5af9a8SCao jin * functions. The first function does not assign the region to a NUMA 736e5af9a8SCao jin * node and it is appropriate for UMA systems. Yet, it is possible to 746e5af9a8SCao jin * use it on NUMA systems as well and assign the region to a NUMA node 756e5af9a8SCao jin * later in the setup process using memblock_set_node(). The 766e5af9a8SCao jin * memblock_add_node() performs such an assignment directly. 773e039c5cSMike Rapoport * 78a2974133SMike Rapoport * Once memblock is setup the memory can be allocated using one of the 79a2974133SMike Rapoport * API variants: 80a2974133SMike Rapoport * 816e5af9a8SCao jin * * memblock_phys_alloc*() - these functions return the **physical** 826e5af9a8SCao jin * address of the allocated memory 836e5af9a8SCao jin * * memblock_alloc*() - these functions return the **virtual** address 846e5af9a8SCao jin * of the allocated memory. 85a2974133SMike Rapoport * 86df1758d9SEthon Paul * Note, that both API variants use implicit assumptions about allowed 87a2974133SMike Rapoport * memory ranges and the fallback methods. Consult the documentation 886e5af9a8SCao jin * of memblock_alloc_internal() and memblock_alloc_range_nid() 896e5af9a8SCao jin * functions for more elaborate description. 903e039c5cSMike Rapoport * 916e5af9a8SCao jin * As the system boot progresses, the architecture specific mem_init() 926e5af9a8SCao jin * function frees all the memory to the buddy page allocator. 933e039c5cSMike Rapoport * 946e5af9a8SCao jin * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the 9577649905SDavid Hildenbrand * memblock data structures (except "physmem") will be discarded after the 9677649905SDavid Hildenbrand * system initialization completes. 973e039c5cSMike Rapoport */ 983e039c5cSMike Rapoport 99a9ee6cf5SMike Rapoport #ifndef CONFIG_NUMA 100bda49a81SMike Rapoport struct pglist_data __refdata contig_page_data; 101bda49a81SMike Rapoport EXPORT_SYMBOL(contig_page_data); 102bda49a81SMike Rapoport #endif 103bda49a81SMike Rapoport 104bda49a81SMike Rapoport unsigned long max_low_pfn; 105bda49a81SMike Rapoport unsigned long min_low_pfn; 106bda49a81SMike Rapoport unsigned long max_pfn; 107bda49a81SMike Rapoport unsigned long long max_possible_pfn; 108bda49a81SMike Rapoport 109*450d0e74SZhou Guanghui static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_MEMORY_REGIONS] __initdata_memblock; 1108a5b403dSArd Biesheuvel static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock; 11170210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 11277649905SDavid Hildenbrand static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS]; 11370210ed9SPhilipp Hachtmann #endif 114fe091c20STejun Heo 115fe091c20STejun Heo struct memblock memblock __initdata_memblock = { 116fe091c20STejun Heo .memory.regions = memblock_memory_init_regions, 117fe091c20STejun Heo .memory.cnt = 1, /* empty dummy entry */ 118*450d0e74SZhou Guanghui .memory.max = INIT_MEMBLOCK_MEMORY_REGIONS, 1190262d9c8SHeiko Carstens .memory.name = "memory", 120fe091c20STejun Heo 121fe091c20STejun Heo .reserved.regions = memblock_reserved_init_regions, 122fe091c20STejun Heo .reserved.cnt = 1, /* empty dummy entry */ 1238a5b403dSArd Biesheuvel .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS, 1240262d9c8SHeiko Carstens .reserved.name = "reserved", 125fe091c20STejun Heo 12679442ed1STang Chen .bottom_up = false, 127fe091c20STejun Heo .current_limit = MEMBLOCK_ALLOC_ANYWHERE, 128fe091c20STejun Heo }; 12995f72d1eSYinghai Lu 13077649905SDavid Hildenbrand #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 13177649905SDavid Hildenbrand struct memblock_type physmem = { 13277649905SDavid Hildenbrand .regions = memblock_physmem_init_regions, 13377649905SDavid Hildenbrand .cnt = 1, /* empty dummy entry */ 13477649905SDavid Hildenbrand .max = INIT_PHYSMEM_REGIONS, 13577649905SDavid Hildenbrand .name = "physmem", 13677649905SDavid Hildenbrand }; 13777649905SDavid Hildenbrand #endif 13877649905SDavid Hildenbrand 1399f3d5eaaSMike Rapoport /* 1409f3d5eaaSMike Rapoport * keep a pointer to &memblock.memory in the text section to use it in 1419f3d5eaaSMike Rapoport * __next_mem_range() and its helpers. 1429f3d5eaaSMike Rapoport * For architectures that do not keep memblock data after init, this 1439f3d5eaaSMike Rapoport * pointer will be reset to NULL at memblock_discard() 1449f3d5eaaSMike Rapoport */ 1459f3d5eaaSMike Rapoport static __refdata struct memblock_type *memblock_memory = &memblock.memory; 1469f3d5eaaSMike Rapoport 147cd991db8SMike Rapoport #define for_each_memblock_type(i, memblock_type, rgn) \ 148cd991db8SMike Rapoport for (i = 0, rgn = &memblock_type->regions[0]; \ 149cd991db8SMike Rapoport i < memblock_type->cnt; \ 150cd991db8SMike Rapoport i++, rgn = &memblock_type->regions[i]) 151cd991db8SMike Rapoport 15287c55870SMike Rapoport #define memblock_dbg(fmt, ...) \ 15387c55870SMike Rapoport do { \ 15487c55870SMike Rapoport if (memblock_debug) \ 15587c55870SMike Rapoport pr_info(fmt, ##__VA_ARGS__); \ 15687c55870SMike Rapoport } while (0) 15787c55870SMike Rapoport 15887c55870SMike Rapoport static int memblock_debug __initdata_memblock; 159a3f5bafcSTony Luck static bool system_has_some_mirror __initdata_memblock = false; 1601aadc056STejun Heo static int memblock_can_resize __initdata_memblock; 161181eb394SGavin Shan static int memblock_memory_in_slab __initdata_memblock = 0; 162181eb394SGavin Shan static int memblock_reserved_in_slab __initdata_memblock = 0; 16395f72d1eSYinghai Lu 164c366ea89SMike Rapoport static enum memblock_flags __init_memblock choose_memblock_flags(void) 165a3f5bafcSTony Luck { 166a3f5bafcSTony Luck return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE; 167a3f5bafcSTony Luck } 168a3f5bafcSTony Luck 169eb18f1b5STejun Heo /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ 170eb18f1b5STejun Heo static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) 171eb18f1b5STejun Heo { 1721c4bc43dSStefan Agner return *size = min(*size, PHYS_ADDR_MAX - base); 173eb18f1b5STejun Heo } 174eb18f1b5STejun Heo 1756ed311b2SBenjamin Herrenschmidt /* 1766ed311b2SBenjamin Herrenschmidt * Address comparison utilities 1776ed311b2SBenjamin Herrenschmidt */ 17810d06439SYinghai Lu static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, 1792898cc4cSBenjamin Herrenschmidt phys_addr_t base2, phys_addr_t size2) 18095f72d1eSYinghai Lu { 18195f72d1eSYinghai Lu return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); 18295f72d1eSYinghai Lu } 18395f72d1eSYinghai Lu 18495cf82ecSTang Chen bool __init_memblock memblock_overlaps_region(struct memblock_type *type, 1852d7d3eb2SH Hartley Sweeten phys_addr_t base, phys_addr_t size) 1866ed311b2SBenjamin Herrenschmidt { 1876ed311b2SBenjamin Herrenschmidt unsigned long i; 1886ed311b2SBenjamin Herrenschmidt 189023accf5SMike Rapoport memblock_cap_size(base, &size); 190023accf5SMike Rapoport 191f14516fbSAlexander Kuleshov for (i = 0; i < type->cnt; i++) 192f14516fbSAlexander Kuleshov if (memblock_addrs_overlap(base, size, type->regions[i].base, 193f14516fbSAlexander Kuleshov type->regions[i].size)) 1946ed311b2SBenjamin Herrenschmidt break; 195c5c5c9d1STang Chen return i < type->cnt; 1966ed311b2SBenjamin Herrenschmidt } 1976ed311b2SBenjamin Herrenschmidt 19847cec443SMike Rapoport /** 19979442ed1STang Chen * __memblock_find_range_bottom_up - find free area utility in bottom-up 20079442ed1STang Chen * @start: start of candidate range 20147cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 20247cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 20379442ed1STang Chen * @size: size of free area to find 20479442ed1STang Chen * @align: alignment of free area to find 205b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 206fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 20779442ed1STang Chen * 20879442ed1STang Chen * Utility called from memblock_find_in_range_node(), find free area bottom-up. 20979442ed1STang Chen * 21047cec443SMike Rapoport * Return: 21179442ed1STang Chen * Found address on success, 0 on failure. 21279442ed1STang Chen */ 21379442ed1STang Chen static phys_addr_t __init_memblock 21479442ed1STang Chen __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, 215fc6daaf9STony Luck phys_addr_t size, phys_addr_t align, int nid, 216e1720feeSMike Rapoport enum memblock_flags flags) 21779442ed1STang Chen { 21879442ed1STang Chen phys_addr_t this_start, this_end, cand; 21979442ed1STang Chen u64 i; 22079442ed1STang Chen 221fc6daaf9STony Luck for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) { 22279442ed1STang Chen this_start = clamp(this_start, start, end); 22379442ed1STang Chen this_end = clamp(this_end, start, end); 22479442ed1STang Chen 22579442ed1STang Chen cand = round_up(this_start, align); 22679442ed1STang Chen if (cand < this_end && this_end - cand >= size) 22779442ed1STang Chen return cand; 22879442ed1STang Chen } 22979442ed1STang Chen 23079442ed1STang Chen return 0; 23179442ed1STang Chen } 23279442ed1STang Chen 2337bd0b0f0STejun Heo /** 2341402899eSTang Chen * __memblock_find_range_top_down - find free area utility, in top-down 2351402899eSTang Chen * @start: start of candidate range 23647cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 23747cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 2381402899eSTang Chen * @size: size of free area to find 2391402899eSTang Chen * @align: alignment of free area to find 240b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 241fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 2421402899eSTang Chen * 2431402899eSTang Chen * Utility called from memblock_find_in_range_node(), find free area top-down. 2441402899eSTang Chen * 24547cec443SMike Rapoport * Return: 24679442ed1STang Chen * Found address on success, 0 on failure. 2471402899eSTang Chen */ 2481402899eSTang Chen static phys_addr_t __init_memblock 2491402899eSTang Chen __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, 250fc6daaf9STony Luck phys_addr_t size, phys_addr_t align, int nid, 251e1720feeSMike Rapoport enum memblock_flags flags) 2521402899eSTang Chen { 2531402899eSTang Chen phys_addr_t this_start, this_end, cand; 2541402899eSTang Chen u64 i; 2551402899eSTang Chen 256fc6daaf9STony Luck for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end, 257fc6daaf9STony Luck NULL) { 2581402899eSTang Chen this_start = clamp(this_start, start, end); 2591402899eSTang Chen this_end = clamp(this_end, start, end); 2601402899eSTang Chen 2611402899eSTang Chen if (this_end < size) 2621402899eSTang Chen continue; 2631402899eSTang Chen 2641402899eSTang Chen cand = round_down(this_end - size, align); 2651402899eSTang Chen if (cand >= this_start) 2661402899eSTang Chen return cand; 2671402899eSTang Chen } 2681402899eSTang Chen 2691402899eSTang Chen return 0; 2701402899eSTang Chen } 2711402899eSTang Chen 2721402899eSTang Chen /** 2737bd0b0f0STejun Heo * memblock_find_in_range_node - find free area in given range and node 2747bd0b0f0STejun Heo * @size: size of free area to find 2757bd0b0f0STejun Heo * @align: alignment of free area to find 27687029ee9SGrygorii Strashko * @start: start of candidate range 27747cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 27847cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 279b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 280fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 2817bd0b0f0STejun Heo * 2827bd0b0f0STejun Heo * Find @size free area aligned to @align in the specified range and node. 2837bd0b0f0STejun Heo * 28447cec443SMike Rapoport * Return: 28579442ed1STang Chen * Found address on success, 0 on failure. 2866ed311b2SBenjamin Herrenschmidt */ 287c366ea89SMike Rapoport static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, 28887029ee9SGrygorii Strashko phys_addr_t align, phys_addr_t start, 289e1720feeSMike Rapoport phys_addr_t end, int nid, 290e1720feeSMike Rapoport enum memblock_flags flags) 291f7210e6cSTang Chen { 292f7210e6cSTang Chen /* pump up @end */ 293fed84c78SQian Cai if (end == MEMBLOCK_ALLOC_ACCESSIBLE || 294c6975d7cSQian Cai end == MEMBLOCK_ALLOC_NOLEAKTRACE) 295f7210e6cSTang Chen end = memblock.current_limit; 296f7210e6cSTang Chen 297f7210e6cSTang Chen /* avoid allocating the first page */ 298f7210e6cSTang Chen start = max_t(phys_addr_t, start, PAGE_SIZE); 299f7210e6cSTang Chen end = max(start, end); 30079442ed1STang Chen 3012dcb3964SRoman Gushchin if (memblock_bottom_up()) 3022dcb3964SRoman Gushchin return __memblock_find_range_bottom_up(start, end, size, align, 3032dcb3964SRoman Gushchin nid, flags); 3042dcb3964SRoman Gushchin else 3052dcb3964SRoman Gushchin return __memblock_find_range_top_down(start, end, size, align, 3062dcb3964SRoman Gushchin nid, flags); 307f7210e6cSTang Chen } 3086ed311b2SBenjamin Herrenschmidt 3097bd0b0f0STejun Heo /** 3107bd0b0f0STejun Heo * memblock_find_in_range - find free area in given range 3117bd0b0f0STejun Heo * @start: start of candidate range 31247cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 31347cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 3147bd0b0f0STejun Heo * @size: size of free area to find 3157bd0b0f0STejun Heo * @align: alignment of free area to find 3167bd0b0f0STejun Heo * 3177bd0b0f0STejun Heo * Find @size free area aligned to @align in the specified range. 3187bd0b0f0STejun Heo * 31947cec443SMike Rapoport * Return: 32079442ed1STang Chen * Found address on success, 0 on failure. 3217bd0b0f0STejun Heo */ 322a7259df7SMike Rapoport static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, 3237bd0b0f0STejun Heo phys_addr_t end, phys_addr_t size, 3247bd0b0f0STejun Heo phys_addr_t align) 3257bd0b0f0STejun Heo { 326a3f5bafcSTony Luck phys_addr_t ret; 327e1720feeSMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 328a3f5bafcSTony Luck 329a3f5bafcSTony Luck again: 330a3f5bafcSTony Luck ret = memblock_find_in_range_node(size, align, start, end, 331a3f5bafcSTony Luck NUMA_NO_NODE, flags); 332a3f5bafcSTony Luck 333a3f5bafcSTony Luck if (!ret && (flags & MEMBLOCK_MIRROR)) { 334a3f5bafcSTony Luck pr_warn("Could not allocate %pap bytes of mirrored memory\n", 335a3f5bafcSTony Luck &size); 336a3f5bafcSTony Luck flags &= ~MEMBLOCK_MIRROR; 337a3f5bafcSTony Luck goto again; 338a3f5bafcSTony Luck } 339a3f5bafcSTony Luck 340a3f5bafcSTony Luck return ret; 3417bd0b0f0STejun Heo } 3427bd0b0f0STejun Heo 34310d06439SYinghai Lu static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) 34495f72d1eSYinghai Lu { 3451440c4e2STejun Heo type->total_size -= type->regions[r].size; 3467c0caeb8STejun Heo memmove(&type->regions[r], &type->regions[r + 1], 3477c0caeb8STejun Heo (type->cnt - (r + 1)) * sizeof(type->regions[r])); 348e3239ff9SBenjamin Herrenschmidt type->cnt--; 34995f72d1eSYinghai Lu 3508f7a6605SBenjamin Herrenschmidt /* Special case for empty arrays */ 3518f7a6605SBenjamin Herrenschmidt if (type->cnt == 0) { 3521440c4e2STejun Heo WARN_ON(type->total_size != 0); 3538f7a6605SBenjamin Herrenschmidt type->cnt = 1; 3548f7a6605SBenjamin Herrenschmidt type->regions[0].base = 0; 3558f7a6605SBenjamin Herrenschmidt type->regions[0].size = 0; 35666a20757STang Chen type->regions[0].flags = 0; 3577c0caeb8STejun Heo memblock_set_region_node(&type->regions[0], MAX_NUMNODES); 3588f7a6605SBenjamin Herrenschmidt } 35995f72d1eSYinghai Lu } 36095f72d1eSYinghai Lu 361350e88baSMike Rapoport #ifndef CONFIG_ARCH_KEEP_MEMBLOCK 3623010f876SPavel Tatashin /** 36347cec443SMike Rapoport * memblock_discard - discard memory and reserved arrays if they were allocated 3643010f876SPavel Tatashin */ 3653010f876SPavel Tatashin void __init memblock_discard(void) 36629f67386SYinghai Lu { 3673010f876SPavel Tatashin phys_addr_t addr, size; 36829f67386SYinghai Lu 3693010f876SPavel Tatashin if (memblock.reserved.regions != memblock_reserved_init_regions) { 3703010f876SPavel Tatashin addr = __pa(memblock.reserved.regions); 3713010f876SPavel Tatashin size = PAGE_ALIGN(sizeof(struct memblock_region) * 37229f67386SYinghai Lu memblock.reserved.max); 373c94afc46SMiaohe Lin if (memblock_reserved_in_slab) 374c94afc46SMiaohe Lin kfree(memblock.reserved.regions); 375c94afc46SMiaohe Lin else 376621d9739SMike Rapoport memblock_free_late(addr, size); 37729f67386SYinghai Lu } 37829f67386SYinghai Lu 37991b540f9SPavel Tatashin if (memblock.memory.regions != memblock_memory_init_regions) { 3803010f876SPavel Tatashin addr = __pa(memblock.memory.regions); 3813010f876SPavel Tatashin size = PAGE_ALIGN(sizeof(struct memblock_region) * 3825e270e25SPhilipp Hachtmann memblock.memory.max); 383c94afc46SMiaohe Lin if (memblock_memory_in_slab) 384c94afc46SMiaohe Lin kfree(memblock.memory.regions); 385c94afc46SMiaohe Lin else 386621d9739SMike Rapoport memblock_free_late(addr, size); 3875e270e25SPhilipp Hachtmann } 3889f3d5eaaSMike Rapoport 3899f3d5eaaSMike Rapoport memblock_memory = NULL; 3903010f876SPavel Tatashin } 3915e270e25SPhilipp Hachtmann #endif 3925e270e25SPhilipp Hachtmann 39348c3b583SGreg Pearson /** 39448c3b583SGreg Pearson * memblock_double_array - double the size of the memblock regions array 39548c3b583SGreg Pearson * @type: memblock type of the regions array being doubled 39648c3b583SGreg Pearson * @new_area_start: starting address of memory range to avoid overlap with 39748c3b583SGreg Pearson * @new_area_size: size of memory range to avoid overlap with 39848c3b583SGreg Pearson * 39948c3b583SGreg Pearson * Double the size of the @type regions array. If memblock is being used to 40048c3b583SGreg Pearson * allocate memory for a new reserved regions array and there is a previously 40148c3b583SGreg Pearson * allocated memory range [@new_area_start, @new_area_start + @new_area_size] 40248c3b583SGreg Pearson * waiting to be reserved, ensure the memory used by the new array does 40348c3b583SGreg Pearson * not overlap. 40448c3b583SGreg Pearson * 40547cec443SMike Rapoport * Return: 40648c3b583SGreg Pearson * 0 on success, -1 on failure. 40748c3b583SGreg Pearson */ 40848c3b583SGreg Pearson static int __init_memblock memblock_double_array(struct memblock_type *type, 40948c3b583SGreg Pearson phys_addr_t new_area_start, 41048c3b583SGreg Pearson phys_addr_t new_area_size) 411142b45a7SBenjamin Herrenschmidt { 412142b45a7SBenjamin Herrenschmidt struct memblock_region *new_array, *old_array; 41329f67386SYinghai Lu phys_addr_t old_alloc_size, new_alloc_size; 414a36aab89SMike Rapoport phys_addr_t old_size, new_size, addr, new_end; 415142b45a7SBenjamin Herrenschmidt int use_slab = slab_is_available(); 416181eb394SGavin Shan int *in_slab; 417142b45a7SBenjamin Herrenschmidt 418142b45a7SBenjamin Herrenschmidt /* We don't allow resizing until we know about the reserved regions 419142b45a7SBenjamin Herrenschmidt * of memory that aren't suitable for allocation 420142b45a7SBenjamin Herrenschmidt */ 421142b45a7SBenjamin Herrenschmidt if (!memblock_can_resize) 422142b45a7SBenjamin Herrenschmidt return -1; 423142b45a7SBenjamin Herrenschmidt 424142b45a7SBenjamin Herrenschmidt /* Calculate new doubled size */ 425142b45a7SBenjamin Herrenschmidt old_size = type->max * sizeof(struct memblock_region); 426142b45a7SBenjamin Herrenschmidt new_size = old_size << 1; 42729f67386SYinghai Lu /* 42829f67386SYinghai Lu * We need to allocated new one align to PAGE_SIZE, 42929f67386SYinghai Lu * so we can free them completely later. 43029f67386SYinghai Lu */ 43129f67386SYinghai Lu old_alloc_size = PAGE_ALIGN(old_size); 43229f67386SYinghai Lu new_alloc_size = PAGE_ALIGN(new_size); 433142b45a7SBenjamin Herrenschmidt 434181eb394SGavin Shan /* Retrieve the slab flag */ 435181eb394SGavin Shan if (type == &memblock.memory) 436181eb394SGavin Shan in_slab = &memblock_memory_in_slab; 437181eb394SGavin Shan else 438181eb394SGavin Shan in_slab = &memblock_reserved_in_slab; 439181eb394SGavin Shan 440a2974133SMike Rapoport /* Try to find some space for it */ 441142b45a7SBenjamin Herrenschmidt if (use_slab) { 442142b45a7SBenjamin Herrenschmidt new_array = kmalloc(new_size, GFP_KERNEL); 4431f5026a7STejun Heo addr = new_array ? __pa(new_array) : 0; 4444e2f0775SGavin Shan } else { 44548c3b583SGreg Pearson /* only exclude range when trying to double reserved.regions */ 44648c3b583SGreg Pearson if (type != &memblock.reserved) 44748c3b583SGreg Pearson new_area_start = new_area_size = 0; 44848c3b583SGreg Pearson 44948c3b583SGreg Pearson addr = memblock_find_in_range(new_area_start + new_area_size, 45048c3b583SGreg Pearson memblock.current_limit, 45129f67386SYinghai Lu new_alloc_size, PAGE_SIZE); 45248c3b583SGreg Pearson if (!addr && new_area_size) 45348c3b583SGreg Pearson addr = memblock_find_in_range(0, 45448c3b583SGreg Pearson min(new_area_start, memblock.current_limit), 45529f67386SYinghai Lu new_alloc_size, PAGE_SIZE); 45648c3b583SGreg Pearson 45715674868SSachin Kamat new_array = addr ? __va(addr) : NULL; 4584e2f0775SGavin Shan } 4591f5026a7STejun Heo if (!addr) { 460142b45a7SBenjamin Herrenschmidt pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 4610262d9c8SHeiko Carstens type->name, type->max, type->max * 2); 462142b45a7SBenjamin Herrenschmidt return -1; 463142b45a7SBenjamin Herrenschmidt } 464142b45a7SBenjamin Herrenschmidt 465a36aab89SMike Rapoport new_end = addr + new_size - 1; 466a36aab89SMike Rapoport memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]", 467a36aab89SMike Rapoport type->name, type->max * 2, &addr, &new_end); 468ea9e4376SYinghai Lu 469fd07383bSAndrew Morton /* 470fd07383bSAndrew Morton * Found space, we now need to move the array over before we add the 471fd07383bSAndrew Morton * reserved region since it may be our reserved array itself that is 472fd07383bSAndrew Morton * full. 473142b45a7SBenjamin Herrenschmidt */ 474142b45a7SBenjamin Herrenschmidt memcpy(new_array, type->regions, old_size); 475142b45a7SBenjamin Herrenschmidt memset(new_array + type->max, 0, old_size); 476142b45a7SBenjamin Herrenschmidt old_array = type->regions; 477142b45a7SBenjamin Herrenschmidt type->regions = new_array; 478142b45a7SBenjamin Herrenschmidt type->max <<= 1; 479142b45a7SBenjamin Herrenschmidt 480fd07383bSAndrew Morton /* Free old array. We needn't free it if the array is the static one */ 481181eb394SGavin Shan if (*in_slab) 482181eb394SGavin Shan kfree(old_array); 483181eb394SGavin Shan else if (old_array != memblock_memory_init_regions && 484142b45a7SBenjamin Herrenschmidt old_array != memblock_reserved_init_regions) 4854421cca0SMike Rapoport memblock_free(old_array, old_alloc_size); 486142b45a7SBenjamin Herrenschmidt 487fd07383bSAndrew Morton /* 488fd07383bSAndrew Morton * Reserve the new array if that comes from the memblock. Otherwise, we 489fd07383bSAndrew Morton * needn't do it 490181eb394SGavin Shan */ 491181eb394SGavin Shan if (!use_slab) 49229f67386SYinghai Lu BUG_ON(memblock_reserve(addr, new_alloc_size)); 493181eb394SGavin Shan 494181eb394SGavin Shan /* Update slab flag */ 495181eb394SGavin Shan *in_slab = use_slab; 496181eb394SGavin Shan 497142b45a7SBenjamin Herrenschmidt return 0; 498142b45a7SBenjamin Herrenschmidt } 499142b45a7SBenjamin Herrenschmidt 500784656f9STejun Heo /** 501784656f9STejun Heo * memblock_merge_regions - merge neighboring compatible regions 502784656f9STejun Heo * @type: memblock type to scan 503784656f9STejun Heo * 504784656f9STejun Heo * Scan @type and merge neighboring compatible regions. 505784656f9STejun Heo */ 506784656f9STejun Heo static void __init_memblock memblock_merge_regions(struct memblock_type *type) 507784656f9STejun Heo { 508784656f9STejun Heo int i = 0; 509784656f9STejun Heo 510784656f9STejun Heo /* cnt never goes below 1 */ 511784656f9STejun Heo while (i < type->cnt - 1) { 512784656f9STejun Heo struct memblock_region *this = &type->regions[i]; 513784656f9STejun Heo struct memblock_region *next = &type->regions[i + 1]; 514784656f9STejun Heo 5157c0caeb8STejun Heo if (this->base + this->size != next->base || 5167c0caeb8STejun Heo memblock_get_region_node(this) != 51766a20757STang Chen memblock_get_region_node(next) || 51866a20757STang Chen this->flags != next->flags) { 519784656f9STejun Heo BUG_ON(this->base + this->size > next->base); 520784656f9STejun Heo i++; 521784656f9STejun Heo continue; 522784656f9STejun Heo } 523784656f9STejun Heo 524784656f9STejun Heo this->size += next->size; 525c0232ae8SLin Feng /* move forward from next + 1, index of which is i + 2 */ 526c0232ae8SLin Feng memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 527784656f9STejun Heo type->cnt--; 528784656f9STejun Heo } 529784656f9STejun Heo } 530784656f9STejun Heo 531784656f9STejun Heo /** 532784656f9STejun Heo * memblock_insert_region - insert new memblock region 533784656f9STejun Heo * @type: memblock type to insert into 534784656f9STejun Heo * @idx: index for the insertion point 535784656f9STejun Heo * @base: base address of the new region 536784656f9STejun Heo * @size: size of the new region 537209ff86dSTang Chen * @nid: node id of the new region 53866a20757STang Chen * @flags: flags of the new region 539784656f9STejun Heo * 540784656f9STejun Heo * Insert new memblock region [@base, @base + @size) into @type at @idx. 541412d0008SAlexander Kuleshov * @type must already have extra room to accommodate the new region. 542784656f9STejun Heo */ 543784656f9STejun Heo static void __init_memblock memblock_insert_region(struct memblock_type *type, 544784656f9STejun Heo int idx, phys_addr_t base, 54566a20757STang Chen phys_addr_t size, 546e1720feeSMike Rapoport int nid, 547e1720feeSMike Rapoport enum memblock_flags flags) 548784656f9STejun Heo { 549784656f9STejun Heo struct memblock_region *rgn = &type->regions[idx]; 550784656f9STejun Heo 551784656f9STejun Heo BUG_ON(type->cnt >= type->max); 552784656f9STejun Heo memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 553784656f9STejun Heo rgn->base = base; 554784656f9STejun Heo rgn->size = size; 55566a20757STang Chen rgn->flags = flags; 5567c0caeb8STejun Heo memblock_set_region_node(rgn, nid); 557784656f9STejun Heo type->cnt++; 5581440c4e2STejun Heo type->total_size += size; 559784656f9STejun Heo } 560784656f9STejun Heo 561784656f9STejun Heo /** 562f1af9d3aSPhilipp Hachtmann * memblock_add_range - add new memblock region 563784656f9STejun Heo * @type: memblock type to add new region into 564784656f9STejun Heo * @base: base address of the new region 565784656f9STejun Heo * @size: size of the new region 5667fb0bc3fSTejun Heo * @nid: nid of the new region 56766a20757STang Chen * @flags: flags of the new region 568784656f9STejun Heo * 569784656f9STejun Heo * Add new memblock region [@base, @base + @size) into @type. The new region 570784656f9STejun Heo * is allowed to overlap with existing ones - overlaps don't affect already 571784656f9STejun Heo * existing regions. @type is guaranteed to be minimal (all neighbouring 572784656f9STejun Heo * compatible regions are merged) after the addition. 573784656f9STejun Heo * 57447cec443SMike Rapoport * Return: 575784656f9STejun Heo * 0 on success, -errno on failure. 576784656f9STejun Heo */ 57702634a44SAnshuman Khandual static int __init_memblock memblock_add_range(struct memblock_type *type, 57866a20757STang Chen phys_addr_t base, phys_addr_t size, 579e1720feeSMike Rapoport int nid, enum memblock_flags flags) 58095f72d1eSYinghai Lu { 581784656f9STejun Heo bool insert = false; 582eb18f1b5STejun Heo phys_addr_t obase = base; 583eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 5848c9c1701SAlexander Kuleshov int idx, nr_new; 5858c9c1701SAlexander Kuleshov struct memblock_region *rgn; 58695f72d1eSYinghai Lu 587b3dc627cSTejun Heo if (!size) 588b3dc627cSTejun Heo return 0; 589b3dc627cSTejun Heo 590784656f9STejun Heo /* special case for empty array */ 591784656f9STejun Heo if (type->regions[0].size == 0) { 5921440c4e2STejun Heo WARN_ON(type->cnt != 1 || type->total_size); 593784656f9STejun Heo type->regions[0].base = base; 594784656f9STejun Heo type->regions[0].size = size; 59566a20757STang Chen type->regions[0].flags = flags; 5967fb0bc3fSTejun Heo memblock_set_region_node(&type->regions[0], nid); 5971440c4e2STejun Heo type->total_size = size; 598784656f9STejun Heo return 0; 599784656f9STejun Heo } 600784656f9STejun Heo repeat: 601784656f9STejun Heo /* 602784656f9STejun Heo * The following is executed twice. Once with %false @insert and 603784656f9STejun Heo * then with %true. The first counts the number of regions needed 604412d0008SAlexander Kuleshov * to accommodate the new area. The second actually inserts them. 605784656f9STejun Heo */ 606784656f9STejun Heo base = obase; 607784656f9STejun Heo nr_new = 0; 608784656f9STejun Heo 60966e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 610784656f9STejun Heo phys_addr_t rbase = rgn->base; 611784656f9STejun Heo phys_addr_t rend = rbase + rgn->size; 6128f7a6605SBenjamin Herrenschmidt 613784656f9STejun Heo if (rbase >= end) 6148f7a6605SBenjamin Herrenschmidt break; 615784656f9STejun Heo if (rend <= base) 616784656f9STejun Heo continue; 617784656f9STejun Heo /* 618784656f9STejun Heo * @rgn overlaps. If it separates the lower part of new 619784656f9STejun Heo * area, insert that portion. 6208f7a6605SBenjamin Herrenschmidt */ 621784656f9STejun Heo if (rbase > base) { 622a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 623c0a29498SWei Yang WARN_ON(nid != memblock_get_region_node(rgn)); 624c0a29498SWei Yang #endif 6254fcab5f4SWei Yang WARN_ON(flags != rgn->flags); 626784656f9STejun Heo nr_new++; 627784656f9STejun Heo if (insert) 6288c9c1701SAlexander Kuleshov memblock_insert_region(type, idx++, base, 62966a20757STang Chen rbase - base, nid, 63066a20757STang Chen flags); 631784656f9STejun Heo } 632784656f9STejun Heo /* area below @rend is dealt with, forget about it */ 633784656f9STejun Heo base = min(rend, end); 6348f7a6605SBenjamin Herrenschmidt } 6358f7a6605SBenjamin Herrenschmidt 636784656f9STejun Heo /* insert the remaining portion */ 637784656f9STejun Heo if (base < end) { 638784656f9STejun Heo nr_new++; 639784656f9STejun Heo if (insert) 6408c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, base, end - base, 64166a20757STang Chen nid, flags); 6428f7a6605SBenjamin Herrenschmidt } 6438f7a6605SBenjamin Herrenschmidt 644ef3cc4dbSnimisolo if (!nr_new) 645ef3cc4dbSnimisolo return 0; 646ef3cc4dbSnimisolo 647784656f9STejun Heo /* 648784656f9STejun Heo * If this was the first round, resize array and repeat for actual 649784656f9STejun Heo * insertions; otherwise, merge and return. 6508f7a6605SBenjamin Herrenschmidt */ 651784656f9STejun Heo if (!insert) { 652784656f9STejun Heo while (type->cnt + nr_new > type->max) 65348c3b583SGreg Pearson if (memblock_double_array(type, obase, size) < 0) 654784656f9STejun Heo return -ENOMEM; 655784656f9STejun Heo insert = true; 656784656f9STejun Heo goto repeat; 65795f72d1eSYinghai Lu } else { 658784656f9STejun Heo memblock_merge_regions(type); 65995f72d1eSYinghai Lu return 0; 66095f72d1eSYinghai Lu } 661784656f9STejun Heo } 66295f72d1eSYinghai Lu 66348a833ccSMike Rapoport /** 66448a833ccSMike Rapoport * memblock_add_node - add new memblock region within a NUMA node 66548a833ccSMike Rapoport * @base: base address of the new region 66648a833ccSMike Rapoport * @size: size of the new region 66748a833ccSMike Rapoport * @nid: nid of the new region 668952eea9bSDavid Hildenbrand * @flags: flags of the new region 66948a833ccSMike Rapoport * 67048a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 67148a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 67248a833ccSMike Rapoport * 67348a833ccSMike Rapoport * Return: 67448a833ccSMike Rapoport * 0 on success, -errno on failure. 67548a833ccSMike Rapoport */ 6767fb0bc3fSTejun Heo int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 677952eea9bSDavid Hildenbrand int nid, enum memblock_flags flags) 6787fb0bc3fSTejun Heo { 67900974b9aSGeert Uytterhoeven phys_addr_t end = base + size - 1; 68000974b9aSGeert Uytterhoeven 681952eea9bSDavid Hildenbrand memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__, 682952eea9bSDavid Hildenbrand &base, &end, nid, flags, (void *)_RET_IP_); 68300974b9aSGeert Uytterhoeven 684952eea9bSDavid Hildenbrand return memblock_add_range(&memblock.memory, base, size, nid, flags); 6857fb0bc3fSTejun Heo } 6867fb0bc3fSTejun Heo 68748a833ccSMike Rapoport /** 68848a833ccSMike Rapoport * memblock_add - add new memblock region 68948a833ccSMike Rapoport * @base: base address of the new region 69048a833ccSMike Rapoport * @size: size of the new region 69148a833ccSMike Rapoport * 69248a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 69348a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 69448a833ccSMike Rapoport * 69548a833ccSMike Rapoport * Return: 69648a833ccSMike Rapoport * 0 on success, -errno on failure. 69748a833ccSMike Rapoport */ 698f705ac4bSAlexander Kuleshov int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 6996a4055bcSAlexander Kuleshov { 7005d63f81cSMiles Chen phys_addr_t end = base + size - 1; 7015d63f81cSMiles Chen 702a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 7035d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 7046a4055bcSAlexander Kuleshov 705f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 70695f72d1eSYinghai Lu } 70795f72d1eSYinghai Lu 7086a9ceb31STejun Heo /** 7096a9ceb31STejun Heo * memblock_isolate_range - isolate given range into disjoint memblocks 7106a9ceb31STejun Heo * @type: memblock type to isolate range for 7116a9ceb31STejun Heo * @base: base of range to isolate 7126a9ceb31STejun Heo * @size: size of range to isolate 7136a9ceb31STejun Heo * @start_rgn: out parameter for the start of isolated region 7146a9ceb31STejun Heo * @end_rgn: out parameter for the end of isolated region 7156a9ceb31STejun Heo * 7166a9ceb31STejun Heo * Walk @type and ensure that regions don't cross the boundaries defined by 7176a9ceb31STejun Heo * [@base, @base + @size). Crossing regions are split at the boundaries, 7186a9ceb31STejun Heo * which may create at most two more regions. The index of the first 7196a9ceb31STejun Heo * region inside the range is returned in *@start_rgn and end in *@end_rgn. 7206a9ceb31STejun Heo * 72147cec443SMike Rapoport * Return: 7226a9ceb31STejun Heo * 0 on success, -errno on failure. 7236a9ceb31STejun Heo */ 7246a9ceb31STejun Heo static int __init_memblock memblock_isolate_range(struct memblock_type *type, 7256a9ceb31STejun Heo phys_addr_t base, phys_addr_t size, 7266a9ceb31STejun Heo int *start_rgn, int *end_rgn) 7276a9ceb31STejun Heo { 728eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 7298c9c1701SAlexander Kuleshov int idx; 7308c9c1701SAlexander Kuleshov struct memblock_region *rgn; 7316a9ceb31STejun Heo 7326a9ceb31STejun Heo *start_rgn = *end_rgn = 0; 7336a9ceb31STejun Heo 734b3dc627cSTejun Heo if (!size) 735b3dc627cSTejun Heo return 0; 736b3dc627cSTejun Heo 7376a9ceb31STejun Heo /* we'll create at most two more regions */ 7386a9ceb31STejun Heo while (type->cnt + 2 > type->max) 73948c3b583SGreg Pearson if (memblock_double_array(type, base, size) < 0) 7406a9ceb31STejun Heo return -ENOMEM; 7416a9ceb31STejun Heo 74266e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 7436a9ceb31STejun Heo phys_addr_t rbase = rgn->base; 7446a9ceb31STejun Heo phys_addr_t rend = rbase + rgn->size; 7456a9ceb31STejun Heo 7466a9ceb31STejun Heo if (rbase >= end) 7476a9ceb31STejun Heo break; 7486a9ceb31STejun Heo if (rend <= base) 7496a9ceb31STejun Heo continue; 7506a9ceb31STejun Heo 7516a9ceb31STejun Heo if (rbase < base) { 7526a9ceb31STejun Heo /* 7536a9ceb31STejun Heo * @rgn intersects from below. Split and continue 7546a9ceb31STejun Heo * to process the next region - the new top half. 7556a9ceb31STejun Heo */ 7566a9ceb31STejun Heo rgn->base = base; 7571440c4e2STejun Heo rgn->size -= base - rbase; 7581440c4e2STejun Heo type->total_size -= base - rbase; 7598c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, rbase, base - rbase, 76066a20757STang Chen memblock_get_region_node(rgn), 76166a20757STang Chen rgn->flags); 7626a9ceb31STejun Heo } else if (rend > end) { 7636a9ceb31STejun Heo /* 7646a9ceb31STejun Heo * @rgn intersects from above. Split and redo the 7656a9ceb31STejun Heo * current region - the new bottom half. 7666a9ceb31STejun Heo */ 7676a9ceb31STejun Heo rgn->base = end; 7681440c4e2STejun Heo rgn->size -= end - rbase; 7691440c4e2STejun Heo type->total_size -= end - rbase; 7708c9c1701SAlexander Kuleshov memblock_insert_region(type, idx--, rbase, end - rbase, 77166a20757STang Chen memblock_get_region_node(rgn), 77266a20757STang Chen rgn->flags); 7736a9ceb31STejun Heo } else { 7746a9ceb31STejun Heo /* @rgn is fully contained, record it */ 7756a9ceb31STejun Heo if (!*end_rgn) 7768c9c1701SAlexander Kuleshov *start_rgn = idx; 7778c9c1701SAlexander Kuleshov *end_rgn = idx + 1; 7786a9ceb31STejun Heo } 7796a9ceb31STejun Heo } 7806a9ceb31STejun Heo 7816a9ceb31STejun Heo return 0; 7826a9ceb31STejun Heo } 7836a9ceb31STejun Heo 78435bd16a2SAlexander Kuleshov static int __init_memblock memblock_remove_range(struct memblock_type *type, 7858f7a6605SBenjamin Herrenschmidt phys_addr_t base, phys_addr_t size) 78695f72d1eSYinghai Lu { 78771936180STejun Heo int start_rgn, end_rgn; 78871936180STejun Heo int i, ret; 78995f72d1eSYinghai Lu 79071936180STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 79171936180STejun Heo if (ret) 79271936180STejun Heo return ret; 79395f72d1eSYinghai Lu 79471936180STejun Heo for (i = end_rgn - 1; i >= start_rgn; i--) 79571936180STejun Heo memblock_remove_region(type, i); 79695f72d1eSYinghai Lu return 0; 79795f72d1eSYinghai Lu } 79895f72d1eSYinghai Lu 799581adcbeSTejun Heo int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 80095f72d1eSYinghai Lu { 80125cf23d7SMinchan Kim phys_addr_t end = base + size - 1; 80225cf23d7SMinchan Kim 803a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 80425cf23d7SMinchan Kim &base, &end, (void *)_RET_IP_); 80525cf23d7SMinchan Kim 806f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.memory, base, size); 80795f72d1eSYinghai Lu } 80895f72d1eSYinghai Lu 8094d72868cSMike Rapoport /** 8104421cca0SMike Rapoport * memblock_free - free boot memory allocation 81177e02cf5SLinus Torvalds * @ptr: starting address of the boot memory allocation 81277e02cf5SLinus Torvalds * @size: size of the boot memory block in bytes 81377e02cf5SLinus Torvalds * 81477e02cf5SLinus Torvalds * Free boot memory block previously allocated by memblock_alloc_xx() API. 81577e02cf5SLinus Torvalds * The freeing memory will not be released to the buddy allocator. 81677e02cf5SLinus Torvalds */ 8174421cca0SMike Rapoport void __init_memblock memblock_free(void *ptr, size_t size) 81877e02cf5SLinus Torvalds { 81977e02cf5SLinus Torvalds if (ptr) 8203ecc6834SMike Rapoport memblock_phys_free(__pa(ptr), size); 82177e02cf5SLinus Torvalds } 82277e02cf5SLinus Torvalds 82377e02cf5SLinus Torvalds /** 8243ecc6834SMike Rapoport * memblock_phys_free - free boot memory block 8254d72868cSMike Rapoport * @base: phys starting address of the boot memory block 8264d72868cSMike Rapoport * @size: size of the boot memory block in bytes 8274d72868cSMike Rapoport * 8284d72868cSMike Rapoport * Free boot memory block previously allocated by memblock_alloc_xx() API. 8294d72868cSMike Rapoport * The freeing memory will not be released to the buddy allocator. 8304d72868cSMike Rapoport */ 8313ecc6834SMike Rapoport int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size) 83295f72d1eSYinghai Lu { 8335d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8345d63f81cSMiles Chen 835a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 8365d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 83724aa0788STejun Heo 8389099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 839f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.reserved, base, size); 84095f72d1eSYinghai Lu } 84195f72d1eSYinghai Lu 842f705ac4bSAlexander Kuleshov int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 84395f72d1eSYinghai Lu { 8445d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8455d63f81cSMiles Chen 846a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 8475d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 84895f72d1eSYinghai Lu 849f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 85095f72d1eSYinghai Lu } 85195f72d1eSYinghai Lu 85202634a44SAnshuman Khandual #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 85302634a44SAnshuman Khandual int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size) 85402634a44SAnshuman Khandual { 85502634a44SAnshuman Khandual phys_addr_t end = base + size - 1; 85602634a44SAnshuman Khandual 85702634a44SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 85802634a44SAnshuman Khandual &base, &end, (void *)_RET_IP_); 85902634a44SAnshuman Khandual 86077649905SDavid Hildenbrand return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0); 86102634a44SAnshuman Khandual } 86202634a44SAnshuman Khandual #endif 86302634a44SAnshuman Khandual 86435fd0808STejun Heo /** 86547cec443SMike Rapoport * memblock_setclr_flag - set or clear flag for a memory region 86647cec443SMike Rapoport * @base: base address of the region 86747cec443SMike Rapoport * @size: size of the region 86847cec443SMike Rapoport * @set: set or clear the flag 8698958b249SHaitao Shi * @flag: the flag to update 87066b16edfSTang Chen * 8714308ce17STony Luck * This function isolates region [@base, @base + @size), and sets/clears flag 87266b16edfSTang Chen * 87347cec443SMike Rapoport * Return: 0 on success, -errno on failure. 87466b16edfSTang Chen */ 8754308ce17STony Luck static int __init_memblock memblock_setclr_flag(phys_addr_t base, 8764308ce17STony Luck phys_addr_t size, int set, int flag) 87766b16edfSTang Chen { 87866b16edfSTang Chen struct memblock_type *type = &memblock.memory; 87966b16edfSTang Chen int i, ret, start_rgn, end_rgn; 88066b16edfSTang Chen 88166b16edfSTang Chen ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 88266b16edfSTang Chen if (ret) 88366b16edfSTang Chen return ret; 88466b16edfSTang Chen 885fe145124SMike Rapoport for (i = start_rgn; i < end_rgn; i++) { 886fe145124SMike Rapoport struct memblock_region *r = &type->regions[i]; 887fe145124SMike Rapoport 8884308ce17STony Luck if (set) 889fe145124SMike Rapoport r->flags |= flag; 8904308ce17STony Luck else 891fe145124SMike Rapoport r->flags &= ~flag; 892fe145124SMike Rapoport } 89366b16edfSTang Chen 89466b16edfSTang Chen memblock_merge_regions(type); 89566b16edfSTang Chen return 0; 89666b16edfSTang Chen } 89766b16edfSTang Chen 89866b16edfSTang Chen /** 8994308ce17STony Luck * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 9004308ce17STony Luck * @base: the base phys addr of the region 9014308ce17STony Luck * @size: the size of the region 9024308ce17STony Luck * 90347cec443SMike Rapoport * Return: 0 on success, -errno on failure. 9044308ce17STony Luck */ 9054308ce17STony Luck int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 9064308ce17STony Luck { 9074308ce17STony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 9084308ce17STony Luck } 9094308ce17STony Luck 9104308ce17STony Luck /** 91166b16edfSTang Chen * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 91266b16edfSTang Chen * @base: the base phys addr of the region 91366b16edfSTang Chen * @size: the size of the region 91466b16edfSTang Chen * 91547cec443SMike Rapoport * Return: 0 on success, -errno on failure. 91666b16edfSTang Chen */ 91766b16edfSTang Chen int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 91866b16edfSTang Chen { 9194308ce17STony Luck return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 92066b16edfSTang Chen } 92166b16edfSTang Chen 92266b16edfSTang Chen /** 923a3f5bafcSTony Luck * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 924a3f5bafcSTony Luck * @base: the base phys addr of the region 925a3f5bafcSTony Luck * @size: the size of the region 926a3f5bafcSTony Luck * 92747cec443SMike Rapoport * Return: 0 on success, -errno on failure. 928a3f5bafcSTony Luck */ 929a3f5bafcSTony Luck int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 930a3f5bafcSTony Luck { 931a3f5bafcSTony Luck system_has_some_mirror = true; 932a3f5bafcSTony Luck 933a3f5bafcSTony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 934a3f5bafcSTony Luck } 935a3f5bafcSTony Luck 936bf3d3cc5SArd Biesheuvel /** 937bf3d3cc5SArd Biesheuvel * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 938bf3d3cc5SArd Biesheuvel * @base: the base phys addr of the region 939bf3d3cc5SArd Biesheuvel * @size: the size of the region 940bf3d3cc5SArd Biesheuvel * 9419092d4f7SMike Rapoport * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the 9429092d4f7SMike Rapoport * direct mapping of the physical memory. These regions will still be 9439092d4f7SMike Rapoport * covered by the memory map. The struct page representing NOMAP memory 9449092d4f7SMike Rapoport * frames in the memory map will be PageReserved() 9459092d4f7SMike Rapoport * 946658aafc8SMike Rapoport * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from 947658aafc8SMike Rapoport * memblock, the caller must inform kmemleak to ignore that memory 948658aafc8SMike Rapoport * 94947cec443SMike Rapoport * Return: 0 on success, -errno on failure. 950bf3d3cc5SArd Biesheuvel */ 951bf3d3cc5SArd Biesheuvel int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 952bf3d3cc5SArd Biesheuvel { 9536c9a5455SMike Rapoport return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 954bf3d3cc5SArd Biesheuvel } 955a3f5bafcSTony Luck 956a3f5bafcSTony Luck /** 9574c546b8aSAKASHI Takahiro * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 9584c546b8aSAKASHI Takahiro * @base: the base phys addr of the region 9594c546b8aSAKASHI Takahiro * @size: the size of the region 9604c546b8aSAKASHI Takahiro * 96147cec443SMike Rapoport * Return: 0 on success, -errno on failure. 9624c546b8aSAKASHI Takahiro */ 9634c546b8aSAKASHI Takahiro int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 9644c546b8aSAKASHI Takahiro { 9654c546b8aSAKASHI Takahiro return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 9664c546b8aSAKASHI Takahiro } 9674c546b8aSAKASHI Takahiro 9689f3d5eaaSMike Rapoport static bool should_skip_region(struct memblock_type *type, 9699f3d5eaaSMike Rapoport struct memblock_region *m, 9709f3d5eaaSMike Rapoport int nid, int flags) 971c9a688a3SMike Rapoport { 972c9a688a3SMike Rapoport int m_nid = memblock_get_region_node(m); 973c9a688a3SMike Rapoport 9749f3d5eaaSMike Rapoport /* we never skip regions when iterating memblock.reserved or physmem */ 9759f3d5eaaSMike Rapoport if (type != memblock_memory) 9769f3d5eaaSMike Rapoport return false; 9779f3d5eaaSMike Rapoport 978c9a688a3SMike Rapoport /* only memory regions are associated with nodes, check it */ 979c9a688a3SMike Rapoport if (nid != NUMA_NO_NODE && nid != m_nid) 980c9a688a3SMike Rapoport return true; 981c9a688a3SMike Rapoport 982c9a688a3SMike Rapoport /* skip hotpluggable memory regions if needed */ 98379e482e9SMike Rapoport if (movable_node_is_enabled() && memblock_is_hotpluggable(m) && 98479e482e9SMike Rapoport !(flags & MEMBLOCK_HOTPLUG)) 985c9a688a3SMike Rapoport return true; 986c9a688a3SMike Rapoport 987c9a688a3SMike Rapoport /* if we want mirror memory skip non-mirror memory regions */ 988c9a688a3SMike Rapoport if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 989c9a688a3SMike Rapoport return true; 990c9a688a3SMike Rapoport 991c9a688a3SMike Rapoport /* skip nomap memory unless we were asked for it explicitly */ 992c9a688a3SMike Rapoport if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 993c9a688a3SMike Rapoport return true; 994c9a688a3SMike Rapoport 995f7892d8eSDavid Hildenbrand /* skip driver-managed memory unless we were asked for it explicitly */ 996f7892d8eSDavid Hildenbrand if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m)) 997f7892d8eSDavid Hildenbrand return true; 998f7892d8eSDavid Hildenbrand 999c9a688a3SMike Rapoport return false; 1000c9a688a3SMike Rapoport } 1001c9a688a3SMike Rapoport 10028e7a7f86SRobin Holt /** 1003a2974133SMike Rapoport * __next_mem_range - next function for for_each_free_mem_range() etc. 100435fd0808STejun Heo * @idx: pointer to u64 loop variable 1005b1154233SGrygorii Strashko * @nid: node selector, %NUMA_NO_NODE for all nodes 1006fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 1007f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 1008f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 1009dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1010dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1011dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 101235fd0808STejun Heo * 1013f1af9d3aSPhilipp Hachtmann * Find the first area from *@idx which matches @nid, fill the out 101435fd0808STejun Heo * parameters, and update *@idx for the next iteration. The lower 32bit of 1015f1af9d3aSPhilipp Hachtmann * *@idx contains index into type_a and the upper 32bit indexes the 1016f1af9d3aSPhilipp Hachtmann * areas before each region in type_b. For example, if type_b regions 101735fd0808STejun Heo * look like the following, 101835fd0808STejun Heo * 101935fd0808STejun Heo * 0:[0-16), 1:[32-48), 2:[128-130) 102035fd0808STejun Heo * 102135fd0808STejun Heo * The upper 32bit indexes the following regions. 102235fd0808STejun Heo * 102335fd0808STejun Heo * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 102435fd0808STejun Heo * 102535fd0808STejun Heo * As both region arrays are sorted, the function advances the two indices 102635fd0808STejun Heo * in lockstep and returns each intersection. 102735fd0808STejun Heo */ 102877649905SDavid Hildenbrand void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags, 1029f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 103077649905SDavid Hildenbrand struct memblock_type *type_b, phys_addr_t *out_start, 103135fd0808STejun Heo phys_addr_t *out_end, int *out_nid) 103235fd0808STejun Heo { 1033f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 1034f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 1035b1154233SGrygorii Strashko 1036f1af9d3aSPhilipp Hachtmann if (WARN_ONCE(nid == MAX_NUMNODES, 1037f1af9d3aSPhilipp Hachtmann "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1038560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 103935fd0808STejun Heo 1040f1af9d3aSPhilipp Hachtmann for (; idx_a < type_a->cnt; idx_a++) { 1041f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1042f1af9d3aSPhilipp Hachtmann 104335fd0808STejun Heo phys_addr_t m_start = m->base; 104435fd0808STejun Heo phys_addr_t m_end = m->base + m->size; 1045f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 104635fd0808STejun Heo 10479f3d5eaaSMike Rapoport if (should_skip_region(type_a, m, nid, flags)) 1048bf3d3cc5SArd Biesheuvel continue; 1049bf3d3cc5SArd Biesheuvel 1050f1af9d3aSPhilipp Hachtmann if (!type_b) { 1051f1af9d3aSPhilipp Hachtmann if (out_start) 1052f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1053f1af9d3aSPhilipp Hachtmann if (out_end) 1054f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1055f1af9d3aSPhilipp Hachtmann if (out_nid) 1056f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1057f1af9d3aSPhilipp Hachtmann idx_a++; 1058f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1059f1af9d3aSPhilipp Hachtmann return; 1060f1af9d3aSPhilipp Hachtmann } 106135fd0808STejun Heo 1062f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1063f1af9d3aSPhilipp Hachtmann for (; idx_b < type_b->cnt + 1; idx_b++) { 1064f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1065f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1066f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1067f1af9d3aSPhilipp Hachtmann 1068f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1069f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1070f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 10711c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1072f1af9d3aSPhilipp Hachtmann 1073f1af9d3aSPhilipp Hachtmann /* 1074f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1075f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1076f1af9d3aSPhilipp Hachtmann */ 107735fd0808STejun Heo if (r_start >= m_end) 107835fd0808STejun Heo break; 107935fd0808STejun Heo /* if the two regions intersect, we're done */ 108035fd0808STejun Heo if (m_start < r_end) { 108135fd0808STejun Heo if (out_start) 1082f1af9d3aSPhilipp Hachtmann *out_start = 1083f1af9d3aSPhilipp Hachtmann max(m_start, r_start); 108435fd0808STejun Heo if (out_end) 108535fd0808STejun Heo *out_end = min(m_end, r_end); 108635fd0808STejun Heo if (out_nid) 1087f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 108835fd0808STejun Heo /* 1089f1af9d3aSPhilipp Hachtmann * The region which ends first is 1090f1af9d3aSPhilipp Hachtmann * advanced for the next iteration. 109135fd0808STejun Heo */ 109235fd0808STejun Heo if (m_end <= r_end) 1093f1af9d3aSPhilipp Hachtmann idx_a++; 109435fd0808STejun Heo else 1095f1af9d3aSPhilipp Hachtmann idx_b++; 1096f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 109735fd0808STejun Heo return; 109835fd0808STejun Heo } 109935fd0808STejun Heo } 110035fd0808STejun Heo } 110135fd0808STejun Heo 110235fd0808STejun Heo /* signal end of iteration */ 110335fd0808STejun Heo *idx = ULLONG_MAX; 110435fd0808STejun Heo } 110535fd0808STejun Heo 11067bd0b0f0STejun Heo /** 1107f1af9d3aSPhilipp Hachtmann * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1108f1af9d3aSPhilipp Hachtmann * 11097bd0b0f0STejun Heo * @idx: pointer to u64 loop variable 1110ad5ea8cdSAlexander Kuleshov * @nid: node selector, %NUMA_NO_NODE for all nodes 1111fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 1112f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 1113f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 1114dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1115dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1116dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 11177bd0b0f0STejun Heo * 111847cec443SMike Rapoport * Finds the next range from type_a which is not marked as unsuitable 111947cec443SMike Rapoport * in type_b. 112047cec443SMike Rapoport * 1121f1af9d3aSPhilipp Hachtmann * Reverse of __next_mem_range(). 11227bd0b0f0STejun Heo */ 1123e1720feeSMike Rapoport void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1124e1720feeSMike Rapoport enum memblock_flags flags, 1125f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 1126f1af9d3aSPhilipp Hachtmann struct memblock_type *type_b, 11277bd0b0f0STejun Heo phys_addr_t *out_start, 11287bd0b0f0STejun Heo phys_addr_t *out_end, int *out_nid) 11297bd0b0f0STejun Heo { 1130f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 1131f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 1132b1154233SGrygorii Strashko 1133560dca27SGrygorii Strashko if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1134560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 11357bd0b0f0STejun Heo 11367bd0b0f0STejun Heo if (*idx == (u64)ULLONG_MAX) { 1137f1af9d3aSPhilipp Hachtmann idx_a = type_a->cnt - 1; 1138e47608abSzijun_hu if (type_b != NULL) 1139f1af9d3aSPhilipp Hachtmann idx_b = type_b->cnt; 1140e47608abSzijun_hu else 1141e47608abSzijun_hu idx_b = 0; 11427bd0b0f0STejun Heo } 11437bd0b0f0STejun Heo 1144f1af9d3aSPhilipp Hachtmann for (; idx_a >= 0; idx_a--) { 1145f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1146f1af9d3aSPhilipp Hachtmann 11477bd0b0f0STejun Heo phys_addr_t m_start = m->base; 11487bd0b0f0STejun Heo phys_addr_t m_end = m->base + m->size; 1149f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 11507bd0b0f0STejun Heo 11519f3d5eaaSMike Rapoport if (should_skip_region(type_a, m, nid, flags)) 1152bf3d3cc5SArd Biesheuvel continue; 1153bf3d3cc5SArd Biesheuvel 1154f1af9d3aSPhilipp Hachtmann if (!type_b) { 1155f1af9d3aSPhilipp Hachtmann if (out_start) 1156f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1157f1af9d3aSPhilipp Hachtmann if (out_end) 1158f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1159f1af9d3aSPhilipp Hachtmann if (out_nid) 1160f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1161fb399b48Szijun_hu idx_a--; 1162f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1163f1af9d3aSPhilipp Hachtmann return; 1164f1af9d3aSPhilipp Hachtmann } 11657bd0b0f0STejun Heo 1166f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1167f1af9d3aSPhilipp Hachtmann for (; idx_b >= 0; idx_b--) { 1168f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1169f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1170f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1171f1af9d3aSPhilipp Hachtmann 1172f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1173f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1174f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 11751c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1176f1af9d3aSPhilipp Hachtmann /* 1177f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1178f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1179f1af9d3aSPhilipp Hachtmann */ 1180f1af9d3aSPhilipp Hachtmann 11817bd0b0f0STejun Heo if (r_end <= m_start) 11827bd0b0f0STejun Heo break; 11837bd0b0f0STejun Heo /* if the two regions intersect, we're done */ 11847bd0b0f0STejun Heo if (m_end > r_start) { 11857bd0b0f0STejun Heo if (out_start) 11867bd0b0f0STejun Heo *out_start = max(m_start, r_start); 11877bd0b0f0STejun Heo if (out_end) 11887bd0b0f0STejun Heo *out_end = min(m_end, r_end); 11897bd0b0f0STejun Heo if (out_nid) 1190f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 11917bd0b0f0STejun Heo if (m_start >= r_start) 1192f1af9d3aSPhilipp Hachtmann idx_a--; 11937bd0b0f0STejun Heo else 1194f1af9d3aSPhilipp Hachtmann idx_b--; 1195f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 11967bd0b0f0STejun Heo return; 11977bd0b0f0STejun Heo } 11987bd0b0f0STejun Heo } 11997bd0b0f0STejun Heo } 1200f1af9d3aSPhilipp Hachtmann /* signal end of iteration */ 12017bd0b0f0STejun Heo *idx = ULLONG_MAX; 12027bd0b0f0STejun Heo } 12037bd0b0f0STejun Heo 12047c0caeb8STejun Heo /* 120545e79815SChen Chang * Common iterator interface used to define for_each_mem_pfn_range(). 12067c0caeb8STejun Heo */ 12077c0caeb8STejun Heo void __init_memblock __next_mem_pfn_range(int *idx, int nid, 12087c0caeb8STejun Heo unsigned long *out_start_pfn, 12097c0caeb8STejun Heo unsigned long *out_end_pfn, int *out_nid) 12107c0caeb8STejun Heo { 12117c0caeb8STejun Heo struct memblock_type *type = &memblock.memory; 12127c0caeb8STejun Heo struct memblock_region *r; 1213d622abf7SMike Rapoport int r_nid; 12147c0caeb8STejun Heo 12157c0caeb8STejun Heo while (++*idx < type->cnt) { 12167c0caeb8STejun Heo r = &type->regions[*idx]; 1217d622abf7SMike Rapoport r_nid = memblock_get_region_node(r); 12187c0caeb8STejun Heo 12197c0caeb8STejun Heo if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 12207c0caeb8STejun Heo continue; 1221d622abf7SMike Rapoport if (nid == MAX_NUMNODES || nid == r_nid) 12227c0caeb8STejun Heo break; 12237c0caeb8STejun Heo } 12247c0caeb8STejun Heo if (*idx >= type->cnt) { 12257c0caeb8STejun Heo *idx = -1; 12267c0caeb8STejun Heo return; 12277c0caeb8STejun Heo } 12287c0caeb8STejun Heo 12297c0caeb8STejun Heo if (out_start_pfn) 12307c0caeb8STejun Heo *out_start_pfn = PFN_UP(r->base); 12317c0caeb8STejun Heo if (out_end_pfn) 12327c0caeb8STejun Heo *out_end_pfn = PFN_DOWN(r->base + r->size); 12337c0caeb8STejun Heo if (out_nid) 1234d622abf7SMike Rapoport *out_nid = r_nid; 12357c0caeb8STejun Heo } 12367c0caeb8STejun Heo 12377c0caeb8STejun Heo /** 12387c0caeb8STejun Heo * memblock_set_node - set node ID on memblock regions 12397c0caeb8STejun Heo * @base: base of area to set node ID for 12407c0caeb8STejun Heo * @size: size of area to set node ID for 1241e7e8de59STang Chen * @type: memblock type to set node ID for 12427c0caeb8STejun Heo * @nid: node ID to set 12437c0caeb8STejun Heo * 1244e7e8de59STang Chen * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 12457c0caeb8STejun Heo * Regions which cross the area boundaries are split as necessary. 12467c0caeb8STejun Heo * 124747cec443SMike Rapoport * Return: 12487c0caeb8STejun Heo * 0 on success, -errno on failure. 12497c0caeb8STejun Heo */ 12507c0caeb8STejun Heo int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1251e7e8de59STang Chen struct memblock_type *type, int nid) 12527c0caeb8STejun Heo { 1253a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 12546a9ceb31STejun Heo int start_rgn, end_rgn; 12556a9ceb31STejun Heo int i, ret; 12567c0caeb8STejun Heo 12576a9ceb31STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 12586a9ceb31STejun Heo if (ret) 12596a9ceb31STejun Heo return ret; 12607c0caeb8STejun Heo 12616a9ceb31STejun Heo for (i = start_rgn; i < end_rgn; i++) 1262e9d24ad3SWanpeng Li memblock_set_region_node(&type->regions[i], nid); 12637c0caeb8STejun Heo 12647c0caeb8STejun Heo memblock_merge_regions(type); 12653f08a302SMike Rapoport #endif 12667c0caeb8STejun Heo return 0; 12677c0caeb8STejun Heo } 12683f08a302SMike Rapoport 1269837566e7SAlexander Duyck #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1270837566e7SAlexander Duyck /** 1271837566e7SAlexander Duyck * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1272837566e7SAlexander Duyck * 1273837566e7SAlexander Duyck * @idx: pointer to u64 loop variable 1274837566e7SAlexander Duyck * @zone: zone in which all of the memory blocks reside 1275837566e7SAlexander Duyck * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1276837566e7SAlexander Duyck * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1277837566e7SAlexander Duyck * 1278837566e7SAlexander Duyck * This function is meant to be a zone/pfn specific wrapper for the 1279837566e7SAlexander Duyck * for_each_mem_range type iterators. Specifically they are used in the 1280837566e7SAlexander Duyck * deferred memory init routines and as such we were duplicating much of 1281837566e7SAlexander Duyck * this logic throughout the code. So instead of having it in multiple 1282837566e7SAlexander Duyck * locations it seemed like it would make more sense to centralize this to 1283837566e7SAlexander Duyck * one new iterator that does everything they need. 1284837566e7SAlexander Duyck */ 1285837566e7SAlexander Duyck void __init_memblock 1286837566e7SAlexander Duyck __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1287837566e7SAlexander Duyck unsigned long *out_spfn, unsigned long *out_epfn) 1288837566e7SAlexander Duyck { 1289837566e7SAlexander Duyck int zone_nid = zone_to_nid(zone); 1290837566e7SAlexander Duyck phys_addr_t spa, epa; 1291837566e7SAlexander Duyck 1292837566e7SAlexander Duyck __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1293837566e7SAlexander Duyck &memblock.memory, &memblock.reserved, 1294f30b002cSMiaohe Lin &spa, &epa, NULL); 1295837566e7SAlexander Duyck 1296837566e7SAlexander Duyck while (*idx != U64_MAX) { 1297837566e7SAlexander Duyck unsigned long epfn = PFN_DOWN(epa); 1298837566e7SAlexander Duyck unsigned long spfn = PFN_UP(spa); 1299837566e7SAlexander Duyck 1300837566e7SAlexander Duyck /* 1301837566e7SAlexander Duyck * Verify the end is at least past the start of the zone and 1302837566e7SAlexander Duyck * that we have at least one PFN to initialize. 1303837566e7SAlexander Duyck */ 1304837566e7SAlexander Duyck if (zone->zone_start_pfn < epfn && spfn < epfn) { 1305837566e7SAlexander Duyck /* if we went too far just stop searching */ 1306837566e7SAlexander Duyck if (zone_end_pfn(zone) <= spfn) { 1307837566e7SAlexander Duyck *idx = U64_MAX; 1308837566e7SAlexander Duyck break; 1309837566e7SAlexander Duyck } 1310837566e7SAlexander Duyck 1311837566e7SAlexander Duyck if (out_spfn) 1312837566e7SAlexander Duyck *out_spfn = max(zone->zone_start_pfn, spfn); 1313837566e7SAlexander Duyck if (out_epfn) 1314837566e7SAlexander Duyck *out_epfn = min(zone_end_pfn(zone), epfn); 1315837566e7SAlexander Duyck 1316837566e7SAlexander Duyck return; 1317837566e7SAlexander Duyck } 1318837566e7SAlexander Duyck 1319837566e7SAlexander Duyck __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1320837566e7SAlexander Duyck &memblock.memory, &memblock.reserved, 1321f30b002cSMiaohe Lin &spa, &epa, NULL); 1322837566e7SAlexander Duyck } 1323837566e7SAlexander Duyck 1324837566e7SAlexander Duyck /* signal end of iteration */ 1325837566e7SAlexander Duyck if (out_spfn) 1326837566e7SAlexander Duyck *out_spfn = ULONG_MAX; 1327837566e7SAlexander Duyck if (out_epfn) 1328837566e7SAlexander Duyck *out_epfn = 0; 1329837566e7SAlexander Duyck } 1330837566e7SAlexander Duyck 1331837566e7SAlexander Duyck #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 13327c0caeb8STejun Heo 133392d12f95SMike Rapoport /** 133492d12f95SMike Rapoport * memblock_alloc_range_nid - allocate boot memory block 133592d12f95SMike Rapoport * @size: size of memory block to be allocated in bytes 133692d12f95SMike Rapoport * @align: alignment of the region and block's size 133792d12f95SMike Rapoport * @start: the lower bound of the memory region to allocate (phys address) 133892d12f95SMike Rapoport * @end: the upper bound of the memory region to allocate (phys address) 133992d12f95SMike Rapoport * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 13400ac398b1SYunfeng Ye * @exact_nid: control the allocation fall back to other nodes 134192d12f95SMike Rapoport * 134292d12f95SMike Rapoport * The allocation is performed from memory region limited by 134395830666SCao jin * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 134492d12f95SMike Rapoport * 13450ac398b1SYunfeng Ye * If the specified node can not hold the requested memory and @exact_nid 13460ac398b1SYunfeng Ye * is false, the allocation falls back to any node in the system. 134792d12f95SMike Rapoport * 134892d12f95SMike Rapoport * For systems with memory mirroring, the allocation is attempted first 134992d12f95SMike Rapoport * from the regions with mirroring enabled and then retried from any 135092d12f95SMike Rapoport * memory region. 135192d12f95SMike Rapoport * 1352c200d900SPatrick Wang * In addition, function using kmemleak_alloc_phys for allocated boot 1353c200d900SPatrick Wang * memory block, it is never reported as leaks. 135492d12f95SMike Rapoport * 135592d12f95SMike Rapoport * Return: 135692d12f95SMike Rapoport * Physical address of allocated memory block on success, %0 on failure. 135792d12f95SMike Rapoport */ 13588676af1fSAslan Bakirov phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 13592bfc2862SAkinobu Mita phys_addr_t align, phys_addr_t start, 13600ac398b1SYunfeng Ye phys_addr_t end, int nid, 13610ac398b1SYunfeng Ye bool exact_nid) 136295f72d1eSYinghai Lu { 136392d12f95SMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 13646ed311b2SBenjamin Herrenschmidt phys_addr_t found; 136595f72d1eSYinghai Lu 136692d12f95SMike Rapoport if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 136792d12f95SMike Rapoport nid = NUMA_NO_NODE; 136892d12f95SMike Rapoport 13692f770806SMike Rapoport if (!align) { 13702f770806SMike Rapoport /* Can't use WARNs this early in boot on powerpc */ 13712f770806SMike Rapoport dump_stack(); 13722f770806SMike Rapoport align = SMP_CACHE_BYTES; 13732f770806SMike Rapoport } 13742f770806SMike Rapoport 137592d12f95SMike Rapoport again: 1376fc6daaf9STony Luck found = memblock_find_in_range_node(size, align, start, end, nid, 1377fc6daaf9STony Luck flags); 137892d12f95SMike Rapoport if (found && !memblock_reserve(found, size)) 137992d12f95SMike Rapoport goto done; 138092d12f95SMike Rapoport 13810ac398b1SYunfeng Ye if (nid != NUMA_NO_NODE && !exact_nid) { 138292d12f95SMike Rapoport found = memblock_find_in_range_node(size, align, start, 138392d12f95SMike Rapoport end, NUMA_NO_NODE, 138492d12f95SMike Rapoport flags); 138592d12f95SMike Rapoport if (found && !memblock_reserve(found, size)) 138692d12f95SMike Rapoport goto done; 138792d12f95SMike Rapoport } 138892d12f95SMike Rapoport 138992d12f95SMike Rapoport if (flags & MEMBLOCK_MIRROR) { 139092d12f95SMike Rapoport flags &= ~MEMBLOCK_MIRROR; 139192d12f95SMike Rapoport pr_warn("Could not allocate %pap bytes of mirrored memory\n", 139292d12f95SMike Rapoport &size); 139392d12f95SMike Rapoport goto again; 139492d12f95SMike Rapoport } 139592d12f95SMike Rapoport 139692d12f95SMike Rapoport return 0; 139792d12f95SMike Rapoport 139892d12f95SMike Rapoport done: 1399c6975d7cSQian Cai /* 1400c6975d7cSQian Cai * Skip kmemleak for those places like kasan_init() and 1401c6975d7cSQian Cai * early_pgtable_alloc() due to high volume. 1402c6975d7cSQian Cai */ 1403c6975d7cSQian Cai if (end != MEMBLOCK_ALLOC_NOLEAKTRACE) 1404aedf95eaSCatalin Marinas /* 1405c200d900SPatrick Wang * Memblock allocated blocks are never reported as 1406c200d900SPatrick Wang * leaks. This is because many of these blocks are 1407c200d900SPatrick Wang * only referred via the physical address which is 1408c200d900SPatrick Wang * not looked up by kmemleak. 1409aedf95eaSCatalin Marinas */ 1410c200d900SPatrick Wang kmemleak_alloc_phys(found, size, 0); 141192d12f95SMike Rapoport 14126ed311b2SBenjamin Herrenschmidt return found; 1413aedf95eaSCatalin Marinas } 141495f72d1eSYinghai Lu 1415a2974133SMike Rapoport /** 1416a2974133SMike Rapoport * memblock_phys_alloc_range - allocate a memory block inside specified range 1417a2974133SMike Rapoport * @size: size of memory block to be allocated in bytes 1418a2974133SMike Rapoport * @align: alignment of the region and block's size 1419a2974133SMike Rapoport * @start: the lower bound of the memory region to allocate (physical address) 1420a2974133SMike Rapoport * @end: the upper bound of the memory region to allocate (physical address) 1421a2974133SMike Rapoport * 1422a2974133SMike Rapoport * Allocate @size bytes in the between @start and @end. 1423a2974133SMike Rapoport * 1424a2974133SMike Rapoport * Return: physical address of the allocated memory block on success, 1425a2974133SMike Rapoport * %0 on failure. 1426a2974133SMike Rapoport */ 14278a770c2aSMike Rapoport phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 14288a770c2aSMike Rapoport phys_addr_t align, 14298a770c2aSMike Rapoport phys_addr_t start, 14308a770c2aSMike Rapoport phys_addr_t end) 14312bfc2862SAkinobu Mita { 1432b5cf2d6cSFaiyaz Mohammed memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1433b5cf2d6cSFaiyaz Mohammed __func__, (u64)size, (u64)align, &start, &end, 1434b5cf2d6cSFaiyaz Mohammed (void *)_RET_IP_); 14350ac398b1SYunfeng Ye return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 14360ac398b1SYunfeng Ye false); 14377bd0b0f0STejun Heo } 14387bd0b0f0STejun Heo 1439a2974133SMike Rapoport /** 144017cbe038SLevi Yun * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1441a2974133SMike Rapoport * @size: size of memory block to be allocated in bytes 1442a2974133SMike Rapoport * @align: alignment of the region and block's size 1443a2974133SMike Rapoport * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1444a2974133SMike Rapoport * 1445a2974133SMike Rapoport * Allocates memory block from the specified NUMA node. If the node 1446a2974133SMike Rapoport * has no available memory, attempts to allocated from any node in the 1447a2974133SMike Rapoport * system. 1448a2974133SMike Rapoport * 1449a2974133SMike Rapoport * Return: physical address of the allocated memory block on success, 1450a2974133SMike Rapoport * %0 on failure. 1451a2974133SMike Rapoport */ 14529a8dd708SMike Rapoport phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 14539d1e2492SBenjamin Herrenschmidt { 145433755574SMike Rapoport return memblock_alloc_range_nid(size, align, 0, 14550ac398b1SYunfeng Ye MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 145695f72d1eSYinghai Lu } 145795f72d1eSYinghai Lu 145826f09e9bSSantosh Shilimkar /** 1459eb31d559SMike Rapoport * memblock_alloc_internal - allocate boot memory block 146026f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 146126f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 146226f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region to allocate (phys address) 146326f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region to allocate (phys address) 146426f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 14650ac398b1SYunfeng Ye * @exact_nid: control the allocation fall back to other nodes 146626f09e9bSSantosh Shilimkar * 146792d12f95SMike Rapoport * Allocates memory block using memblock_alloc_range_nid() and 146892d12f95SMike Rapoport * converts the returned physical address to virtual. 146992d12f95SMike Rapoport * 147026f09e9bSSantosh Shilimkar * The @min_addr limit is dropped if it can not be satisfied and the allocation 147192d12f95SMike Rapoport * will fall back to memory below @min_addr. Other constraints, such 147292d12f95SMike Rapoport * as node and mirrored memory will be handled again in 147392d12f95SMike Rapoport * memblock_alloc_range_nid(). 147426f09e9bSSantosh Shilimkar * 147547cec443SMike Rapoport * Return: 147626f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 147726f09e9bSSantosh Shilimkar */ 1478eb31d559SMike Rapoport static void * __init memblock_alloc_internal( 147926f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 148026f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 14810ac398b1SYunfeng Ye int nid, bool exact_nid) 148226f09e9bSSantosh Shilimkar { 148326f09e9bSSantosh Shilimkar phys_addr_t alloc; 148426f09e9bSSantosh Shilimkar 148526f09e9bSSantosh Shilimkar /* 148626f09e9bSSantosh Shilimkar * Detect any accidental use of these APIs after slab is ready, as at 148726f09e9bSSantosh Shilimkar * this moment memblock may be deinitialized already and its 1488c6ffc5caSMike Rapoport * internal data may be destroyed (after execution of memblock_free_all) 148926f09e9bSSantosh Shilimkar */ 149026f09e9bSSantosh Shilimkar if (WARN_ON_ONCE(slab_is_available())) 149126f09e9bSSantosh Shilimkar return kzalloc_node(size, GFP_NOWAIT, nid); 149226f09e9bSSantosh Shilimkar 1493f3057ad7SMike Rapoport if (max_addr > memblock.current_limit) 1494f3057ad7SMike Rapoport max_addr = memblock.current_limit; 1495f3057ad7SMike Rapoport 14960ac398b1SYunfeng Ye alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 14970ac398b1SYunfeng Ye exact_nid); 14982f770806SMike Rapoport 149992d12f95SMike Rapoport /* retry allocation without lower limit */ 150092d12f95SMike Rapoport if (!alloc && min_addr) 15010ac398b1SYunfeng Ye alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 15020ac398b1SYunfeng Ye exact_nid); 150326f09e9bSSantosh Shilimkar 150492d12f95SMike Rapoport if (!alloc) 1505a3f5bafcSTony Luck return NULL; 150626f09e9bSSantosh Shilimkar 150792d12f95SMike Rapoport return phys_to_virt(alloc); 150826f09e9bSSantosh Shilimkar } 150926f09e9bSSantosh Shilimkar 151026f09e9bSSantosh Shilimkar /** 15110ac398b1SYunfeng Ye * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 15120ac398b1SYunfeng Ye * without zeroing memory 15130ac398b1SYunfeng Ye * @size: size of memory block to be allocated in bytes 15140ac398b1SYunfeng Ye * @align: alignment of the region and block's size 15150ac398b1SYunfeng Ye * @min_addr: the lower bound of the memory region from where the allocation 15160ac398b1SYunfeng Ye * is preferred (phys address) 15170ac398b1SYunfeng Ye * @max_addr: the upper bound of the memory region from where the allocation 15180ac398b1SYunfeng Ye * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 15190ac398b1SYunfeng Ye * allocate only from memory limited by memblock.current_limit value 15200ac398b1SYunfeng Ye * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 15210ac398b1SYunfeng Ye * 15220ac398b1SYunfeng Ye * Public function, provides additional debug information (including caller 15230ac398b1SYunfeng Ye * info), if enabled. Does not zero allocated memory. 15240ac398b1SYunfeng Ye * 15250ac398b1SYunfeng Ye * Return: 15260ac398b1SYunfeng Ye * Virtual address of allocated memory block on success, NULL on failure. 15270ac398b1SYunfeng Ye */ 15280ac398b1SYunfeng Ye void * __init memblock_alloc_exact_nid_raw( 15290ac398b1SYunfeng Ye phys_addr_t size, phys_addr_t align, 15300ac398b1SYunfeng Ye phys_addr_t min_addr, phys_addr_t max_addr, 15310ac398b1SYunfeng Ye int nid) 15320ac398b1SYunfeng Ye { 15330ac398b1SYunfeng Ye memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 15340ac398b1SYunfeng Ye __func__, (u64)size, (u64)align, nid, &min_addr, 15350ac398b1SYunfeng Ye &max_addr, (void *)_RET_IP_); 15360ac398b1SYunfeng Ye 153708678804SMike Rapoport return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 153808678804SMike Rapoport true); 15390ac398b1SYunfeng Ye } 15400ac398b1SYunfeng Ye 15410ac398b1SYunfeng Ye /** 1542eb31d559SMike Rapoport * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1543ea1f5f37SPavel Tatashin * memory and without panicking 1544ea1f5f37SPavel Tatashin * @size: size of memory block to be allocated in bytes 1545ea1f5f37SPavel Tatashin * @align: alignment of the region and block's size 1546ea1f5f37SPavel Tatashin * @min_addr: the lower bound of the memory region from where the allocation 1547ea1f5f37SPavel Tatashin * is preferred (phys address) 1548ea1f5f37SPavel Tatashin * @max_addr: the upper bound of the memory region from where the allocation 154997ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1550ea1f5f37SPavel Tatashin * allocate only from memory limited by memblock.current_limit value 1551ea1f5f37SPavel Tatashin * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1552ea1f5f37SPavel Tatashin * 1553ea1f5f37SPavel Tatashin * Public function, provides additional debug information (including caller 1554ea1f5f37SPavel Tatashin * info), if enabled. Does not zero allocated memory, does not panic if request 1555ea1f5f37SPavel Tatashin * cannot be satisfied. 1556ea1f5f37SPavel Tatashin * 155747cec443SMike Rapoport * Return: 1558ea1f5f37SPavel Tatashin * Virtual address of allocated memory block on success, NULL on failure. 1559ea1f5f37SPavel Tatashin */ 1560eb31d559SMike Rapoport void * __init memblock_alloc_try_nid_raw( 1561ea1f5f37SPavel Tatashin phys_addr_t size, phys_addr_t align, 1562ea1f5f37SPavel Tatashin phys_addr_t min_addr, phys_addr_t max_addr, 1563ea1f5f37SPavel Tatashin int nid) 1564ea1f5f37SPavel Tatashin { 1565d75f773cSSakari Ailus memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1566a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1567a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1568ea1f5f37SPavel Tatashin 156908678804SMike Rapoport return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 157008678804SMike Rapoport false); 1571ea1f5f37SPavel Tatashin } 1572ea1f5f37SPavel Tatashin 1573ea1f5f37SPavel Tatashin /** 1574c0dbe825SMike Rapoport * memblock_alloc_try_nid - allocate boot memory block 157526f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 157626f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 157726f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region from where the allocation 157826f09e9bSSantosh Shilimkar * is preferred (phys address) 157926f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region from where the allocation 158097ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 158126f09e9bSSantosh Shilimkar * allocate only from memory limited by memblock.current_limit value 158226f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 158326f09e9bSSantosh Shilimkar * 1584c0dbe825SMike Rapoport * Public function, provides additional debug information (including caller 1585c0dbe825SMike Rapoport * info), if enabled. This function zeroes the allocated memory. 158626f09e9bSSantosh Shilimkar * 158747cec443SMike Rapoport * Return: 158826f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 158926f09e9bSSantosh Shilimkar */ 1590eb31d559SMike Rapoport void * __init memblock_alloc_try_nid( 159126f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 159226f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 159326f09e9bSSantosh Shilimkar int nid) 159426f09e9bSSantosh Shilimkar { 159526f09e9bSSantosh Shilimkar void *ptr; 159626f09e9bSSantosh Shilimkar 1597d75f773cSSakari Ailus memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1598a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1599a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1600eb31d559SMike Rapoport ptr = memblock_alloc_internal(size, align, 16010ac398b1SYunfeng Ye min_addr, max_addr, nid, false); 1602c0dbe825SMike Rapoport if (ptr) 1603ea1f5f37SPavel Tatashin memset(ptr, 0, size); 160426f09e9bSSantosh Shilimkar 1605c0dbe825SMike Rapoport return ptr; 160626f09e9bSSantosh Shilimkar } 160726f09e9bSSantosh Shilimkar 160826f09e9bSSantosh Shilimkar /** 1609621d9739SMike Rapoport * memblock_free_late - free pages directly to buddy allocator 161048a833ccSMike Rapoport * @base: phys starting address of the boot memory block 161126f09e9bSSantosh Shilimkar * @size: size of the boot memory block in bytes 161226f09e9bSSantosh Shilimkar * 1613a2974133SMike Rapoport * This is only useful when the memblock allocator has already been torn 161426f09e9bSSantosh Shilimkar * down, but we are still initializing the system. Pages are released directly 1615a2974133SMike Rapoport * to the buddy allocator. 161626f09e9bSSantosh Shilimkar */ 1617621d9739SMike Rapoport void __init memblock_free_late(phys_addr_t base, phys_addr_t size) 161826f09e9bSSantosh Shilimkar { 1619a36aab89SMike Rapoport phys_addr_t cursor, end; 162026f09e9bSSantosh Shilimkar 1621a36aab89SMike Rapoport end = base + size - 1; 1622d75f773cSSakari Ailus memblock_dbg("%s: [%pa-%pa] %pS\n", 1623a36aab89SMike Rapoport __func__, &base, &end, (void *)_RET_IP_); 16249099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 162526f09e9bSSantosh Shilimkar cursor = PFN_UP(base); 162626f09e9bSSantosh Shilimkar end = PFN_DOWN(base + size); 162726f09e9bSSantosh Shilimkar 162826f09e9bSSantosh Shilimkar for (; cursor < end; cursor++) { 16297c2ee349SMike Rapoport memblock_free_pages(pfn_to_page(cursor), cursor, 0); 1630ca79b0c2SArun KS totalram_pages_inc(); 163126f09e9bSSantosh Shilimkar } 163226f09e9bSSantosh Shilimkar } 16339d1e2492SBenjamin Herrenschmidt 16349d1e2492SBenjamin Herrenschmidt /* 16359d1e2492SBenjamin Herrenschmidt * Remaining API functions 16369d1e2492SBenjamin Herrenschmidt */ 16379d1e2492SBenjamin Herrenschmidt 16381f1ffb8aSDavid Gibson phys_addr_t __init_memblock memblock_phys_mem_size(void) 163995f72d1eSYinghai Lu { 16401440c4e2STejun Heo return memblock.memory.total_size; 164195f72d1eSYinghai Lu } 164295f72d1eSYinghai Lu 16438907de5dSSrikar Dronamraju phys_addr_t __init_memblock memblock_reserved_size(void) 16448907de5dSSrikar Dronamraju { 16458907de5dSSrikar Dronamraju return memblock.reserved.total_size; 16468907de5dSSrikar Dronamraju } 16478907de5dSSrikar Dronamraju 16480a93ebefSSam Ravnborg /* lowest address */ 16490a93ebefSSam Ravnborg phys_addr_t __init_memblock memblock_start_of_DRAM(void) 16500a93ebefSSam Ravnborg { 16510a93ebefSSam Ravnborg return memblock.memory.regions[0].base; 16520a93ebefSSam Ravnborg } 16530a93ebefSSam Ravnborg 165410d06439SYinghai Lu phys_addr_t __init_memblock memblock_end_of_DRAM(void) 165595f72d1eSYinghai Lu { 165695f72d1eSYinghai Lu int idx = memblock.memory.cnt - 1; 165795f72d1eSYinghai Lu 1658e3239ff9SBenjamin Herrenschmidt return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 165995f72d1eSYinghai Lu } 166095f72d1eSYinghai Lu 1661a571d4ebSDennis Chen static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 166295f72d1eSYinghai Lu { 16631c4bc43dSStefan Agner phys_addr_t max_addr = PHYS_ADDR_MAX; 1664136199f0SEmil Medve struct memblock_region *r; 166595f72d1eSYinghai Lu 1666a571d4ebSDennis Chen /* 1667a571d4ebSDennis Chen * translate the memory @limit size into the max address within one of 1668a571d4ebSDennis Chen * the memory memblock regions, if the @limit exceeds the total size 16691c4bc43dSStefan Agner * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1670a571d4ebSDennis Chen */ 1671cc6de168SMike Rapoport for_each_mem_region(r) { 1672c0ce8fefSTejun Heo if (limit <= r->size) { 1673c0ce8fefSTejun Heo max_addr = r->base + limit; 167495f72d1eSYinghai Lu break; 167595f72d1eSYinghai Lu } 1676c0ce8fefSTejun Heo limit -= r->size; 167795f72d1eSYinghai Lu } 1678c0ce8fefSTejun Heo 1679a571d4ebSDennis Chen return max_addr; 1680a571d4ebSDennis Chen } 1681a571d4ebSDennis Chen 1682a571d4ebSDennis Chen void __init memblock_enforce_memory_limit(phys_addr_t limit) 1683a571d4ebSDennis Chen { 168449aef717SColin Ian King phys_addr_t max_addr; 1685a571d4ebSDennis Chen 1686a571d4ebSDennis Chen if (!limit) 1687a571d4ebSDennis Chen return; 1688a571d4ebSDennis Chen 1689a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1690a571d4ebSDennis Chen 1691a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 16921c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1693a571d4ebSDennis Chen return; 1694a571d4ebSDennis Chen 1695c0ce8fefSTejun Heo /* truncate both memory and reserved regions */ 1696f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.memory, max_addr, 16971c4bc43dSStefan Agner PHYS_ADDR_MAX); 1698f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.reserved, max_addr, 16991c4bc43dSStefan Agner PHYS_ADDR_MAX); 170095f72d1eSYinghai Lu } 170195f72d1eSYinghai Lu 1702c9ca9b4eSAKASHI Takahiro void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1703c9ca9b4eSAKASHI Takahiro { 1704c9ca9b4eSAKASHI Takahiro int start_rgn, end_rgn; 1705c9ca9b4eSAKASHI Takahiro int i, ret; 1706c9ca9b4eSAKASHI Takahiro 1707c9ca9b4eSAKASHI Takahiro if (!size) 1708c9ca9b4eSAKASHI Takahiro return; 1709c9ca9b4eSAKASHI Takahiro 17105173ed72SPeng Fan if (!memblock_memory->total_size) { 1711e888fa7bSGeert Uytterhoeven pr_warn("%s: No memory registered yet\n", __func__); 1712e888fa7bSGeert Uytterhoeven return; 1713e888fa7bSGeert Uytterhoeven } 1714e888fa7bSGeert Uytterhoeven 1715c9ca9b4eSAKASHI Takahiro ret = memblock_isolate_range(&memblock.memory, base, size, 1716c9ca9b4eSAKASHI Takahiro &start_rgn, &end_rgn); 1717c9ca9b4eSAKASHI Takahiro if (ret) 1718c9ca9b4eSAKASHI Takahiro return; 1719c9ca9b4eSAKASHI Takahiro 1720c9ca9b4eSAKASHI Takahiro /* remove all the MAP regions */ 1721c9ca9b4eSAKASHI Takahiro for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1722c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1723c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1724c9ca9b4eSAKASHI Takahiro 1725c9ca9b4eSAKASHI Takahiro for (i = start_rgn - 1; i >= 0; i--) 1726c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1727c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1728c9ca9b4eSAKASHI Takahiro 1729c9ca9b4eSAKASHI Takahiro /* truncate the reserved regions */ 1730c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 0, base); 1731c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 17321c4bc43dSStefan Agner base + size, PHYS_ADDR_MAX); 1733c9ca9b4eSAKASHI Takahiro } 1734c9ca9b4eSAKASHI Takahiro 1735a571d4ebSDennis Chen void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1736a571d4ebSDennis Chen { 1737a571d4ebSDennis Chen phys_addr_t max_addr; 1738a571d4ebSDennis Chen 1739a571d4ebSDennis Chen if (!limit) 1740a571d4ebSDennis Chen return; 1741a571d4ebSDennis Chen 1742a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1743a571d4ebSDennis Chen 1744a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 17451c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1746a571d4ebSDennis Chen return; 1747a571d4ebSDennis Chen 1748c9ca9b4eSAKASHI Takahiro memblock_cap_memory_range(0, max_addr); 1749a571d4ebSDennis Chen } 1750a571d4ebSDennis Chen 1751cd79481dSYinghai Lu static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 175272d4b0b4SBenjamin Herrenschmidt { 175372d4b0b4SBenjamin Herrenschmidt unsigned int left = 0, right = type->cnt; 175472d4b0b4SBenjamin Herrenschmidt 175572d4b0b4SBenjamin Herrenschmidt do { 175672d4b0b4SBenjamin Herrenschmidt unsigned int mid = (right + left) / 2; 175772d4b0b4SBenjamin Herrenschmidt 175872d4b0b4SBenjamin Herrenschmidt if (addr < type->regions[mid].base) 175972d4b0b4SBenjamin Herrenschmidt right = mid; 176072d4b0b4SBenjamin Herrenschmidt else if (addr >= (type->regions[mid].base + 176172d4b0b4SBenjamin Herrenschmidt type->regions[mid].size)) 176272d4b0b4SBenjamin Herrenschmidt left = mid + 1; 176372d4b0b4SBenjamin Herrenschmidt else 176472d4b0b4SBenjamin Herrenschmidt return mid; 176572d4b0b4SBenjamin Herrenschmidt } while (left < right); 176672d4b0b4SBenjamin Herrenschmidt return -1; 176772d4b0b4SBenjamin Herrenschmidt } 176872d4b0b4SBenjamin Herrenschmidt 1769f5a222dcSYueyi Li bool __init_memblock memblock_is_reserved(phys_addr_t addr) 177095f72d1eSYinghai Lu { 177172d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.reserved, addr) != -1; 177295f72d1eSYinghai Lu } 177372d4b0b4SBenjamin Herrenschmidt 1774b4ad0c7eSYaowei Bai bool __init_memblock memblock_is_memory(phys_addr_t addr) 177572d4b0b4SBenjamin Herrenschmidt { 177672d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.memory, addr) != -1; 177772d4b0b4SBenjamin Herrenschmidt } 177872d4b0b4SBenjamin Herrenschmidt 1779937f0c26SYaowei Bai bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1780bf3d3cc5SArd Biesheuvel { 1781bf3d3cc5SArd Biesheuvel int i = memblock_search(&memblock.memory, addr); 1782bf3d3cc5SArd Biesheuvel 1783bf3d3cc5SArd Biesheuvel if (i == -1) 1784bf3d3cc5SArd Biesheuvel return false; 1785bf3d3cc5SArd Biesheuvel return !memblock_is_nomap(&memblock.memory.regions[i]); 1786bf3d3cc5SArd Biesheuvel } 1787bf3d3cc5SArd Biesheuvel 1788e76b63f8SYinghai Lu int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1789e76b63f8SYinghai Lu unsigned long *start_pfn, unsigned long *end_pfn) 1790e76b63f8SYinghai Lu { 1791e76b63f8SYinghai Lu struct memblock_type *type = &memblock.memory; 179216763230SFabian Frederick int mid = memblock_search(type, PFN_PHYS(pfn)); 1793e76b63f8SYinghai Lu 1794e76b63f8SYinghai Lu if (mid == -1) 1795e76b63f8SYinghai Lu return -1; 1796e76b63f8SYinghai Lu 1797f7e2f7e8SFabian Frederick *start_pfn = PFN_DOWN(type->regions[mid].base); 1798f7e2f7e8SFabian Frederick *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1799e76b63f8SYinghai Lu 1800d622abf7SMike Rapoport return memblock_get_region_node(&type->regions[mid]); 1801e76b63f8SYinghai Lu } 1802e76b63f8SYinghai Lu 1803eab30949SStephen Boyd /** 1804eab30949SStephen Boyd * memblock_is_region_memory - check if a region is a subset of memory 1805eab30949SStephen Boyd * @base: base of region to check 1806eab30949SStephen Boyd * @size: size of region to check 1807eab30949SStephen Boyd * 1808eab30949SStephen Boyd * Check if the region [@base, @base + @size) is a subset of a memory block. 1809eab30949SStephen Boyd * 181047cec443SMike Rapoport * Return: 1811eab30949SStephen Boyd * 0 if false, non-zero if true 1812eab30949SStephen Boyd */ 1813937f0c26SYaowei Bai bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 181472d4b0b4SBenjamin Herrenschmidt { 1815abb65272STomi Valkeinen int idx = memblock_search(&memblock.memory, base); 1816eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 181772d4b0b4SBenjamin Herrenschmidt 181872d4b0b4SBenjamin Herrenschmidt if (idx == -1) 1819937f0c26SYaowei Bai return false; 1820ef415ef4SWei Yang return (memblock.memory.regions[idx].base + 1821eb18f1b5STejun Heo memblock.memory.regions[idx].size) >= end; 182295f72d1eSYinghai Lu } 182395f72d1eSYinghai Lu 1824eab30949SStephen Boyd /** 1825eab30949SStephen Boyd * memblock_is_region_reserved - check if a region intersects reserved memory 1826eab30949SStephen Boyd * @base: base of region to check 1827eab30949SStephen Boyd * @size: size of region to check 1828eab30949SStephen Boyd * 182947cec443SMike Rapoport * Check if the region [@base, @base + @size) intersects a reserved 183047cec443SMike Rapoport * memory block. 1831eab30949SStephen Boyd * 183247cec443SMike Rapoport * Return: 1833c5c5c9d1STang Chen * True if they intersect, false if not. 1834eab30949SStephen Boyd */ 1835c5c5c9d1STang Chen bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 183695f72d1eSYinghai Lu { 1837c5c5c9d1STang Chen return memblock_overlaps_region(&memblock.reserved, base, size); 183895f72d1eSYinghai Lu } 183995f72d1eSYinghai Lu 18406ede1fd3SYinghai Lu void __init_memblock memblock_trim_memory(phys_addr_t align) 18416ede1fd3SYinghai Lu { 18426ede1fd3SYinghai Lu phys_addr_t start, end, orig_start, orig_end; 1843136199f0SEmil Medve struct memblock_region *r; 18446ede1fd3SYinghai Lu 1845cc6de168SMike Rapoport for_each_mem_region(r) { 1846136199f0SEmil Medve orig_start = r->base; 1847136199f0SEmil Medve orig_end = r->base + r->size; 18486ede1fd3SYinghai Lu start = round_up(orig_start, align); 18496ede1fd3SYinghai Lu end = round_down(orig_end, align); 18506ede1fd3SYinghai Lu 18516ede1fd3SYinghai Lu if (start == orig_start && end == orig_end) 18526ede1fd3SYinghai Lu continue; 18536ede1fd3SYinghai Lu 18546ede1fd3SYinghai Lu if (start < end) { 1855136199f0SEmil Medve r->base = start; 1856136199f0SEmil Medve r->size = end - start; 18576ede1fd3SYinghai Lu } else { 1858136199f0SEmil Medve memblock_remove_region(&memblock.memory, 1859136199f0SEmil Medve r - memblock.memory.regions); 1860136199f0SEmil Medve r--; 18616ede1fd3SYinghai Lu } 18626ede1fd3SYinghai Lu } 18636ede1fd3SYinghai Lu } 1864e63075a3SBenjamin Herrenschmidt 18653661ca66SYinghai Lu void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1866e63075a3SBenjamin Herrenschmidt { 1867e63075a3SBenjamin Herrenschmidt memblock.current_limit = limit; 1868e63075a3SBenjamin Herrenschmidt } 1869e63075a3SBenjamin Herrenschmidt 1870fec51014SLaura Abbott phys_addr_t __init_memblock memblock_get_current_limit(void) 1871fec51014SLaura Abbott { 1872fec51014SLaura Abbott return memblock.current_limit; 1873fec51014SLaura Abbott } 1874fec51014SLaura Abbott 18750262d9c8SHeiko Carstens static void __init_memblock memblock_dump(struct memblock_type *type) 18766ed311b2SBenjamin Herrenschmidt { 18775d63f81cSMiles Chen phys_addr_t base, end, size; 1878e1720feeSMike Rapoport enum memblock_flags flags; 18798c9c1701SAlexander Kuleshov int idx; 18808c9c1701SAlexander Kuleshov struct memblock_region *rgn; 18816ed311b2SBenjamin Herrenschmidt 18820262d9c8SHeiko Carstens pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 18836ed311b2SBenjamin Herrenschmidt 188466e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 18857c0caeb8STejun Heo char nid_buf[32] = ""; 18866ed311b2SBenjamin Herrenschmidt 18877c0caeb8STejun Heo base = rgn->base; 18887c0caeb8STejun Heo size = rgn->size; 18895d63f81cSMiles Chen end = base + size - 1; 189066a20757STang Chen flags = rgn->flags; 1891a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 18927c0caeb8STejun Heo if (memblock_get_region_node(rgn) != MAX_NUMNODES) 18937c0caeb8STejun Heo snprintf(nid_buf, sizeof(nid_buf), " on node %d", 18947c0caeb8STejun Heo memblock_get_region_node(rgn)); 18957c0caeb8STejun Heo #endif 1896e1720feeSMike Rapoport pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 18970262d9c8SHeiko Carstens type->name, idx, &base, &end, &size, nid_buf, flags); 18986ed311b2SBenjamin Herrenschmidt } 18996ed311b2SBenjamin Herrenschmidt } 19006ed311b2SBenjamin Herrenschmidt 190187c55870SMike Rapoport static void __init_memblock __memblock_dump_all(void) 19026ed311b2SBenjamin Herrenschmidt { 19036ed311b2SBenjamin Herrenschmidt pr_info("MEMBLOCK configuration:\n"); 19045d63f81cSMiles Chen pr_info(" memory size = %pa reserved size = %pa\n", 19055d63f81cSMiles Chen &memblock.memory.total_size, 19065d63f81cSMiles Chen &memblock.reserved.total_size); 19076ed311b2SBenjamin Herrenschmidt 19080262d9c8SHeiko Carstens memblock_dump(&memblock.memory); 19090262d9c8SHeiko Carstens memblock_dump(&memblock.reserved); 1910409efd4cSHeiko Carstens #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 191177649905SDavid Hildenbrand memblock_dump(&physmem); 1912409efd4cSHeiko Carstens #endif 19136ed311b2SBenjamin Herrenschmidt } 19146ed311b2SBenjamin Herrenschmidt 191587c55870SMike Rapoport void __init_memblock memblock_dump_all(void) 191687c55870SMike Rapoport { 191787c55870SMike Rapoport if (memblock_debug) 191887c55870SMike Rapoport __memblock_dump_all(); 191987c55870SMike Rapoport } 192087c55870SMike Rapoport 19211aadc056STejun Heo void __init memblock_allow_resize(void) 19226ed311b2SBenjamin Herrenschmidt { 1923142b45a7SBenjamin Herrenschmidt memblock_can_resize = 1; 19246ed311b2SBenjamin Herrenschmidt } 19256ed311b2SBenjamin Herrenschmidt 19266ed311b2SBenjamin Herrenschmidt static int __init early_memblock(char *p) 19276ed311b2SBenjamin Herrenschmidt { 19286ed311b2SBenjamin Herrenschmidt if (p && strstr(p, "debug")) 19296ed311b2SBenjamin Herrenschmidt memblock_debug = 1; 19306ed311b2SBenjamin Herrenschmidt return 0; 19316ed311b2SBenjamin Herrenschmidt } 19326ed311b2SBenjamin Herrenschmidt early_param("memblock", early_memblock); 19336ed311b2SBenjamin Herrenschmidt 19344f5b0c17SMike Rapoport static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 19354f5b0c17SMike Rapoport { 19364f5b0c17SMike Rapoport struct page *start_pg, *end_pg; 19374f5b0c17SMike Rapoport phys_addr_t pg, pgend; 19384f5b0c17SMike Rapoport 19394f5b0c17SMike Rapoport /* 19404f5b0c17SMike Rapoport * Convert start_pfn/end_pfn to a struct page pointer. 19414f5b0c17SMike Rapoport */ 19424f5b0c17SMike Rapoport start_pg = pfn_to_page(start_pfn - 1) + 1; 19434f5b0c17SMike Rapoport end_pg = pfn_to_page(end_pfn - 1) + 1; 19444f5b0c17SMike Rapoport 19454f5b0c17SMike Rapoport /* 19464f5b0c17SMike Rapoport * Convert to physical addresses, and round start upwards and end 19474f5b0c17SMike Rapoport * downwards. 19484f5b0c17SMike Rapoport */ 19494f5b0c17SMike Rapoport pg = PAGE_ALIGN(__pa(start_pg)); 19504f5b0c17SMike Rapoport pgend = __pa(end_pg) & PAGE_MASK; 19514f5b0c17SMike Rapoport 19524f5b0c17SMike Rapoport /* 19534f5b0c17SMike Rapoport * If there are free pages between these, free the section of the 19544f5b0c17SMike Rapoport * memmap array. 19554f5b0c17SMike Rapoport */ 19564f5b0c17SMike Rapoport if (pg < pgend) 19573ecc6834SMike Rapoport memblock_phys_free(pg, pgend - pg); 19584f5b0c17SMike Rapoport } 19594f5b0c17SMike Rapoport 19604f5b0c17SMike Rapoport /* 19614f5b0c17SMike Rapoport * The mem_map array can get very big. Free the unused area of the memory map. 19624f5b0c17SMike Rapoport */ 19634f5b0c17SMike Rapoport static void __init free_unused_memmap(void) 19644f5b0c17SMike Rapoport { 19654f5b0c17SMike Rapoport unsigned long start, end, prev_end = 0; 19664f5b0c17SMike Rapoport int i; 19674f5b0c17SMike Rapoport 19684f5b0c17SMike Rapoport if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 19694f5b0c17SMike Rapoport IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 19704f5b0c17SMike Rapoport return; 19714f5b0c17SMike Rapoport 19724f5b0c17SMike Rapoport /* 19734f5b0c17SMike Rapoport * This relies on each bank being in address order. 19744f5b0c17SMike Rapoport * The banks are sorted previously in bootmem_init(). 19754f5b0c17SMike Rapoport */ 19764f5b0c17SMike Rapoport for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 19774f5b0c17SMike Rapoport #ifdef CONFIG_SPARSEMEM 19784f5b0c17SMike Rapoport /* 19794f5b0c17SMike Rapoport * Take care not to free memmap entries that don't exist 19804f5b0c17SMike Rapoport * due to SPARSEMEM sections which aren't present. 19814f5b0c17SMike Rapoport */ 19824f5b0c17SMike Rapoport start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 19834f5b0c17SMike Rapoport #endif 19844f5b0c17SMike Rapoport /* 1985e2a86800SMike Rapoport * Align down here since many operations in VM subsystem 1986e2a86800SMike Rapoport * presume that there are no holes in the memory map inside 1987e2a86800SMike Rapoport * a pageblock 19884f5b0c17SMike Rapoport */ 1989e2a86800SMike Rapoport start = round_down(start, pageblock_nr_pages); 19904f5b0c17SMike Rapoport 19914f5b0c17SMike Rapoport /* 19924f5b0c17SMike Rapoport * If we had a previous bank, and there is a space 19934f5b0c17SMike Rapoport * between the current bank and the previous, free it. 19944f5b0c17SMike Rapoport */ 19954f5b0c17SMike Rapoport if (prev_end && prev_end < start) 19964f5b0c17SMike Rapoport free_memmap(prev_end, start); 19974f5b0c17SMike Rapoport 19984f5b0c17SMike Rapoport /* 1999e2a86800SMike Rapoport * Align up here since many operations in VM subsystem 2000e2a86800SMike Rapoport * presume that there are no holes in the memory map inside 2001e2a86800SMike Rapoport * a pageblock 20024f5b0c17SMike Rapoport */ 2003e2a86800SMike Rapoport prev_end = ALIGN(end, pageblock_nr_pages); 20044f5b0c17SMike Rapoport } 20054f5b0c17SMike Rapoport 20064f5b0c17SMike Rapoport #ifdef CONFIG_SPARSEMEM 2007f921f53eSMike Rapoport if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 2008f921f53eSMike Rapoport prev_end = ALIGN(end, pageblock_nr_pages); 20094f5b0c17SMike Rapoport free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 2010f921f53eSMike Rapoport } 20114f5b0c17SMike Rapoport #endif 20124f5b0c17SMike Rapoport } 20134f5b0c17SMike Rapoport 2014bda49a81SMike Rapoport static void __init __free_pages_memory(unsigned long start, unsigned long end) 2015bda49a81SMike Rapoport { 2016bda49a81SMike Rapoport int order; 2017bda49a81SMike Rapoport 2018bda49a81SMike Rapoport while (start < end) { 2019bda49a81SMike Rapoport order = min(MAX_ORDER - 1UL, __ffs(start)); 2020bda49a81SMike Rapoport 2021bda49a81SMike Rapoport while (start + (1UL << order) > end) 2022bda49a81SMike Rapoport order--; 2023bda49a81SMike Rapoport 2024bda49a81SMike Rapoport memblock_free_pages(pfn_to_page(start), start, order); 2025bda49a81SMike Rapoport 2026bda49a81SMike Rapoport start += (1UL << order); 2027bda49a81SMike Rapoport } 2028bda49a81SMike Rapoport } 2029bda49a81SMike Rapoport 2030bda49a81SMike Rapoport static unsigned long __init __free_memory_core(phys_addr_t start, 2031bda49a81SMike Rapoport phys_addr_t end) 2032bda49a81SMike Rapoport { 2033bda49a81SMike Rapoport unsigned long start_pfn = PFN_UP(start); 2034bda49a81SMike Rapoport unsigned long end_pfn = min_t(unsigned long, 2035bda49a81SMike Rapoport PFN_DOWN(end), max_low_pfn); 2036bda49a81SMike Rapoport 2037bda49a81SMike Rapoport if (start_pfn >= end_pfn) 2038bda49a81SMike Rapoport return 0; 2039bda49a81SMike Rapoport 2040bda49a81SMike Rapoport __free_pages_memory(start_pfn, end_pfn); 2041bda49a81SMike Rapoport 2042bda49a81SMike Rapoport return end_pfn - start_pfn; 2043bda49a81SMike Rapoport } 2044bda49a81SMike Rapoport 20459092d4f7SMike Rapoport static void __init memmap_init_reserved_pages(void) 20469092d4f7SMike Rapoport { 20479092d4f7SMike Rapoport struct memblock_region *region; 20489092d4f7SMike Rapoport phys_addr_t start, end; 20499092d4f7SMike Rapoport u64 i; 20509092d4f7SMike Rapoport 20519092d4f7SMike Rapoport /* initialize struct pages for the reserved regions */ 20529092d4f7SMike Rapoport for_each_reserved_mem_range(i, &start, &end) 20539092d4f7SMike Rapoport reserve_bootmem_region(start, end); 20549092d4f7SMike Rapoport 20559092d4f7SMike Rapoport /* and also treat struct pages for the NOMAP regions as PageReserved */ 20569092d4f7SMike Rapoport for_each_mem_region(region) { 20579092d4f7SMike Rapoport if (memblock_is_nomap(region)) { 20589092d4f7SMike Rapoport start = region->base; 20599092d4f7SMike Rapoport end = start + region->size; 20609092d4f7SMike Rapoport reserve_bootmem_region(start, end); 20619092d4f7SMike Rapoport } 20629092d4f7SMike Rapoport } 20639092d4f7SMike Rapoport } 20649092d4f7SMike Rapoport 2065bda49a81SMike Rapoport static unsigned long __init free_low_memory_core_early(void) 2066bda49a81SMike Rapoport { 2067bda49a81SMike Rapoport unsigned long count = 0; 2068bda49a81SMike Rapoport phys_addr_t start, end; 2069bda49a81SMike Rapoport u64 i; 2070bda49a81SMike Rapoport 2071bda49a81SMike Rapoport memblock_clear_hotplug(0, -1); 2072bda49a81SMike Rapoport 20739092d4f7SMike Rapoport memmap_init_reserved_pages(); 2074bda49a81SMike Rapoport 2075bda49a81SMike Rapoport /* 2076bda49a81SMike Rapoport * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2077bda49a81SMike Rapoport * because in some case like Node0 doesn't have RAM installed 2078bda49a81SMike Rapoport * low ram will be on Node1 2079bda49a81SMike Rapoport */ 2080bda49a81SMike Rapoport for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2081bda49a81SMike Rapoport NULL) 2082bda49a81SMike Rapoport count += __free_memory_core(start, end); 2083bda49a81SMike Rapoport 2084bda49a81SMike Rapoport return count; 2085bda49a81SMike Rapoport } 2086bda49a81SMike Rapoport 2087bda49a81SMike Rapoport static int reset_managed_pages_done __initdata; 2088bda49a81SMike Rapoport 2089bda49a81SMike Rapoport void reset_node_managed_pages(pg_data_t *pgdat) 2090bda49a81SMike Rapoport { 2091bda49a81SMike Rapoport struct zone *z; 2092bda49a81SMike Rapoport 2093bda49a81SMike Rapoport for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 20949705bea5SArun KS atomic_long_set(&z->managed_pages, 0); 2095bda49a81SMike Rapoport } 2096bda49a81SMike Rapoport 2097bda49a81SMike Rapoport void __init reset_all_zones_managed_pages(void) 2098bda49a81SMike Rapoport { 2099bda49a81SMike Rapoport struct pglist_data *pgdat; 2100bda49a81SMike Rapoport 2101bda49a81SMike Rapoport if (reset_managed_pages_done) 2102bda49a81SMike Rapoport return; 2103bda49a81SMike Rapoport 2104bda49a81SMike Rapoport for_each_online_pgdat(pgdat) 2105bda49a81SMike Rapoport reset_node_managed_pages(pgdat); 2106bda49a81SMike Rapoport 2107bda49a81SMike Rapoport reset_managed_pages_done = 1; 2108bda49a81SMike Rapoport } 2109bda49a81SMike Rapoport 2110bda49a81SMike Rapoport /** 2111bda49a81SMike Rapoport * memblock_free_all - release free pages to the buddy allocator 2112bda49a81SMike Rapoport */ 2113097d43d8SDaeseok Youn void __init memblock_free_all(void) 2114bda49a81SMike Rapoport { 2115bda49a81SMike Rapoport unsigned long pages; 2116bda49a81SMike Rapoport 21174f5b0c17SMike Rapoport free_unused_memmap(); 2118bda49a81SMike Rapoport reset_all_zones_managed_pages(); 2119bda49a81SMike Rapoport 2120bda49a81SMike Rapoport pages = free_low_memory_core_early(); 2121ca79b0c2SArun KS totalram_pages_add(pages); 2122bda49a81SMike Rapoport } 2123bda49a81SMike Rapoport 2124350e88baSMike Rapoport #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 21256d03b885SBenjamin Herrenschmidt 21266d03b885SBenjamin Herrenschmidt static int memblock_debug_show(struct seq_file *m, void *private) 21276d03b885SBenjamin Herrenschmidt { 21286d03b885SBenjamin Herrenschmidt struct memblock_type *type = m->private; 21296d03b885SBenjamin Herrenschmidt struct memblock_region *reg; 21306d03b885SBenjamin Herrenschmidt int i; 21315d63f81cSMiles Chen phys_addr_t end; 21326d03b885SBenjamin Herrenschmidt 21336d03b885SBenjamin Herrenschmidt for (i = 0; i < type->cnt; i++) { 21346d03b885SBenjamin Herrenschmidt reg = &type->regions[i]; 21355d63f81cSMiles Chen end = reg->base + reg->size - 1; 21366d03b885SBenjamin Herrenschmidt 21375d63f81cSMiles Chen seq_printf(m, "%4d: ", i); 21385d63f81cSMiles Chen seq_printf(m, "%pa..%pa\n", ®->base, &end); 21396d03b885SBenjamin Herrenschmidt } 21406d03b885SBenjamin Herrenschmidt return 0; 21416d03b885SBenjamin Herrenschmidt } 21425ad35093SAndy Shevchenko DEFINE_SHOW_ATTRIBUTE(memblock_debug); 21436d03b885SBenjamin Herrenschmidt 21446d03b885SBenjamin Herrenschmidt static int __init memblock_init_debugfs(void) 21456d03b885SBenjamin Herrenschmidt { 21466d03b885SBenjamin Herrenschmidt struct dentry *root = debugfs_create_dir("memblock", NULL); 2147d9f7979cSGreg Kroah-Hartman 21480825a6f9SJoe Perches debugfs_create_file("memory", 0444, root, 21490825a6f9SJoe Perches &memblock.memory, &memblock_debug_fops); 21500825a6f9SJoe Perches debugfs_create_file("reserved", 0444, root, 21510825a6f9SJoe Perches &memblock.reserved, &memblock_debug_fops); 215270210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 215377649905SDavid Hildenbrand debugfs_create_file("physmem", 0444, root, &physmem, 215477649905SDavid Hildenbrand &memblock_debug_fops); 215570210ed9SPhilipp Hachtmann #endif 21566d03b885SBenjamin Herrenschmidt 21576d03b885SBenjamin Herrenschmidt return 0; 21586d03b885SBenjamin Herrenschmidt } 21596d03b885SBenjamin Herrenschmidt __initcall(memblock_init_debugfs); 21606d03b885SBenjamin Herrenschmidt 21616d03b885SBenjamin Herrenschmidt #endif /* CONFIG_DEBUG_FS */ 2162