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 323e039c5cSMike Rapoport /** 333e039c5cSMike Rapoport * DOC: memblock overview 343e039c5cSMike Rapoport * 353e039c5cSMike Rapoport * Memblock is a method of managing memory regions during the early 363e039c5cSMike Rapoport * boot period when the usual kernel memory allocators are not up and 373e039c5cSMike Rapoport * running. 383e039c5cSMike Rapoport * 393e039c5cSMike Rapoport * Memblock views the system memory as collections of contiguous 403e039c5cSMike Rapoport * regions. There are several types of these collections: 413e039c5cSMike Rapoport * 423e039c5cSMike Rapoport * * ``memory`` - describes the physical memory available to the 433e039c5cSMike Rapoport * kernel; this may differ from the actual physical memory installed 443e039c5cSMike Rapoport * in the system, for instance when the memory is restricted with 453e039c5cSMike Rapoport * ``mem=`` command line parameter 463e039c5cSMike Rapoport * * ``reserved`` - describes the regions that were allocated 4777649905SDavid Hildenbrand * * ``physmem`` - describes the actual physical memory available during 4877649905SDavid Hildenbrand * boot regardless of the possible restrictions and memory hot(un)plug; 4977649905SDavid Hildenbrand * the ``physmem`` type is only available on some architectures. 503e039c5cSMike Rapoport * 519303c9d5SMauro Carvalho Chehab * Each region is represented by struct memblock_region that 523e039c5cSMike Rapoport * defines the region extents, its attributes and NUMA node id on NUMA 531bf162e4SMauro Carvalho Chehab * systems. Every memory type is described by the struct memblock_type 541bf162e4SMauro Carvalho Chehab * which contains an array of memory regions along with 5577649905SDavid Hildenbrand * the allocator metadata. The "memory" and "reserved" types are nicely 569303c9d5SMauro Carvalho Chehab * wrapped with struct memblock. This structure is statically 5777649905SDavid Hildenbrand * initialized at build time. The region arrays are initially sized to 5877649905SDavid Hildenbrand * %INIT_MEMBLOCK_REGIONS for "memory" and %INIT_MEMBLOCK_RESERVED_REGIONS 5977649905SDavid Hildenbrand * for "reserved". The region array for "physmem" is initially sized to 6077649905SDavid Hildenbrand * %INIT_PHYSMEM_REGIONS. 616e5af9a8SCao jin * The memblock_allow_resize() enables automatic resizing of the region 626e5af9a8SCao jin * arrays during addition of new regions. This feature should be used 636e5af9a8SCao jin * with care so that memory allocated for the region array will not 646e5af9a8SCao jin * overlap with areas that should be reserved, for example initrd. 653e039c5cSMike Rapoport * 663e039c5cSMike Rapoport * The early architecture setup should tell memblock what the physical 676e5af9a8SCao jin * memory layout is by using memblock_add() or memblock_add_node() 686e5af9a8SCao jin * functions. The first function does not assign the region to a NUMA 696e5af9a8SCao jin * node and it is appropriate for UMA systems. Yet, it is possible to 706e5af9a8SCao jin * use it on NUMA systems as well and assign the region to a NUMA node 716e5af9a8SCao jin * later in the setup process using memblock_set_node(). The 726e5af9a8SCao jin * memblock_add_node() performs such an assignment directly. 733e039c5cSMike Rapoport * 74a2974133SMike Rapoport * Once memblock is setup the memory can be allocated using one of the 75a2974133SMike Rapoport * API variants: 76a2974133SMike Rapoport * 776e5af9a8SCao jin * * memblock_phys_alloc*() - these functions return the **physical** 786e5af9a8SCao jin * address of the allocated memory 796e5af9a8SCao jin * * memblock_alloc*() - these functions return the **virtual** address 806e5af9a8SCao jin * of the allocated memory. 81a2974133SMike Rapoport * 82df1758d9SEthon Paul * Note, that both API variants use implicit assumptions about allowed 83a2974133SMike Rapoport * memory ranges and the fallback methods. Consult the documentation 846e5af9a8SCao jin * of memblock_alloc_internal() and memblock_alloc_range_nid() 856e5af9a8SCao jin * functions for more elaborate description. 863e039c5cSMike Rapoport * 876e5af9a8SCao jin * As the system boot progresses, the architecture specific mem_init() 886e5af9a8SCao jin * function frees all the memory to the buddy page allocator. 893e039c5cSMike Rapoport * 906e5af9a8SCao jin * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the 9177649905SDavid Hildenbrand * memblock data structures (except "physmem") will be discarded after the 9277649905SDavid Hildenbrand * system initialization completes. 933e039c5cSMike Rapoport */ 943e039c5cSMike Rapoport 95a9ee6cf5SMike Rapoport #ifndef CONFIG_NUMA 96bda49a81SMike Rapoport struct pglist_data __refdata contig_page_data; 97bda49a81SMike Rapoport EXPORT_SYMBOL(contig_page_data); 98bda49a81SMike Rapoport #endif 99bda49a81SMike Rapoport 100bda49a81SMike Rapoport unsigned long max_low_pfn; 101bda49a81SMike Rapoport unsigned long min_low_pfn; 102bda49a81SMike Rapoport unsigned long max_pfn; 103bda49a81SMike Rapoport unsigned long long max_possible_pfn; 104bda49a81SMike Rapoport 105fe091c20STejun Heo static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; 1068a5b403dSArd Biesheuvel static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock; 10770210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 10877649905SDavid Hildenbrand static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS]; 10970210ed9SPhilipp Hachtmann #endif 110fe091c20STejun Heo 111fe091c20STejun Heo struct memblock memblock __initdata_memblock = { 112fe091c20STejun Heo .memory.regions = memblock_memory_init_regions, 113fe091c20STejun Heo .memory.cnt = 1, /* empty dummy entry */ 114fe091c20STejun Heo .memory.max = INIT_MEMBLOCK_REGIONS, 1150262d9c8SHeiko Carstens .memory.name = "memory", 116fe091c20STejun Heo 117fe091c20STejun Heo .reserved.regions = memblock_reserved_init_regions, 118fe091c20STejun Heo .reserved.cnt = 1, /* empty dummy entry */ 1198a5b403dSArd Biesheuvel .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS, 1200262d9c8SHeiko Carstens .reserved.name = "reserved", 121fe091c20STejun Heo 12279442ed1STang Chen .bottom_up = false, 123fe091c20STejun Heo .current_limit = MEMBLOCK_ALLOC_ANYWHERE, 124fe091c20STejun Heo }; 12595f72d1eSYinghai Lu 12677649905SDavid Hildenbrand #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 12777649905SDavid Hildenbrand struct memblock_type physmem = { 12877649905SDavid Hildenbrand .regions = memblock_physmem_init_regions, 12977649905SDavid Hildenbrand .cnt = 1, /* empty dummy entry */ 13077649905SDavid Hildenbrand .max = INIT_PHYSMEM_REGIONS, 13177649905SDavid Hildenbrand .name = "physmem", 13277649905SDavid Hildenbrand }; 13377649905SDavid Hildenbrand #endif 13477649905SDavid Hildenbrand 1359f3d5eaaSMike Rapoport /* 1369f3d5eaaSMike Rapoport * keep a pointer to &memblock.memory in the text section to use it in 1379f3d5eaaSMike Rapoport * __next_mem_range() and its helpers. 1389f3d5eaaSMike Rapoport * For architectures that do not keep memblock data after init, this 1399f3d5eaaSMike Rapoport * pointer will be reset to NULL at memblock_discard() 1409f3d5eaaSMike Rapoport */ 1419f3d5eaaSMike Rapoport static __refdata struct memblock_type *memblock_memory = &memblock.memory; 1429f3d5eaaSMike Rapoport 143cd991db8SMike Rapoport #define for_each_memblock_type(i, memblock_type, rgn) \ 144cd991db8SMike Rapoport for (i = 0, rgn = &memblock_type->regions[0]; \ 145cd991db8SMike Rapoport i < memblock_type->cnt; \ 146cd991db8SMike Rapoport i++, rgn = &memblock_type->regions[i]) 147cd991db8SMike Rapoport 14887c55870SMike Rapoport #define memblock_dbg(fmt, ...) \ 14987c55870SMike Rapoport do { \ 15087c55870SMike Rapoport if (memblock_debug) \ 15187c55870SMike Rapoport pr_info(fmt, ##__VA_ARGS__); \ 15287c55870SMike Rapoport } while (0) 15387c55870SMike Rapoport 15487c55870SMike Rapoport static int memblock_debug __initdata_memblock; 155a3f5bafcSTony Luck static bool system_has_some_mirror __initdata_memblock = false; 1561aadc056STejun Heo static int memblock_can_resize __initdata_memblock; 157181eb394SGavin Shan static int memblock_memory_in_slab __initdata_memblock = 0; 158181eb394SGavin Shan static int memblock_reserved_in_slab __initdata_memblock = 0; 15995f72d1eSYinghai Lu 160c366ea89SMike Rapoport static enum memblock_flags __init_memblock choose_memblock_flags(void) 161a3f5bafcSTony Luck { 162a3f5bafcSTony Luck return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE; 163a3f5bafcSTony Luck } 164a3f5bafcSTony Luck 165eb18f1b5STejun Heo /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ 166eb18f1b5STejun Heo static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) 167eb18f1b5STejun Heo { 1681c4bc43dSStefan Agner return *size = min(*size, PHYS_ADDR_MAX - base); 169eb18f1b5STejun Heo } 170eb18f1b5STejun Heo 1716ed311b2SBenjamin Herrenschmidt /* 1726ed311b2SBenjamin Herrenschmidt * Address comparison utilities 1736ed311b2SBenjamin Herrenschmidt */ 17410d06439SYinghai Lu static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, 1752898cc4cSBenjamin Herrenschmidt phys_addr_t base2, phys_addr_t size2) 17695f72d1eSYinghai Lu { 17795f72d1eSYinghai Lu return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); 17895f72d1eSYinghai Lu } 17995f72d1eSYinghai Lu 18095cf82ecSTang Chen bool __init_memblock memblock_overlaps_region(struct memblock_type *type, 1812d7d3eb2SH Hartley Sweeten phys_addr_t base, phys_addr_t size) 1826ed311b2SBenjamin Herrenschmidt { 1836ed311b2SBenjamin Herrenschmidt unsigned long i; 1846ed311b2SBenjamin Herrenschmidt 185023accf5SMike Rapoport memblock_cap_size(base, &size); 186023accf5SMike Rapoport 187f14516fbSAlexander Kuleshov for (i = 0; i < type->cnt; i++) 188f14516fbSAlexander Kuleshov if (memblock_addrs_overlap(base, size, type->regions[i].base, 189f14516fbSAlexander Kuleshov type->regions[i].size)) 1906ed311b2SBenjamin Herrenschmidt break; 191c5c5c9d1STang Chen return i < type->cnt; 1926ed311b2SBenjamin Herrenschmidt } 1936ed311b2SBenjamin Herrenschmidt 19447cec443SMike Rapoport /** 19579442ed1STang Chen * __memblock_find_range_bottom_up - find free area utility in bottom-up 19679442ed1STang Chen * @start: start of candidate range 19747cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 19847cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 19979442ed1STang Chen * @size: size of free area to find 20079442ed1STang Chen * @align: alignment of free area to find 201b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 202fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 20379442ed1STang Chen * 20479442ed1STang Chen * Utility called from memblock_find_in_range_node(), find free area bottom-up. 20579442ed1STang Chen * 20647cec443SMike Rapoport * Return: 20779442ed1STang Chen * Found address on success, 0 on failure. 20879442ed1STang Chen */ 20979442ed1STang Chen static phys_addr_t __init_memblock 21079442ed1STang Chen __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, 211fc6daaf9STony Luck phys_addr_t size, phys_addr_t align, int nid, 212e1720feeSMike Rapoport enum memblock_flags flags) 21379442ed1STang Chen { 21479442ed1STang Chen phys_addr_t this_start, this_end, cand; 21579442ed1STang Chen u64 i; 21679442ed1STang Chen 217fc6daaf9STony Luck for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) { 21879442ed1STang Chen this_start = clamp(this_start, start, end); 21979442ed1STang Chen this_end = clamp(this_end, start, end); 22079442ed1STang Chen 22179442ed1STang Chen cand = round_up(this_start, align); 22279442ed1STang Chen if (cand < this_end && this_end - cand >= size) 22379442ed1STang Chen return cand; 22479442ed1STang Chen } 22579442ed1STang Chen 22679442ed1STang Chen return 0; 22779442ed1STang Chen } 22879442ed1STang Chen 2297bd0b0f0STejun Heo /** 2301402899eSTang Chen * __memblock_find_range_top_down - find free area utility, in top-down 2311402899eSTang Chen * @start: start of candidate range 23247cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 23347cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 2341402899eSTang Chen * @size: size of free area to find 2351402899eSTang Chen * @align: alignment of free area to find 236b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 237fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 2381402899eSTang Chen * 2391402899eSTang Chen * Utility called from memblock_find_in_range_node(), find free area top-down. 2401402899eSTang Chen * 24147cec443SMike Rapoport * Return: 24279442ed1STang Chen * Found address on success, 0 on failure. 2431402899eSTang Chen */ 2441402899eSTang Chen static phys_addr_t __init_memblock 2451402899eSTang Chen __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, 246fc6daaf9STony Luck phys_addr_t size, phys_addr_t align, int nid, 247e1720feeSMike Rapoport enum memblock_flags flags) 2481402899eSTang Chen { 2491402899eSTang Chen phys_addr_t this_start, this_end, cand; 2501402899eSTang Chen u64 i; 2511402899eSTang Chen 252fc6daaf9STony Luck for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end, 253fc6daaf9STony Luck NULL) { 2541402899eSTang Chen this_start = clamp(this_start, start, end); 2551402899eSTang Chen this_end = clamp(this_end, start, end); 2561402899eSTang Chen 2571402899eSTang Chen if (this_end < size) 2581402899eSTang Chen continue; 2591402899eSTang Chen 2601402899eSTang Chen cand = round_down(this_end - size, align); 2611402899eSTang Chen if (cand >= this_start) 2621402899eSTang Chen return cand; 2631402899eSTang Chen } 2641402899eSTang Chen 2651402899eSTang Chen return 0; 2661402899eSTang Chen } 2671402899eSTang Chen 2681402899eSTang Chen /** 2697bd0b0f0STejun Heo * memblock_find_in_range_node - find free area in given range and node 2707bd0b0f0STejun Heo * @size: size of free area to find 2717bd0b0f0STejun Heo * @align: alignment of free area to find 27287029ee9SGrygorii Strashko * @start: start of candidate range 27347cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 27447cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 275b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 276fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 2777bd0b0f0STejun Heo * 2787bd0b0f0STejun Heo * Find @size free area aligned to @align in the specified range and node. 2797bd0b0f0STejun Heo * 28047cec443SMike Rapoport * Return: 28179442ed1STang Chen * Found address on success, 0 on failure. 2826ed311b2SBenjamin Herrenschmidt */ 283c366ea89SMike Rapoport static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, 28487029ee9SGrygorii Strashko phys_addr_t align, phys_addr_t start, 285e1720feeSMike Rapoport phys_addr_t end, int nid, 286e1720feeSMike Rapoport enum memblock_flags flags) 287f7210e6cSTang Chen { 288f7210e6cSTang Chen /* pump up @end */ 289fed84c78SQian Cai if (end == MEMBLOCK_ALLOC_ACCESSIBLE || 290fed84c78SQian Cai end == MEMBLOCK_ALLOC_KASAN) 291f7210e6cSTang Chen end = memblock.current_limit; 292f7210e6cSTang Chen 293f7210e6cSTang Chen /* avoid allocating the first page */ 294f7210e6cSTang Chen start = max_t(phys_addr_t, start, PAGE_SIZE); 295f7210e6cSTang Chen end = max(start, end); 29679442ed1STang Chen 2972dcb3964SRoman Gushchin if (memblock_bottom_up()) 2982dcb3964SRoman Gushchin return __memblock_find_range_bottom_up(start, end, size, align, 2992dcb3964SRoman Gushchin nid, flags); 3002dcb3964SRoman Gushchin else 3012dcb3964SRoman Gushchin return __memblock_find_range_top_down(start, end, size, align, 3022dcb3964SRoman Gushchin nid, flags); 303f7210e6cSTang Chen } 3046ed311b2SBenjamin Herrenschmidt 3057bd0b0f0STejun Heo /** 3067bd0b0f0STejun Heo * memblock_find_in_range - find free area in given range 3077bd0b0f0STejun Heo * @start: start of candidate range 30847cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 30947cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 3107bd0b0f0STejun Heo * @size: size of free area to find 3117bd0b0f0STejun Heo * @align: alignment of free area to find 3127bd0b0f0STejun Heo * 3137bd0b0f0STejun Heo * Find @size free area aligned to @align in the specified range. 3147bd0b0f0STejun Heo * 31547cec443SMike Rapoport * Return: 31679442ed1STang Chen * Found address on success, 0 on failure. 3177bd0b0f0STejun Heo */ 318*a7259df7SMike Rapoport static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, 3197bd0b0f0STejun Heo phys_addr_t end, phys_addr_t size, 3207bd0b0f0STejun Heo phys_addr_t align) 3217bd0b0f0STejun Heo { 322a3f5bafcSTony Luck phys_addr_t ret; 323e1720feeSMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 324a3f5bafcSTony Luck 325a3f5bafcSTony Luck again: 326a3f5bafcSTony Luck ret = memblock_find_in_range_node(size, align, start, end, 327a3f5bafcSTony Luck NUMA_NO_NODE, flags); 328a3f5bafcSTony Luck 329a3f5bafcSTony Luck if (!ret && (flags & MEMBLOCK_MIRROR)) { 330a3f5bafcSTony Luck pr_warn("Could not allocate %pap bytes of mirrored memory\n", 331a3f5bafcSTony Luck &size); 332a3f5bafcSTony Luck flags &= ~MEMBLOCK_MIRROR; 333a3f5bafcSTony Luck goto again; 334a3f5bafcSTony Luck } 335a3f5bafcSTony Luck 336a3f5bafcSTony Luck return ret; 3377bd0b0f0STejun Heo } 3387bd0b0f0STejun Heo 33910d06439SYinghai Lu static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) 34095f72d1eSYinghai Lu { 3411440c4e2STejun Heo type->total_size -= type->regions[r].size; 3427c0caeb8STejun Heo memmove(&type->regions[r], &type->regions[r + 1], 3437c0caeb8STejun Heo (type->cnt - (r + 1)) * sizeof(type->regions[r])); 344e3239ff9SBenjamin Herrenschmidt type->cnt--; 34595f72d1eSYinghai Lu 3468f7a6605SBenjamin Herrenschmidt /* Special case for empty arrays */ 3478f7a6605SBenjamin Herrenschmidt if (type->cnt == 0) { 3481440c4e2STejun Heo WARN_ON(type->total_size != 0); 3498f7a6605SBenjamin Herrenschmidt type->cnt = 1; 3508f7a6605SBenjamin Herrenschmidt type->regions[0].base = 0; 3518f7a6605SBenjamin Herrenschmidt type->regions[0].size = 0; 35266a20757STang Chen type->regions[0].flags = 0; 3537c0caeb8STejun Heo memblock_set_region_node(&type->regions[0], MAX_NUMNODES); 3548f7a6605SBenjamin Herrenschmidt } 35595f72d1eSYinghai Lu } 35695f72d1eSYinghai Lu 357350e88baSMike Rapoport #ifndef CONFIG_ARCH_KEEP_MEMBLOCK 3583010f876SPavel Tatashin /** 35947cec443SMike Rapoport * memblock_discard - discard memory and reserved arrays if they were allocated 3603010f876SPavel Tatashin */ 3613010f876SPavel Tatashin void __init memblock_discard(void) 36229f67386SYinghai Lu { 3633010f876SPavel Tatashin phys_addr_t addr, size; 36429f67386SYinghai Lu 3653010f876SPavel Tatashin if (memblock.reserved.regions != memblock_reserved_init_regions) { 3663010f876SPavel Tatashin addr = __pa(memblock.reserved.regions); 3673010f876SPavel Tatashin size = PAGE_ALIGN(sizeof(struct memblock_region) * 36829f67386SYinghai Lu memblock.reserved.max); 3693010f876SPavel Tatashin __memblock_free_late(addr, size); 37029f67386SYinghai Lu } 37129f67386SYinghai Lu 37291b540f9SPavel Tatashin if (memblock.memory.regions != memblock_memory_init_regions) { 3733010f876SPavel Tatashin addr = __pa(memblock.memory.regions); 3743010f876SPavel Tatashin size = PAGE_ALIGN(sizeof(struct memblock_region) * 3755e270e25SPhilipp Hachtmann memblock.memory.max); 3763010f876SPavel Tatashin __memblock_free_late(addr, size); 3775e270e25SPhilipp Hachtmann } 3789f3d5eaaSMike Rapoport 3799f3d5eaaSMike Rapoport memblock_memory = NULL; 3803010f876SPavel Tatashin } 3815e270e25SPhilipp Hachtmann #endif 3825e270e25SPhilipp Hachtmann 38348c3b583SGreg Pearson /** 38448c3b583SGreg Pearson * memblock_double_array - double the size of the memblock regions array 38548c3b583SGreg Pearson * @type: memblock type of the regions array being doubled 38648c3b583SGreg Pearson * @new_area_start: starting address of memory range to avoid overlap with 38748c3b583SGreg Pearson * @new_area_size: size of memory range to avoid overlap with 38848c3b583SGreg Pearson * 38948c3b583SGreg Pearson * Double the size of the @type regions array. If memblock is being used to 39048c3b583SGreg Pearson * allocate memory for a new reserved regions array and there is a previously 39148c3b583SGreg Pearson * allocated memory range [@new_area_start, @new_area_start + @new_area_size] 39248c3b583SGreg Pearson * waiting to be reserved, ensure the memory used by the new array does 39348c3b583SGreg Pearson * not overlap. 39448c3b583SGreg Pearson * 39547cec443SMike Rapoport * Return: 39648c3b583SGreg Pearson * 0 on success, -1 on failure. 39748c3b583SGreg Pearson */ 39848c3b583SGreg Pearson static int __init_memblock memblock_double_array(struct memblock_type *type, 39948c3b583SGreg Pearson phys_addr_t new_area_start, 40048c3b583SGreg Pearson phys_addr_t new_area_size) 401142b45a7SBenjamin Herrenschmidt { 402142b45a7SBenjamin Herrenschmidt struct memblock_region *new_array, *old_array; 40329f67386SYinghai Lu phys_addr_t old_alloc_size, new_alloc_size; 404a36aab89SMike Rapoport phys_addr_t old_size, new_size, addr, new_end; 405142b45a7SBenjamin Herrenschmidt int use_slab = slab_is_available(); 406181eb394SGavin Shan int *in_slab; 407142b45a7SBenjamin Herrenschmidt 408142b45a7SBenjamin Herrenschmidt /* We don't allow resizing until we know about the reserved regions 409142b45a7SBenjamin Herrenschmidt * of memory that aren't suitable for allocation 410142b45a7SBenjamin Herrenschmidt */ 411142b45a7SBenjamin Herrenschmidt if (!memblock_can_resize) 412142b45a7SBenjamin Herrenschmidt return -1; 413142b45a7SBenjamin Herrenschmidt 414142b45a7SBenjamin Herrenschmidt /* Calculate new doubled size */ 415142b45a7SBenjamin Herrenschmidt old_size = type->max * sizeof(struct memblock_region); 416142b45a7SBenjamin Herrenschmidt new_size = old_size << 1; 41729f67386SYinghai Lu /* 41829f67386SYinghai Lu * We need to allocated new one align to PAGE_SIZE, 41929f67386SYinghai Lu * so we can free them completely later. 42029f67386SYinghai Lu */ 42129f67386SYinghai Lu old_alloc_size = PAGE_ALIGN(old_size); 42229f67386SYinghai Lu new_alloc_size = PAGE_ALIGN(new_size); 423142b45a7SBenjamin Herrenschmidt 424181eb394SGavin Shan /* Retrieve the slab flag */ 425181eb394SGavin Shan if (type == &memblock.memory) 426181eb394SGavin Shan in_slab = &memblock_memory_in_slab; 427181eb394SGavin Shan else 428181eb394SGavin Shan in_slab = &memblock_reserved_in_slab; 429181eb394SGavin Shan 430a2974133SMike Rapoport /* Try to find some space for it */ 431142b45a7SBenjamin Herrenschmidt if (use_slab) { 432142b45a7SBenjamin Herrenschmidt new_array = kmalloc(new_size, GFP_KERNEL); 4331f5026a7STejun Heo addr = new_array ? __pa(new_array) : 0; 4344e2f0775SGavin Shan } else { 43548c3b583SGreg Pearson /* only exclude range when trying to double reserved.regions */ 43648c3b583SGreg Pearson if (type != &memblock.reserved) 43748c3b583SGreg Pearson new_area_start = new_area_size = 0; 43848c3b583SGreg Pearson 43948c3b583SGreg Pearson addr = memblock_find_in_range(new_area_start + new_area_size, 44048c3b583SGreg Pearson memblock.current_limit, 44129f67386SYinghai Lu new_alloc_size, PAGE_SIZE); 44248c3b583SGreg Pearson if (!addr && new_area_size) 44348c3b583SGreg Pearson addr = memblock_find_in_range(0, 44448c3b583SGreg Pearson min(new_area_start, memblock.current_limit), 44529f67386SYinghai Lu new_alloc_size, PAGE_SIZE); 44648c3b583SGreg Pearson 44715674868SSachin Kamat new_array = addr ? __va(addr) : NULL; 4484e2f0775SGavin Shan } 4491f5026a7STejun Heo if (!addr) { 450142b45a7SBenjamin Herrenschmidt pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 4510262d9c8SHeiko Carstens type->name, type->max, type->max * 2); 452142b45a7SBenjamin Herrenschmidt return -1; 453142b45a7SBenjamin Herrenschmidt } 454142b45a7SBenjamin Herrenschmidt 455a36aab89SMike Rapoport new_end = addr + new_size - 1; 456a36aab89SMike Rapoport memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]", 457a36aab89SMike Rapoport type->name, type->max * 2, &addr, &new_end); 458ea9e4376SYinghai Lu 459fd07383bSAndrew Morton /* 460fd07383bSAndrew Morton * Found space, we now need to move the array over before we add the 461fd07383bSAndrew Morton * reserved region since it may be our reserved array itself that is 462fd07383bSAndrew Morton * full. 463142b45a7SBenjamin Herrenschmidt */ 464142b45a7SBenjamin Herrenschmidt memcpy(new_array, type->regions, old_size); 465142b45a7SBenjamin Herrenschmidt memset(new_array + type->max, 0, old_size); 466142b45a7SBenjamin Herrenschmidt old_array = type->regions; 467142b45a7SBenjamin Herrenschmidt type->regions = new_array; 468142b45a7SBenjamin Herrenschmidt type->max <<= 1; 469142b45a7SBenjamin Herrenschmidt 470fd07383bSAndrew Morton /* Free old array. We needn't free it if the array is the static one */ 471181eb394SGavin Shan if (*in_slab) 472181eb394SGavin Shan kfree(old_array); 473181eb394SGavin Shan else if (old_array != memblock_memory_init_regions && 474142b45a7SBenjamin Herrenschmidt old_array != memblock_reserved_init_regions) 47529f67386SYinghai Lu memblock_free(__pa(old_array), old_alloc_size); 476142b45a7SBenjamin Herrenschmidt 477fd07383bSAndrew Morton /* 478fd07383bSAndrew Morton * Reserve the new array if that comes from the memblock. Otherwise, we 479fd07383bSAndrew Morton * needn't do it 480181eb394SGavin Shan */ 481181eb394SGavin Shan if (!use_slab) 48229f67386SYinghai Lu BUG_ON(memblock_reserve(addr, new_alloc_size)); 483181eb394SGavin Shan 484181eb394SGavin Shan /* Update slab flag */ 485181eb394SGavin Shan *in_slab = use_slab; 486181eb394SGavin Shan 487142b45a7SBenjamin Herrenschmidt return 0; 488142b45a7SBenjamin Herrenschmidt } 489142b45a7SBenjamin Herrenschmidt 490784656f9STejun Heo /** 491784656f9STejun Heo * memblock_merge_regions - merge neighboring compatible regions 492784656f9STejun Heo * @type: memblock type to scan 493784656f9STejun Heo * 494784656f9STejun Heo * Scan @type and merge neighboring compatible regions. 495784656f9STejun Heo */ 496784656f9STejun Heo static void __init_memblock memblock_merge_regions(struct memblock_type *type) 497784656f9STejun Heo { 498784656f9STejun Heo int i = 0; 499784656f9STejun Heo 500784656f9STejun Heo /* cnt never goes below 1 */ 501784656f9STejun Heo while (i < type->cnt - 1) { 502784656f9STejun Heo struct memblock_region *this = &type->regions[i]; 503784656f9STejun Heo struct memblock_region *next = &type->regions[i + 1]; 504784656f9STejun Heo 5057c0caeb8STejun Heo if (this->base + this->size != next->base || 5067c0caeb8STejun Heo memblock_get_region_node(this) != 50766a20757STang Chen memblock_get_region_node(next) || 50866a20757STang Chen this->flags != next->flags) { 509784656f9STejun Heo BUG_ON(this->base + this->size > next->base); 510784656f9STejun Heo i++; 511784656f9STejun Heo continue; 512784656f9STejun Heo } 513784656f9STejun Heo 514784656f9STejun Heo this->size += next->size; 515c0232ae8SLin Feng /* move forward from next + 1, index of which is i + 2 */ 516c0232ae8SLin Feng memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 517784656f9STejun Heo type->cnt--; 518784656f9STejun Heo } 519784656f9STejun Heo } 520784656f9STejun Heo 521784656f9STejun Heo /** 522784656f9STejun Heo * memblock_insert_region - insert new memblock region 523784656f9STejun Heo * @type: memblock type to insert into 524784656f9STejun Heo * @idx: index for the insertion point 525784656f9STejun Heo * @base: base address of the new region 526784656f9STejun Heo * @size: size of the new region 527209ff86dSTang Chen * @nid: node id of the new region 52866a20757STang Chen * @flags: flags of the new region 529784656f9STejun Heo * 530784656f9STejun Heo * Insert new memblock region [@base, @base + @size) into @type at @idx. 531412d0008SAlexander Kuleshov * @type must already have extra room to accommodate the new region. 532784656f9STejun Heo */ 533784656f9STejun Heo static void __init_memblock memblock_insert_region(struct memblock_type *type, 534784656f9STejun Heo int idx, phys_addr_t base, 53566a20757STang Chen phys_addr_t size, 536e1720feeSMike Rapoport int nid, 537e1720feeSMike Rapoport enum memblock_flags flags) 538784656f9STejun Heo { 539784656f9STejun Heo struct memblock_region *rgn = &type->regions[idx]; 540784656f9STejun Heo 541784656f9STejun Heo BUG_ON(type->cnt >= type->max); 542784656f9STejun Heo memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 543784656f9STejun Heo rgn->base = base; 544784656f9STejun Heo rgn->size = size; 54566a20757STang Chen rgn->flags = flags; 5467c0caeb8STejun Heo memblock_set_region_node(rgn, nid); 547784656f9STejun Heo type->cnt++; 5481440c4e2STejun Heo type->total_size += size; 549784656f9STejun Heo } 550784656f9STejun Heo 551784656f9STejun Heo /** 552f1af9d3aSPhilipp Hachtmann * memblock_add_range - add new memblock region 553784656f9STejun Heo * @type: memblock type to add new region into 554784656f9STejun Heo * @base: base address of the new region 555784656f9STejun Heo * @size: size of the new region 5567fb0bc3fSTejun Heo * @nid: nid of the new region 55766a20757STang Chen * @flags: flags of the new region 558784656f9STejun Heo * 559784656f9STejun Heo * Add new memblock region [@base, @base + @size) into @type. The new region 560784656f9STejun Heo * is allowed to overlap with existing ones - overlaps don't affect already 561784656f9STejun Heo * existing regions. @type is guaranteed to be minimal (all neighbouring 562784656f9STejun Heo * compatible regions are merged) after the addition. 563784656f9STejun Heo * 56447cec443SMike Rapoport * Return: 565784656f9STejun Heo * 0 on success, -errno on failure. 566784656f9STejun Heo */ 56702634a44SAnshuman Khandual static int __init_memblock memblock_add_range(struct memblock_type *type, 56866a20757STang Chen phys_addr_t base, phys_addr_t size, 569e1720feeSMike Rapoport int nid, enum memblock_flags flags) 57095f72d1eSYinghai Lu { 571784656f9STejun Heo bool insert = false; 572eb18f1b5STejun Heo phys_addr_t obase = base; 573eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 5748c9c1701SAlexander Kuleshov int idx, nr_new; 5758c9c1701SAlexander Kuleshov struct memblock_region *rgn; 57695f72d1eSYinghai Lu 577b3dc627cSTejun Heo if (!size) 578b3dc627cSTejun Heo return 0; 579b3dc627cSTejun Heo 580784656f9STejun Heo /* special case for empty array */ 581784656f9STejun Heo if (type->regions[0].size == 0) { 5821440c4e2STejun Heo WARN_ON(type->cnt != 1 || type->total_size); 583784656f9STejun Heo type->regions[0].base = base; 584784656f9STejun Heo type->regions[0].size = size; 58566a20757STang Chen type->regions[0].flags = flags; 5867fb0bc3fSTejun Heo memblock_set_region_node(&type->regions[0], nid); 5871440c4e2STejun Heo type->total_size = size; 588784656f9STejun Heo return 0; 589784656f9STejun Heo } 590784656f9STejun Heo repeat: 591784656f9STejun Heo /* 592784656f9STejun Heo * The following is executed twice. Once with %false @insert and 593784656f9STejun Heo * then with %true. The first counts the number of regions needed 594412d0008SAlexander Kuleshov * to accommodate the new area. The second actually inserts them. 595784656f9STejun Heo */ 596784656f9STejun Heo base = obase; 597784656f9STejun Heo nr_new = 0; 598784656f9STejun Heo 59966e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 600784656f9STejun Heo phys_addr_t rbase = rgn->base; 601784656f9STejun Heo phys_addr_t rend = rbase + rgn->size; 6028f7a6605SBenjamin Herrenschmidt 603784656f9STejun Heo if (rbase >= end) 6048f7a6605SBenjamin Herrenschmidt break; 605784656f9STejun Heo if (rend <= base) 606784656f9STejun Heo continue; 607784656f9STejun Heo /* 608784656f9STejun Heo * @rgn overlaps. If it separates the lower part of new 609784656f9STejun Heo * area, insert that portion. 6108f7a6605SBenjamin Herrenschmidt */ 611784656f9STejun Heo if (rbase > base) { 612a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 613c0a29498SWei Yang WARN_ON(nid != memblock_get_region_node(rgn)); 614c0a29498SWei Yang #endif 6154fcab5f4SWei Yang WARN_ON(flags != rgn->flags); 616784656f9STejun Heo nr_new++; 617784656f9STejun Heo if (insert) 6188c9c1701SAlexander Kuleshov memblock_insert_region(type, idx++, base, 61966a20757STang Chen rbase - base, nid, 62066a20757STang Chen flags); 621784656f9STejun Heo } 622784656f9STejun Heo /* area below @rend is dealt with, forget about it */ 623784656f9STejun Heo base = min(rend, end); 6248f7a6605SBenjamin Herrenschmidt } 6258f7a6605SBenjamin Herrenschmidt 626784656f9STejun Heo /* insert the remaining portion */ 627784656f9STejun Heo if (base < end) { 628784656f9STejun Heo nr_new++; 629784656f9STejun Heo if (insert) 6308c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, base, end - base, 63166a20757STang Chen nid, flags); 6328f7a6605SBenjamin Herrenschmidt } 6338f7a6605SBenjamin Herrenschmidt 634ef3cc4dbSnimisolo if (!nr_new) 635ef3cc4dbSnimisolo return 0; 636ef3cc4dbSnimisolo 637784656f9STejun Heo /* 638784656f9STejun Heo * If this was the first round, resize array and repeat for actual 639784656f9STejun Heo * insertions; otherwise, merge and return. 6408f7a6605SBenjamin Herrenschmidt */ 641784656f9STejun Heo if (!insert) { 642784656f9STejun Heo while (type->cnt + nr_new > type->max) 64348c3b583SGreg Pearson if (memblock_double_array(type, obase, size) < 0) 644784656f9STejun Heo return -ENOMEM; 645784656f9STejun Heo insert = true; 646784656f9STejun Heo goto repeat; 64795f72d1eSYinghai Lu } else { 648784656f9STejun Heo memblock_merge_regions(type); 64995f72d1eSYinghai Lu return 0; 65095f72d1eSYinghai Lu } 651784656f9STejun Heo } 65295f72d1eSYinghai Lu 65348a833ccSMike Rapoport /** 65448a833ccSMike Rapoport * memblock_add_node - add new memblock region within a NUMA node 65548a833ccSMike Rapoport * @base: base address of the new region 65648a833ccSMike Rapoport * @size: size of the new region 65748a833ccSMike Rapoport * @nid: nid of the new region 65848a833ccSMike Rapoport * 65948a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 66048a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 66148a833ccSMike Rapoport * 66248a833ccSMike Rapoport * Return: 66348a833ccSMike Rapoport * 0 on success, -errno on failure. 66448a833ccSMike Rapoport */ 6657fb0bc3fSTejun Heo int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 6667fb0bc3fSTejun Heo int nid) 6677fb0bc3fSTejun Heo { 668f1af9d3aSPhilipp Hachtmann return memblock_add_range(&memblock.memory, base, size, nid, 0); 6697fb0bc3fSTejun Heo } 6707fb0bc3fSTejun Heo 67148a833ccSMike Rapoport /** 67248a833ccSMike Rapoport * memblock_add - add new memblock region 67348a833ccSMike Rapoport * @base: base address of the new region 67448a833ccSMike Rapoport * @size: size of the new region 67548a833ccSMike Rapoport * 67648a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 67748a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 67848a833ccSMike Rapoport * 67948a833ccSMike Rapoport * Return: 68048a833ccSMike Rapoport * 0 on success, -errno on failure. 68148a833ccSMike Rapoport */ 682f705ac4bSAlexander Kuleshov int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 6836a4055bcSAlexander Kuleshov { 6845d63f81cSMiles Chen phys_addr_t end = base + size - 1; 6855d63f81cSMiles Chen 686a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 6875d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 6886a4055bcSAlexander Kuleshov 689f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 69095f72d1eSYinghai Lu } 69195f72d1eSYinghai Lu 6926a9ceb31STejun Heo /** 6936a9ceb31STejun Heo * memblock_isolate_range - isolate given range into disjoint memblocks 6946a9ceb31STejun Heo * @type: memblock type to isolate range for 6956a9ceb31STejun Heo * @base: base of range to isolate 6966a9ceb31STejun Heo * @size: size of range to isolate 6976a9ceb31STejun Heo * @start_rgn: out parameter for the start of isolated region 6986a9ceb31STejun Heo * @end_rgn: out parameter for the end of isolated region 6996a9ceb31STejun Heo * 7006a9ceb31STejun Heo * Walk @type and ensure that regions don't cross the boundaries defined by 7016a9ceb31STejun Heo * [@base, @base + @size). Crossing regions are split at the boundaries, 7026a9ceb31STejun Heo * which may create at most two more regions. The index of the first 7036a9ceb31STejun Heo * region inside the range is returned in *@start_rgn and end in *@end_rgn. 7046a9ceb31STejun Heo * 70547cec443SMike Rapoport * Return: 7066a9ceb31STejun Heo * 0 on success, -errno on failure. 7076a9ceb31STejun Heo */ 7086a9ceb31STejun Heo static int __init_memblock memblock_isolate_range(struct memblock_type *type, 7096a9ceb31STejun Heo phys_addr_t base, phys_addr_t size, 7106a9ceb31STejun Heo int *start_rgn, int *end_rgn) 7116a9ceb31STejun Heo { 712eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 7138c9c1701SAlexander Kuleshov int idx; 7148c9c1701SAlexander Kuleshov struct memblock_region *rgn; 7156a9ceb31STejun Heo 7166a9ceb31STejun Heo *start_rgn = *end_rgn = 0; 7176a9ceb31STejun Heo 718b3dc627cSTejun Heo if (!size) 719b3dc627cSTejun Heo return 0; 720b3dc627cSTejun Heo 7216a9ceb31STejun Heo /* we'll create at most two more regions */ 7226a9ceb31STejun Heo while (type->cnt + 2 > type->max) 72348c3b583SGreg Pearson if (memblock_double_array(type, base, size) < 0) 7246a9ceb31STejun Heo return -ENOMEM; 7256a9ceb31STejun Heo 72666e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 7276a9ceb31STejun Heo phys_addr_t rbase = rgn->base; 7286a9ceb31STejun Heo phys_addr_t rend = rbase + rgn->size; 7296a9ceb31STejun Heo 7306a9ceb31STejun Heo if (rbase >= end) 7316a9ceb31STejun Heo break; 7326a9ceb31STejun Heo if (rend <= base) 7336a9ceb31STejun Heo continue; 7346a9ceb31STejun Heo 7356a9ceb31STejun Heo if (rbase < base) { 7366a9ceb31STejun Heo /* 7376a9ceb31STejun Heo * @rgn intersects from below. Split and continue 7386a9ceb31STejun Heo * to process the next region - the new top half. 7396a9ceb31STejun Heo */ 7406a9ceb31STejun Heo rgn->base = base; 7411440c4e2STejun Heo rgn->size -= base - rbase; 7421440c4e2STejun Heo type->total_size -= base - rbase; 7438c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, rbase, base - rbase, 74466a20757STang Chen memblock_get_region_node(rgn), 74566a20757STang Chen rgn->flags); 7466a9ceb31STejun Heo } else if (rend > end) { 7476a9ceb31STejun Heo /* 7486a9ceb31STejun Heo * @rgn intersects from above. Split and redo the 7496a9ceb31STejun Heo * current region - the new bottom half. 7506a9ceb31STejun Heo */ 7516a9ceb31STejun Heo rgn->base = end; 7521440c4e2STejun Heo rgn->size -= end - rbase; 7531440c4e2STejun Heo type->total_size -= end - rbase; 7548c9c1701SAlexander Kuleshov memblock_insert_region(type, idx--, rbase, end - rbase, 75566a20757STang Chen memblock_get_region_node(rgn), 75666a20757STang Chen rgn->flags); 7576a9ceb31STejun Heo } else { 7586a9ceb31STejun Heo /* @rgn is fully contained, record it */ 7596a9ceb31STejun Heo if (!*end_rgn) 7608c9c1701SAlexander Kuleshov *start_rgn = idx; 7618c9c1701SAlexander Kuleshov *end_rgn = idx + 1; 7626a9ceb31STejun Heo } 7636a9ceb31STejun Heo } 7646a9ceb31STejun Heo 7656a9ceb31STejun Heo return 0; 7666a9ceb31STejun Heo } 7676a9ceb31STejun Heo 76835bd16a2SAlexander Kuleshov static int __init_memblock memblock_remove_range(struct memblock_type *type, 7698f7a6605SBenjamin Herrenschmidt phys_addr_t base, phys_addr_t size) 77095f72d1eSYinghai Lu { 77171936180STejun Heo int start_rgn, end_rgn; 77271936180STejun Heo int i, ret; 77395f72d1eSYinghai Lu 77471936180STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 77571936180STejun Heo if (ret) 77671936180STejun Heo return ret; 77795f72d1eSYinghai Lu 77871936180STejun Heo for (i = end_rgn - 1; i >= start_rgn; i--) 77971936180STejun Heo memblock_remove_region(type, i); 78095f72d1eSYinghai Lu return 0; 78195f72d1eSYinghai Lu } 78295f72d1eSYinghai Lu 783581adcbeSTejun Heo int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 78495f72d1eSYinghai Lu { 78525cf23d7SMinchan Kim phys_addr_t end = base + size - 1; 78625cf23d7SMinchan Kim 787a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 78825cf23d7SMinchan Kim &base, &end, (void *)_RET_IP_); 78925cf23d7SMinchan Kim 790f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.memory, base, size); 79195f72d1eSYinghai Lu } 79295f72d1eSYinghai Lu 7934d72868cSMike Rapoport /** 7944d72868cSMike Rapoport * memblock_free - free boot memory block 7954d72868cSMike Rapoport * @base: phys starting address of the boot memory block 7964d72868cSMike Rapoport * @size: size of the boot memory block in bytes 7974d72868cSMike Rapoport * 7984d72868cSMike Rapoport * Free boot memory block previously allocated by memblock_alloc_xx() API. 7994d72868cSMike Rapoport * The freeing memory will not be released to the buddy allocator. 8004d72868cSMike Rapoport */ 801581adcbeSTejun Heo int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) 80295f72d1eSYinghai Lu { 8035d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8045d63f81cSMiles Chen 805a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 8065d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 80724aa0788STejun Heo 8089099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 809f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.reserved, base, size); 81095f72d1eSYinghai Lu } 81195f72d1eSYinghai Lu 812f705ac4bSAlexander Kuleshov int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 81395f72d1eSYinghai Lu { 8145d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8155d63f81cSMiles Chen 816a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 8175d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 81895f72d1eSYinghai Lu 819f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 82095f72d1eSYinghai Lu } 82195f72d1eSYinghai Lu 82202634a44SAnshuman Khandual #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 82302634a44SAnshuman Khandual int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size) 82402634a44SAnshuman Khandual { 82502634a44SAnshuman Khandual phys_addr_t end = base + size - 1; 82602634a44SAnshuman Khandual 82702634a44SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 82802634a44SAnshuman Khandual &base, &end, (void *)_RET_IP_); 82902634a44SAnshuman Khandual 83077649905SDavid Hildenbrand return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0); 83102634a44SAnshuman Khandual } 83202634a44SAnshuman Khandual #endif 83302634a44SAnshuman Khandual 83435fd0808STejun Heo /** 83547cec443SMike Rapoport * memblock_setclr_flag - set or clear flag for a memory region 83647cec443SMike Rapoport * @base: base address of the region 83747cec443SMike Rapoport * @size: size of the region 83847cec443SMike Rapoport * @set: set or clear the flag 8398958b249SHaitao Shi * @flag: the flag to update 84066b16edfSTang Chen * 8414308ce17STony Luck * This function isolates region [@base, @base + @size), and sets/clears flag 84266b16edfSTang Chen * 84347cec443SMike Rapoport * Return: 0 on success, -errno on failure. 84466b16edfSTang Chen */ 8454308ce17STony Luck static int __init_memblock memblock_setclr_flag(phys_addr_t base, 8464308ce17STony Luck phys_addr_t size, int set, int flag) 84766b16edfSTang Chen { 84866b16edfSTang Chen struct memblock_type *type = &memblock.memory; 84966b16edfSTang Chen int i, ret, start_rgn, end_rgn; 85066b16edfSTang Chen 85166b16edfSTang Chen ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 85266b16edfSTang Chen if (ret) 85366b16edfSTang Chen return ret; 85466b16edfSTang Chen 855fe145124SMike Rapoport for (i = start_rgn; i < end_rgn; i++) { 856fe145124SMike Rapoport struct memblock_region *r = &type->regions[i]; 857fe145124SMike Rapoport 8584308ce17STony Luck if (set) 859fe145124SMike Rapoport r->flags |= flag; 8604308ce17STony Luck else 861fe145124SMike Rapoport r->flags &= ~flag; 862fe145124SMike Rapoport } 86366b16edfSTang Chen 86466b16edfSTang Chen memblock_merge_regions(type); 86566b16edfSTang Chen return 0; 86666b16edfSTang Chen } 86766b16edfSTang Chen 86866b16edfSTang Chen /** 8694308ce17STony Luck * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 8704308ce17STony Luck * @base: the base phys addr of the region 8714308ce17STony Luck * @size: the size of the region 8724308ce17STony Luck * 87347cec443SMike Rapoport * Return: 0 on success, -errno on failure. 8744308ce17STony Luck */ 8754308ce17STony Luck int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 8764308ce17STony Luck { 8774308ce17STony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 8784308ce17STony Luck } 8794308ce17STony Luck 8804308ce17STony Luck /** 88166b16edfSTang Chen * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 88266b16edfSTang Chen * @base: the base phys addr of the region 88366b16edfSTang Chen * @size: the size of the region 88466b16edfSTang Chen * 88547cec443SMike Rapoport * Return: 0 on success, -errno on failure. 88666b16edfSTang Chen */ 88766b16edfSTang Chen int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 88866b16edfSTang Chen { 8894308ce17STony Luck return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 89066b16edfSTang Chen } 89166b16edfSTang Chen 89266b16edfSTang Chen /** 893a3f5bafcSTony Luck * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 894a3f5bafcSTony Luck * @base: the base phys addr of the region 895a3f5bafcSTony Luck * @size: the size of the region 896a3f5bafcSTony Luck * 89747cec443SMike Rapoport * Return: 0 on success, -errno on failure. 898a3f5bafcSTony Luck */ 899a3f5bafcSTony Luck int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 900a3f5bafcSTony Luck { 901a3f5bafcSTony Luck system_has_some_mirror = true; 902a3f5bafcSTony Luck 903a3f5bafcSTony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 904a3f5bafcSTony Luck } 905a3f5bafcSTony Luck 906bf3d3cc5SArd Biesheuvel /** 907bf3d3cc5SArd Biesheuvel * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 908bf3d3cc5SArd Biesheuvel * @base: the base phys addr of the region 909bf3d3cc5SArd Biesheuvel * @size: the size of the region 910bf3d3cc5SArd Biesheuvel * 9119092d4f7SMike Rapoport * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the 9129092d4f7SMike Rapoport * direct mapping of the physical memory. These regions will still be 9139092d4f7SMike Rapoport * covered by the memory map. The struct page representing NOMAP memory 9149092d4f7SMike Rapoport * frames in the memory map will be PageReserved() 9159092d4f7SMike Rapoport * 91647cec443SMike Rapoport * Return: 0 on success, -errno on failure. 917bf3d3cc5SArd Biesheuvel */ 918bf3d3cc5SArd Biesheuvel int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 919bf3d3cc5SArd Biesheuvel { 920bf3d3cc5SArd Biesheuvel return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 921bf3d3cc5SArd Biesheuvel } 922a3f5bafcSTony Luck 923a3f5bafcSTony Luck /** 9244c546b8aSAKASHI Takahiro * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 9254c546b8aSAKASHI Takahiro * @base: the base phys addr of the region 9264c546b8aSAKASHI Takahiro * @size: the size of the region 9274c546b8aSAKASHI Takahiro * 92847cec443SMike Rapoport * Return: 0 on success, -errno on failure. 9294c546b8aSAKASHI Takahiro */ 9304c546b8aSAKASHI Takahiro int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 9314c546b8aSAKASHI Takahiro { 9324c546b8aSAKASHI Takahiro return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 9334c546b8aSAKASHI Takahiro } 9344c546b8aSAKASHI Takahiro 9359f3d5eaaSMike Rapoport static bool should_skip_region(struct memblock_type *type, 9369f3d5eaaSMike Rapoport struct memblock_region *m, 9379f3d5eaaSMike Rapoport int nid, int flags) 938c9a688a3SMike Rapoport { 939c9a688a3SMike Rapoport int m_nid = memblock_get_region_node(m); 940c9a688a3SMike Rapoport 9419f3d5eaaSMike Rapoport /* we never skip regions when iterating memblock.reserved or physmem */ 9429f3d5eaaSMike Rapoport if (type != memblock_memory) 9439f3d5eaaSMike Rapoport return false; 9449f3d5eaaSMike Rapoport 945c9a688a3SMike Rapoport /* only memory regions are associated with nodes, check it */ 946c9a688a3SMike Rapoport if (nid != NUMA_NO_NODE && nid != m_nid) 947c9a688a3SMike Rapoport return true; 948c9a688a3SMike Rapoport 949c9a688a3SMike Rapoport /* skip hotpluggable memory regions if needed */ 95079e482e9SMike Rapoport if (movable_node_is_enabled() && memblock_is_hotpluggable(m) && 95179e482e9SMike Rapoport !(flags & MEMBLOCK_HOTPLUG)) 952c9a688a3SMike Rapoport return true; 953c9a688a3SMike Rapoport 954c9a688a3SMike Rapoport /* if we want mirror memory skip non-mirror memory regions */ 955c9a688a3SMike Rapoport if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 956c9a688a3SMike Rapoport return true; 957c9a688a3SMike Rapoport 958c9a688a3SMike Rapoport /* skip nomap memory unless we were asked for it explicitly */ 959c9a688a3SMike Rapoport if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 960c9a688a3SMike Rapoport return true; 961c9a688a3SMike Rapoport 962c9a688a3SMike Rapoport return false; 963c9a688a3SMike Rapoport } 964c9a688a3SMike Rapoport 9658e7a7f86SRobin Holt /** 966a2974133SMike Rapoport * __next_mem_range - next function for for_each_free_mem_range() etc. 96735fd0808STejun Heo * @idx: pointer to u64 loop variable 968b1154233SGrygorii Strashko * @nid: node selector, %NUMA_NO_NODE for all nodes 969fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 970f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 971f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 972dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 973dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 974dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 97535fd0808STejun Heo * 976f1af9d3aSPhilipp Hachtmann * Find the first area from *@idx which matches @nid, fill the out 97735fd0808STejun Heo * parameters, and update *@idx for the next iteration. The lower 32bit of 978f1af9d3aSPhilipp Hachtmann * *@idx contains index into type_a and the upper 32bit indexes the 979f1af9d3aSPhilipp Hachtmann * areas before each region in type_b. For example, if type_b regions 98035fd0808STejun Heo * look like the following, 98135fd0808STejun Heo * 98235fd0808STejun Heo * 0:[0-16), 1:[32-48), 2:[128-130) 98335fd0808STejun Heo * 98435fd0808STejun Heo * The upper 32bit indexes the following regions. 98535fd0808STejun Heo * 98635fd0808STejun Heo * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 98735fd0808STejun Heo * 98835fd0808STejun Heo * As both region arrays are sorted, the function advances the two indices 98935fd0808STejun Heo * in lockstep and returns each intersection. 99035fd0808STejun Heo */ 99177649905SDavid Hildenbrand void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags, 992f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 99377649905SDavid Hildenbrand struct memblock_type *type_b, phys_addr_t *out_start, 99435fd0808STejun Heo phys_addr_t *out_end, int *out_nid) 99535fd0808STejun Heo { 996f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 997f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 998b1154233SGrygorii Strashko 999f1af9d3aSPhilipp Hachtmann if (WARN_ONCE(nid == MAX_NUMNODES, 1000f1af9d3aSPhilipp Hachtmann "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1001560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 100235fd0808STejun Heo 1003f1af9d3aSPhilipp Hachtmann for (; idx_a < type_a->cnt; idx_a++) { 1004f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1005f1af9d3aSPhilipp Hachtmann 100635fd0808STejun Heo phys_addr_t m_start = m->base; 100735fd0808STejun Heo phys_addr_t m_end = m->base + m->size; 1008f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 100935fd0808STejun Heo 10109f3d5eaaSMike Rapoport if (should_skip_region(type_a, m, nid, flags)) 1011bf3d3cc5SArd Biesheuvel continue; 1012bf3d3cc5SArd Biesheuvel 1013f1af9d3aSPhilipp Hachtmann if (!type_b) { 1014f1af9d3aSPhilipp Hachtmann if (out_start) 1015f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1016f1af9d3aSPhilipp Hachtmann if (out_end) 1017f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1018f1af9d3aSPhilipp Hachtmann if (out_nid) 1019f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1020f1af9d3aSPhilipp Hachtmann idx_a++; 1021f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1022f1af9d3aSPhilipp Hachtmann return; 1023f1af9d3aSPhilipp Hachtmann } 102435fd0808STejun Heo 1025f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1026f1af9d3aSPhilipp Hachtmann for (; idx_b < type_b->cnt + 1; idx_b++) { 1027f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1028f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1029f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1030f1af9d3aSPhilipp Hachtmann 1031f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1032f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1033f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 10341c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1035f1af9d3aSPhilipp Hachtmann 1036f1af9d3aSPhilipp Hachtmann /* 1037f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1038f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1039f1af9d3aSPhilipp Hachtmann */ 104035fd0808STejun Heo if (r_start >= m_end) 104135fd0808STejun Heo break; 104235fd0808STejun Heo /* if the two regions intersect, we're done */ 104335fd0808STejun Heo if (m_start < r_end) { 104435fd0808STejun Heo if (out_start) 1045f1af9d3aSPhilipp Hachtmann *out_start = 1046f1af9d3aSPhilipp Hachtmann max(m_start, r_start); 104735fd0808STejun Heo if (out_end) 104835fd0808STejun Heo *out_end = min(m_end, r_end); 104935fd0808STejun Heo if (out_nid) 1050f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 105135fd0808STejun Heo /* 1052f1af9d3aSPhilipp Hachtmann * The region which ends first is 1053f1af9d3aSPhilipp Hachtmann * advanced for the next iteration. 105435fd0808STejun Heo */ 105535fd0808STejun Heo if (m_end <= r_end) 1056f1af9d3aSPhilipp Hachtmann idx_a++; 105735fd0808STejun Heo else 1058f1af9d3aSPhilipp Hachtmann idx_b++; 1059f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 106035fd0808STejun Heo return; 106135fd0808STejun Heo } 106235fd0808STejun Heo } 106335fd0808STejun Heo } 106435fd0808STejun Heo 106535fd0808STejun Heo /* signal end of iteration */ 106635fd0808STejun Heo *idx = ULLONG_MAX; 106735fd0808STejun Heo } 106835fd0808STejun Heo 10697bd0b0f0STejun Heo /** 1070f1af9d3aSPhilipp Hachtmann * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1071f1af9d3aSPhilipp Hachtmann * 10727bd0b0f0STejun Heo * @idx: pointer to u64 loop variable 1073ad5ea8cdSAlexander Kuleshov * @nid: node selector, %NUMA_NO_NODE for all nodes 1074fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 1075f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 1076f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 1077dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1078dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1079dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 10807bd0b0f0STejun Heo * 108147cec443SMike Rapoport * Finds the next range from type_a which is not marked as unsuitable 108247cec443SMike Rapoport * in type_b. 108347cec443SMike Rapoport * 1084f1af9d3aSPhilipp Hachtmann * Reverse of __next_mem_range(). 10857bd0b0f0STejun Heo */ 1086e1720feeSMike Rapoport void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1087e1720feeSMike Rapoport enum memblock_flags flags, 1088f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 1089f1af9d3aSPhilipp Hachtmann struct memblock_type *type_b, 10907bd0b0f0STejun Heo phys_addr_t *out_start, 10917bd0b0f0STejun Heo phys_addr_t *out_end, int *out_nid) 10927bd0b0f0STejun Heo { 1093f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 1094f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 1095b1154233SGrygorii Strashko 1096560dca27SGrygorii Strashko if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1097560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 10987bd0b0f0STejun Heo 10997bd0b0f0STejun Heo if (*idx == (u64)ULLONG_MAX) { 1100f1af9d3aSPhilipp Hachtmann idx_a = type_a->cnt - 1; 1101e47608abSzijun_hu if (type_b != NULL) 1102f1af9d3aSPhilipp Hachtmann idx_b = type_b->cnt; 1103e47608abSzijun_hu else 1104e47608abSzijun_hu idx_b = 0; 11057bd0b0f0STejun Heo } 11067bd0b0f0STejun Heo 1107f1af9d3aSPhilipp Hachtmann for (; idx_a >= 0; idx_a--) { 1108f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1109f1af9d3aSPhilipp Hachtmann 11107bd0b0f0STejun Heo phys_addr_t m_start = m->base; 11117bd0b0f0STejun Heo phys_addr_t m_end = m->base + m->size; 1112f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 11137bd0b0f0STejun Heo 11149f3d5eaaSMike Rapoport if (should_skip_region(type_a, m, nid, flags)) 1115bf3d3cc5SArd Biesheuvel continue; 1116bf3d3cc5SArd Biesheuvel 1117f1af9d3aSPhilipp Hachtmann if (!type_b) { 1118f1af9d3aSPhilipp Hachtmann if (out_start) 1119f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1120f1af9d3aSPhilipp Hachtmann if (out_end) 1121f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1122f1af9d3aSPhilipp Hachtmann if (out_nid) 1123f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1124fb399b48Szijun_hu idx_a--; 1125f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1126f1af9d3aSPhilipp Hachtmann return; 1127f1af9d3aSPhilipp Hachtmann } 11287bd0b0f0STejun Heo 1129f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1130f1af9d3aSPhilipp Hachtmann for (; idx_b >= 0; idx_b--) { 1131f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1132f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1133f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1134f1af9d3aSPhilipp Hachtmann 1135f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1136f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1137f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 11381c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1139f1af9d3aSPhilipp Hachtmann /* 1140f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1141f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1142f1af9d3aSPhilipp Hachtmann */ 1143f1af9d3aSPhilipp Hachtmann 11447bd0b0f0STejun Heo if (r_end <= m_start) 11457bd0b0f0STejun Heo break; 11467bd0b0f0STejun Heo /* if the two regions intersect, we're done */ 11477bd0b0f0STejun Heo if (m_end > r_start) { 11487bd0b0f0STejun Heo if (out_start) 11497bd0b0f0STejun Heo *out_start = max(m_start, r_start); 11507bd0b0f0STejun Heo if (out_end) 11517bd0b0f0STejun Heo *out_end = min(m_end, r_end); 11527bd0b0f0STejun Heo if (out_nid) 1153f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 11547bd0b0f0STejun Heo if (m_start >= r_start) 1155f1af9d3aSPhilipp Hachtmann idx_a--; 11567bd0b0f0STejun Heo else 1157f1af9d3aSPhilipp Hachtmann idx_b--; 1158f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 11597bd0b0f0STejun Heo return; 11607bd0b0f0STejun Heo } 11617bd0b0f0STejun Heo } 11627bd0b0f0STejun Heo } 1163f1af9d3aSPhilipp Hachtmann /* signal end of iteration */ 11647bd0b0f0STejun Heo *idx = ULLONG_MAX; 11657bd0b0f0STejun Heo } 11667bd0b0f0STejun Heo 11677c0caeb8STejun Heo /* 116845e79815SChen Chang * Common iterator interface used to define for_each_mem_pfn_range(). 11697c0caeb8STejun Heo */ 11707c0caeb8STejun Heo void __init_memblock __next_mem_pfn_range(int *idx, int nid, 11717c0caeb8STejun Heo unsigned long *out_start_pfn, 11727c0caeb8STejun Heo unsigned long *out_end_pfn, int *out_nid) 11737c0caeb8STejun Heo { 11747c0caeb8STejun Heo struct memblock_type *type = &memblock.memory; 11757c0caeb8STejun Heo struct memblock_region *r; 1176d622abf7SMike Rapoport int r_nid; 11777c0caeb8STejun Heo 11787c0caeb8STejun Heo while (++*idx < type->cnt) { 11797c0caeb8STejun Heo r = &type->regions[*idx]; 1180d622abf7SMike Rapoport r_nid = memblock_get_region_node(r); 11817c0caeb8STejun Heo 11827c0caeb8STejun Heo if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 11837c0caeb8STejun Heo continue; 1184d622abf7SMike Rapoport if (nid == MAX_NUMNODES || nid == r_nid) 11857c0caeb8STejun Heo break; 11867c0caeb8STejun Heo } 11877c0caeb8STejun Heo if (*idx >= type->cnt) { 11887c0caeb8STejun Heo *idx = -1; 11897c0caeb8STejun Heo return; 11907c0caeb8STejun Heo } 11917c0caeb8STejun Heo 11927c0caeb8STejun Heo if (out_start_pfn) 11937c0caeb8STejun Heo *out_start_pfn = PFN_UP(r->base); 11947c0caeb8STejun Heo if (out_end_pfn) 11957c0caeb8STejun Heo *out_end_pfn = PFN_DOWN(r->base + r->size); 11967c0caeb8STejun Heo if (out_nid) 1197d622abf7SMike Rapoport *out_nid = r_nid; 11987c0caeb8STejun Heo } 11997c0caeb8STejun Heo 12007c0caeb8STejun Heo /** 12017c0caeb8STejun Heo * memblock_set_node - set node ID on memblock regions 12027c0caeb8STejun Heo * @base: base of area to set node ID for 12037c0caeb8STejun Heo * @size: size of area to set node ID for 1204e7e8de59STang Chen * @type: memblock type to set node ID for 12057c0caeb8STejun Heo * @nid: node ID to set 12067c0caeb8STejun Heo * 1207e7e8de59STang Chen * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 12087c0caeb8STejun Heo * Regions which cross the area boundaries are split as necessary. 12097c0caeb8STejun Heo * 121047cec443SMike Rapoport * Return: 12117c0caeb8STejun Heo * 0 on success, -errno on failure. 12127c0caeb8STejun Heo */ 12137c0caeb8STejun Heo int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1214e7e8de59STang Chen struct memblock_type *type, int nid) 12157c0caeb8STejun Heo { 1216a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 12176a9ceb31STejun Heo int start_rgn, end_rgn; 12186a9ceb31STejun Heo int i, ret; 12197c0caeb8STejun Heo 12206a9ceb31STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 12216a9ceb31STejun Heo if (ret) 12226a9ceb31STejun Heo return ret; 12237c0caeb8STejun Heo 12246a9ceb31STejun Heo for (i = start_rgn; i < end_rgn; i++) 1225e9d24ad3SWanpeng Li memblock_set_region_node(&type->regions[i], nid); 12267c0caeb8STejun Heo 12277c0caeb8STejun Heo memblock_merge_regions(type); 12283f08a302SMike Rapoport #endif 12297c0caeb8STejun Heo return 0; 12307c0caeb8STejun Heo } 12313f08a302SMike Rapoport 1232837566e7SAlexander Duyck #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1233837566e7SAlexander Duyck /** 1234837566e7SAlexander Duyck * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1235837566e7SAlexander Duyck * 1236837566e7SAlexander Duyck * @idx: pointer to u64 loop variable 1237837566e7SAlexander Duyck * @zone: zone in which all of the memory blocks reside 1238837566e7SAlexander Duyck * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1239837566e7SAlexander Duyck * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1240837566e7SAlexander Duyck * 1241837566e7SAlexander Duyck * This function is meant to be a zone/pfn specific wrapper for the 1242837566e7SAlexander Duyck * for_each_mem_range type iterators. Specifically they are used in the 1243837566e7SAlexander Duyck * deferred memory init routines and as such we were duplicating much of 1244837566e7SAlexander Duyck * this logic throughout the code. So instead of having it in multiple 1245837566e7SAlexander Duyck * locations it seemed like it would make more sense to centralize this to 1246837566e7SAlexander Duyck * one new iterator that does everything they need. 1247837566e7SAlexander Duyck */ 1248837566e7SAlexander Duyck void __init_memblock 1249837566e7SAlexander Duyck __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1250837566e7SAlexander Duyck unsigned long *out_spfn, unsigned long *out_epfn) 1251837566e7SAlexander Duyck { 1252837566e7SAlexander Duyck int zone_nid = zone_to_nid(zone); 1253837566e7SAlexander Duyck phys_addr_t spa, epa; 1254837566e7SAlexander Duyck int nid; 1255837566e7SAlexander Duyck 1256837566e7SAlexander Duyck __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1257837566e7SAlexander Duyck &memblock.memory, &memblock.reserved, 1258837566e7SAlexander Duyck &spa, &epa, &nid); 1259837566e7SAlexander Duyck 1260837566e7SAlexander Duyck while (*idx != U64_MAX) { 1261837566e7SAlexander Duyck unsigned long epfn = PFN_DOWN(epa); 1262837566e7SAlexander Duyck unsigned long spfn = PFN_UP(spa); 1263837566e7SAlexander Duyck 1264837566e7SAlexander Duyck /* 1265837566e7SAlexander Duyck * Verify the end is at least past the start of the zone and 1266837566e7SAlexander Duyck * that we have at least one PFN to initialize. 1267837566e7SAlexander Duyck */ 1268837566e7SAlexander Duyck if (zone->zone_start_pfn < epfn && spfn < epfn) { 1269837566e7SAlexander Duyck /* if we went too far just stop searching */ 1270837566e7SAlexander Duyck if (zone_end_pfn(zone) <= spfn) { 1271837566e7SAlexander Duyck *idx = U64_MAX; 1272837566e7SAlexander Duyck break; 1273837566e7SAlexander Duyck } 1274837566e7SAlexander Duyck 1275837566e7SAlexander Duyck if (out_spfn) 1276837566e7SAlexander Duyck *out_spfn = max(zone->zone_start_pfn, spfn); 1277837566e7SAlexander Duyck if (out_epfn) 1278837566e7SAlexander Duyck *out_epfn = min(zone_end_pfn(zone), epfn); 1279837566e7SAlexander Duyck 1280837566e7SAlexander Duyck return; 1281837566e7SAlexander Duyck } 1282837566e7SAlexander Duyck 1283837566e7SAlexander Duyck __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1284837566e7SAlexander Duyck &memblock.memory, &memblock.reserved, 1285837566e7SAlexander Duyck &spa, &epa, &nid); 1286837566e7SAlexander Duyck } 1287837566e7SAlexander Duyck 1288837566e7SAlexander Duyck /* signal end of iteration */ 1289837566e7SAlexander Duyck if (out_spfn) 1290837566e7SAlexander Duyck *out_spfn = ULONG_MAX; 1291837566e7SAlexander Duyck if (out_epfn) 1292837566e7SAlexander Duyck *out_epfn = 0; 1293837566e7SAlexander Duyck } 1294837566e7SAlexander Duyck 1295837566e7SAlexander Duyck #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 12967c0caeb8STejun Heo 129792d12f95SMike Rapoport /** 129892d12f95SMike Rapoport * memblock_alloc_range_nid - allocate boot memory block 129992d12f95SMike Rapoport * @size: size of memory block to be allocated in bytes 130092d12f95SMike Rapoport * @align: alignment of the region and block's size 130192d12f95SMike Rapoport * @start: the lower bound of the memory region to allocate (phys address) 130292d12f95SMike Rapoport * @end: the upper bound of the memory region to allocate (phys address) 130392d12f95SMike Rapoport * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 13040ac398b1SYunfeng Ye * @exact_nid: control the allocation fall back to other nodes 130592d12f95SMike Rapoport * 130692d12f95SMike Rapoport * The allocation is performed from memory region limited by 130795830666SCao jin * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 130892d12f95SMike Rapoport * 13090ac398b1SYunfeng Ye * If the specified node can not hold the requested memory and @exact_nid 13100ac398b1SYunfeng Ye * is false, the allocation falls back to any node in the system. 131192d12f95SMike Rapoport * 131292d12f95SMike Rapoport * For systems with memory mirroring, the allocation is attempted first 131392d12f95SMike Rapoport * from the regions with mirroring enabled and then retried from any 131492d12f95SMike Rapoport * memory region. 131592d12f95SMike Rapoport * 131692d12f95SMike Rapoport * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for 131792d12f95SMike Rapoport * allocated boot memory block, so that it is never reported as leaks. 131892d12f95SMike Rapoport * 131992d12f95SMike Rapoport * Return: 132092d12f95SMike Rapoport * Physical address of allocated memory block on success, %0 on failure. 132192d12f95SMike Rapoport */ 13228676af1fSAslan Bakirov phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 13232bfc2862SAkinobu Mita phys_addr_t align, phys_addr_t start, 13240ac398b1SYunfeng Ye phys_addr_t end, int nid, 13250ac398b1SYunfeng Ye bool exact_nid) 132695f72d1eSYinghai Lu { 132792d12f95SMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 13286ed311b2SBenjamin Herrenschmidt phys_addr_t found; 132995f72d1eSYinghai Lu 133092d12f95SMike Rapoport if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 133192d12f95SMike Rapoport nid = NUMA_NO_NODE; 133292d12f95SMike Rapoport 13332f770806SMike Rapoport if (!align) { 13342f770806SMike Rapoport /* Can't use WARNs this early in boot on powerpc */ 13352f770806SMike Rapoport dump_stack(); 13362f770806SMike Rapoport align = SMP_CACHE_BYTES; 13372f770806SMike Rapoport } 13382f770806SMike Rapoport 133992d12f95SMike Rapoport again: 1340fc6daaf9STony Luck found = memblock_find_in_range_node(size, align, start, end, nid, 1341fc6daaf9STony Luck flags); 134292d12f95SMike Rapoport if (found && !memblock_reserve(found, size)) 134392d12f95SMike Rapoport goto done; 134492d12f95SMike Rapoport 13450ac398b1SYunfeng Ye if (nid != NUMA_NO_NODE && !exact_nid) { 134692d12f95SMike Rapoport found = memblock_find_in_range_node(size, align, start, 134792d12f95SMike Rapoport end, NUMA_NO_NODE, 134892d12f95SMike Rapoport flags); 134992d12f95SMike Rapoport if (found && !memblock_reserve(found, size)) 135092d12f95SMike Rapoport goto done; 135192d12f95SMike Rapoport } 135292d12f95SMike Rapoport 135392d12f95SMike Rapoport if (flags & MEMBLOCK_MIRROR) { 135492d12f95SMike Rapoport flags &= ~MEMBLOCK_MIRROR; 135592d12f95SMike Rapoport pr_warn("Could not allocate %pap bytes of mirrored memory\n", 135692d12f95SMike Rapoport &size); 135792d12f95SMike Rapoport goto again; 135892d12f95SMike Rapoport } 135992d12f95SMike Rapoport 136092d12f95SMike Rapoport return 0; 136192d12f95SMike Rapoport 136292d12f95SMike Rapoport done: 136392d12f95SMike Rapoport /* Skip kmemleak for kasan_init() due to high volume. */ 136492d12f95SMike Rapoport if (end != MEMBLOCK_ALLOC_KASAN) 1365aedf95eaSCatalin Marinas /* 136692d12f95SMike Rapoport * The min_count is set to 0 so that memblock allocated 136792d12f95SMike Rapoport * blocks are never reported as leaks. This is because many 136892d12f95SMike Rapoport * of these blocks are only referred via the physical 136992d12f95SMike Rapoport * address which is not looked up by kmemleak. 1370aedf95eaSCatalin Marinas */ 13719099daedSCatalin Marinas kmemleak_alloc_phys(found, size, 0, 0); 137292d12f95SMike Rapoport 13736ed311b2SBenjamin Herrenschmidt return found; 1374aedf95eaSCatalin Marinas } 137595f72d1eSYinghai Lu 1376a2974133SMike Rapoport /** 1377a2974133SMike Rapoport * memblock_phys_alloc_range - allocate a memory block inside specified range 1378a2974133SMike Rapoport * @size: size of memory block to be allocated in bytes 1379a2974133SMike Rapoport * @align: alignment of the region and block's size 1380a2974133SMike Rapoport * @start: the lower bound of the memory region to allocate (physical address) 1381a2974133SMike Rapoport * @end: the upper bound of the memory region to allocate (physical address) 1382a2974133SMike Rapoport * 1383a2974133SMike Rapoport * Allocate @size bytes in the between @start and @end. 1384a2974133SMike Rapoport * 1385a2974133SMike Rapoport * Return: physical address of the allocated memory block on success, 1386a2974133SMike Rapoport * %0 on failure. 1387a2974133SMike Rapoport */ 13888a770c2aSMike Rapoport phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 13898a770c2aSMike Rapoport phys_addr_t align, 13908a770c2aSMike Rapoport phys_addr_t start, 13918a770c2aSMike Rapoport phys_addr_t end) 13922bfc2862SAkinobu Mita { 1393b5cf2d6cSFaiyaz Mohammed memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1394b5cf2d6cSFaiyaz Mohammed __func__, (u64)size, (u64)align, &start, &end, 1395b5cf2d6cSFaiyaz Mohammed (void *)_RET_IP_); 13960ac398b1SYunfeng Ye return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 13970ac398b1SYunfeng Ye false); 13987bd0b0f0STejun Heo } 13997bd0b0f0STejun Heo 1400a2974133SMike Rapoport /** 140117cbe038SLevi Yun * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1402a2974133SMike Rapoport * @size: size of memory block to be allocated in bytes 1403a2974133SMike Rapoport * @align: alignment of the region and block's size 1404a2974133SMike Rapoport * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1405a2974133SMike Rapoport * 1406a2974133SMike Rapoport * Allocates memory block from the specified NUMA node. If the node 1407a2974133SMike Rapoport * has no available memory, attempts to allocated from any node in the 1408a2974133SMike Rapoport * system. 1409a2974133SMike Rapoport * 1410a2974133SMike Rapoport * Return: physical address of the allocated memory block on success, 1411a2974133SMike Rapoport * %0 on failure. 1412a2974133SMike Rapoport */ 14139a8dd708SMike Rapoport phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 14149d1e2492SBenjamin Herrenschmidt { 141533755574SMike Rapoport return memblock_alloc_range_nid(size, align, 0, 14160ac398b1SYunfeng Ye MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 141795f72d1eSYinghai Lu } 141895f72d1eSYinghai Lu 141926f09e9bSSantosh Shilimkar /** 1420eb31d559SMike Rapoport * memblock_alloc_internal - allocate boot memory block 142126f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 142226f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 142326f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region to allocate (phys address) 142426f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region to allocate (phys address) 142526f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 14260ac398b1SYunfeng Ye * @exact_nid: control the allocation fall back to other nodes 142726f09e9bSSantosh Shilimkar * 142892d12f95SMike Rapoport * Allocates memory block using memblock_alloc_range_nid() and 142992d12f95SMike Rapoport * converts the returned physical address to virtual. 143092d12f95SMike Rapoport * 143126f09e9bSSantosh Shilimkar * The @min_addr limit is dropped if it can not be satisfied and the allocation 143292d12f95SMike Rapoport * will fall back to memory below @min_addr. Other constraints, such 143392d12f95SMike Rapoport * as node and mirrored memory will be handled again in 143492d12f95SMike Rapoport * memblock_alloc_range_nid(). 143526f09e9bSSantosh Shilimkar * 143647cec443SMike Rapoport * Return: 143726f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 143826f09e9bSSantosh Shilimkar */ 1439eb31d559SMike Rapoport static void * __init memblock_alloc_internal( 144026f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 144126f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 14420ac398b1SYunfeng Ye int nid, bool exact_nid) 144326f09e9bSSantosh Shilimkar { 144426f09e9bSSantosh Shilimkar phys_addr_t alloc; 144526f09e9bSSantosh Shilimkar 144626f09e9bSSantosh Shilimkar /* 144726f09e9bSSantosh Shilimkar * Detect any accidental use of these APIs after slab is ready, as at 144826f09e9bSSantosh Shilimkar * this moment memblock may be deinitialized already and its 1449c6ffc5caSMike Rapoport * internal data may be destroyed (after execution of memblock_free_all) 145026f09e9bSSantosh Shilimkar */ 145126f09e9bSSantosh Shilimkar if (WARN_ON_ONCE(slab_is_available())) 145226f09e9bSSantosh Shilimkar return kzalloc_node(size, GFP_NOWAIT, nid); 145326f09e9bSSantosh Shilimkar 1454f3057ad7SMike Rapoport if (max_addr > memblock.current_limit) 1455f3057ad7SMike Rapoport max_addr = memblock.current_limit; 1456f3057ad7SMike Rapoport 14570ac398b1SYunfeng Ye alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 14580ac398b1SYunfeng Ye exact_nid); 14592f770806SMike Rapoport 146092d12f95SMike Rapoport /* retry allocation without lower limit */ 146192d12f95SMike Rapoport if (!alloc && min_addr) 14620ac398b1SYunfeng Ye alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 14630ac398b1SYunfeng Ye exact_nid); 146426f09e9bSSantosh Shilimkar 146592d12f95SMike Rapoport if (!alloc) 1466a3f5bafcSTony Luck return NULL; 146726f09e9bSSantosh Shilimkar 146892d12f95SMike Rapoport return phys_to_virt(alloc); 146926f09e9bSSantosh Shilimkar } 147026f09e9bSSantosh Shilimkar 147126f09e9bSSantosh Shilimkar /** 14720ac398b1SYunfeng Ye * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 14730ac398b1SYunfeng Ye * without zeroing memory 14740ac398b1SYunfeng Ye * @size: size of memory block to be allocated in bytes 14750ac398b1SYunfeng Ye * @align: alignment of the region and block's size 14760ac398b1SYunfeng Ye * @min_addr: the lower bound of the memory region from where the allocation 14770ac398b1SYunfeng Ye * is preferred (phys address) 14780ac398b1SYunfeng Ye * @max_addr: the upper bound of the memory region from where the allocation 14790ac398b1SYunfeng Ye * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 14800ac398b1SYunfeng Ye * allocate only from memory limited by memblock.current_limit value 14810ac398b1SYunfeng Ye * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 14820ac398b1SYunfeng Ye * 14830ac398b1SYunfeng Ye * Public function, provides additional debug information (including caller 14840ac398b1SYunfeng Ye * info), if enabled. Does not zero allocated memory. 14850ac398b1SYunfeng Ye * 14860ac398b1SYunfeng Ye * Return: 14870ac398b1SYunfeng Ye * Virtual address of allocated memory block on success, NULL on failure. 14880ac398b1SYunfeng Ye */ 14890ac398b1SYunfeng Ye void * __init memblock_alloc_exact_nid_raw( 14900ac398b1SYunfeng Ye phys_addr_t size, phys_addr_t align, 14910ac398b1SYunfeng Ye phys_addr_t min_addr, phys_addr_t max_addr, 14920ac398b1SYunfeng Ye int nid) 14930ac398b1SYunfeng Ye { 14940ac398b1SYunfeng Ye memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 14950ac398b1SYunfeng Ye __func__, (u64)size, (u64)align, nid, &min_addr, 14960ac398b1SYunfeng Ye &max_addr, (void *)_RET_IP_); 14970ac398b1SYunfeng Ye 149808678804SMike Rapoport return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 149908678804SMike Rapoport true); 15000ac398b1SYunfeng Ye } 15010ac398b1SYunfeng Ye 15020ac398b1SYunfeng Ye /** 1503eb31d559SMike Rapoport * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1504ea1f5f37SPavel Tatashin * memory and without panicking 1505ea1f5f37SPavel Tatashin * @size: size of memory block to be allocated in bytes 1506ea1f5f37SPavel Tatashin * @align: alignment of the region and block's size 1507ea1f5f37SPavel Tatashin * @min_addr: the lower bound of the memory region from where the allocation 1508ea1f5f37SPavel Tatashin * is preferred (phys address) 1509ea1f5f37SPavel Tatashin * @max_addr: the upper bound of the memory region from where the allocation 151097ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1511ea1f5f37SPavel Tatashin * allocate only from memory limited by memblock.current_limit value 1512ea1f5f37SPavel Tatashin * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1513ea1f5f37SPavel Tatashin * 1514ea1f5f37SPavel Tatashin * Public function, provides additional debug information (including caller 1515ea1f5f37SPavel Tatashin * info), if enabled. Does not zero allocated memory, does not panic if request 1516ea1f5f37SPavel Tatashin * cannot be satisfied. 1517ea1f5f37SPavel Tatashin * 151847cec443SMike Rapoport * Return: 1519ea1f5f37SPavel Tatashin * Virtual address of allocated memory block on success, NULL on failure. 1520ea1f5f37SPavel Tatashin */ 1521eb31d559SMike Rapoport void * __init memblock_alloc_try_nid_raw( 1522ea1f5f37SPavel Tatashin phys_addr_t size, phys_addr_t align, 1523ea1f5f37SPavel Tatashin phys_addr_t min_addr, phys_addr_t max_addr, 1524ea1f5f37SPavel Tatashin int nid) 1525ea1f5f37SPavel Tatashin { 1526d75f773cSSakari Ailus memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1527a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1528a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1529ea1f5f37SPavel Tatashin 153008678804SMike Rapoport return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 153108678804SMike Rapoport false); 1532ea1f5f37SPavel Tatashin } 1533ea1f5f37SPavel Tatashin 1534ea1f5f37SPavel Tatashin /** 1535c0dbe825SMike Rapoport * memblock_alloc_try_nid - allocate boot memory block 153626f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 153726f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 153826f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region from where the allocation 153926f09e9bSSantosh Shilimkar * is preferred (phys address) 154026f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region from where the allocation 154197ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 154226f09e9bSSantosh Shilimkar * allocate only from memory limited by memblock.current_limit value 154326f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 154426f09e9bSSantosh Shilimkar * 1545c0dbe825SMike Rapoport * Public function, provides additional debug information (including caller 1546c0dbe825SMike Rapoport * info), if enabled. This function zeroes the allocated memory. 154726f09e9bSSantosh Shilimkar * 154847cec443SMike Rapoport * Return: 154926f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 155026f09e9bSSantosh Shilimkar */ 1551eb31d559SMike Rapoport void * __init memblock_alloc_try_nid( 155226f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 155326f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 155426f09e9bSSantosh Shilimkar int nid) 155526f09e9bSSantosh Shilimkar { 155626f09e9bSSantosh Shilimkar void *ptr; 155726f09e9bSSantosh Shilimkar 1558d75f773cSSakari Ailus memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1559a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1560a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1561eb31d559SMike Rapoport ptr = memblock_alloc_internal(size, align, 15620ac398b1SYunfeng Ye min_addr, max_addr, nid, false); 1563c0dbe825SMike Rapoport if (ptr) 1564ea1f5f37SPavel Tatashin memset(ptr, 0, size); 156526f09e9bSSantosh Shilimkar 1566c0dbe825SMike Rapoport return ptr; 156726f09e9bSSantosh Shilimkar } 156826f09e9bSSantosh Shilimkar 156926f09e9bSSantosh Shilimkar /** 1570a2974133SMike Rapoport * __memblock_free_late - free pages directly to buddy allocator 157148a833ccSMike Rapoport * @base: phys starting address of the boot memory block 157226f09e9bSSantosh Shilimkar * @size: size of the boot memory block in bytes 157326f09e9bSSantosh Shilimkar * 1574a2974133SMike Rapoport * This is only useful when the memblock allocator has already been torn 157526f09e9bSSantosh Shilimkar * down, but we are still initializing the system. Pages are released directly 1576a2974133SMike Rapoport * to the buddy allocator. 157726f09e9bSSantosh Shilimkar */ 157826f09e9bSSantosh Shilimkar void __init __memblock_free_late(phys_addr_t base, phys_addr_t size) 157926f09e9bSSantosh Shilimkar { 1580a36aab89SMike Rapoport phys_addr_t cursor, end; 158126f09e9bSSantosh Shilimkar 1582a36aab89SMike Rapoport end = base + size - 1; 1583d75f773cSSakari Ailus memblock_dbg("%s: [%pa-%pa] %pS\n", 1584a36aab89SMike Rapoport __func__, &base, &end, (void *)_RET_IP_); 15859099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 158626f09e9bSSantosh Shilimkar cursor = PFN_UP(base); 158726f09e9bSSantosh Shilimkar end = PFN_DOWN(base + size); 158826f09e9bSSantosh Shilimkar 158926f09e9bSSantosh Shilimkar for (; cursor < end; cursor++) { 15907c2ee349SMike Rapoport memblock_free_pages(pfn_to_page(cursor), cursor, 0); 1591ca79b0c2SArun KS totalram_pages_inc(); 159226f09e9bSSantosh Shilimkar } 159326f09e9bSSantosh Shilimkar } 15949d1e2492SBenjamin Herrenschmidt 15959d1e2492SBenjamin Herrenschmidt /* 15969d1e2492SBenjamin Herrenschmidt * Remaining API functions 15979d1e2492SBenjamin Herrenschmidt */ 15989d1e2492SBenjamin Herrenschmidt 15991f1ffb8aSDavid Gibson phys_addr_t __init_memblock memblock_phys_mem_size(void) 160095f72d1eSYinghai Lu { 16011440c4e2STejun Heo return memblock.memory.total_size; 160295f72d1eSYinghai Lu } 160395f72d1eSYinghai Lu 16048907de5dSSrikar Dronamraju phys_addr_t __init_memblock memblock_reserved_size(void) 16058907de5dSSrikar Dronamraju { 16068907de5dSSrikar Dronamraju return memblock.reserved.total_size; 16078907de5dSSrikar Dronamraju } 16088907de5dSSrikar Dronamraju 16090a93ebefSSam Ravnborg /* lowest address */ 16100a93ebefSSam Ravnborg phys_addr_t __init_memblock memblock_start_of_DRAM(void) 16110a93ebefSSam Ravnborg { 16120a93ebefSSam Ravnborg return memblock.memory.regions[0].base; 16130a93ebefSSam Ravnborg } 16140a93ebefSSam Ravnborg 161510d06439SYinghai Lu phys_addr_t __init_memblock memblock_end_of_DRAM(void) 161695f72d1eSYinghai Lu { 161795f72d1eSYinghai Lu int idx = memblock.memory.cnt - 1; 161895f72d1eSYinghai Lu 1619e3239ff9SBenjamin Herrenschmidt return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 162095f72d1eSYinghai Lu } 162195f72d1eSYinghai Lu 1622a571d4ebSDennis Chen static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 162395f72d1eSYinghai Lu { 16241c4bc43dSStefan Agner phys_addr_t max_addr = PHYS_ADDR_MAX; 1625136199f0SEmil Medve struct memblock_region *r; 162695f72d1eSYinghai Lu 1627a571d4ebSDennis Chen /* 1628a571d4ebSDennis Chen * translate the memory @limit size into the max address within one of 1629a571d4ebSDennis Chen * the memory memblock regions, if the @limit exceeds the total size 16301c4bc43dSStefan Agner * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1631a571d4ebSDennis Chen */ 1632cc6de168SMike Rapoport for_each_mem_region(r) { 1633c0ce8fefSTejun Heo if (limit <= r->size) { 1634c0ce8fefSTejun Heo max_addr = r->base + limit; 163595f72d1eSYinghai Lu break; 163695f72d1eSYinghai Lu } 1637c0ce8fefSTejun Heo limit -= r->size; 163895f72d1eSYinghai Lu } 1639c0ce8fefSTejun Heo 1640a571d4ebSDennis Chen return max_addr; 1641a571d4ebSDennis Chen } 1642a571d4ebSDennis Chen 1643a571d4ebSDennis Chen void __init memblock_enforce_memory_limit(phys_addr_t limit) 1644a571d4ebSDennis Chen { 164549aef717SColin Ian King phys_addr_t max_addr; 1646a571d4ebSDennis Chen 1647a571d4ebSDennis Chen if (!limit) 1648a571d4ebSDennis Chen return; 1649a571d4ebSDennis Chen 1650a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1651a571d4ebSDennis Chen 1652a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 16531c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1654a571d4ebSDennis Chen return; 1655a571d4ebSDennis Chen 1656c0ce8fefSTejun Heo /* truncate both memory and reserved regions */ 1657f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.memory, max_addr, 16581c4bc43dSStefan Agner PHYS_ADDR_MAX); 1659f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.reserved, max_addr, 16601c4bc43dSStefan Agner PHYS_ADDR_MAX); 166195f72d1eSYinghai Lu } 166295f72d1eSYinghai Lu 1663c9ca9b4eSAKASHI Takahiro void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1664c9ca9b4eSAKASHI Takahiro { 1665c9ca9b4eSAKASHI Takahiro int start_rgn, end_rgn; 1666c9ca9b4eSAKASHI Takahiro int i, ret; 1667c9ca9b4eSAKASHI Takahiro 1668c9ca9b4eSAKASHI Takahiro if (!size) 1669c9ca9b4eSAKASHI Takahiro return; 1670c9ca9b4eSAKASHI Takahiro 1671c9ca9b4eSAKASHI Takahiro ret = memblock_isolate_range(&memblock.memory, base, size, 1672c9ca9b4eSAKASHI Takahiro &start_rgn, &end_rgn); 1673c9ca9b4eSAKASHI Takahiro if (ret) 1674c9ca9b4eSAKASHI Takahiro return; 1675c9ca9b4eSAKASHI Takahiro 1676c9ca9b4eSAKASHI Takahiro /* remove all the MAP regions */ 1677c9ca9b4eSAKASHI Takahiro for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1678c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1679c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1680c9ca9b4eSAKASHI Takahiro 1681c9ca9b4eSAKASHI Takahiro for (i = start_rgn - 1; i >= 0; i--) 1682c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1683c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1684c9ca9b4eSAKASHI Takahiro 1685c9ca9b4eSAKASHI Takahiro /* truncate the reserved regions */ 1686c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 0, base); 1687c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 16881c4bc43dSStefan Agner base + size, PHYS_ADDR_MAX); 1689c9ca9b4eSAKASHI Takahiro } 1690c9ca9b4eSAKASHI Takahiro 1691a571d4ebSDennis Chen void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1692a571d4ebSDennis Chen { 1693a571d4ebSDennis Chen phys_addr_t max_addr; 1694a571d4ebSDennis Chen 1695a571d4ebSDennis Chen if (!limit) 1696a571d4ebSDennis Chen return; 1697a571d4ebSDennis Chen 1698a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1699a571d4ebSDennis Chen 1700a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 17011c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1702a571d4ebSDennis Chen return; 1703a571d4ebSDennis Chen 1704c9ca9b4eSAKASHI Takahiro memblock_cap_memory_range(0, max_addr); 1705a571d4ebSDennis Chen } 1706a571d4ebSDennis Chen 1707cd79481dSYinghai Lu static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 170872d4b0b4SBenjamin Herrenschmidt { 170972d4b0b4SBenjamin Herrenschmidt unsigned int left = 0, right = type->cnt; 171072d4b0b4SBenjamin Herrenschmidt 171172d4b0b4SBenjamin Herrenschmidt do { 171272d4b0b4SBenjamin Herrenschmidt unsigned int mid = (right + left) / 2; 171372d4b0b4SBenjamin Herrenschmidt 171472d4b0b4SBenjamin Herrenschmidt if (addr < type->regions[mid].base) 171572d4b0b4SBenjamin Herrenschmidt right = mid; 171672d4b0b4SBenjamin Herrenschmidt else if (addr >= (type->regions[mid].base + 171772d4b0b4SBenjamin Herrenschmidt type->regions[mid].size)) 171872d4b0b4SBenjamin Herrenschmidt left = mid + 1; 171972d4b0b4SBenjamin Herrenschmidt else 172072d4b0b4SBenjamin Herrenschmidt return mid; 172172d4b0b4SBenjamin Herrenschmidt } while (left < right); 172272d4b0b4SBenjamin Herrenschmidt return -1; 172372d4b0b4SBenjamin Herrenschmidt } 172472d4b0b4SBenjamin Herrenschmidt 1725f5a222dcSYueyi Li bool __init_memblock memblock_is_reserved(phys_addr_t addr) 172695f72d1eSYinghai Lu { 172772d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.reserved, addr) != -1; 172895f72d1eSYinghai Lu } 172972d4b0b4SBenjamin Herrenschmidt 1730b4ad0c7eSYaowei Bai bool __init_memblock memblock_is_memory(phys_addr_t addr) 173172d4b0b4SBenjamin Herrenschmidt { 173272d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.memory, addr) != -1; 173372d4b0b4SBenjamin Herrenschmidt } 173472d4b0b4SBenjamin Herrenschmidt 1735937f0c26SYaowei Bai bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1736bf3d3cc5SArd Biesheuvel { 1737bf3d3cc5SArd Biesheuvel int i = memblock_search(&memblock.memory, addr); 1738bf3d3cc5SArd Biesheuvel 1739bf3d3cc5SArd Biesheuvel if (i == -1) 1740bf3d3cc5SArd Biesheuvel return false; 1741bf3d3cc5SArd Biesheuvel return !memblock_is_nomap(&memblock.memory.regions[i]); 1742bf3d3cc5SArd Biesheuvel } 1743bf3d3cc5SArd Biesheuvel 1744e76b63f8SYinghai Lu int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1745e76b63f8SYinghai Lu unsigned long *start_pfn, unsigned long *end_pfn) 1746e76b63f8SYinghai Lu { 1747e76b63f8SYinghai Lu struct memblock_type *type = &memblock.memory; 174816763230SFabian Frederick int mid = memblock_search(type, PFN_PHYS(pfn)); 1749e76b63f8SYinghai Lu 1750e76b63f8SYinghai Lu if (mid == -1) 1751e76b63f8SYinghai Lu return -1; 1752e76b63f8SYinghai Lu 1753f7e2f7e8SFabian Frederick *start_pfn = PFN_DOWN(type->regions[mid].base); 1754f7e2f7e8SFabian Frederick *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1755e76b63f8SYinghai Lu 1756d622abf7SMike Rapoport return memblock_get_region_node(&type->regions[mid]); 1757e76b63f8SYinghai Lu } 1758e76b63f8SYinghai Lu 1759eab30949SStephen Boyd /** 1760eab30949SStephen Boyd * memblock_is_region_memory - check if a region is a subset of memory 1761eab30949SStephen Boyd * @base: base of region to check 1762eab30949SStephen Boyd * @size: size of region to check 1763eab30949SStephen Boyd * 1764eab30949SStephen Boyd * Check if the region [@base, @base + @size) is a subset of a memory block. 1765eab30949SStephen Boyd * 176647cec443SMike Rapoport * Return: 1767eab30949SStephen Boyd * 0 if false, non-zero if true 1768eab30949SStephen Boyd */ 1769937f0c26SYaowei Bai bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 177072d4b0b4SBenjamin Herrenschmidt { 1771abb65272STomi Valkeinen int idx = memblock_search(&memblock.memory, base); 1772eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 177372d4b0b4SBenjamin Herrenschmidt 177472d4b0b4SBenjamin Herrenschmidt if (idx == -1) 1775937f0c26SYaowei Bai return false; 1776ef415ef4SWei Yang return (memblock.memory.regions[idx].base + 1777eb18f1b5STejun Heo memblock.memory.regions[idx].size) >= end; 177895f72d1eSYinghai Lu } 177995f72d1eSYinghai Lu 1780eab30949SStephen Boyd /** 1781eab30949SStephen Boyd * memblock_is_region_reserved - check if a region intersects reserved memory 1782eab30949SStephen Boyd * @base: base of region to check 1783eab30949SStephen Boyd * @size: size of region to check 1784eab30949SStephen Boyd * 178547cec443SMike Rapoport * Check if the region [@base, @base + @size) intersects a reserved 178647cec443SMike Rapoport * memory block. 1787eab30949SStephen Boyd * 178847cec443SMike Rapoport * Return: 1789c5c5c9d1STang Chen * True if they intersect, false if not. 1790eab30949SStephen Boyd */ 1791c5c5c9d1STang Chen bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 179295f72d1eSYinghai Lu { 1793c5c5c9d1STang Chen return memblock_overlaps_region(&memblock.reserved, base, size); 179495f72d1eSYinghai Lu } 179595f72d1eSYinghai Lu 17966ede1fd3SYinghai Lu void __init_memblock memblock_trim_memory(phys_addr_t align) 17976ede1fd3SYinghai Lu { 17986ede1fd3SYinghai Lu phys_addr_t start, end, orig_start, orig_end; 1799136199f0SEmil Medve struct memblock_region *r; 18006ede1fd3SYinghai Lu 1801cc6de168SMike Rapoport for_each_mem_region(r) { 1802136199f0SEmil Medve orig_start = r->base; 1803136199f0SEmil Medve orig_end = r->base + r->size; 18046ede1fd3SYinghai Lu start = round_up(orig_start, align); 18056ede1fd3SYinghai Lu end = round_down(orig_end, align); 18066ede1fd3SYinghai Lu 18076ede1fd3SYinghai Lu if (start == orig_start && end == orig_end) 18086ede1fd3SYinghai Lu continue; 18096ede1fd3SYinghai Lu 18106ede1fd3SYinghai Lu if (start < end) { 1811136199f0SEmil Medve r->base = start; 1812136199f0SEmil Medve r->size = end - start; 18136ede1fd3SYinghai Lu } else { 1814136199f0SEmil Medve memblock_remove_region(&memblock.memory, 1815136199f0SEmil Medve r - memblock.memory.regions); 1816136199f0SEmil Medve r--; 18176ede1fd3SYinghai Lu } 18186ede1fd3SYinghai Lu } 18196ede1fd3SYinghai Lu } 1820e63075a3SBenjamin Herrenschmidt 18213661ca66SYinghai Lu void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1822e63075a3SBenjamin Herrenschmidt { 1823e63075a3SBenjamin Herrenschmidt memblock.current_limit = limit; 1824e63075a3SBenjamin Herrenschmidt } 1825e63075a3SBenjamin Herrenschmidt 1826fec51014SLaura Abbott phys_addr_t __init_memblock memblock_get_current_limit(void) 1827fec51014SLaura Abbott { 1828fec51014SLaura Abbott return memblock.current_limit; 1829fec51014SLaura Abbott } 1830fec51014SLaura Abbott 18310262d9c8SHeiko Carstens static void __init_memblock memblock_dump(struct memblock_type *type) 18326ed311b2SBenjamin Herrenschmidt { 18335d63f81cSMiles Chen phys_addr_t base, end, size; 1834e1720feeSMike Rapoport enum memblock_flags flags; 18358c9c1701SAlexander Kuleshov int idx; 18368c9c1701SAlexander Kuleshov struct memblock_region *rgn; 18376ed311b2SBenjamin Herrenschmidt 18380262d9c8SHeiko Carstens pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 18396ed311b2SBenjamin Herrenschmidt 184066e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 18417c0caeb8STejun Heo char nid_buf[32] = ""; 18426ed311b2SBenjamin Herrenschmidt 18437c0caeb8STejun Heo base = rgn->base; 18447c0caeb8STejun Heo size = rgn->size; 18455d63f81cSMiles Chen end = base + size - 1; 184666a20757STang Chen flags = rgn->flags; 1847a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 18487c0caeb8STejun Heo if (memblock_get_region_node(rgn) != MAX_NUMNODES) 18497c0caeb8STejun Heo snprintf(nid_buf, sizeof(nid_buf), " on node %d", 18507c0caeb8STejun Heo memblock_get_region_node(rgn)); 18517c0caeb8STejun Heo #endif 1852e1720feeSMike Rapoport pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 18530262d9c8SHeiko Carstens type->name, idx, &base, &end, &size, nid_buf, flags); 18546ed311b2SBenjamin Herrenschmidt } 18556ed311b2SBenjamin Herrenschmidt } 18566ed311b2SBenjamin Herrenschmidt 185787c55870SMike Rapoport static void __init_memblock __memblock_dump_all(void) 18586ed311b2SBenjamin Herrenschmidt { 18596ed311b2SBenjamin Herrenschmidt pr_info("MEMBLOCK configuration:\n"); 18605d63f81cSMiles Chen pr_info(" memory size = %pa reserved size = %pa\n", 18615d63f81cSMiles Chen &memblock.memory.total_size, 18625d63f81cSMiles Chen &memblock.reserved.total_size); 18636ed311b2SBenjamin Herrenschmidt 18640262d9c8SHeiko Carstens memblock_dump(&memblock.memory); 18650262d9c8SHeiko Carstens memblock_dump(&memblock.reserved); 1866409efd4cSHeiko Carstens #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 186777649905SDavid Hildenbrand memblock_dump(&physmem); 1868409efd4cSHeiko Carstens #endif 18696ed311b2SBenjamin Herrenschmidt } 18706ed311b2SBenjamin Herrenschmidt 187187c55870SMike Rapoport void __init_memblock memblock_dump_all(void) 187287c55870SMike Rapoport { 187387c55870SMike Rapoport if (memblock_debug) 187487c55870SMike Rapoport __memblock_dump_all(); 187587c55870SMike Rapoport } 187687c55870SMike Rapoport 18771aadc056STejun Heo void __init memblock_allow_resize(void) 18786ed311b2SBenjamin Herrenschmidt { 1879142b45a7SBenjamin Herrenschmidt memblock_can_resize = 1; 18806ed311b2SBenjamin Herrenschmidt } 18816ed311b2SBenjamin Herrenschmidt 18826ed311b2SBenjamin Herrenschmidt static int __init early_memblock(char *p) 18836ed311b2SBenjamin Herrenschmidt { 18846ed311b2SBenjamin Herrenschmidt if (p && strstr(p, "debug")) 18856ed311b2SBenjamin Herrenschmidt memblock_debug = 1; 18866ed311b2SBenjamin Herrenschmidt return 0; 18876ed311b2SBenjamin Herrenschmidt } 18886ed311b2SBenjamin Herrenschmidt early_param("memblock", early_memblock); 18896ed311b2SBenjamin Herrenschmidt 18904f5b0c17SMike Rapoport static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 18914f5b0c17SMike Rapoport { 18924f5b0c17SMike Rapoport struct page *start_pg, *end_pg; 18934f5b0c17SMike Rapoport phys_addr_t pg, pgend; 18944f5b0c17SMike Rapoport 18954f5b0c17SMike Rapoport /* 18964f5b0c17SMike Rapoport * Convert start_pfn/end_pfn to a struct page pointer. 18974f5b0c17SMike Rapoport */ 18984f5b0c17SMike Rapoport start_pg = pfn_to_page(start_pfn - 1) + 1; 18994f5b0c17SMike Rapoport end_pg = pfn_to_page(end_pfn - 1) + 1; 19004f5b0c17SMike Rapoport 19014f5b0c17SMike Rapoport /* 19024f5b0c17SMike Rapoport * Convert to physical addresses, and round start upwards and end 19034f5b0c17SMike Rapoport * downwards. 19044f5b0c17SMike Rapoport */ 19054f5b0c17SMike Rapoport pg = PAGE_ALIGN(__pa(start_pg)); 19064f5b0c17SMike Rapoport pgend = __pa(end_pg) & PAGE_MASK; 19074f5b0c17SMike Rapoport 19084f5b0c17SMike Rapoport /* 19094f5b0c17SMike Rapoport * If there are free pages between these, free the section of the 19104f5b0c17SMike Rapoport * memmap array. 19114f5b0c17SMike Rapoport */ 19124f5b0c17SMike Rapoport if (pg < pgend) 19134f5b0c17SMike Rapoport memblock_free(pg, pgend - pg); 19144f5b0c17SMike Rapoport } 19154f5b0c17SMike Rapoport 19164f5b0c17SMike Rapoport /* 19174f5b0c17SMike Rapoport * The mem_map array can get very big. Free the unused area of the memory map. 19184f5b0c17SMike Rapoport */ 19194f5b0c17SMike Rapoport static void __init free_unused_memmap(void) 19204f5b0c17SMike Rapoport { 19214f5b0c17SMike Rapoport unsigned long start, end, prev_end = 0; 19224f5b0c17SMike Rapoport int i; 19234f5b0c17SMike Rapoport 19244f5b0c17SMike Rapoport if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 19254f5b0c17SMike Rapoport IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 19264f5b0c17SMike Rapoport return; 19274f5b0c17SMike Rapoport 19284f5b0c17SMike Rapoport /* 19294f5b0c17SMike Rapoport * This relies on each bank being in address order. 19304f5b0c17SMike Rapoport * The banks are sorted previously in bootmem_init(). 19314f5b0c17SMike Rapoport */ 19324f5b0c17SMike Rapoport for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 19334f5b0c17SMike Rapoport #ifdef CONFIG_SPARSEMEM 19344f5b0c17SMike Rapoport /* 19354f5b0c17SMike Rapoport * Take care not to free memmap entries that don't exist 19364f5b0c17SMike Rapoport * due to SPARSEMEM sections which aren't present. 19374f5b0c17SMike Rapoport */ 19384f5b0c17SMike Rapoport start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 19394f5b0c17SMike Rapoport #endif 19404f5b0c17SMike Rapoport /* 1941e2a86800SMike Rapoport * Align down here since many operations in VM subsystem 1942e2a86800SMike Rapoport * presume that there are no holes in the memory map inside 1943e2a86800SMike Rapoport * a pageblock 19444f5b0c17SMike Rapoport */ 1945e2a86800SMike Rapoport start = round_down(start, pageblock_nr_pages); 19464f5b0c17SMike Rapoport 19474f5b0c17SMike Rapoport /* 19484f5b0c17SMike Rapoport * If we had a previous bank, and there is a space 19494f5b0c17SMike Rapoport * between the current bank and the previous, free it. 19504f5b0c17SMike Rapoport */ 19514f5b0c17SMike Rapoport if (prev_end && prev_end < start) 19524f5b0c17SMike Rapoport free_memmap(prev_end, start); 19534f5b0c17SMike Rapoport 19544f5b0c17SMike Rapoport /* 1955e2a86800SMike Rapoport * Align up here since many operations in VM subsystem 1956e2a86800SMike Rapoport * presume that there are no holes in the memory map inside 1957e2a86800SMike Rapoport * a pageblock 19584f5b0c17SMike Rapoport */ 1959e2a86800SMike Rapoport prev_end = ALIGN(end, pageblock_nr_pages); 19604f5b0c17SMike Rapoport } 19614f5b0c17SMike Rapoport 19624f5b0c17SMike Rapoport #ifdef CONFIG_SPARSEMEM 1963f921f53eSMike Rapoport if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 1964f921f53eSMike Rapoport prev_end = ALIGN(end, pageblock_nr_pages); 19654f5b0c17SMike Rapoport free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 1966f921f53eSMike Rapoport } 19674f5b0c17SMike Rapoport #endif 19684f5b0c17SMike Rapoport } 19694f5b0c17SMike Rapoport 1970bda49a81SMike Rapoport static void __init __free_pages_memory(unsigned long start, unsigned long end) 1971bda49a81SMike Rapoport { 1972bda49a81SMike Rapoport int order; 1973bda49a81SMike Rapoport 1974bda49a81SMike Rapoport while (start < end) { 1975bda49a81SMike Rapoport order = min(MAX_ORDER - 1UL, __ffs(start)); 1976bda49a81SMike Rapoport 1977bda49a81SMike Rapoport while (start + (1UL << order) > end) 1978bda49a81SMike Rapoport order--; 1979bda49a81SMike Rapoport 1980bda49a81SMike Rapoport memblock_free_pages(pfn_to_page(start), start, order); 1981bda49a81SMike Rapoport 1982bda49a81SMike Rapoport start += (1UL << order); 1983bda49a81SMike Rapoport } 1984bda49a81SMike Rapoport } 1985bda49a81SMike Rapoport 1986bda49a81SMike Rapoport static unsigned long __init __free_memory_core(phys_addr_t start, 1987bda49a81SMike Rapoport phys_addr_t end) 1988bda49a81SMike Rapoport { 1989bda49a81SMike Rapoport unsigned long start_pfn = PFN_UP(start); 1990bda49a81SMike Rapoport unsigned long end_pfn = min_t(unsigned long, 1991bda49a81SMike Rapoport PFN_DOWN(end), max_low_pfn); 1992bda49a81SMike Rapoport 1993bda49a81SMike Rapoport if (start_pfn >= end_pfn) 1994bda49a81SMike Rapoport return 0; 1995bda49a81SMike Rapoport 1996bda49a81SMike Rapoport __free_pages_memory(start_pfn, end_pfn); 1997bda49a81SMike Rapoport 1998bda49a81SMike Rapoport return end_pfn - start_pfn; 1999bda49a81SMike Rapoport } 2000bda49a81SMike Rapoport 20019092d4f7SMike Rapoport static void __init memmap_init_reserved_pages(void) 20029092d4f7SMike Rapoport { 20039092d4f7SMike Rapoport struct memblock_region *region; 20049092d4f7SMike Rapoport phys_addr_t start, end; 20059092d4f7SMike Rapoport u64 i; 20069092d4f7SMike Rapoport 20079092d4f7SMike Rapoport /* initialize struct pages for the reserved regions */ 20089092d4f7SMike Rapoport for_each_reserved_mem_range(i, &start, &end) 20099092d4f7SMike Rapoport reserve_bootmem_region(start, end); 20109092d4f7SMike Rapoport 20119092d4f7SMike Rapoport /* and also treat struct pages for the NOMAP regions as PageReserved */ 20129092d4f7SMike Rapoport for_each_mem_region(region) { 20139092d4f7SMike Rapoport if (memblock_is_nomap(region)) { 20149092d4f7SMike Rapoport start = region->base; 20159092d4f7SMike Rapoport end = start + region->size; 20169092d4f7SMike Rapoport reserve_bootmem_region(start, end); 20179092d4f7SMike Rapoport } 20189092d4f7SMike Rapoport } 20199092d4f7SMike Rapoport } 20209092d4f7SMike Rapoport 2021bda49a81SMike Rapoport static unsigned long __init free_low_memory_core_early(void) 2022bda49a81SMike Rapoport { 2023bda49a81SMike Rapoport unsigned long count = 0; 2024bda49a81SMike Rapoport phys_addr_t start, end; 2025bda49a81SMike Rapoport u64 i; 2026bda49a81SMike Rapoport 2027bda49a81SMike Rapoport memblock_clear_hotplug(0, -1); 2028bda49a81SMike Rapoport 20299092d4f7SMike Rapoport memmap_init_reserved_pages(); 2030bda49a81SMike Rapoport 2031bda49a81SMike Rapoport /* 2032bda49a81SMike Rapoport * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2033bda49a81SMike Rapoport * because in some case like Node0 doesn't have RAM installed 2034bda49a81SMike Rapoport * low ram will be on Node1 2035bda49a81SMike Rapoport */ 2036bda49a81SMike Rapoport for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2037bda49a81SMike Rapoport NULL) 2038bda49a81SMike Rapoport count += __free_memory_core(start, end); 2039bda49a81SMike Rapoport 2040bda49a81SMike Rapoport return count; 2041bda49a81SMike Rapoport } 2042bda49a81SMike Rapoport 2043bda49a81SMike Rapoport static int reset_managed_pages_done __initdata; 2044bda49a81SMike Rapoport 2045bda49a81SMike Rapoport void reset_node_managed_pages(pg_data_t *pgdat) 2046bda49a81SMike Rapoport { 2047bda49a81SMike Rapoport struct zone *z; 2048bda49a81SMike Rapoport 2049bda49a81SMike Rapoport for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 20509705bea5SArun KS atomic_long_set(&z->managed_pages, 0); 2051bda49a81SMike Rapoport } 2052bda49a81SMike Rapoport 2053bda49a81SMike Rapoport void __init reset_all_zones_managed_pages(void) 2054bda49a81SMike Rapoport { 2055bda49a81SMike Rapoport struct pglist_data *pgdat; 2056bda49a81SMike Rapoport 2057bda49a81SMike Rapoport if (reset_managed_pages_done) 2058bda49a81SMike Rapoport return; 2059bda49a81SMike Rapoport 2060bda49a81SMike Rapoport for_each_online_pgdat(pgdat) 2061bda49a81SMike Rapoport reset_node_managed_pages(pgdat); 2062bda49a81SMike Rapoport 2063bda49a81SMike Rapoport reset_managed_pages_done = 1; 2064bda49a81SMike Rapoport } 2065bda49a81SMike Rapoport 2066bda49a81SMike Rapoport /** 2067bda49a81SMike Rapoport * memblock_free_all - release free pages to the buddy allocator 2068bda49a81SMike Rapoport */ 2069097d43d8SDaeseok Youn void __init memblock_free_all(void) 2070bda49a81SMike Rapoport { 2071bda49a81SMike Rapoport unsigned long pages; 2072bda49a81SMike Rapoport 20734f5b0c17SMike Rapoport free_unused_memmap(); 2074bda49a81SMike Rapoport reset_all_zones_managed_pages(); 2075bda49a81SMike Rapoport 2076bda49a81SMike Rapoport pages = free_low_memory_core_early(); 2077ca79b0c2SArun KS totalram_pages_add(pages); 2078bda49a81SMike Rapoport } 2079bda49a81SMike Rapoport 2080350e88baSMike Rapoport #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 20816d03b885SBenjamin Herrenschmidt 20826d03b885SBenjamin Herrenschmidt static int memblock_debug_show(struct seq_file *m, void *private) 20836d03b885SBenjamin Herrenschmidt { 20846d03b885SBenjamin Herrenschmidt struct memblock_type *type = m->private; 20856d03b885SBenjamin Herrenschmidt struct memblock_region *reg; 20866d03b885SBenjamin Herrenschmidt int i; 20875d63f81cSMiles Chen phys_addr_t end; 20886d03b885SBenjamin Herrenschmidt 20896d03b885SBenjamin Herrenschmidt for (i = 0; i < type->cnt; i++) { 20906d03b885SBenjamin Herrenschmidt reg = &type->regions[i]; 20915d63f81cSMiles Chen end = reg->base + reg->size - 1; 20926d03b885SBenjamin Herrenschmidt 20935d63f81cSMiles Chen seq_printf(m, "%4d: ", i); 20945d63f81cSMiles Chen seq_printf(m, "%pa..%pa\n", ®->base, &end); 20956d03b885SBenjamin Herrenschmidt } 20966d03b885SBenjamin Herrenschmidt return 0; 20976d03b885SBenjamin Herrenschmidt } 20985ad35093SAndy Shevchenko DEFINE_SHOW_ATTRIBUTE(memblock_debug); 20996d03b885SBenjamin Herrenschmidt 21006d03b885SBenjamin Herrenschmidt static int __init memblock_init_debugfs(void) 21016d03b885SBenjamin Herrenschmidt { 21026d03b885SBenjamin Herrenschmidt struct dentry *root = debugfs_create_dir("memblock", NULL); 2103d9f7979cSGreg Kroah-Hartman 21040825a6f9SJoe Perches debugfs_create_file("memory", 0444, root, 21050825a6f9SJoe Perches &memblock.memory, &memblock_debug_fops); 21060825a6f9SJoe Perches debugfs_create_file("reserved", 0444, root, 21070825a6f9SJoe Perches &memblock.reserved, &memblock_debug_fops); 210870210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 210977649905SDavid Hildenbrand debugfs_create_file("physmem", 0444, root, &physmem, 211077649905SDavid Hildenbrand &memblock_debug_fops); 211170210ed9SPhilipp Hachtmann #endif 21126d03b885SBenjamin Herrenschmidt 21136d03b885SBenjamin Herrenschmidt return 0; 21146d03b885SBenjamin Herrenschmidt } 21156d03b885SBenjamin Herrenschmidt __initcall(memblock_init_debugfs); 21166d03b885SBenjamin Herrenschmidt 21176d03b885SBenjamin Herrenschmidt #endif /* CONFIG_DEBUG_FS */ 2118