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 32450d0e74SZhou Guanghui #ifndef INIT_MEMBLOCK_MEMORY_REGIONS 33450d0e74SZhou Guanghui #define INIT_MEMBLOCK_MEMORY_REGIONS INIT_MEMBLOCK_REGIONS 34450d0e74SZhou Guanghui #endif 35450d0e74SZhou 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 62450d0e74SZhou Guanghui * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and 63450d0e74SZhou Guanghui * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array 64450d0e74SZhou 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 109450d0e74SZhou 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 */ 118450d0e74SZhou 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)) { 33414d9a675SMa Wupeng pr_warn_ratelimited("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 5032fe03412SPeng Zhang * @start_rgn: start scanning from (@start_rgn - 1) 5042fe03412SPeng Zhang * @end_rgn: end scanning at (@end_rgn - 1) 5052fe03412SPeng Zhang * Scan @type and merge neighboring compatible regions in [@start_rgn - 1, @end_rgn) 506784656f9STejun Heo */ 5072fe03412SPeng Zhang static void __init_memblock memblock_merge_regions(struct memblock_type *type, 5082fe03412SPeng Zhang unsigned long start_rgn, 5092fe03412SPeng Zhang unsigned long end_rgn) 510784656f9STejun Heo { 511784656f9STejun Heo int i = 0; 5122fe03412SPeng Zhang if (start_rgn) 5132fe03412SPeng Zhang i = start_rgn - 1; 5142fe03412SPeng Zhang end_rgn = min(end_rgn, type->cnt - 1); 5152fe03412SPeng Zhang while (i < end_rgn) { 516784656f9STejun Heo struct memblock_region *this = &type->regions[i]; 517784656f9STejun Heo struct memblock_region *next = &type->regions[i + 1]; 518784656f9STejun Heo 5197c0caeb8STejun Heo if (this->base + this->size != next->base || 5207c0caeb8STejun Heo memblock_get_region_node(this) != 52166a20757STang Chen memblock_get_region_node(next) || 52266a20757STang Chen this->flags != next->flags) { 523784656f9STejun Heo BUG_ON(this->base + this->size > next->base); 524784656f9STejun Heo i++; 525784656f9STejun Heo continue; 526784656f9STejun Heo } 527784656f9STejun Heo 528784656f9STejun Heo this->size += next->size; 529c0232ae8SLin Feng /* move forward from next + 1, index of which is i + 2 */ 530c0232ae8SLin Feng memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 531784656f9STejun Heo type->cnt--; 5322fe03412SPeng Zhang end_rgn--; 533784656f9STejun Heo } 534784656f9STejun Heo } 535784656f9STejun Heo 536784656f9STejun Heo /** 537784656f9STejun Heo * memblock_insert_region - insert new memblock region 538784656f9STejun Heo * @type: memblock type to insert into 539784656f9STejun Heo * @idx: index for the insertion point 540784656f9STejun Heo * @base: base address of the new region 541784656f9STejun Heo * @size: size of the new region 542209ff86dSTang Chen * @nid: node id of the new region 54366a20757STang Chen * @flags: flags of the new region 544784656f9STejun Heo * 545784656f9STejun Heo * Insert new memblock region [@base, @base + @size) into @type at @idx. 546412d0008SAlexander Kuleshov * @type must already have extra room to accommodate the new region. 547784656f9STejun Heo */ 548784656f9STejun Heo static void __init_memblock memblock_insert_region(struct memblock_type *type, 549784656f9STejun Heo int idx, phys_addr_t base, 55066a20757STang Chen phys_addr_t size, 551e1720feeSMike Rapoport int nid, 552e1720feeSMike Rapoport enum memblock_flags flags) 553784656f9STejun Heo { 554784656f9STejun Heo struct memblock_region *rgn = &type->regions[idx]; 555784656f9STejun Heo 556784656f9STejun Heo BUG_ON(type->cnt >= type->max); 557784656f9STejun Heo memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 558784656f9STejun Heo rgn->base = base; 559784656f9STejun Heo rgn->size = size; 56066a20757STang Chen rgn->flags = flags; 5617c0caeb8STejun Heo memblock_set_region_node(rgn, nid); 562784656f9STejun Heo type->cnt++; 5631440c4e2STejun Heo type->total_size += size; 564784656f9STejun Heo } 565784656f9STejun Heo 566784656f9STejun Heo /** 567f1af9d3aSPhilipp Hachtmann * memblock_add_range - add new memblock region 568784656f9STejun Heo * @type: memblock type to add new region into 569784656f9STejun Heo * @base: base address of the new region 570784656f9STejun Heo * @size: size of the new region 5717fb0bc3fSTejun Heo * @nid: nid of the new region 57266a20757STang Chen * @flags: flags of the new region 573784656f9STejun Heo * 574784656f9STejun Heo * Add new memblock region [@base, @base + @size) into @type. The new region 575784656f9STejun Heo * is allowed to overlap with existing ones - overlaps don't affect already 576784656f9STejun Heo * existing regions. @type is guaranteed to be minimal (all neighbouring 577784656f9STejun Heo * compatible regions are merged) after the addition. 578784656f9STejun Heo * 57947cec443SMike Rapoport * Return: 580784656f9STejun Heo * 0 on success, -errno on failure. 581784656f9STejun Heo */ 58202634a44SAnshuman Khandual static int __init_memblock memblock_add_range(struct memblock_type *type, 58366a20757STang Chen phys_addr_t base, phys_addr_t size, 584e1720feeSMike Rapoport int nid, enum memblock_flags flags) 58595f72d1eSYinghai Lu { 586784656f9STejun Heo bool insert = false; 587eb18f1b5STejun Heo phys_addr_t obase = base; 588eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 5892fe03412SPeng Zhang int idx, nr_new, start_rgn = -1, end_rgn; 5908c9c1701SAlexander Kuleshov struct memblock_region *rgn; 59195f72d1eSYinghai Lu 592b3dc627cSTejun Heo if (!size) 593b3dc627cSTejun Heo return 0; 594b3dc627cSTejun Heo 595784656f9STejun Heo /* special case for empty array */ 596784656f9STejun Heo if (type->regions[0].size == 0) { 5971440c4e2STejun Heo WARN_ON(type->cnt != 1 || type->total_size); 598784656f9STejun Heo type->regions[0].base = base; 599784656f9STejun Heo type->regions[0].size = size; 60066a20757STang Chen type->regions[0].flags = flags; 6017fb0bc3fSTejun Heo memblock_set_region_node(&type->regions[0], nid); 6021440c4e2STejun Heo type->total_size = size; 603784656f9STejun Heo return 0; 604784656f9STejun Heo } 60528e1a8f4SJinyu Tang 60628e1a8f4SJinyu Tang /* 60728e1a8f4SJinyu Tang * The worst case is when new range overlaps all existing regions, 60828e1a8f4SJinyu Tang * then we'll need type->cnt + 1 empty regions in @type. So if 609ad500fb2SPeng Zhang * type->cnt * 2 + 1 is less than or equal to type->max, we know 61028e1a8f4SJinyu Tang * that there is enough empty regions in @type, and we can insert 61128e1a8f4SJinyu Tang * regions directly. 61228e1a8f4SJinyu Tang */ 613ad500fb2SPeng Zhang if (type->cnt * 2 + 1 <= type->max) 61428e1a8f4SJinyu Tang insert = true; 61528e1a8f4SJinyu Tang 616784656f9STejun Heo repeat: 617784656f9STejun Heo /* 618784656f9STejun Heo * The following is executed twice. Once with %false @insert and 619784656f9STejun Heo * then with %true. The first counts the number of regions needed 620412d0008SAlexander Kuleshov * to accommodate the new area. The second actually inserts them. 621784656f9STejun Heo */ 622784656f9STejun Heo base = obase; 623784656f9STejun Heo nr_new = 0; 624784656f9STejun Heo 62566e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 626784656f9STejun Heo phys_addr_t rbase = rgn->base; 627784656f9STejun Heo phys_addr_t rend = rbase + rgn->size; 6288f7a6605SBenjamin Herrenschmidt 629784656f9STejun Heo if (rbase >= end) 6308f7a6605SBenjamin Herrenschmidt break; 631784656f9STejun Heo if (rend <= base) 632784656f9STejun Heo continue; 633784656f9STejun Heo /* 634784656f9STejun Heo * @rgn overlaps. If it separates the lower part of new 635784656f9STejun Heo * area, insert that portion. 6368f7a6605SBenjamin Herrenschmidt */ 637784656f9STejun Heo if (rbase > base) { 638a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 639c0a29498SWei Yang WARN_ON(nid != memblock_get_region_node(rgn)); 640c0a29498SWei Yang #endif 6414fcab5f4SWei Yang WARN_ON(flags != rgn->flags); 642784656f9STejun Heo nr_new++; 6432fe03412SPeng Zhang if (insert) { 6442fe03412SPeng Zhang if (start_rgn == -1) 6452fe03412SPeng Zhang start_rgn = idx; 6462fe03412SPeng Zhang end_rgn = idx + 1; 6478c9c1701SAlexander Kuleshov memblock_insert_region(type, idx++, base, 64866a20757STang Chen rbase - base, nid, 64966a20757STang Chen flags); 650784656f9STejun Heo } 6512fe03412SPeng Zhang } 652784656f9STejun Heo /* area below @rend is dealt with, forget about it */ 653784656f9STejun Heo base = min(rend, end); 6548f7a6605SBenjamin Herrenschmidt } 6558f7a6605SBenjamin Herrenschmidt 656784656f9STejun Heo /* insert the remaining portion */ 657784656f9STejun Heo if (base < end) { 658784656f9STejun Heo nr_new++; 6592fe03412SPeng Zhang if (insert) { 6602fe03412SPeng Zhang if (start_rgn == -1) 6612fe03412SPeng Zhang start_rgn = idx; 6622fe03412SPeng Zhang end_rgn = idx + 1; 6638c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, base, end - base, 66466a20757STang Chen nid, flags); 6658f7a6605SBenjamin Herrenschmidt } 6662fe03412SPeng Zhang } 6678f7a6605SBenjamin Herrenschmidt 668ef3cc4dbSnimisolo if (!nr_new) 669ef3cc4dbSnimisolo return 0; 670ef3cc4dbSnimisolo 671784656f9STejun Heo /* 672784656f9STejun Heo * If this was the first round, resize array and repeat for actual 673784656f9STejun Heo * insertions; otherwise, merge and return. 6748f7a6605SBenjamin Herrenschmidt */ 675784656f9STejun Heo if (!insert) { 676784656f9STejun Heo while (type->cnt + nr_new > type->max) 67748c3b583SGreg Pearson if (memblock_double_array(type, obase, size) < 0) 678784656f9STejun Heo return -ENOMEM; 679784656f9STejun Heo insert = true; 680784656f9STejun Heo goto repeat; 68195f72d1eSYinghai Lu } else { 6822fe03412SPeng Zhang memblock_merge_regions(type, start_rgn, end_rgn); 68395f72d1eSYinghai Lu return 0; 68495f72d1eSYinghai Lu } 685784656f9STejun Heo } 68695f72d1eSYinghai Lu 68748a833ccSMike Rapoport /** 68848a833ccSMike Rapoport * memblock_add_node - add new memblock region within a NUMA node 68948a833ccSMike Rapoport * @base: base address of the new region 69048a833ccSMike Rapoport * @size: size of the new region 69148a833ccSMike Rapoport * @nid: nid of the new region 692952eea9bSDavid Hildenbrand * @flags: flags of the new region 69348a833ccSMike Rapoport * 69448a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 69548a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 69648a833ccSMike Rapoport * 69748a833ccSMike Rapoport * Return: 69848a833ccSMike Rapoport * 0 on success, -errno on failure. 69948a833ccSMike Rapoport */ 7007fb0bc3fSTejun Heo int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 701952eea9bSDavid Hildenbrand int nid, enum memblock_flags flags) 7027fb0bc3fSTejun Heo { 70300974b9aSGeert Uytterhoeven phys_addr_t end = base + size - 1; 70400974b9aSGeert Uytterhoeven 705952eea9bSDavid Hildenbrand memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__, 706952eea9bSDavid Hildenbrand &base, &end, nid, flags, (void *)_RET_IP_); 70700974b9aSGeert Uytterhoeven 708952eea9bSDavid Hildenbrand return memblock_add_range(&memblock.memory, base, size, nid, flags); 7097fb0bc3fSTejun Heo } 7107fb0bc3fSTejun Heo 71148a833ccSMike Rapoport /** 71248a833ccSMike Rapoport * memblock_add - add new memblock region 71348a833ccSMike Rapoport * @base: base address of the new region 71448a833ccSMike Rapoport * @size: size of the new region 71548a833ccSMike Rapoport * 71648a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 71748a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 71848a833ccSMike Rapoport * 71948a833ccSMike Rapoport * Return: 72048a833ccSMike Rapoport * 0 on success, -errno on failure. 72148a833ccSMike Rapoport */ 722f705ac4bSAlexander Kuleshov int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 7236a4055bcSAlexander Kuleshov { 7245d63f81cSMiles Chen phys_addr_t end = base + size - 1; 7255d63f81cSMiles Chen 726a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 7275d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 7286a4055bcSAlexander Kuleshov 729f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 73095f72d1eSYinghai Lu } 73195f72d1eSYinghai Lu 7326a9ceb31STejun Heo /** 7336a9ceb31STejun Heo * memblock_isolate_range - isolate given range into disjoint memblocks 7346a9ceb31STejun Heo * @type: memblock type to isolate range for 7356a9ceb31STejun Heo * @base: base of range to isolate 7366a9ceb31STejun Heo * @size: size of range to isolate 7376a9ceb31STejun Heo * @start_rgn: out parameter for the start of isolated region 7386a9ceb31STejun Heo * @end_rgn: out parameter for the end of isolated region 7396a9ceb31STejun Heo * 7406a9ceb31STejun Heo * Walk @type and ensure that regions don't cross the boundaries defined by 7416a9ceb31STejun Heo * [@base, @base + @size). Crossing regions are split at the boundaries, 7426a9ceb31STejun Heo * which may create at most two more regions. The index of the first 7436a9ceb31STejun Heo * region inside the range is returned in *@start_rgn and end in *@end_rgn. 7446a9ceb31STejun Heo * 74547cec443SMike Rapoport * Return: 7466a9ceb31STejun Heo * 0 on success, -errno on failure. 7476a9ceb31STejun Heo */ 7486a9ceb31STejun Heo static int __init_memblock memblock_isolate_range(struct memblock_type *type, 7496a9ceb31STejun Heo phys_addr_t base, phys_addr_t size, 7506a9ceb31STejun Heo int *start_rgn, int *end_rgn) 7516a9ceb31STejun Heo { 752eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 7538c9c1701SAlexander Kuleshov int idx; 7548c9c1701SAlexander Kuleshov struct memblock_region *rgn; 7556a9ceb31STejun Heo 7566a9ceb31STejun Heo *start_rgn = *end_rgn = 0; 7576a9ceb31STejun Heo 758b3dc627cSTejun Heo if (!size) 759b3dc627cSTejun Heo return 0; 760b3dc627cSTejun Heo 7616a9ceb31STejun Heo /* we'll create at most two more regions */ 7626a9ceb31STejun Heo while (type->cnt + 2 > type->max) 76348c3b583SGreg Pearson if (memblock_double_array(type, base, size) < 0) 7646a9ceb31STejun Heo return -ENOMEM; 7656a9ceb31STejun Heo 76666e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 7676a9ceb31STejun Heo phys_addr_t rbase = rgn->base; 7686a9ceb31STejun Heo phys_addr_t rend = rbase + rgn->size; 7696a9ceb31STejun Heo 7706a9ceb31STejun Heo if (rbase >= end) 7716a9ceb31STejun Heo break; 7726a9ceb31STejun Heo if (rend <= base) 7736a9ceb31STejun Heo continue; 7746a9ceb31STejun Heo 7756a9ceb31STejun Heo if (rbase < base) { 7766a9ceb31STejun Heo /* 7776a9ceb31STejun Heo * @rgn intersects from below. Split and continue 7786a9ceb31STejun Heo * to process the next region - the new top half. 7796a9ceb31STejun Heo */ 7806a9ceb31STejun Heo rgn->base = base; 7811440c4e2STejun Heo rgn->size -= base - rbase; 7821440c4e2STejun Heo type->total_size -= base - rbase; 7838c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, rbase, base - rbase, 78466a20757STang Chen memblock_get_region_node(rgn), 78566a20757STang Chen rgn->flags); 7866a9ceb31STejun Heo } else if (rend > end) { 7876a9ceb31STejun Heo /* 7886a9ceb31STejun Heo * @rgn intersects from above. Split and redo the 7896a9ceb31STejun Heo * current region - the new bottom half. 7906a9ceb31STejun Heo */ 7916a9ceb31STejun Heo rgn->base = end; 7921440c4e2STejun Heo rgn->size -= end - rbase; 7931440c4e2STejun Heo type->total_size -= end - rbase; 7948c9c1701SAlexander Kuleshov memblock_insert_region(type, idx--, rbase, end - rbase, 79566a20757STang Chen memblock_get_region_node(rgn), 79666a20757STang Chen rgn->flags); 7976a9ceb31STejun Heo } else { 7986a9ceb31STejun Heo /* @rgn is fully contained, record it */ 7996a9ceb31STejun Heo if (!*end_rgn) 8008c9c1701SAlexander Kuleshov *start_rgn = idx; 8018c9c1701SAlexander Kuleshov *end_rgn = idx + 1; 8026a9ceb31STejun Heo } 8036a9ceb31STejun Heo } 8046a9ceb31STejun Heo 8056a9ceb31STejun Heo return 0; 8066a9ceb31STejun Heo } 8076a9ceb31STejun Heo 80835bd16a2SAlexander Kuleshov static int __init_memblock memblock_remove_range(struct memblock_type *type, 8098f7a6605SBenjamin Herrenschmidt phys_addr_t base, phys_addr_t size) 81095f72d1eSYinghai Lu { 81171936180STejun Heo int start_rgn, end_rgn; 81271936180STejun Heo int i, ret; 81395f72d1eSYinghai Lu 81471936180STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 81571936180STejun Heo if (ret) 81671936180STejun Heo return ret; 81795f72d1eSYinghai Lu 81871936180STejun Heo for (i = end_rgn - 1; i >= start_rgn; i--) 81971936180STejun Heo memblock_remove_region(type, i); 82095f72d1eSYinghai Lu return 0; 82195f72d1eSYinghai Lu } 82295f72d1eSYinghai Lu 823581adcbeSTejun Heo int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 82495f72d1eSYinghai Lu { 82525cf23d7SMinchan Kim phys_addr_t end = base + size - 1; 82625cf23d7SMinchan Kim 827a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 82825cf23d7SMinchan Kim &base, &end, (void *)_RET_IP_); 82925cf23d7SMinchan Kim 830f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.memory, base, size); 83195f72d1eSYinghai Lu } 83295f72d1eSYinghai Lu 8334d72868cSMike Rapoport /** 8344421cca0SMike Rapoport * memblock_free - free boot memory allocation 83577e02cf5SLinus Torvalds * @ptr: starting address of the boot memory allocation 83677e02cf5SLinus Torvalds * @size: size of the boot memory block in bytes 83777e02cf5SLinus Torvalds * 83877e02cf5SLinus Torvalds * Free boot memory block previously allocated by memblock_alloc_xx() API. 83977e02cf5SLinus Torvalds * The freeing memory will not be released to the buddy allocator. 84077e02cf5SLinus Torvalds */ 8414421cca0SMike Rapoport void __init_memblock memblock_free(void *ptr, size_t size) 84277e02cf5SLinus Torvalds { 84377e02cf5SLinus Torvalds if (ptr) 8443ecc6834SMike Rapoport memblock_phys_free(__pa(ptr), size); 84577e02cf5SLinus Torvalds } 84677e02cf5SLinus Torvalds 84777e02cf5SLinus Torvalds /** 8483ecc6834SMike Rapoport * memblock_phys_free - free boot memory block 8494d72868cSMike Rapoport * @base: phys starting address of the boot memory block 8504d72868cSMike Rapoport * @size: size of the boot memory block in bytes 8514d72868cSMike Rapoport * 852fa81ab49SMiaoqian Lin * Free boot memory block previously allocated by memblock_phys_alloc_xx() API. 8534d72868cSMike Rapoport * The freeing memory will not be released to the buddy allocator. 8544d72868cSMike Rapoport */ 8553ecc6834SMike Rapoport int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size) 85695f72d1eSYinghai Lu { 8575d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8585d63f81cSMiles Chen 859a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 8605d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 86124aa0788STejun Heo 8629099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 863f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.reserved, base, size); 86495f72d1eSYinghai Lu } 86595f72d1eSYinghai Lu 866f705ac4bSAlexander Kuleshov int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 86795f72d1eSYinghai Lu { 8685d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8695d63f81cSMiles Chen 870a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 8715d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 87295f72d1eSYinghai Lu 873f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 87495f72d1eSYinghai Lu } 87595f72d1eSYinghai Lu 87602634a44SAnshuman Khandual #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 87702634a44SAnshuman Khandual int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size) 87802634a44SAnshuman Khandual { 87902634a44SAnshuman Khandual phys_addr_t end = base + size - 1; 88002634a44SAnshuman Khandual 88102634a44SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 88202634a44SAnshuman Khandual &base, &end, (void *)_RET_IP_); 88302634a44SAnshuman Khandual 88477649905SDavid Hildenbrand return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0); 88502634a44SAnshuman Khandual } 88602634a44SAnshuman Khandual #endif 88702634a44SAnshuman Khandual 88835fd0808STejun Heo /** 88947cec443SMike Rapoport * memblock_setclr_flag - set or clear flag for a memory region 89047cec443SMike Rapoport * @base: base address of the region 89147cec443SMike Rapoport * @size: size of the region 89247cec443SMike Rapoport * @set: set or clear the flag 8938958b249SHaitao Shi * @flag: the flag to update 89466b16edfSTang Chen * 8954308ce17STony Luck * This function isolates region [@base, @base + @size), and sets/clears flag 89666b16edfSTang Chen * 89747cec443SMike Rapoport * Return: 0 on success, -errno on failure. 89866b16edfSTang Chen */ 8994308ce17STony Luck static int __init_memblock memblock_setclr_flag(phys_addr_t base, 9004308ce17STony Luck phys_addr_t size, int set, int flag) 90166b16edfSTang Chen { 90266b16edfSTang Chen struct memblock_type *type = &memblock.memory; 90366b16edfSTang Chen int i, ret, start_rgn, end_rgn; 90466b16edfSTang Chen 90566b16edfSTang Chen ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 90666b16edfSTang Chen if (ret) 90766b16edfSTang Chen return ret; 90866b16edfSTang Chen 909fe145124SMike Rapoport for (i = start_rgn; i < end_rgn; i++) { 910fe145124SMike Rapoport struct memblock_region *r = &type->regions[i]; 911fe145124SMike Rapoport 9124308ce17STony Luck if (set) 913fe145124SMike Rapoport r->flags |= flag; 9144308ce17STony Luck else 915fe145124SMike Rapoport r->flags &= ~flag; 916fe145124SMike Rapoport } 91766b16edfSTang Chen 9182fe03412SPeng Zhang memblock_merge_regions(type, start_rgn, end_rgn); 91966b16edfSTang Chen return 0; 92066b16edfSTang Chen } 92166b16edfSTang Chen 92266b16edfSTang Chen /** 9234308ce17STony Luck * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 9244308ce17STony Luck * @base: the base phys addr of the region 9254308ce17STony Luck * @size: the size of the region 9264308ce17STony Luck * 92747cec443SMike Rapoport * Return: 0 on success, -errno on failure. 9284308ce17STony Luck */ 9294308ce17STony Luck int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 9304308ce17STony Luck { 9314308ce17STony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 9324308ce17STony Luck } 9334308ce17STony Luck 9344308ce17STony Luck /** 93566b16edfSTang Chen * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 93666b16edfSTang Chen * @base: the base phys addr of the region 93766b16edfSTang Chen * @size: the size of the region 93866b16edfSTang Chen * 93947cec443SMike Rapoport * Return: 0 on success, -errno on failure. 94066b16edfSTang Chen */ 94166b16edfSTang Chen int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 94266b16edfSTang Chen { 9434308ce17STony Luck return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 94466b16edfSTang Chen } 94566b16edfSTang Chen 94666b16edfSTang Chen /** 947a3f5bafcSTony Luck * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 948a3f5bafcSTony Luck * @base: the base phys addr of the region 949a3f5bafcSTony Luck * @size: the size of the region 950a3f5bafcSTony Luck * 95147cec443SMike Rapoport * Return: 0 on success, -errno on failure. 952a3f5bafcSTony Luck */ 953a3f5bafcSTony Luck int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 954a3f5bafcSTony Luck { 955902c2d91SMa Wupeng if (!mirrored_kernelcore) 956902c2d91SMa Wupeng return 0; 957902c2d91SMa Wupeng 958a3f5bafcSTony Luck system_has_some_mirror = true; 959a3f5bafcSTony Luck 960a3f5bafcSTony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 961a3f5bafcSTony Luck } 962a3f5bafcSTony Luck 963bf3d3cc5SArd Biesheuvel /** 964bf3d3cc5SArd Biesheuvel * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 965bf3d3cc5SArd Biesheuvel * @base: the base phys addr of the region 966bf3d3cc5SArd Biesheuvel * @size: the size of the region 967bf3d3cc5SArd Biesheuvel * 9689092d4f7SMike Rapoport * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the 9699092d4f7SMike Rapoport * direct mapping of the physical memory. These regions will still be 9709092d4f7SMike Rapoport * covered by the memory map. The struct page representing NOMAP memory 9719092d4f7SMike Rapoport * frames in the memory map will be PageReserved() 9729092d4f7SMike Rapoport * 973658aafc8SMike Rapoport * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from 974658aafc8SMike Rapoport * memblock, the caller must inform kmemleak to ignore that memory 975658aafc8SMike Rapoport * 97647cec443SMike Rapoport * Return: 0 on success, -errno on failure. 977bf3d3cc5SArd Biesheuvel */ 978bf3d3cc5SArd Biesheuvel int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 979bf3d3cc5SArd Biesheuvel { 9806c9a5455SMike Rapoport return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 981bf3d3cc5SArd Biesheuvel } 982a3f5bafcSTony Luck 983a3f5bafcSTony Luck /** 9844c546b8aSAKASHI Takahiro * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 9854c546b8aSAKASHI Takahiro * @base: the base phys addr of the region 9864c546b8aSAKASHI Takahiro * @size: the size of the region 9874c546b8aSAKASHI Takahiro * 98847cec443SMike Rapoport * Return: 0 on success, -errno on failure. 9894c546b8aSAKASHI Takahiro */ 9904c546b8aSAKASHI Takahiro int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 9914c546b8aSAKASHI Takahiro { 9924c546b8aSAKASHI Takahiro return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 9934c546b8aSAKASHI Takahiro } 9944c546b8aSAKASHI Takahiro 9959f3d5eaaSMike Rapoport static bool should_skip_region(struct memblock_type *type, 9969f3d5eaaSMike Rapoport struct memblock_region *m, 9979f3d5eaaSMike Rapoport int nid, int flags) 998c9a688a3SMike Rapoport { 999c9a688a3SMike Rapoport int m_nid = memblock_get_region_node(m); 1000c9a688a3SMike Rapoport 10019f3d5eaaSMike Rapoport /* we never skip regions when iterating memblock.reserved or physmem */ 10029f3d5eaaSMike Rapoport if (type != memblock_memory) 10039f3d5eaaSMike Rapoport return false; 10049f3d5eaaSMike Rapoport 1005c9a688a3SMike Rapoport /* only memory regions are associated with nodes, check it */ 1006c9a688a3SMike Rapoport if (nid != NUMA_NO_NODE && nid != m_nid) 1007c9a688a3SMike Rapoport return true; 1008c9a688a3SMike Rapoport 1009c9a688a3SMike Rapoport /* skip hotpluggable memory regions if needed */ 101079e482e9SMike Rapoport if (movable_node_is_enabled() && memblock_is_hotpluggable(m) && 101179e482e9SMike Rapoport !(flags & MEMBLOCK_HOTPLUG)) 1012c9a688a3SMike Rapoport return true; 1013c9a688a3SMike Rapoport 1014c9a688a3SMike Rapoport /* if we want mirror memory skip non-mirror memory regions */ 1015c9a688a3SMike Rapoport if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1016c9a688a3SMike Rapoport return true; 1017c9a688a3SMike Rapoport 1018c9a688a3SMike Rapoport /* skip nomap memory unless we were asked for it explicitly */ 1019c9a688a3SMike Rapoport if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1020c9a688a3SMike Rapoport return true; 1021c9a688a3SMike Rapoport 1022f7892d8eSDavid Hildenbrand /* skip driver-managed memory unless we were asked for it explicitly */ 1023f7892d8eSDavid Hildenbrand if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m)) 1024f7892d8eSDavid Hildenbrand return true; 1025f7892d8eSDavid Hildenbrand 1026c9a688a3SMike Rapoport return false; 1027c9a688a3SMike Rapoport } 1028c9a688a3SMike Rapoport 10298e7a7f86SRobin Holt /** 1030a2974133SMike Rapoport * __next_mem_range - next function for for_each_free_mem_range() etc. 103135fd0808STejun Heo * @idx: pointer to u64 loop variable 1032b1154233SGrygorii Strashko * @nid: node selector, %NUMA_NO_NODE for all nodes 1033fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 1034f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 1035f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 1036dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1037dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1038dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 103935fd0808STejun Heo * 1040f1af9d3aSPhilipp Hachtmann * Find the first area from *@idx which matches @nid, fill the out 104135fd0808STejun Heo * parameters, and update *@idx for the next iteration. The lower 32bit of 1042f1af9d3aSPhilipp Hachtmann * *@idx contains index into type_a and the upper 32bit indexes the 1043f1af9d3aSPhilipp Hachtmann * areas before each region in type_b. For example, if type_b regions 104435fd0808STejun Heo * look like the following, 104535fd0808STejun Heo * 104635fd0808STejun Heo * 0:[0-16), 1:[32-48), 2:[128-130) 104735fd0808STejun Heo * 104835fd0808STejun Heo * The upper 32bit indexes the following regions. 104935fd0808STejun Heo * 105035fd0808STejun Heo * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 105135fd0808STejun Heo * 105235fd0808STejun Heo * As both region arrays are sorted, the function advances the two indices 105335fd0808STejun Heo * in lockstep and returns each intersection. 105435fd0808STejun Heo */ 105577649905SDavid Hildenbrand void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags, 1056f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 105777649905SDavid Hildenbrand struct memblock_type *type_b, phys_addr_t *out_start, 105835fd0808STejun Heo phys_addr_t *out_end, int *out_nid) 105935fd0808STejun Heo { 1060f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 1061f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 1062b1154233SGrygorii Strashko 1063f1af9d3aSPhilipp Hachtmann if (WARN_ONCE(nid == MAX_NUMNODES, 1064f1af9d3aSPhilipp Hachtmann "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1065560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 106635fd0808STejun Heo 1067f1af9d3aSPhilipp Hachtmann for (; idx_a < type_a->cnt; idx_a++) { 1068f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1069f1af9d3aSPhilipp Hachtmann 107035fd0808STejun Heo phys_addr_t m_start = m->base; 107135fd0808STejun Heo phys_addr_t m_end = m->base + m->size; 1072f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 107335fd0808STejun Heo 10749f3d5eaaSMike Rapoport if (should_skip_region(type_a, m, nid, flags)) 1075bf3d3cc5SArd Biesheuvel continue; 1076bf3d3cc5SArd Biesheuvel 1077f1af9d3aSPhilipp Hachtmann if (!type_b) { 1078f1af9d3aSPhilipp Hachtmann if (out_start) 1079f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1080f1af9d3aSPhilipp Hachtmann if (out_end) 1081f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1082f1af9d3aSPhilipp Hachtmann if (out_nid) 1083f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1084f1af9d3aSPhilipp Hachtmann idx_a++; 1085f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1086f1af9d3aSPhilipp Hachtmann return; 1087f1af9d3aSPhilipp Hachtmann } 108835fd0808STejun Heo 1089f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1090f1af9d3aSPhilipp Hachtmann for (; idx_b < type_b->cnt + 1; idx_b++) { 1091f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1092f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1093f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1094f1af9d3aSPhilipp Hachtmann 1095f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1096f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1097f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 10981c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1099f1af9d3aSPhilipp Hachtmann 1100f1af9d3aSPhilipp Hachtmann /* 1101f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1102f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1103f1af9d3aSPhilipp Hachtmann */ 110435fd0808STejun Heo if (r_start >= m_end) 110535fd0808STejun Heo break; 110635fd0808STejun Heo /* if the two regions intersect, we're done */ 110735fd0808STejun Heo if (m_start < r_end) { 110835fd0808STejun Heo if (out_start) 1109f1af9d3aSPhilipp Hachtmann *out_start = 1110f1af9d3aSPhilipp Hachtmann max(m_start, r_start); 111135fd0808STejun Heo if (out_end) 111235fd0808STejun Heo *out_end = min(m_end, r_end); 111335fd0808STejun Heo if (out_nid) 1114f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 111535fd0808STejun Heo /* 1116f1af9d3aSPhilipp Hachtmann * The region which ends first is 1117f1af9d3aSPhilipp Hachtmann * advanced for the next iteration. 111835fd0808STejun Heo */ 111935fd0808STejun Heo if (m_end <= r_end) 1120f1af9d3aSPhilipp Hachtmann idx_a++; 112135fd0808STejun Heo else 1122f1af9d3aSPhilipp Hachtmann idx_b++; 1123f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 112435fd0808STejun Heo return; 112535fd0808STejun Heo } 112635fd0808STejun Heo } 112735fd0808STejun Heo } 112835fd0808STejun Heo 112935fd0808STejun Heo /* signal end of iteration */ 113035fd0808STejun Heo *idx = ULLONG_MAX; 113135fd0808STejun Heo } 113235fd0808STejun Heo 11337bd0b0f0STejun Heo /** 1134f1af9d3aSPhilipp Hachtmann * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1135f1af9d3aSPhilipp Hachtmann * 11367bd0b0f0STejun Heo * @idx: pointer to u64 loop variable 1137ad5ea8cdSAlexander Kuleshov * @nid: node selector, %NUMA_NO_NODE for all nodes 1138fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 1139f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 1140f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 1141dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1142dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1143dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 11447bd0b0f0STejun Heo * 114547cec443SMike Rapoport * Finds the next range from type_a which is not marked as unsuitable 114647cec443SMike Rapoport * in type_b. 114747cec443SMike Rapoport * 1148f1af9d3aSPhilipp Hachtmann * Reverse of __next_mem_range(). 11497bd0b0f0STejun Heo */ 1150e1720feeSMike Rapoport void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1151e1720feeSMike Rapoport enum memblock_flags flags, 1152f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 1153f1af9d3aSPhilipp Hachtmann struct memblock_type *type_b, 11547bd0b0f0STejun Heo phys_addr_t *out_start, 11557bd0b0f0STejun Heo phys_addr_t *out_end, int *out_nid) 11567bd0b0f0STejun Heo { 1157f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 1158f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 1159b1154233SGrygorii Strashko 1160560dca27SGrygorii Strashko if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1161560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 11627bd0b0f0STejun Heo 11637bd0b0f0STejun Heo if (*idx == (u64)ULLONG_MAX) { 1164f1af9d3aSPhilipp Hachtmann idx_a = type_a->cnt - 1; 1165e47608abSzijun_hu if (type_b != NULL) 1166f1af9d3aSPhilipp Hachtmann idx_b = type_b->cnt; 1167e47608abSzijun_hu else 1168e47608abSzijun_hu idx_b = 0; 11697bd0b0f0STejun Heo } 11707bd0b0f0STejun Heo 1171f1af9d3aSPhilipp Hachtmann for (; idx_a >= 0; idx_a--) { 1172f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1173f1af9d3aSPhilipp Hachtmann 11747bd0b0f0STejun Heo phys_addr_t m_start = m->base; 11757bd0b0f0STejun Heo phys_addr_t m_end = m->base + m->size; 1176f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 11777bd0b0f0STejun Heo 11789f3d5eaaSMike Rapoport if (should_skip_region(type_a, m, nid, flags)) 1179bf3d3cc5SArd Biesheuvel continue; 1180bf3d3cc5SArd Biesheuvel 1181f1af9d3aSPhilipp Hachtmann if (!type_b) { 1182f1af9d3aSPhilipp Hachtmann if (out_start) 1183f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1184f1af9d3aSPhilipp Hachtmann if (out_end) 1185f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1186f1af9d3aSPhilipp Hachtmann if (out_nid) 1187f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1188fb399b48Szijun_hu idx_a--; 1189f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1190f1af9d3aSPhilipp Hachtmann return; 1191f1af9d3aSPhilipp Hachtmann } 11927bd0b0f0STejun Heo 1193f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1194f1af9d3aSPhilipp Hachtmann for (; idx_b >= 0; idx_b--) { 1195f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1196f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1197f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1198f1af9d3aSPhilipp Hachtmann 1199f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1200f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1201f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 12021c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1203f1af9d3aSPhilipp Hachtmann /* 1204f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1205f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1206f1af9d3aSPhilipp Hachtmann */ 1207f1af9d3aSPhilipp Hachtmann 12087bd0b0f0STejun Heo if (r_end <= m_start) 12097bd0b0f0STejun Heo break; 12107bd0b0f0STejun Heo /* if the two regions intersect, we're done */ 12117bd0b0f0STejun Heo if (m_end > r_start) { 12127bd0b0f0STejun Heo if (out_start) 12137bd0b0f0STejun Heo *out_start = max(m_start, r_start); 12147bd0b0f0STejun Heo if (out_end) 12157bd0b0f0STejun Heo *out_end = min(m_end, r_end); 12167bd0b0f0STejun Heo if (out_nid) 1217f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 12187bd0b0f0STejun Heo if (m_start >= r_start) 1219f1af9d3aSPhilipp Hachtmann idx_a--; 12207bd0b0f0STejun Heo else 1221f1af9d3aSPhilipp Hachtmann idx_b--; 1222f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 12237bd0b0f0STejun Heo return; 12247bd0b0f0STejun Heo } 12257bd0b0f0STejun Heo } 12267bd0b0f0STejun Heo } 1227f1af9d3aSPhilipp Hachtmann /* signal end of iteration */ 12287bd0b0f0STejun Heo *idx = ULLONG_MAX; 12297bd0b0f0STejun Heo } 12307bd0b0f0STejun Heo 12317c0caeb8STejun Heo /* 123245e79815SChen Chang * Common iterator interface used to define for_each_mem_pfn_range(). 12337c0caeb8STejun Heo */ 12347c0caeb8STejun Heo void __init_memblock __next_mem_pfn_range(int *idx, int nid, 12357c0caeb8STejun Heo unsigned long *out_start_pfn, 12367c0caeb8STejun Heo unsigned long *out_end_pfn, int *out_nid) 12377c0caeb8STejun Heo { 12387c0caeb8STejun Heo struct memblock_type *type = &memblock.memory; 12397c0caeb8STejun Heo struct memblock_region *r; 1240d622abf7SMike Rapoport int r_nid; 12417c0caeb8STejun Heo 12427c0caeb8STejun Heo while (++*idx < type->cnt) { 12437c0caeb8STejun Heo r = &type->regions[*idx]; 1244d622abf7SMike Rapoport r_nid = memblock_get_region_node(r); 12457c0caeb8STejun Heo 12467c0caeb8STejun Heo if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 12477c0caeb8STejun Heo continue; 1248d622abf7SMike Rapoport if (nid == MAX_NUMNODES || nid == r_nid) 12497c0caeb8STejun Heo break; 12507c0caeb8STejun Heo } 12517c0caeb8STejun Heo if (*idx >= type->cnt) { 12527c0caeb8STejun Heo *idx = -1; 12537c0caeb8STejun Heo return; 12547c0caeb8STejun Heo } 12557c0caeb8STejun Heo 12567c0caeb8STejun Heo if (out_start_pfn) 12577c0caeb8STejun Heo *out_start_pfn = PFN_UP(r->base); 12587c0caeb8STejun Heo if (out_end_pfn) 12597c0caeb8STejun Heo *out_end_pfn = PFN_DOWN(r->base + r->size); 12607c0caeb8STejun Heo if (out_nid) 1261d622abf7SMike Rapoport *out_nid = r_nid; 12627c0caeb8STejun Heo } 12637c0caeb8STejun Heo 12647c0caeb8STejun Heo /** 12657c0caeb8STejun Heo * memblock_set_node - set node ID on memblock regions 12667c0caeb8STejun Heo * @base: base of area to set node ID for 12677c0caeb8STejun Heo * @size: size of area to set node ID for 1268e7e8de59STang Chen * @type: memblock type to set node ID for 12697c0caeb8STejun Heo * @nid: node ID to set 12707c0caeb8STejun Heo * 1271e7e8de59STang Chen * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 12727c0caeb8STejun Heo * Regions which cross the area boundaries are split as necessary. 12737c0caeb8STejun Heo * 127447cec443SMike Rapoport * Return: 12757c0caeb8STejun Heo * 0 on success, -errno on failure. 12767c0caeb8STejun Heo */ 12777c0caeb8STejun Heo int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1278e7e8de59STang Chen struct memblock_type *type, int nid) 12797c0caeb8STejun Heo { 1280a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 12816a9ceb31STejun Heo int start_rgn, end_rgn; 12826a9ceb31STejun Heo int i, ret; 12837c0caeb8STejun Heo 12846a9ceb31STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 12856a9ceb31STejun Heo if (ret) 12866a9ceb31STejun Heo return ret; 12877c0caeb8STejun Heo 12886a9ceb31STejun Heo for (i = start_rgn; i < end_rgn; i++) 1289e9d24ad3SWanpeng Li memblock_set_region_node(&type->regions[i], nid); 12907c0caeb8STejun Heo 12912fe03412SPeng Zhang memblock_merge_regions(type, start_rgn, end_rgn); 12923f08a302SMike Rapoport #endif 12937c0caeb8STejun Heo return 0; 12947c0caeb8STejun Heo } 12953f08a302SMike Rapoport 1296837566e7SAlexander Duyck #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1297837566e7SAlexander Duyck /** 1298837566e7SAlexander Duyck * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1299837566e7SAlexander Duyck * 1300837566e7SAlexander Duyck * @idx: pointer to u64 loop variable 1301837566e7SAlexander Duyck * @zone: zone in which all of the memory blocks reside 1302837566e7SAlexander Duyck * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1303837566e7SAlexander Duyck * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1304837566e7SAlexander Duyck * 1305837566e7SAlexander Duyck * This function is meant to be a zone/pfn specific wrapper for the 1306837566e7SAlexander Duyck * for_each_mem_range type iterators. Specifically they are used in the 1307837566e7SAlexander Duyck * deferred memory init routines and as such we were duplicating much of 1308837566e7SAlexander Duyck * this logic throughout the code. So instead of having it in multiple 1309837566e7SAlexander Duyck * locations it seemed like it would make more sense to centralize this to 1310837566e7SAlexander Duyck * one new iterator that does everything they need. 1311837566e7SAlexander Duyck */ 1312837566e7SAlexander Duyck void __init_memblock 1313837566e7SAlexander Duyck __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1314837566e7SAlexander Duyck unsigned long *out_spfn, unsigned long *out_epfn) 1315837566e7SAlexander Duyck { 1316837566e7SAlexander Duyck int zone_nid = zone_to_nid(zone); 1317837566e7SAlexander Duyck phys_addr_t spa, epa; 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 while (*idx != U64_MAX) { 1324837566e7SAlexander Duyck unsigned long epfn = PFN_DOWN(epa); 1325837566e7SAlexander Duyck unsigned long spfn = PFN_UP(spa); 1326837566e7SAlexander Duyck 1327837566e7SAlexander Duyck /* 1328837566e7SAlexander Duyck * Verify the end is at least past the start of the zone and 1329837566e7SAlexander Duyck * that we have at least one PFN to initialize. 1330837566e7SAlexander Duyck */ 1331837566e7SAlexander Duyck if (zone->zone_start_pfn < epfn && spfn < epfn) { 1332837566e7SAlexander Duyck /* if we went too far just stop searching */ 1333837566e7SAlexander Duyck if (zone_end_pfn(zone) <= spfn) { 1334837566e7SAlexander Duyck *idx = U64_MAX; 1335837566e7SAlexander Duyck break; 1336837566e7SAlexander Duyck } 1337837566e7SAlexander Duyck 1338837566e7SAlexander Duyck if (out_spfn) 1339837566e7SAlexander Duyck *out_spfn = max(zone->zone_start_pfn, spfn); 1340837566e7SAlexander Duyck if (out_epfn) 1341837566e7SAlexander Duyck *out_epfn = min(zone_end_pfn(zone), epfn); 1342837566e7SAlexander Duyck 1343837566e7SAlexander Duyck return; 1344837566e7SAlexander Duyck } 1345837566e7SAlexander Duyck 1346837566e7SAlexander Duyck __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1347837566e7SAlexander Duyck &memblock.memory, &memblock.reserved, 1348f30b002cSMiaohe Lin &spa, &epa, NULL); 1349837566e7SAlexander Duyck } 1350837566e7SAlexander Duyck 1351837566e7SAlexander Duyck /* signal end of iteration */ 1352837566e7SAlexander Duyck if (out_spfn) 1353837566e7SAlexander Duyck *out_spfn = ULONG_MAX; 1354837566e7SAlexander Duyck if (out_epfn) 1355837566e7SAlexander Duyck *out_epfn = 0; 1356837566e7SAlexander Duyck } 1357837566e7SAlexander Duyck 1358837566e7SAlexander Duyck #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 13597c0caeb8STejun Heo 136092d12f95SMike Rapoport /** 136192d12f95SMike Rapoport * memblock_alloc_range_nid - allocate boot memory block 136292d12f95SMike Rapoport * @size: size of memory block to be allocated in bytes 136392d12f95SMike Rapoport * @align: alignment of the region and block's size 136492d12f95SMike Rapoport * @start: the lower bound of the memory region to allocate (phys address) 136592d12f95SMike Rapoport * @end: the upper bound of the memory region to allocate (phys address) 136692d12f95SMike Rapoport * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 13670ac398b1SYunfeng Ye * @exact_nid: control the allocation fall back to other nodes 136892d12f95SMike Rapoport * 136992d12f95SMike Rapoport * The allocation is performed from memory region limited by 137095830666SCao jin * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 137192d12f95SMike Rapoport * 13720ac398b1SYunfeng Ye * If the specified node can not hold the requested memory and @exact_nid 13730ac398b1SYunfeng Ye * is false, the allocation falls back to any node in the system. 137492d12f95SMike Rapoport * 137592d12f95SMike Rapoport * For systems with memory mirroring, the allocation is attempted first 137692d12f95SMike Rapoport * from the regions with mirroring enabled and then retried from any 137792d12f95SMike Rapoport * memory region. 137892d12f95SMike Rapoport * 1379c200d900SPatrick Wang * In addition, function using kmemleak_alloc_phys for allocated boot 1380c200d900SPatrick Wang * memory block, it is never reported as leaks. 138192d12f95SMike Rapoport * 138292d12f95SMike Rapoport * Return: 138392d12f95SMike Rapoport * Physical address of allocated memory block on success, %0 on failure. 138492d12f95SMike Rapoport */ 13858676af1fSAslan Bakirov phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 13862bfc2862SAkinobu Mita phys_addr_t align, phys_addr_t start, 13870ac398b1SYunfeng Ye phys_addr_t end, int nid, 13880ac398b1SYunfeng Ye bool exact_nid) 138995f72d1eSYinghai Lu { 139092d12f95SMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 13916ed311b2SBenjamin Herrenschmidt phys_addr_t found; 139295f72d1eSYinghai Lu 139392d12f95SMike Rapoport if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 139492d12f95SMike Rapoport nid = NUMA_NO_NODE; 139592d12f95SMike Rapoport 13962f770806SMike Rapoport if (!align) { 13972f770806SMike Rapoport /* Can't use WARNs this early in boot on powerpc */ 13982f770806SMike Rapoport dump_stack(); 13992f770806SMike Rapoport align = SMP_CACHE_BYTES; 14002f770806SMike Rapoport } 14012f770806SMike Rapoport 140292d12f95SMike Rapoport again: 1403fc6daaf9STony Luck found = memblock_find_in_range_node(size, align, start, end, nid, 1404fc6daaf9STony Luck flags); 140592d12f95SMike Rapoport if (found && !memblock_reserve(found, size)) 140692d12f95SMike Rapoport goto done; 140792d12f95SMike Rapoport 14080ac398b1SYunfeng Ye if (nid != NUMA_NO_NODE && !exact_nid) { 140992d12f95SMike Rapoport found = memblock_find_in_range_node(size, align, start, 141092d12f95SMike Rapoport end, NUMA_NO_NODE, 141192d12f95SMike Rapoport flags); 141292d12f95SMike Rapoport if (found && !memblock_reserve(found, size)) 141392d12f95SMike Rapoport goto done; 141492d12f95SMike Rapoport } 141592d12f95SMike Rapoport 141692d12f95SMike Rapoport if (flags & MEMBLOCK_MIRROR) { 141792d12f95SMike Rapoport flags &= ~MEMBLOCK_MIRROR; 141814d9a675SMa Wupeng pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 141992d12f95SMike Rapoport &size); 142092d12f95SMike Rapoport goto again; 142192d12f95SMike Rapoport } 142292d12f95SMike Rapoport 142392d12f95SMike Rapoport return 0; 142492d12f95SMike Rapoport 142592d12f95SMike Rapoport done: 1426c6975d7cSQian Cai /* 1427c6975d7cSQian Cai * Skip kmemleak for those places like kasan_init() and 1428c6975d7cSQian Cai * early_pgtable_alloc() due to high volume. 1429c6975d7cSQian Cai */ 1430c6975d7cSQian Cai if (end != MEMBLOCK_ALLOC_NOLEAKTRACE) 1431aedf95eaSCatalin Marinas /* 1432c200d900SPatrick Wang * Memblock allocated blocks are never reported as 1433c200d900SPatrick Wang * leaks. This is because many of these blocks are 1434c200d900SPatrick Wang * only referred via the physical address which is 1435c200d900SPatrick Wang * not looked up by kmemleak. 1436aedf95eaSCatalin Marinas */ 1437c200d900SPatrick Wang kmemleak_alloc_phys(found, size, 0); 143892d12f95SMike Rapoport 14396ed311b2SBenjamin Herrenschmidt return found; 1440aedf95eaSCatalin Marinas } 144195f72d1eSYinghai Lu 1442a2974133SMike Rapoport /** 1443a2974133SMike Rapoport * memblock_phys_alloc_range - allocate a memory block inside specified range 1444a2974133SMike Rapoport * @size: size of memory block to be allocated in bytes 1445a2974133SMike Rapoport * @align: alignment of the region and block's size 1446a2974133SMike Rapoport * @start: the lower bound of the memory region to allocate (physical address) 1447a2974133SMike Rapoport * @end: the upper bound of the memory region to allocate (physical address) 1448a2974133SMike Rapoport * 1449a2974133SMike Rapoport * Allocate @size bytes in the between @start and @end. 1450a2974133SMike Rapoport * 1451a2974133SMike Rapoport * Return: physical address of the allocated memory block on success, 1452a2974133SMike Rapoport * %0 on failure. 1453a2974133SMike Rapoport */ 14548a770c2aSMike Rapoport phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 14558a770c2aSMike Rapoport phys_addr_t align, 14568a770c2aSMike Rapoport phys_addr_t start, 14578a770c2aSMike Rapoport phys_addr_t end) 14582bfc2862SAkinobu Mita { 1459b5cf2d6cSFaiyaz Mohammed memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1460b5cf2d6cSFaiyaz Mohammed __func__, (u64)size, (u64)align, &start, &end, 1461b5cf2d6cSFaiyaz Mohammed (void *)_RET_IP_); 14620ac398b1SYunfeng Ye return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 14630ac398b1SYunfeng Ye false); 14647bd0b0f0STejun Heo } 14657bd0b0f0STejun Heo 1466a2974133SMike Rapoport /** 146717cbe038SLevi Yun * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1468a2974133SMike Rapoport * @size: size of memory block to be allocated in bytes 1469a2974133SMike Rapoport * @align: alignment of the region and block's size 1470a2974133SMike Rapoport * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1471a2974133SMike Rapoport * 1472a2974133SMike Rapoport * Allocates memory block from the specified NUMA node. If the node 1473a2974133SMike Rapoport * has no available memory, attempts to allocated from any node in the 1474a2974133SMike Rapoport * system. 1475a2974133SMike Rapoport * 1476a2974133SMike Rapoport * Return: physical address of the allocated memory block on success, 1477a2974133SMike Rapoport * %0 on failure. 1478a2974133SMike Rapoport */ 14799a8dd708SMike Rapoport phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 14809d1e2492SBenjamin Herrenschmidt { 148133755574SMike Rapoport return memblock_alloc_range_nid(size, align, 0, 14820ac398b1SYunfeng Ye MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 148395f72d1eSYinghai Lu } 148495f72d1eSYinghai Lu 148526f09e9bSSantosh Shilimkar /** 1486eb31d559SMike Rapoport * memblock_alloc_internal - allocate boot memory block 148726f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 148826f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 148926f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region to allocate (phys address) 149026f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region to allocate (phys address) 149126f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 14920ac398b1SYunfeng Ye * @exact_nid: control the allocation fall back to other nodes 149326f09e9bSSantosh Shilimkar * 149492d12f95SMike Rapoport * Allocates memory block using memblock_alloc_range_nid() and 149592d12f95SMike Rapoport * converts the returned physical address to virtual. 149692d12f95SMike Rapoport * 149726f09e9bSSantosh Shilimkar * The @min_addr limit is dropped if it can not be satisfied and the allocation 149892d12f95SMike Rapoport * will fall back to memory below @min_addr. Other constraints, such 149992d12f95SMike Rapoport * as node and mirrored memory will be handled again in 150092d12f95SMike Rapoport * memblock_alloc_range_nid(). 150126f09e9bSSantosh Shilimkar * 150247cec443SMike Rapoport * Return: 150326f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 150426f09e9bSSantosh Shilimkar */ 1505eb31d559SMike Rapoport static void * __init memblock_alloc_internal( 150626f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 150726f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 15080ac398b1SYunfeng Ye int nid, bool exact_nid) 150926f09e9bSSantosh Shilimkar { 151026f09e9bSSantosh Shilimkar phys_addr_t alloc; 151126f09e9bSSantosh Shilimkar 151226f09e9bSSantosh Shilimkar /* 151326f09e9bSSantosh Shilimkar * Detect any accidental use of these APIs after slab is ready, as at 151426f09e9bSSantosh Shilimkar * this moment memblock may be deinitialized already and its 1515c6ffc5caSMike Rapoport * internal data may be destroyed (after execution of memblock_free_all) 151626f09e9bSSantosh Shilimkar */ 151726f09e9bSSantosh Shilimkar if (WARN_ON_ONCE(slab_is_available())) 151826f09e9bSSantosh Shilimkar return kzalloc_node(size, GFP_NOWAIT, nid); 151926f09e9bSSantosh Shilimkar 1520f3057ad7SMike Rapoport if (max_addr > memblock.current_limit) 1521f3057ad7SMike Rapoport max_addr = memblock.current_limit; 1522f3057ad7SMike Rapoport 15230ac398b1SYunfeng Ye alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 15240ac398b1SYunfeng Ye exact_nid); 15252f770806SMike Rapoport 152692d12f95SMike Rapoport /* retry allocation without lower limit */ 152792d12f95SMike Rapoport if (!alloc && min_addr) 15280ac398b1SYunfeng Ye alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 15290ac398b1SYunfeng Ye exact_nid); 153026f09e9bSSantosh Shilimkar 153192d12f95SMike Rapoport if (!alloc) 1532a3f5bafcSTony Luck return NULL; 153326f09e9bSSantosh Shilimkar 153492d12f95SMike Rapoport return phys_to_virt(alloc); 153526f09e9bSSantosh Shilimkar } 153626f09e9bSSantosh Shilimkar 153726f09e9bSSantosh Shilimkar /** 15380ac398b1SYunfeng Ye * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 15390ac398b1SYunfeng Ye * without zeroing memory 15400ac398b1SYunfeng Ye * @size: size of memory block to be allocated in bytes 15410ac398b1SYunfeng Ye * @align: alignment of the region and block's size 15420ac398b1SYunfeng Ye * @min_addr: the lower bound of the memory region from where the allocation 15430ac398b1SYunfeng Ye * is preferred (phys address) 15440ac398b1SYunfeng Ye * @max_addr: the upper bound of the memory region from where the allocation 15450ac398b1SYunfeng Ye * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 15460ac398b1SYunfeng Ye * allocate only from memory limited by memblock.current_limit value 15470ac398b1SYunfeng Ye * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 15480ac398b1SYunfeng Ye * 15490ac398b1SYunfeng Ye * Public function, provides additional debug information (including caller 15500ac398b1SYunfeng Ye * info), if enabled. Does not zero allocated memory. 15510ac398b1SYunfeng Ye * 15520ac398b1SYunfeng Ye * Return: 15530ac398b1SYunfeng Ye * Virtual address of allocated memory block on success, NULL on failure. 15540ac398b1SYunfeng Ye */ 15550ac398b1SYunfeng Ye void * __init memblock_alloc_exact_nid_raw( 15560ac398b1SYunfeng Ye phys_addr_t size, phys_addr_t align, 15570ac398b1SYunfeng Ye phys_addr_t min_addr, phys_addr_t max_addr, 15580ac398b1SYunfeng Ye int nid) 15590ac398b1SYunfeng Ye { 15600ac398b1SYunfeng Ye memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 15610ac398b1SYunfeng Ye __func__, (u64)size, (u64)align, nid, &min_addr, 15620ac398b1SYunfeng Ye &max_addr, (void *)_RET_IP_); 15630ac398b1SYunfeng Ye 156408678804SMike Rapoport return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 156508678804SMike Rapoport true); 15660ac398b1SYunfeng Ye } 15670ac398b1SYunfeng Ye 15680ac398b1SYunfeng Ye /** 1569eb31d559SMike Rapoport * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1570ea1f5f37SPavel Tatashin * memory and without panicking 1571ea1f5f37SPavel Tatashin * @size: size of memory block to be allocated in bytes 1572ea1f5f37SPavel Tatashin * @align: alignment of the region and block's size 1573ea1f5f37SPavel Tatashin * @min_addr: the lower bound of the memory region from where the allocation 1574ea1f5f37SPavel Tatashin * is preferred (phys address) 1575ea1f5f37SPavel Tatashin * @max_addr: the upper bound of the memory region from where the allocation 157697ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1577ea1f5f37SPavel Tatashin * allocate only from memory limited by memblock.current_limit value 1578ea1f5f37SPavel Tatashin * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1579ea1f5f37SPavel Tatashin * 1580ea1f5f37SPavel Tatashin * Public function, provides additional debug information (including caller 1581ea1f5f37SPavel Tatashin * info), if enabled. Does not zero allocated memory, does not panic if request 1582ea1f5f37SPavel Tatashin * cannot be satisfied. 1583ea1f5f37SPavel Tatashin * 158447cec443SMike Rapoport * Return: 1585ea1f5f37SPavel Tatashin * Virtual address of allocated memory block on success, NULL on failure. 1586ea1f5f37SPavel Tatashin */ 1587eb31d559SMike Rapoport void * __init memblock_alloc_try_nid_raw( 1588ea1f5f37SPavel Tatashin phys_addr_t size, phys_addr_t align, 1589ea1f5f37SPavel Tatashin phys_addr_t min_addr, phys_addr_t max_addr, 1590ea1f5f37SPavel Tatashin int nid) 1591ea1f5f37SPavel Tatashin { 1592d75f773cSSakari Ailus memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1593a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1594a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1595ea1f5f37SPavel Tatashin 159608678804SMike Rapoport return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 159708678804SMike Rapoport false); 1598ea1f5f37SPavel Tatashin } 1599ea1f5f37SPavel Tatashin 1600ea1f5f37SPavel Tatashin /** 1601c0dbe825SMike Rapoport * memblock_alloc_try_nid - allocate boot memory block 160226f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 160326f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 160426f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region from where the allocation 160526f09e9bSSantosh Shilimkar * is preferred (phys address) 160626f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region from where the allocation 160797ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 160826f09e9bSSantosh Shilimkar * allocate only from memory limited by memblock.current_limit value 160926f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 161026f09e9bSSantosh Shilimkar * 1611c0dbe825SMike Rapoport * Public function, provides additional debug information (including caller 1612c0dbe825SMike Rapoport * info), if enabled. This function zeroes the allocated memory. 161326f09e9bSSantosh Shilimkar * 161447cec443SMike Rapoport * Return: 161526f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 161626f09e9bSSantosh Shilimkar */ 1617eb31d559SMike Rapoport void * __init memblock_alloc_try_nid( 161826f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 161926f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 162026f09e9bSSantosh Shilimkar int nid) 162126f09e9bSSantosh Shilimkar { 162226f09e9bSSantosh Shilimkar void *ptr; 162326f09e9bSSantosh Shilimkar 1624d75f773cSSakari Ailus memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1625a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1626a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1627eb31d559SMike Rapoport ptr = memblock_alloc_internal(size, align, 16280ac398b1SYunfeng Ye min_addr, max_addr, nid, false); 1629c0dbe825SMike Rapoport if (ptr) 1630ea1f5f37SPavel Tatashin memset(ptr, 0, size); 163126f09e9bSSantosh Shilimkar 1632c0dbe825SMike Rapoport return ptr; 163326f09e9bSSantosh Shilimkar } 163426f09e9bSSantosh Shilimkar 163526f09e9bSSantosh Shilimkar /** 1636621d9739SMike Rapoport * memblock_free_late - free pages directly to buddy allocator 163748a833ccSMike Rapoport * @base: phys starting address of the boot memory block 163826f09e9bSSantosh Shilimkar * @size: size of the boot memory block in bytes 163926f09e9bSSantosh Shilimkar * 1640a2974133SMike Rapoport * This is only useful when the memblock allocator has already been torn 164126f09e9bSSantosh Shilimkar * down, but we are still initializing the system. Pages are released directly 1642a2974133SMike Rapoport * to the buddy allocator. 164326f09e9bSSantosh Shilimkar */ 1644621d9739SMike Rapoport void __init memblock_free_late(phys_addr_t base, phys_addr_t size) 164526f09e9bSSantosh Shilimkar { 1646a36aab89SMike Rapoport phys_addr_t cursor, end; 164726f09e9bSSantosh Shilimkar 1648a36aab89SMike Rapoport end = base + size - 1; 1649d75f773cSSakari Ailus memblock_dbg("%s: [%pa-%pa] %pS\n", 1650a36aab89SMike Rapoport __func__, &base, &end, (void *)_RET_IP_); 16519099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 165226f09e9bSSantosh Shilimkar cursor = PFN_UP(base); 165326f09e9bSSantosh Shilimkar end = PFN_DOWN(base + size); 165426f09e9bSSantosh Shilimkar 165526f09e9bSSantosh Shilimkar for (; cursor < end; cursor++) { 1656647037adSAaron Thompson memblock_free_pages(pfn_to_page(cursor), cursor, 0); 1657ca79b0c2SArun KS totalram_pages_inc(); 165826f09e9bSSantosh Shilimkar } 165926f09e9bSSantosh Shilimkar } 16609d1e2492SBenjamin Herrenschmidt 16619d1e2492SBenjamin Herrenschmidt /* 16629d1e2492SBenjamin Herrenschmidt * Remaining API functions 16639d1e2492SBenjamin Herrenschmidt */ 16649d1e2492SBenjamin Herrenschmidt 16651f1ffb8aSDavid Gibson phys_addr_t __init_memblock memblock_phys_mem_size(void) 166695f72d1eSYinghai Lu { 16671440c4e2STejun Heo return memblock.memory.total_size; 166895f72d1eSYinghai Lu } 166995f72d1eSYinghai Lu 16708907de5dSSrikar Dronamraju phys_addr_t __init_memblock memblock_reserved_size(void) 16718907de5dSSrikar Dronamraju { 16728907de5dSSrikar Dronamraju return memblock.reserved.total_size; 16738907de5dSSrikar Dronamraju } 16748907de5dSSrikar Dronamraju 16750a93ebefSSam Ravnborg /* lowest address */ 16760a93ebefSSam Ravnborg phys_addr_t __init_memblock memblock_start_of_DRAM(void) 16770a93ebefSSam Ravnborg { 16780a93ebefSSam Ravnborg return memblock.memory.regions[0].base; 16790a93ebefSSam Ravnborg } 16800a93ebefSSam Ravnborg 168110d06439SYinghai Lu phys_addr_t __init_memblock memblock_end_of_DRAM(void) 168295f72d1eSYinghai Lu { 168395f72d1eSYinghai Lu int idx = memblock.memory.cnt - 1; 168495f72d1eSYinghai Lu 1685e3239ff9SBenjamin Herrenschmidt return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 168695f72d1eSYinghai Lu } 168795f72d1eSYinghai Lu 1688a571d4ebSDennis Chen static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 168995f72d1eSYinghai Lu { 16901c4bc43dSStefan Agner phys_addr_t max_addr = PHYS_ADDR_MAX; 1691136199f0SEmil Medve struct memblock_region *r; 169295f72d1eSYinghai Lu 1693a571d4ebSDennis Chen /* 1694a571d4ebSDennis Chen * translate the memory @limit size into the max address within one of 1695a571d4ebSDennis Chen * the memory memblock regions, if the @limit exceeds the total size 16961c4bc43dSStefan Agner * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1697a571d4ebSDennis Chen */ 1698cc6de168SMike Rapoport for_each_mem_region(r) { 1699c0ce8fefSTejun Heo if (limit <= r->size) { 1700c0ce8fefSTejun Heo max_addr = r->base + limit; 170195f72d1eSYinghai Lu break; 170295f72d1eSYinghai Lu } 1703c0ce8fefSTejun Heo limit -= r->size; 170495f72d1eSYinghai Lu } 1705c0ce8fefSTejun Heo 1706a571d4ebSDennis Chen return max_addr; 1707a571d4ebSDennis Chen } 1708a571d4ebSDennis Chen 1709a571d4ebSDennis Chen void __init memblock_enforce_memory_limit(phys_addr_t limit) 1710a571d4ebSDennis Chen { 171149aef717SColin Ian King phys_addr_t max_addr; 1712a571d4ebSDennis Chen 1713a571d4ebSDennis Chen if (!limit) 1714a571d4ebSDennis Chen return; 1715a571d4ebSDennis Chen 1716a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1717a571d4ebSDennis Chen 1718a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 17191c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1720a571d4ebSDennis Chen return; 1721a571d4ebSDennis Chen 1722c0ce8fefSTejun Heo /* truncate both memory and reserved regions */ 1723f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.memory, max_addr, 17241c4bc43dSStefan Agner PHYS_ADDR_MAX); 1725f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.reserved, max_addr, 17261c4bc43dSStefan Agner PHYS_ADDR_MAX); 172795f72d1eSYinghai Lu } 172895f72d1eSYinghai Lu 1729c9ca9b4eSAKASHI Takahiro void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1730c9ca9b4eSAKASHI Takahiro { 1731c9ca9b4eSAKASHI Takahiro int start_rgn, end_rgn; 1732c9ca9b4eSAKASHI Takahiro int i, ret; 1733c9ca9b4eSAKASHI Takahiro 1734c9ca9b4eSAKASHI Takahiro if (!size) 1735c9ca9b4eSAKASHI Takahiro return; 1736c9ca9b4eSAKASHI Takahiro 17375173ed72SPeng Fan if (!memblock_memory->total_size) { 1738e888fa7bSGeert Uytterhoeven pr_warn("%s: No memory registered yet\n", __func__); 1739e888fa7bSGeert Uytterhoeven return; 1740e888fa7bSGeert Uytterhoeven } 1741e888fa7bSGeert Uytterhoeven 1742c9ca9b4eSAKASHI Takahiro ret = memblock_isolate_range(&memblock.memory, base, size, 1743c9ca9b4eSAKASHI Takahiro &start_rgn, &end_rgn); 1744c9ca9b4eSAKASHI Takahiro if (ret) 1745c9ca9b4eSAKASHI Takahiro return; 1746c9ca9b4eSAKASHI Takahiro 1747c9ca9b4eSAKASHI Takahiro /* remove all the MAP regions */ 1748c9ca9b4eSAKASHI Takahiro for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1749c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1750c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1751c9ca9b4eSAKASHI Takahiro 1752c9ca9b4eSAKASHI Takahiro for (i = start_rgn - 1; i >= 0; i--) 1753c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1754c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1755c9ca9b4eSAKASHI Takahiro 1756c9ca9b4eSAKASHI Takahiro /* truncate the reserved regions */ 1757c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 0, base); 1758c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 17591c4bc43dSStefan Agner base + size, PHYS_ADDR_MAX); 1760c9ca9b4eSAKASHI Takahiro } 1761c9ca9b4eSAKASHI Takahiro 1762a571d4ebSDennis Chen void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1763a571d4ebSDennis Chen { 1764a571d4ebSDennis Chen phys_addr_t max_addr; 1765a571d4ebSDennis Chen 1766a571d4ebSDennis Chen if (!limit) 1767a571d4ebSDennis Chen return; 1768a571d4ebSDennis Chen 1769a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1770a571d4ebSDennis Chen 1771a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 17721c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1773a571d4ebSDennis Chen return; 1774a571d4ebSDennis Chen 1775c9ca9b4eSAKASHI Takahiro memblock_cap_memory_range(0, max_addr); 1776a571d4ebSDennis Chen } 1777a571d4ebSDennis Chen 1778cd79481dSYinghai Lu static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 177972d4b0b4SBenjamin Herrenschmidt { 178072d4b0b4SBenjamin Herrenschmidt unsigned int left = 0, right = type->cnt; 178172d4b0b4SBenjamin Herrenschmidt 178272d4b0b4SBenjamin Herrenschmidt do { 178372d4b0b4SBenjamin Herrenschmidt unsigned int mid = (right + left) / 2; 178472d4b0b4SBenjamin Herrenschmidt 178572d4b0b4SBenjamin Herrenschmidt if (addr < type->regions[mid].base) 178672d4b0b4SBenjamin Herrenschmidt right = mid; 178772d4b0b4SBenjamin Herrenschmidt else if (addr >= (type->regions[mid].base + 178872d4b0b4SBenjamin Herrenschmidt type->regions[mid].size)) 178972d4b0b4SBenjamin Herrenschmidt left = mid + 1; 179072d4b0b4SBenjamin Herrenschmidt else 179172d4b0b4SBenjamin Herrenschmidt return mid; 179272d4b0b4SBenjamin Herrenschmidt } while (left < right); 179372d4b0b4SBenjamin Herrenschmidt return -1; 179472d4b0b4SBenjamin Herrenschmidt } 179572d4b0b4SBenjamin Herrenschmidt 1796f5a222dcSYueyi Li bool __init_memblock memblock_is_reserved(phys_addr_t addr) 179795f72d1eSYinghai Lu { 179872d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.reserved, addr) != -1; 179995f72d1eSYinghai Lu } 180072d4b0b4SBenjamin Herrenschmidt 1801b4ad0c7eSYaowei Bai bool __init_memblock memblock_is_memory(phys_addr_t addr) 180272d4b0b4SBenjamin Herrenschmidt { 180372d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.memory, addr) != -1; 180472d4b0b4SBenjamin Herrenschmidt } 180572d4b0b4SBenjamin Herrenschmidt 1806937f0c26SYaowei Bai bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1807bf3d3cc5SArd Biesheuvel { 1808bf3d3cc5SArd Biesheuvel int i = memblock_search(&memblock.memory, addr); 1809bf3d3cc5SArd Biesheuvel 1810bf3d3cc5SArd Biesheuvel if (i == -1) 1811bf3d3cc5SArd Biesheuvel return false; 1812bf3d3cc5SArd Biesheuvel return !memblock_is_nomap(&memblock.memory.regions[i]); 1813bf3d3cc5SArd Biesheuvel } 1814bf3d3cc5SArd Biesheuvel 1815e76b63f8SYinghai Lu int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1816e76b63f8SYinghai Lu unsigned long *start_pfn, unsigned long *end_pfn) 1817e76b63f8SYinghai Lu { 1818e76b63f8SYinghai Lu struct memblock_type *type = &memblock.memory; 181916763230SFabian Frederick int mid = memblock_search(type, PFN_PHYS(pfn)); 1820e76b63f8SYinghai Lu 1821e76b63f8SYinghai Lu if (mid == -1) 1822e76b63f8SYinghai Lu return -1; 1823e76b63f8SYinghai Lu 1824f7e2f7e8SFabian Frederick *start_pfn = PFN_DOWN(type->regions[mid].base); 1825f7e2f7e8SFabian Frederick *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1826e76b63f8SYinghai Lu 1827d622abf7SMike Rapoport return memblock_get_region_node(&type->regions[mid]); 1828e76b63f8SYinghai Lu } 1829e76b63f8SYinghai Lu 1830eab30949SStephen Boyd /** 1831eab30949SStephen Boyd * memblock_is_region_memory - check if a region is a subset of memory 1832eab30949SStephen Boyd * @base: base of region to check 1833eab30949SStephen Boyd * @size: size of region to check 1834eab30949SStephen Boyd * 1835eab30949SStephen Boyd * Check if the region [@base, @base + @size) is a subset of a memory block. 1836eab30949SStephen Boyd * 183747cec443SMike Rapoport * Return: 1838eab30949SStephen Boyd * 0 if false, non-zero if true 1839eab30949SStephen Boyd */ 1840937f0c26SYaowei Bai bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 184172d4b0b4SBenjamin Herrenschmidt { 1842abb65272STomi Valkeinen int idx = memblock_search(&memblock.memory, base); 1843eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 184472d4b0b4SBenjamin Herrenschmidt 184572d4b0b4SBenjamin Herrenschmidt if (idx == -1) 1846937f0c26SYaowei Bai return false; 1847ef415ef4SWei Yang return (memblock.memory.regions[idx].base + 1848eb18f1b5STejun Heo memblock.memory.regions[idx].size) >= end; 184995f72d1eSYinghai Lu } 185095f72d1eSYinghai Lu 1851eab30949SStephen Boyd /** 1852eab30949SStephen Boyd * memblock_is_region_reserved - check if a region intersects reserved memory 1853eab30949SStephen Boyd * @base: base of region to check 1854eab30949SStephen Boyd * @size: size of region to check 1855eab30949SStephen Boyd * 185647cec443SMike Rapoport * Check if the region [@base, @base + @size) intersects a reserved 185747cec443SMike Rapoport * memory block. 1858eab30949SStephen Boyd * 185947cec443SMike Rapoport * Return: 1860c5c5c9d1STang Chen * True if they intersect, false if not. 1861eab30949SStephen Boyd */ 1862c5c5c9d1STang Chen bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 186395f72d1eSYinghai Lu { 1864c5c5c9d1STang Chen return memblock_overlaps_region(&memblock.reserved, base, size); 186595f72d1eSYinghai Lu } 186695f72d1eSYinghai Lu 18676ede1fd3SYinghai Lu void __init_memblock memblock_trim_memory(phys_addr_t align) 18686ede1fd3SYinghai Lu { 18696ede1fd3SYinghai Lu phys_addr_t start, end, orig_start, orig_end; 1870136199f0SEmil Medve struct memblock_region *r; 18716ede1fd3SYinghai Lu 1872cc6de168SMike Rapoport for_each_mem_region(r) { 1873136199f0SEmil Medve orig_start = r->base; 1874136199f0SEmil Medve orig_end = r->base + r->size; 18756ede1fd3SYinghai Lu start = round_up(orig_start, align); 18766ede1fd3SYinghai Lu end = round_down(orig_end, align); 18776ede1fd3SYinghai Lu 18786ede1fd3SYinghai Lu if (start == orig_start && end == orig_end) 18796ede1fd3SYinghai Lu continue; 18806ede1fd3SYinghai Lu 18816ede1fd3SYinghai Lu if (start < end) { 1882136199f0SEmil Medve r->base = start; 1883136199f0SEmil Medve r->size = end - start; 18846ede1fd3SYinghai Lu } else { 1885136199f0SEmil Medve memblock_remove_region(&memblock.memory, 1886136199f0SEmil Medve r - memblock.memory.regions); 1887136199f0SEmil Medve r--; 18886ede1fd3SYinghai Lu } 18896ede1fd3SYinghai Lu } 18906ede1fd3SYinghai Lu } 1891e63075a3SBenjamin Herrenschmidt 18923661ca66SYinghai Lu void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1893e63075a3SBenjamin Herrenschmidt { 1894e63075a3SBenjamin Herrenschmidt memblock.current_limit = limit; 1895e63075a3SBenjamin Herrenschmidt } 1896e63075a3SBenjamin Herrenschmidt 1897fec51014SLaura Abbott phys_addr_t __init_memblock memblock_get_current_limit(void) 1898fec51014SLaura Abbott { 1899fec51014SLaura Abbott return memblock.current_limit; 1900fec51014SLaura Abbott } 1901fec51014SLaura Abbott 19020262d9c8SHeiko Carstens static void __init_memblock memblock_dump(struct memblock_type *type) 19036ed311b2SBenjamin Herrenschmidt { 19045d63f81cSMiles Chen phys_addr_t base, end, size; 1905e1720feeSMike Rapoport enum memblock_flags flags; 19068c9c1701SAlexander Kuleshov int idx; 19078c9c1701SAlexander Kuleshov struct memblock_region *rgn; 19086ed311b2SBenjamin Herrenschmidt 19090262d9c8SHeiko Carstens pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 19106ed311b2SBenjamin Herrenschmidt 191166e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 19127c0caeb8STejun Heo char nid_buf[32] = ""; 19136ed311b2SBenjamin Herrenschmidt 19147c0caeb8STejun Heo base = rgn->base; 19157c0caeb8STejun Heo size = rgn->size; 19165d63f81cSMiles Chen end = base + size - 1; 191766a20757STang Chen flags = rgn->flags; 1918a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 19197c0caeb8STejun Heo if (memblock_get_region_node(rgn) != MAX_NUMNODES) 19207c0caeb8STejun Heo snprintf(nid_buf, sizeof(nid_buf), " on node %d", 19217c0caeb8STejun Heo memblock_get_region_node(rgn)); 19227c0caeb8STejun Heo #endif 1923e1720feeSMike Rapoport pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 19240262d9c8SHeiko Carstens type->name, idx, &base, &end, &size, nid_buf, flags); 19256ed311b2SBenjamin Herrenschmidt } 19266ed311b2SBenjamin Herrenschmidt } 19276ed311b2SBenjamin Herrenschmidt 192887c55870SMike Rapoport static void __init_memblock __memblock_dump_all(void) 19296ed311b2SBenjamin Herrenschmidt { 19306ed311b2SBenjamin Herrenschmidt pr_info("MEMBLOCK configuration:\n"); 19315d63f81cSMiles Chen pr_info(" memory size = %pa reserved size = %pa\n", 19325d63f81cSMiles Chen &memblock.memory.total_size, 19335d63f81cSMiles Chen &memblock.reserved.total_size); 19346ed311b2SBenjamin Herrenschmidt 19350262d9c8SHeiko Carstens memblock_dump(&memblock.memory); 19360262d9c8SHeiko Carstens memblock_dump(&memblock.reserved); 1937409efd4cSHeiko Carstens #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 193877649905SDavid Hildenbrand memblock_dump(&physmem); 1939409efd4cSHeiko Carstens #endif 19406ed311b2SBenjamin Herrenschmidt } 19416ed311b2SBenjamin Herrenschmidt 194287c55870SMike Rapoport void __init_memblock memblock_dump_all(void) 194387c55870SMike Rapoport { 194487c55870SMike Rapoport if (memblock_debug) 194587c55870SMike Rapoport __memblock_dump_all(); 194687c55870SMike Rapoport } 194787c55870SMike Rapoport 19481aadc056STejun Heo void __init memblock_allow_resize(void) 19496ed311b2SBenjamin Herrenschmidt { 1950142b45a7SBenjamin Herrenschmidt memblock_can_resize = 1; 19516ed311b2SBenjamin Herrenschmidt } 19526ed311b2SBenjamin Herrenschmidt 19536ed311b2SBenjamin Herrenschmidt static int __init early_memblock(char *p) 19546ed311b2SBenjamin Herrenschmidt { 19556ed311b2SBenjamin Herrenschmidt if (p && strstr(p, "debug")) 19566ed311b2SBenjamin Herrenschmidt memblock_debug = 1; 19576ed311b2SBenjamin Herrenschmidt return 0; 19586ed311b2SBenjamin Herrenschmidt } 19596ed311b2SBenjamin Herrenschmidt early_param("memblock", early_memblock); 19606ed311b2SBenjamin Herrenschmidt 19614f5b0c17SMike Rapoport static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 19624f5b0c17SMike Rapoport { 19634f5b0c17SMike Rapoport struct page *start_pg, *end_pg; 19644f5b0c17SMike Rapoport phys_addr_t pg, pgend; 19654f5b0c17SMike Rapoport 19664f5b0c17SMike Rapoport /* 19674f5b0c17SMike Rapoport * Convert start_pfn/end_pfn to a struct page pointer. 19684f5b0c17SMike Rapoport */ 19694f5b0c17SMike Rapoport start_pg = pfn_to_page(start_pfn - 1) + 1; 19704f5b0c17SMike Rapoport end_pg = pfn_to_page(end_pfn - 1) + 1; 19714f5b0c17SMike Rapoport 19724f5b0c17SMike Rapoport /* 19734f5b0c17SMike Rapoport * Convert to physical addresses, and round start upwards and end 19744f5b0c17SMike Rapoport * downwards. 19754f5b0c17SMike Rapoport */ 19764f5b0c17SMike Rapoport pg = PAGE_ALIGN(__pa(start_pg)); 19774f5b0c17SMike Rapoport pgend = __pa(end_pg) & PAGE_MASK; 19784f5b0c17SMike Rapoport 19794f5b0c17SMike Rapoport /* 19804f5b0c17SMike Rapoport * If there are free pages between these, free the section of the 19814f5b0c17SMike Rapoport * memmap array. 19824f5b0c17SMike Rapoport */ 19834f5b0c17SMike Rapoport if (pg < pgend) 19843ecc6834SMike Rapoport memblock_phys_free(pg, pgend - pg); 19854f5b0c17SMike Rapoport } 19864f5b0c17SMike Rapoport 19874f5b0c17SMike Rapoport /* 19884f5b0c17SMike Rapoport * The mem_map array can get very big. Free the unused area of the memory map. 19894f5b0c17SMike Rapoport */ 19904f5b0c17SMike Rapoport static void __init free_unused_memmap(void) 19914f5b0c17SMike Rapoport { 19924f5b0c17SMike Rapoport unsigned long start, end, prev_end = 0; 19934f5b0c17SMike Rapoport int i; 19944f5b0c17SMike Rapoport 19954f5b0c17SMike Rapoport if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 19964f5b0c17SMike Rapoport IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 19974f5b0c17SMike Rapoport return; 19984f5b0c17SMike Rapoport 19994f5b0c17SMike Rapoport /* 20004f5b0c17SMike Rapoport * This relies on each bank being in address order. 20014f5b0c17SMike Rapoport * The banks are sorted previously in bootmem_init(). 20024f5b0c17SMike Rapoport */ 20034f5b0c17SMike Rapoport for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 20044f5b0c17SMike Rapoport #ifdef CONFIG_SPARSEMEM 20054f5b0c17SMike Rapoport /* 20064f5b0c17SMike Rapoport * Take care not to free memmap entries that don't exist 20074f5b0c17SMike Rapoport * due to SPARSEMEM sections which aren't present. 20084f5b0c17SMike Rapoport */ 20094f5b0c17SMike Rapoport start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 20104f5b0c17SMike Rapoport #endif 20114f5b0c17SMike Rapoport /* 2012e2a86800SMike Rapoport * Align down here since many operations in VM subsystem 2013e2a86800SMike Rapoport * presume that there are no holes in the memory map inside 2014e2a86800SMike Rapoport * a pageblock 20154f5b0c17SMike Rapoport */ 20164f9bc69aSKefeng Wang start = pageblock_start_pfn(start); 20174f5b0c17SMike Rapoport 20184f5b0c17SMike Rapoport /* 20194f5b0c17SMike Rapoport * If we had a previous bank, and there is a space 20204f5b0c17SMike Rapoport * between the current bank and the previous, free it. 20214f5b0c17SMike Rapoport */ 20224f5b0c17SMike Rapoport if (prev_end && prev_end < start) 20234f5b0c17SMike Rapoport free_memmap(prev_end, start); 20244f5b0c17SMike Rapoport 20254f5b0c17SMike Rapoport /* 2026e2a86800SMike Rapoport * Align up here since many operations in VM subsystem 2027e2a86800SMike Rapoport * presume that there are no holes in the memory map inside 2028e2a86800SMike Rapoport * a pageblock 20294f5b0c17SMike Rapoport */ 20305f7fa13fSKefeng Wang prev_end = pageblock_align(end); 20314f5b0c17SMike Rapoport } 20324f5b0c17SMike Rapoport 20334f5b0c17SMike Rapoport #ifdef CONFIG_SPARSEMEM 2034f921f53eSMike Rapoport if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 20355f7fa13fSKefeng Wang prev_end = pageblock_align(end); 20364f5b0c17SMike Rapoport free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 2037f921f53eSMike Rapoport } 20384f5b0c17SMike Rapoport #endif 20394f5b0c17SMike Rapoport } 20404f5b0c17SMike Rapoport 2041bda49a81SMike Rapoport static void __init __free_pages_memory(unsigned long start, unsigned long end) 2042bda49a81SMike Rapoport { 2043bda49a81SMike Rapoport int order; 2044bda49a81SMike Rapoport 2045bda49a81SMike Rapoport while (start < end) { 204659f876fbSKirill A. Shutemov /* 204759f876fbSKirill A. Shutemov * Free the pages in the largest chunks alignment allows. 204859f876fbSKirill A. Shutemov * 204959f876fbSKirill A. Shutemov * __ffs() behaviour is undefined for 0. start == 0 is 205059f876fbSKirill A. Shutemov * MAX_ORDER-aligned, set order to MAX_ORDER for the case. 205159f876fbSKirill A. Shutemov */ 205259f876fbSKirill A. Shutemov if (start) 205323baf831SKirill A. Shutemov order = min_t(int, MAX_ORDER, __ffs(start)); 205459f876fbSKirill A. Shutemov else 205559f876fbSKirill A. Shutemov order = MAX_ORDER; 2056bda49a81SMike Rapoport 2057bda49a81SMike Rapoport while (start + (1UL << order) > end) 2058bda49a81SMike Rapoport order--; 2059bda49a81SMike Rapoport 2060bda49a81SMike Rapoport memblock_free_pages(pfn_to_page(start), start, order); 2061bda49a81SMike Rapoport 2062bda49a81SMike Rapoport start += (1UL << order); 2063bda49a81SMike Rapoport } 2064bda49a81SMike Rapoport } 2065bda49a81SMike Rapoport 2066bda49a81SMike Rapoport static unsigned long __init __free_memory_core(phys_addr_t start, 2067bda49a81SMike Rapoport phys_addr_t end) 2068bda49a81SMike Rapoport { 2069bda49a81SMike Rapoport unsigned long start_pfn = PFN_UP(start); 2070bda49a81SMike Rapoport unsigned long end_pfn = min_t(unsigned long, 2071bda49a81SMike Rapoport PFN_DOWN(end), max_low_pfn); 2072bda49a81SMike Rapoport 2073bda49a81SMike Rapoport if (start_pfn >= end_pfn) 2074bda49a81SMike Rapoport return 0; 2075bda49a81SMike Rapoport 2076bda49a81SMike Rapoport __free_pages_memory(start_pfn, end_pfn); 2077bda49a81SMike Rapoport 2078bda49a81SMike Rapoport return end_pfn - start_pfn; 2079bda49a81SMike Rapoport } 2080bda49a81SMike Rapoport 20819092d4f7SMike Rapoport static void __init memmap_init_reserved_pages(void) 20829092d4f7SMike Rapoport { 20839092d4f7SMike Rapoport struct memblock_region *region; 20849092d4f7SMike Rapoport phys_addr_t start, end; 2085*61167ad5SYajun Deng int nid; 20869092d4f7SMike Rapoport 2087*61167ad5SYajun Deng /* 2088*61167ad5SYajun Deng * set nid on all reserved pages and also treat struct 2089*61167ad5SYajun Deng * pages for the NOMAP regions as PageReserved 2090*61167ad5SYajun Deng */ 20919092d4f7SMike Rapoport for_each_mem_region(region) { 2092*61167ad5SYajun Deng nid = memblock_get_region_node(region); 20939092d4f7SMike Rapoport start = region->base; 20949092d4f7SMike Rapoport end = start + region->size; 2095*61167ad5SYajun Deng 2096*61167ad5SYajun Deng if (memblock_is_nomap(region)) 2097*61167ad5SYajun Deng reserve_bootmem_region(start, end, nid); 2098*61167ad5SYajun Deng 2099*61167ad5SYajun Deng memblock_set_node(start, end, &memblock.reserved, nid); 21009092d4f7SMike Rapoport } 2101*61167ad5SYajun Deng 2102*61167ad5SYajun Deng /* initialize struct pages for the reserved regions */ 2103*61167ad5SYajun Deng for_each_reserved_mem_region(region) { 2104*61167ad5SYajun Deng nid = memblock_get_region_node(region); 2105*61167ad5SYajun Deng start = region->base; 2106*61167ad5SYajun Deng end = start + region->size; 2107*61167ad5SYajun Deng 2108*61167ad5SYajun Deng reserve_bootmem_region(start, end, nid); 21099092d4f7SMike Rapoport } 21109092d4f7SMike Rapoport } 21119092d4f7SMike Rapoport 2112bda49a81SMike Rapoport static unsigned long __init free_low_memory_core_early(void) 2113bda49a81SMike Rapoport { 2114bda49a81SMike Rapoport unsigned long count = 0; 2115bda49a81SMike Rapoport phys_addr_t start, end; 2116bda49a81SMike Rapoport u64 i; 2117bda49a81SMike Rapoport 2118bda49a81SMike Rapoport memblock_clear_hotplug(0, -1); 2119bda49a81SMike Rapoport 21209092d4f7SMike Rapoport memmap_init_reserved_pages(); 2121bda49a81SMike Rapoport 2122bda49a81SMike Rapoport /* 2123bda49a81SMike Rapoport * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2124bda49a81SMike Rapoport * because in some case like Node0 doesn't have RAM installed 2125bda49a81SMike Rapoport * low ram will be on Node1 2126bda49a81SMike Rapoport */ 2127bda49a81SMike Rapoport for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2128bda49a81SMike Rapoport NULL) 2129bda49a81SMike Rapoport count += __free_memory_core(start, end); 2130bda49a81SMike Rapoport 2131bda49a81SMike Rapoport return count; 2132bda49a81SMike Rapoport } 2133bda49a81SMike Rapoport 2134bda49a81SMike Rapoport static int reset_managed_pages_done __initdata; 2135bda49a81SMike Rapoport 2136a668968fSHaifeng Xu static void __init reset_node_managed_pages(pg_data_t *pgdat) 2137bda49a81SMike Rapoport { 2138bda49a81SMike Rapoport struct zone *z; 2139bda49a81SMike Rapoport 2140bda49a81SMike Rapoport for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 21419705bea5SArun KS atomic_long_set(&z->managed_pages, 0); 2142bda49a81SMike Rapoport } 2143bda49a81SMike Rapoport 2144bda49a81SMike Rapoport void __init reset_all_zones_managed_pages(void) 2145bda49a81SMike Rapoport { 2146bda49a81SMike Rapoport struct pglist_data *pgdat; 2147bda49a81SMike Rapoport 2148bda49a81SMike Rapoport if (reset_managed_pages_done) 2149bda49a81SMike Rapoport return; 2150bda49a81SMike Rapoport 2151bda49a81SMike Rapoport for_each_online_pgdat(pgdat) 2152bda49a81SMike Rapoport reset_node_managed_pages(pgdat); 2153bda49a81SMike Rapoport 2154bda49a81SMike Rapoport reset_managed_pages_done = 1; 2155bda49a81SMike Rapoport } 2156bda49a81SMike Rapoport 2157bda49a81SMike Rapoport /** 2158bda49a81SMike Rapoport * memblock_free_all - release free pages to the buddy allocator 2159bda49a81SMike Rapoport */ 2160097d43d8SDaeseok Youn void __init memblock_free_all(void) 2161bda49a81SMike Rapoport { 2162bda49a81SMike Rapoport unsigned long pages; 2163bda49a81SMike Rapoport 21644f5b0c17SMike Rapoport free_unused_memmap(); 2165bda49a81SMike Rapoport reset_all_zones_managed_pages(); 2166bda49a81SMike Rapoport 2167bda49a81SMike Rapoport pages = free_low_memory_core_early(); 2168ca79b0c2SArun KS totalram_pages_add(pages); 2169bda49a81SMike Rapoport } 2170bda49a81SMike Rapoport 2171350e88baSMike Rapoport #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 21726d03b885SBenjamin Herrenschmidt 21736d03b885SBenjamin Herrenschmidt static int memblock_debug_show(struct seq_file *m, void *private) 21746d03b885SBenjamin Herrenschmidt { 21756d03b885SBenjamin Herrenschmidt struct memblock_type *type = m->private; 21766d03b885SBenjamin Herrenschmidt struct memblock_region *reg; 21776d03b885SBenjamin Herrenschmidt int i; 21785d63f81cSMiles Chen phys_addr_t end; 21796d03b885SBenjamin Herrenschmidt 21806d03b885SBenjamin Herrenschmidt for (i = 0; i < type->cnt; i++) { 21816d03b885SBenjamin Herrenschmidt reg = &type->regions[i]; 21825d63f81cSMiles Chen end = reg->base + reg->size - 1; 21836d03b885SBenjamin Herrenschmidt 21845d63f81cSMiles Chen seq_printf(m, "%4d: ", i); 21855d63f81cSMiles Chen seq_printf(m, "%pa..%pa\n", ®->base, &end); 21866d03b885SBenjamin Herrenschmidt } 21876d03b885SBenjamin Herrenschmidt return 0; 21886d03b885SBenjamin Herrenschmidt } 21895ad35093SAndy Shevchenko DEFINE_SHOW_ATTRIBUTE(memblock_debug); 21906d03b885SBenjamin Herrenschmidt 21916d03b885SBenjamin Herrenschmidt static int __init memblock_init_debugfs(void) 21926d03b885SBenjamin Herrenschmidt { 21936d03b885SBenjamin Herrenschmidt struct dentry *root = debugfs_create_dir("memblock", NULL); 2194d9f7979cSGreg Kroah-Hartman 21950825a6f9SJoe Perches debugfs_create_file("memory", 0444, root, 21960825a6f9SJoe Perches &memblock.memory, &memblock_debug_fops); 21970825a6f9SJoe Perches debugfs_create_file("reserved", 0444, root, 21980825a6f9SJoe Perches &memblock.reserved, &memblock_debug_fops); 219970210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 220077649905SDavid Hildenbrand debugfs_create_file("physmem", 0444, root, &physmem, 220177649905SDavid Hildenbrand &memblock_debug_fops); 220270210ed9SPhilipp Hachtmann #endif 22036d03b885SBenjamin Herrenschmidt 22046d03b885SBenjamin Herrenschmidt return 0; 22056d03b885SBenjamin Herrenschmidt } 22066d03b885SBenjamin Herrenschmidt __initcall(memblock_init_debugfs); 22076d03b885SBenjamin Herrenschmidt 22086d03b885SBenjamin Herrenschmidt #endif /* CONFIG_DEBUG_FS */ 2209