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 || 290c6975d7cSQian Cai end == MEMBLOCK_ALLOC_NOLEAKTRACE) 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 */ 318a7259df7SMike 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); 369c94afc46SMiaohe Lin if (memblock_reserved_in_slab) 370c94afc46SMiaohe Lin kfree(memblock.reserved.regions); 371c94afc46SMiaohe Lin else 372621d9739SMike Rapoport memblock_free_late(addr, size); 37329f67386SYinghai Lu } 37429f67386SYinghai Lu 37591b540f9SPavel Tatashin if (memblock.memory.regions != memblock_memory_init_regions) { 3763010f876SPavel Tatashin addr = __pa(memblock.memory.regions); 3773010f876SPavel Tatashin size = PAGE_ALIGN(sizeof(struct memblock_region) * 3785e270e25SPhilipp Hachtmann memblock.memory.max); 379c94afc46SMiaohe Lin if (memblock_memory_in_slab) 380c94afc46SMiaohe Lin kfree(memblock.memory.regions); 381c94afc46SMiaohe Lin else 382621d9739SMike Rapoport memblock_free_late(addr, size); 3835e270e25SPhilipp Hachtmann } 3849f3d5eaaSMike Rapoport 3859f3d5eaaSMike Rapoport memblock_memory = NULL; 3863010f876SPavel Tatashin } 3875e270e25SPhilipp Hachtmann #endif 3885e270e25SPhilipp Hachtmann 38948c3b583SGreg Pearson /** 39048c3b583SGreg Pearson * memblock_double_array - double the size of the memblock regions array 39148c3b583SGreg Pearson * @type: memblock type of the regions array being doubled 39248c3b583SGreg Pearson * @new_area_start: starting address of memory range to avoid overlap with 39348c3b583SGreg Pearson * @new_area_size: size of memory range to avoid overlap with 39448c3b583SGreg Pearson * 39548c3b583SGreg Pearson * Double the size of the @type regions array. If memblock is being used to 39648c3b583SGreg Pearson * allocate memory for a new reserved regions array and there is a previously 39748c3b583SGreg Pearson * allocated memory range [@new_area_start, @new_area_start + @new_area_size] 39848c3b583SGreg Pearson * waiting to be reserved, ensure the memory used by the new array does 39948c3b583SGreg Pearson * not overlap. 40048c3b583SGreg Pearson * 40147cec443SMike Rapoport * Return: 40248c3b583SGreg Pearson * 0 on success, -1 on failure. 40348c3b583SGreg Pearson */ 40448c3b583SGreg Pearson static int __init_memblock memblock_double_array(struct memblock_type *type, 40548c3b583SGreg Pearson phys_addr_t new_area_start, 40648c3b583SGreg Pearson phys_addr_t new_area_size) 407142b45a7SBenjamin Herrenschmidt { 408142b45a7SBenjamin Herrenschmidt struct memblock_region *new_array, *old_array; 40929f67386SYinghai Lu phys_addr_t old_alloc_size, new_alloc_size; 410a36aab89SMike Rapoport phys_addr_t old_size, new_size, addr, new_end; 411142b45a7SBenjamin Herrenschmidt int use_slab = slab_is_available(); 412181eb394SGavin Shan int *in_slab; 413142b45a7SBenjamin Herrenschmidt 414142b45a7SBenjamin Herrenschmidt /* We don't allow resizing until we know about the reserved regions 415142b45a7SBenjamin Herrenschmidt * of memory that aren't suitable for allocation 416142b45a7SBenjamin Herrenschmidt */ 417142b45a7SBenjamin Herrenschmidt if (!memblock_can_resize) 418142b45a7SBenjamin Herrenschmidt return -1; 419142b45a7SBenjamin Herrenschmidt 420142b45a7SBenjamin Herrenschmidt /* Calculate new doubled size */ 421142b45a7SBenjamin Herrenschmidt old_size = type->max * sizeof(struct memblock_region); 422142b45a7SBenjamin Herrenschmidt new_size = old_size << 1; 42329f67386SYinghai Lu /* 42429f67386SYinghai Lu * We need to allocated new one align to PAGE_SIZE, 42529f67386SYinghai Lu * so we can free them completely later. 42629f67386SYinghai Lu */ 42729f67386SYinghai Lu old_alloc_size = PAGE_ALIGN(old_size); 42829f67386SYinghai Lu new_alloc_size = PAGE_ALIGN(new_size); 429142b45a7SBenjamin Herrenschmidt 430181eb394SGavin Shan /* Retrieve the slab flag */ 431181eb394SGavin Shan if (type == &memblock.memory) 432181eb394SGavin Shan in_slab = &memblock_memory_in_slab; 433181eb394SGavin Shan else 434181eb394SGavin Shan in_slab = &memblock_reserved_in_slab; 435181eb394SGavin Shan 436a2974133SMike Rapoport /* Try to find some space for it */ 437142b45a7SBenjamin Herrenschmidt if (use_slab) { 438142b45a7SBenjamin Herrenschmidt new_array = kmalloc(new_size, GFP_KERNEL); 4391f5026a7STejun Heo addr = new_array ? __pa(new_array) : 0; 4404e2f0775SGavin Shan } else { 44148c3b583SGreg Pearson /* only exclude range when trying to double reserved.regions */ 44248c3b583SGreg Pearson if (type != &memblock.reserved) 44348c3b583SGreg Pearson new_area_start = new_area_size = 0; 44448c3b583SGreg Pearson 44548c3b583SGreg Pearson addr = memblock_find_in_range(new_area_start + new_area_size, 44648c3b583SGreg Pearson memblock.current_limit, 44729f67386SYinghai Lu new_alloc_size, PAGE_SIZE); 44848c3b583SGreg Pearson if (!addr && new_area_size) 44948c3b583SGreg Pearson addr = memblock_find_in_range(0, 45048c3b583SGreg Pearson min(new_area_start, memblock.current_limit), 45129f67386SYinghai Lu new_alloc_size, PAGE_SIZE); 45248c3b583SGreg Pearson 45315674868SSachin Kamat new_array = addr ? __va(addr) : NULL; 4544e2f0775SGavin Shan } 4551f5026a7STejun Heo if (!addr) { 456142b45a7SBenjamin Herrenschmidt pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 4570262d9c8SHeiko Carstens type->name, type->max, type->max * 2); 458142b45a7SBenjamin Herrenschmidt return -1; 459142b45a7SBenjamin Herrenschmidt } 460142b45a7SBenjamin Herrenschmidt 461a36aab89SMike Rapoport new_end = addr + new_size - 1; 462a36aab89SMike Rapoport memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]", 463a36aab89SMike Rapoport type->name, type->max * 2, &addr, &new_end); 464ea9e4376SYinghai Lu 465fd07383bSAndrew Morton /* 466fd07383bSAndrew Morton * Found space, we now need to move the array over before we add the 467fd07383bSAndrew Morton * reserved region since it may be our reserved array itself that is 468fd07383bSAndrew Morton * full. 469142b45a7SBenjamin Herrenschmidt */ 470142b45a7SBenjamin Herrenschmidt memcpy(new_array, type->regions, old_size); 471142b45a7SBenjamin Herrenschmidt memset(new_array + type->max, 0, old_size); 472142b45a7SBenjamin Herrenschmidt old_array = type->regions; 473142b45a7SBenjamin Herrenschmidt type->regions = new_array; 474142b45a7SBenjamin Herrenschmidt type->max <<= 1; 475142b45a7SBenjamin Herrenschmidt 476fd07383bSAndrew Morton /* Free old array. We needn't free it if the array is the static one */ 477181eb394SGavin Shan if (*in_slab) 478181eb394SGavin Shan kfree(old_array); 479181eb394SGavin Shan else if (old_array != memblock_memory_init_regions && 480142b45a7SBenjamin Herrenschmidt old_array != memblock_reserved_init_regions) 4814421cca0SMike Rapoport memblock_free(old_array, old_alloc_size); 482142b45a7SBenjamin Herrenschmidt 483fd07383bSAndrew Morton /* 484fd07383bSAndrew Morton * Reserve the new array if that comes from the memblock. Otherwise, we 485fd07383bSAndrew Morton * needn't do it 486181eb394SGavin Shan */ 487181eb394SGavin Shan if (!use_slab) 48829f67386SYinghai Lu BUG_ON(memblock_reserve(addr, new_alloc_size)); 489181eb394SGavin Shan 490181eb394SGavin Shan /* Update slab flag */ 491181eb394SGavin Shan *in_slab = use_slab; 492181eb394SGavin Shan 493142b45a7SBenjamin Herrenschmidt return 0; 494142b45a7SBenjamin Herrenschmidt } 495142b45a7SBenjamin Herrenschmidt 496784656f9STejun Heo /** 497784656f9STejun Heo * memblock_merge_regions - merge neighboring compatible regions 498784656f9STejun Heo * @type: memblock type to scan 499784656f9STejun Heo * 500784656f9STejun Heo * Scan @type and merge neighboring compatible regions. 501784656f9STejun Heo */ 502784656f9STejun Heo static void __init_memblock memblock_merge_regions(struct memblock_type *type) 503784656f9STejun Heo { 504784656f9STejun Heo int i = 0; 505784656f9STejun Heo 506784656f9STejun Heo /* cnt never goes below 1 */ 507784656f9STejun Heo while (i < type->cnt - 1) { 508784656f9STejun Heo struct memblock_region *this = &type->regions[i]; 509784656f9STejun Heo struct memblock_region *next = &type->regions[i + 1]; 510784656f9STejun Heo 5117c0caeb8STejun Heo if (this->base + this->size != next->base || 5127c0caeb8STejun Heo memblock_get_region_node(this) != 51366a20757STang Chen memblock_get_region_node(next) || 51466a20757STang Chen this->flags != next->flags) { 515784656f9STejun Heo BUG_ON(this->base + this->size > next->base); 516784656f9STejun Heo i++; 517784656f9STejun Heo continue; 518784656f9STejun Heo } 519784656f9STejun Heo 520784656f9STejun Heo this->size += next->size; 521c0232ae8SLin Feng /* move forward from next + 1, index of which is i + 2 */ 522c0232ae8SLin Feng memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 523784656f9STejun Heo type->cnt--; 524784656f9STejun Heo } 525784656f9STejun Heo } 526784656f9STejun Heo 527784656f9STejun Heo /** 528784656f9STejun Heo * memblock_insert_region - insert new memblock region 529784656f9STejun Heo * @type: memblock type to insert into 530784656f9STejun Heo * @idx: index for the insertion point 531784656f9STejun Heo * @base: base address of the new region 532784656f9STejun Heo * @size: size of the new region 533209ff86dSTang Chen * @nid: node id of the new region 53466a20757STang Chen * @flags: flags of the new region 535784656f9STejun Heo * 536784656f9STejun Heo * Insert new memblock region [@base, @base + @size) into @type at @idx. 537412d0008SAlexander Kuleshov * @type must already have extra room to accommodate the new region. 538784656f9STejun Heo */ 539784656f9STejun Heo static void __init_memblock memblock_insert_region(struct memblock_type *type, 540784656f9STejun Heo int idx, phys_addr_t base, 54166a20757STang Chen phys_addr_t size, 542e1720feeSMike Rapoport int nid, 543e1720feeSMike Rapoport enum memblock_flags flags) 544784656f9STejun Heo { 545784656f9STejun Heo struct memblock_region *rgn = &type->regions[idx]; 546784656f9STejun Heo 547784656f9STejun Heo BUG_ON(type->cnt >= type->max); 548784656f9STejun Heo memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 549784656f9STejun Heo rgn->base = base; 550784656f9STejun Heo rgn->size = size; 55166a20757STang Chen rgn->flags = flags; 5527c0caeb8STejun Heo memblock_set_region_node(rgn, nid); 553784656f9STejun Heo type->cnt++; 5541440c4e2STejun Heo type->total_size += size; 555784656f9STejun Heo } 556784656f9STejun Heo 557784656f9STejun Heo /** 558f1af9d3aSPhilipp Hachtmann * memblock_add_range - add new memblock region 559784656f9STejun Heo * @type: memblock type to add new region into 560784656f9STejun Heo * @base: base address of the new region 561784656f9STejun Heo * @size: size of the new region 5627fb0bc3fSTejun Heo * @nid: nid of the new region 56366a20757STang Chen * @flags: flags of the new region 564784656f9STejun Heo * 565784656f9STejun Heo * Add new memblock region [@base, @base + @size) into @type. The new region 566784656f9STejun Heo * is allowed to overlap with existing ones - overlaps don't affect already 567784656f9STejun Heo * existing regions. @type is guaranteed to be minimal (all neighbouring 568784656f9STejun Heo * compatible regions are merged) after the addition. 569784656f9STejun Heo * 57047cec443SMike Rapoport * Return: 571784656f9STejun Heo * 0 on success, -errno on failure. 572784656f9STejun Heo */ 57302634a44SAnshuman Khandual static int __init_memblock memblock_add_range(struct memblock_type *type, 57466a20757STang Chen phys_addr_t base, phys_addr_t size, 575e1720feeSMike Rapoport int nid, enum memblock_flags flags) 57695f72d1eSYinghai Lu { 577784656f9STejun Heo bool insert = false; 578eb18f1b5STejun Heo phys_addr_t obase = base; 579eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 5808c9c1701SAlexander Kuleshov int idx, nr_new; 5818c9c1701SAlexander Kuleshov struct memblock_region *rgn; 58295f72d1eSYinghai Lu 583b3dc627cSTejun Heo if (!size) 584b3dc627cSTejun Heo return 0; 585b3dc627cSTejun Heo 586784656f9STejun Heo /* special case for empty array */ 587784656f9STejun Heo if (type->regions[0].size == 0) { 5881440c4e2STejun Heo WARN_ON(type->cnt != 1 || type->total_size); 589784656f9STejun Heo type->regions[0].base = base; 590784656f9STejun Heo type->regions[0].size = size; 59166a20757STang Chen type->regions[0].flags = flags; 5927fb0bc3fSTejun Heo memblock_set_region_node(&type->regions[0], nid); 5931440c4e2STejun Heo type->total_size = size; 594784656f9STejun Heo return 0; 595784656f9STejun Heo } 596784656f9STejun Heo repeat: 597784656f9STejun Heo /* 598784656f9STejun Heo * The following is executed twice. Once with %false @insert and 599784656f9STejun Heo * then with %true. The first counts the number of regions needed 600412d0008SAlexander Kuleshov * to accommodate the new area. The second actually inserts them. 601784656f9STejun Heo */ 602784656f9STejun Heo base = obase; 603784656f9STejun Heo nr_new = 0; 604784656f9STejun Heo 60566e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 606784656f9STejun Heo phys_addr_t rbase = rgn->base; 607784656f9STejun Heo phys_addr_t rend = rbase + rgn->size; 6088f7a6605SBenjamin Herrenschmidt 609784656f9STejun Heo if (rbase >= end) 6108f7a6605SBenjamin Herrenschmidt break; 611784656f9STejun Heo if (rend <= base) 612784656f9STejun Heo continue; 613784656f9STejun Heo /* 614784656f9STejun Heo * @rgn overlaps. If it separates the lower part of new 615784656f9STejun Heo * area, insert that portion. 6168f7a6605SBenjamin Herrenschmidt */ 617784656f9STejun Heo if (rbase > base) { 618a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 619c0a29498SWei Yang WARN_ON(nid != memblock_get_region_node(rgn)); 620c0a29498SWei Yang #endif 6214fcab5f4SWei Yang WARN_ON(flags != rgn->flags); 622784656f9STejun Heo nr_new++; 623784656f9STejun Heo if (insert) 6248c9c1701SAlexander Kuleshov memblock_insert_region(type, idx++, base, 62566a20757STang Chen rbase - base, nid, 62666a20757STang Chen flags); 627784656f9STejun Heo } 628784656f9STejun Heo /* area below @rend is dealt with, forget about it */ 629784656f9STejun Heo base = min(rend, end); 6308f7a6605SBenjamin Herrenschmidt } 6318f7a6605SBenjamin Herrenschmidt 632784656f9STejun Heo /* insert the remaining portion */ 633784656f9STejun Heo if (base < end) { 634784656f9STejun Heo nr_new++; 635784656f9STejun Heo if (insert) 6368c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, base, end - base, 63766a20757STang Chen nid, flags); 6388f7a6605SBenjamin Herrenschmidt } 6398f7a6605SBenjamin Herrenschmidt 640ef3cc4dbSnimisolo if (!nr_new) 641ef3cc4dbSnimisolo return 0; 642ef3cc4dbSnimisolo 643784656f9STejun Heo /* 644784656f9STejun Heo * If this was the first round, resize array and repeat for actual 645784656f9STejun Heo * insertions; otherwise, merge and return. 6468f7a6605SBenjamin Herrenschmidt */ 647784656f9STejun Heo if (!insert) { 648784656f9STejun Heo while (type->cnt + nr_new > type->max) 64948c3b583SGreg Pearson if (memblock_double_array(type, obase, size) < 0) 650784656f9STejun Heo return -ENOMEM; 651784656f9STejun Heo insert = true; 652784656f9STejun Heo goto repeat; 65395f72d1eSYinghai Lu } else { 654784656f9STejun Heo memblock_merge_regions(type); 65595f72d1eSYinghai Lu return 0; 65695f72d1eSYinghai Lu } 657784656f9STejun Heo } 65895f72d1eSYinghai Lu 65948a833ccSMike Rapoport /** 66048a833ccSMike Rapoport * memblock_add_node - add new memblock region within a NUMA node 66148a833ccSMike Rapoport * @base: base address of the new region 66248a833ccSMike Rapoport * @size: size of the new region 66348a833ccSMike Rapoport * @nid: nid of the new region 664952eea9bSDavid Hildenbrand * @flags: flags of the new region 66548a833ccSMike Rapoport * 66648a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 66748a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 66848a833ccSMike Rapoport * 66948a833ccSMike Rapoport * Return: 67048a833ccSMike Rapoport * 0 on success, -errno on failure. 67148a833ccSMike Rapoport */ 6727fb0bc3fSTejun Heo int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 673952eea9bSDavid Hildenbrand int nid, enum memblock_flags flags) 6747fb0bc3fSTejun Heo { 67500974b9aSGeert Uytterhoeven phys_addr_t end = base + size - 1; 67600974b9aSGeert Uytterhoeven 677952eea9bSDavid Hildenbrand memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__, 678952eea9bSDavid Hildenbrand &base, &end, nid, flags, (void *)_RET_IP_); 67900974b9aSGeert Uytterhoeven 680952eea9bSDavid Hildenbrand return memblock_add_range(&memblock.memory, base, size, nid, flags); 6817fb0bc3fSTejun Heo } 6827fb0bc3fSTejun Heo 68348a833ccSMike Rapoport /** 68448a833ccSMike Rapoport * memblock_add - add new memblock region 68548a833ccSMike Rapoport * @base: base address of the new region 68648a833ccSMike Rapoport * @size: size of the new region 68748a833ccSMike Rapoport * 68848a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 68948a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 69048a833ccSMike Rapoport * 69148a833ccSMike Rapoport * Return: 69248a833ccSMike Rapoport * 0 on success, -errno on failure. 69348a833ccSMike Rapoport */ 694f705ac4bSAlexander Kuleshov int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 6956a4055bcSAlexander Kuleshov { 6965d63f81cSMiles Chen phys_addr_t end = base + size - 1; 6975d63f81cSMiles Chen 698a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 6995d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 7006a4055bcSAlexander Kuleshov 701f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 70295f72d1eSYinghai Lu } 70395f72d1eSYinghai Lu 7046a9ceb31STejun Heo /** 7056a9ceb31STejun Heo * memblock_isolate_range - isolate given range into disjoint memblocks 7066a9ceb31STejun Heo * @type: memblock type to isolate range for 7076a9ceb31STejun Heo * @base: base of range to isolate 7086a9ceb31STejun Heo * @size: size of range to isolate 7096a9ceb31STejun Heo * @start_rgn: out parameter for the start of isolated region 7106a9ceb31STejun Heo * @end_rgn: out parameter for the end of isolated region 7116a9ceb31STejun Heo * 7126a9ceb31STejun Heo * Walk @type and ensure that regions don't cross the boundaries defined by 7136a9ceb31STejun Heo * [@base, @base + @size). Crossing regions are split at the boundaries, 7146a9ceb31STejun Heo * which may create at most two more regions. The index of the first 7156a9ceb31STejun Heo * region inside the range is returned in *@start_rgn and end in *@end_rgn. 7166a9ceb31STejun Heo * 71747cec443SMike Rapoport * Return: 7186a9ceb31STejun Heo * 0 on success, -errno on failure. 7196a9ceb31STejun Heo */ 7206a9ceb31STejun Heo static int __init_memblock memblock_isolate_range(struct memblock_type *type, 7216a9ceb31STejun Heo phys_addr_t base, phys_addr_t size, 7226a9ceb31STejun Heo int *start_rgn, int *end_rgn) 7236a9ceb31STejun Heo { 724eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 7258c9c1701SAlexander Kuleshov int idx; 7268c9c1701SAlexander Kuleshov struct memblock_region *rgn; 7276a9ceb31STejun Heo 7286a9ceb31STejun Heo *start_rgn = *end_rgn = 0; 7296a9ceb31STejun Heo 730b3dc627cSTejun Heo if (!size) 731b3dc627cSTejun Heo return 0; 732b3dc627cSTejun Heo 7336a9ceb31STejun Heo /* we'll create at most two more regions */ 7346a9ceb31STejun Heo while (type->cnt + 2 > type->max) 73548c3b583SGreg Pearson if (memblock_double_array(type, base, size) < 0) 7366a9ceb31STejun Heo return -ENOMEM; 7376a9ceb31STejun Heo 73866e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 7396a9ceb31STejun Heo phys_addr_t rbase = rgn->base; 7406a9ceb31STejun Heo phys_addr_t rend = rbase + rgn->size; 7416a9ceb31STejun Heo 7426a9ceb31STejun Heo if (rbase >= end) 7436a9ceb31STejun Heo break; 7446a9ceb31STejun Heo if (rend <= base) 7456a9ceb31STejun Heo continue; 7466a9ceb31STejun Heo 7476a9ceb31STejun Heo if (rbase < base) { 7486a9ceb31STejun Heo /* 7496a9ceb31STejun Heo * @rgn intersects from below. Split and continue 7506a9ceb31STejun Heo * to process the next region - the new top half. 7516a9ceb31STejun Heo */ 7526a9ceb31STejun Heo rgn->base = base; 7531440c4e2STejun Heo rgn->size -= base - rbase; 7541440c4e2STejun Heo type->total_size -= base - rbase; 7558c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, rbase, base - rbase, 75666a20757STang Chen memblock_get_region_node(rgn), 75766a20757STang Chen rgn->flags); 7586a9ceb31STejun Heo } else if (rend > end) { 7596a9ceb31STejun Heo /* 7606a9ceb31STejun Heo * @rgn intersects from above. Split and redo the 7616a9ceb31STejun Heo * current region - the new bottom half. 7626a9ceb31STejun Heo */ 7636a9ceb31STejun Heo rgn->base = end; 7641440c4e2STejun Heo rgn->size -= end - rbase; 7651440c4e2STejun Heo type->total_size -= end - rbase; 7668c9c1701SAlexander Kuleshov memblock_insert_region(type, idx--, rbase, end - rbase, 76766a20757STang Chen memblock_get_region_node(rgn), 76866a20757STang Chen rgn->flags); 7696a9ceb31STejun Heo } else { 7706a9ceb31STejun Heo /* @rgn is fully contained, record it */ 7716a9ceb31STejun Heo if (!*end_rgn) 7728c9c1701SAlexander Kuleshov *start_rgn = idx; 7738c9c1701SAlexander Kuleshov *end_rgn = idx + 1; 7746a9ceb31STejun Heo } 7756a9ceb31STejun Heo } 7766a9ceb31STejun Heo 7776a9ceb31STejun Heo return 0; 7786a9ceb31STejun Heo } 7796a9ceb31STejun Heo 78035bd16a2SAlexander Kuleshov static int __init_memblock memblock_remove_range(struct memblock_type *type, 7818f7a6605SBenjamin Herrenschmidt phys_addr_t base, phys_addr_t size) 78295f72d1eSYinghai Lu { 78371936180STejun Heo int start_rgn, end_rgn; 78471936180STejun Heo int i, ret; 78595f72d1eSYinghai Lu 78671936180STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 78771936180STejun Heo if (ret) 78871936180STejun Heo return ret; 78995f72d1eSYinghai Lu 79071936180STejun Heo for (i = end_rgn - 1; i >= start_rgn; i--) 79171936180STejun Heo memblock_remove_region(type, i); 79295f72d1eSYinghai Lu return 0; 79395f72d1eSYinghai Lu } 79495f72d1eSYinghai Lu 795581adcbeSTejun Heo int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 79695f72d1eSYinghai Lu { 79725cf23d7SMinchan Kim phys_addr_t end = base + size - 1; 79825cf23d7SMinchan Kim 799a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 80025cf23d7SMinchan Kim &base, &end, (void *)_RET_IP_); 80125cf23d7SMinchan Kim 802f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.memory, base, size); 80395f72d1eSYinghai Lu } 80495f72d1eSYinghai Lu 8054d72868cSMike Rapoport /** 8064421cca0SMike Rapoport * memblock_free - free boot memory allocation 80777e02cf5SLinus Torvalds * @ptr: starting address of the boot memory allocation 80877e02cf5SLinus Torvalds * @size: size of the boot memory block in bytes 80977e02cf5SLinus Torvalds * 81077e02cf5SLinus Torvalds * Free boot memory block previously allocated by memblock_alloc_xx() API. 81177e02cf5SLinus Torvalds * The freeing memory will not be released to the buddy allocator. 81277e02cf5SLinus Torvalds */ 8134421cca0SMike Rapoport void __init_memblock memblock_free(void *ptr, size_t size) 81477e02cf5SLinus Torvalds { 81577e02cf5SLinus Torvalds if (ptr) 8163ecc6834SMike Rapoport memblock_phys_free(__pa(ptr), size); 81777e02cf5SLinus Torvalds } 81877e02cf5SLinus Torvalds 81977e02cf5SLinus Torvalds /** 8203ecc6834SMike Rapoport * memblock_phys_free - free boot memory block 8214d72868cSMike Rapoport * @base: phys starting address of the boot memory block 8224d72868cSMike Rapoport * @size: size of the boot memory block in bytes 8234d72868cSMike Rapoport * 8244d72868cSMike Rapoport * Free boot memory block previously allocated by memblock_alloc_xx() API. 8254d72868cSMike Rapoport * The freeing memory will not be released to the buddy allocator. 8264d72868cSMike Rapoport */ 8273ecc6834SMike Rapoport int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size) 82895f72d1eSYinghai Lu { 8295d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8305d63f81cSMiles Chen 831a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 8325d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 83324aa0788STejun Heo 8349099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 835f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.reserved, base, size); 83695f72d1eSYinghai Lu } 83795f72d1eSYinghai Lu 838f705ac4bSAlexander Kuleshov int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 83995f72d1eSYinghai Lu { 8405d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8415d63f81cSMiles Chen 842a090d711SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 8435d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 84495f72d1eSYinghai Lu 845f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 84695f72d1eSYinghai Lu } 84795f72d1eSYinghai Lu 84802634a44SAnshuman Khandual #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 84902634a44SAnshuman Khandual int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size) 85002634a44SAnshuman Khandual { 85102634a44SAnshuman Khandual phys_addr_t end = base + size - 1; 85202634a44SAnshuman Khandual 85302634a44SAnshuman Khandual memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 85402634a44SAnshuman Khandual &base, &end, (void *)_RET_IP_); 85502634a44SAnshuman Khandual 85677649905SDavid Hildenbrand return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0); 85702634a44SAnshuman Khandual } 85802634a44SAnshuman Khandual #endif 85902634a44SAnshuman Khandual 86035fd0808STejun Heo /** 86147cec443SMike Rapoport * memblock_setclr_flag - set or clear flag for a memory region 86247cec443SMike Rapoport * @base: base address of the region 86347cec443SMike Rapoport * @size: size of the region 86447cec443SMike Rapoport * @set: set or clear the flag 8658958b249SHaitao Shi * @flag: the flag to update 86666b16edfSTang Chen * 8674308ce17STony Luck * This function isolates region [@base, @base + @size), and sets/clears flag 86866b16edfSTang Chen * 86947cec443SMike Rapoport * Return: 0 on success, -errno on failure. 87066b16edfSTang Chen */ 8714308ce17STony Luck static int __init_memblock memblock_setclr_flag(phys_addr_t base, 8724308ce17STony Luck phys_addr_t size, int set, int flag) 87366b16edfSTang Chen { 87466b16edfSTang Chen struct memblock_type *type = &memblock.memory; 87566b16edfSTang Chen int i, ret, start_rgn, end_rgn; 87666b16edfSTang Chen 87766b16edfSTang Chen ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 87866b16edfSTang Chen if (ret) 87966b16edfSTang Chen return ret; 88066b16edfSTang Chen 881fe145124SMike Rapoport for (i = start_rgn; i < end_rgn; i++) { 882fe145124SMike Rapoport struct memblock_region *r = &type->regions[i]; 883fe145124SMike Rapoport 8844308ce17STony Luck if (set) 885fe145124SMike Rapoport r->flags |= flag; 8864308ce17STony Luck else 887fe145124SMike Rapoport r->flags &= ~flag; 888fe145124SMike Rapoport } 88966b16edfSTang Chen 89066b16edfSTang Chen memblock_merge_regions(type); 89166b16edfSTang Chen return 0; 89266b16edfSTang Chen } 89366b16edfSTang Chen 89466b16edfSTang Chen /** 8954308ce17STony Luck * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 8964308ce17STony Luck * @base: the base phys addr of the region 8974308ce17STony Luck * @size: the size of the region 8984308ce17STony Luck * 89947cec443SMike Rapoport * Return: 0 on success, -errno on failure. 9004308ce17STony Luck */ 9014308ce17STony Luck int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 9024308ce17STony Luck { 9034308ce17STony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 9044308ce17STony Luck } 9054308ce17STony Luck 9064308ce17STony Luck /** 90766b16edfSTang Chen * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 90866b16edfSTang Chen * @base: the base phys addr of the region 90966b16edfSTang Chen * @size: the size of the region 91066b16edfSTang Chen * 91147cec443SMike Rapoport * Return: 0 on success, -errno on failure. 91266b16edfSTang Chen */ 91366b16edfSTang Chen int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 91466b16edfSTang Chen { 9154308ce17STony Luck return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 91666b16edfSTang Chen } 91766b16edfSTang Chen 91866b16edfSTang Chen /** 919a3f5bafcSTony Luck * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 920a3f5bafcSTony Luck * @base: the base phys addr of the region 921a3f5bafcSTony Luck * @size: the size of the region 922a3f5bafcSTony Luck * 92347cec443SMike Rapoport * Return: 0 on success, -errno on failure. 924a3f5bafcSTony Luck */ 925a3f5bafcSTony Luck int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 926a3f5bafcSTony Luck { 927a3f5bafcSTony Luck system_has_some_mirror = true; 928a3f5bafcSTony Luck 929a3f5bafcSTony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 930a3f5bafcSTony Luck } 931a3f5bafcSTony Luck 932bf3d3cc5SArd Biesheuvel /** 933bf3d3cc5SArd Biesheuvel * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 934bf3d3cc5SArd Biesheuvel * @base: the base phys addr of the region 935bf3d3cc5SArd Biesheuvel * @size: the size of the region 936bf3d3cc5SArd Biesheuvel * 9379092d4f7SMike Rapoport * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the 9389092d4f7SMike Rapoport * direct mapping of the physical memory. These regions will still be 9399092d4f7SMike Rapoport * covered by the memory map. The struct page representing NOMAP memory 9409092d4f7SMike Rapoport * frames in the memory map will be PageReserved() 9419092d4f7SMike Rapoport * 942658aafc8SMike Rapoport * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from 943658aafc8SMike Rapoport * memblock, the caller must inform kmemleak to ignore that memory 944658aafc8SMike Rapoport * 94547cec443SMike Rapoport * Return: 0 on success, -errno on failure. 946bf3d3cc5SArd Biesheuvel */ 947bf3d3cc5SArd Biesheuvel int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 948bf3d3cc5SArd Biesheuvel { 9496c9a5455SMike Rapoport return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 950bf3d3cc5SArd Biesheuvel } 951a3f5bafcSTony Luck 952a3f5bafcSTony Luck /** 9534c546b8aSAKASHI Takahiro * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 9544c546b8aSAKASHI Takahiro * @base: the base phys addr of the region 9554c546b8aSAKASHI Takahiro * @size: the size of the region 9564c546b8aSAKASHI Takahiro * 95747cec443SMike Rapoport * Return: 0 on success, -errno on failure. 9584c546b8aSAKASHI Takahiro */ 9594c546b8aSAKASHI Takahiro int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 9604c546b8aSAKASHI Takahiro { 9614c546b8aSAKASHI Takahiro return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 9624c546b8aSAKASHI Takahiro } 9634c546b8aSAKASHI Takahiro 9649f3d5eaaSMike Rapoport static bool should_skip_region(struct memblock_type *type, 9659f3d5eaaSMike Rapoport struct memblock_region *m, 9669f3d5eaaSMike Rapoport int nid, int flags) 967c9a688a3SMike Rapoport { 968c9a688a3SMike Rapoport int m_nid = memblock_get_region_node(m); 969c9a688a3SMike Rapoport 9709f3d5eaaSMike Rapoport /* we never skip regions when iterating memblock.reserved or physmem */ 9719f3d5eaaSMike Rapoport if (type != memblock_memory) 9729f3d5eaaSMike Rapoport return false; 9739f3d5eaaSMike Rapoport 974c9a688a3SMike Rapoport /* only memory regions are associated with nodes, check it */ 975c9a688a3SMike Rapoport if (nid != NUMA_NO_NODE && nid != m_nid) 976c9a688a3SMike Rapoport return true; 977c9a688a3SMike Rapoport 978c9a688a3SMike Rapoport /* skip hotpluggable memory regions if needed */ 97979e482e9SMike Rapoport if (movable_node_is_enabled() && memblock_is_hotpluggable(m) && 98079e482e9SMike Rapoport !(flags & MEMBLOCK_HOTPLUG)) 981c9a688a3SMike Rapoport return true; 982c9a688a3SMike Rapoport 983c9a688a3SMike Rapoport /* if we want mirror memory skip non-mirror memory regions */ 984c9a688a3SMike Rapoport if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 985c9a688a3SMike Rapoport return true; 986c9a688a3SMike Rapoport 987c9a688a3SMike Rapoport /* skip nomap memory unless we were asked for it explicitly */ 988c9a688a3SMike Rapoport if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 989c9a688a3SMike Rapoport return true; 990c9a688a3SMike Rapoport 991f7892d8eSDavid Hildenbrand /* skip driver-managed memory unless we were asked for it explicitly */ 992f7892d8eSDavid Hildenbrand if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m)) 993f7892d8eSDavid Hildenbrand return true; 994f7892d8eSDavid Hildenbrand 995c9a688a3SMike Rapoport return false; 996c9a688a3SMike Rapoport } 997c9a688a3SMike Rapoport 9988e7a7f86SRobin Holt /** 999a2974133SMike Rapoport * __next_mem_range - next function for for_each_free_mem_range() etc. 100035fd0808STejun Heo * @idx: pointer to u64 loop variable 1001b1154233SGrygorii Strashko * @nid: node selector, %NUMA_NO_NODE for all nodes 1002fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 1003f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 1004f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 1005dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1006dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1007dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 100835fd0808STejun Heo * 1009f1af9d3aSPhilipp Hachtmann * Find the first area from *@idx which matches @nid, fill the out 101035fd0808STejun Heo * parameters, and update *@idx for the next iteration. The lower 32bit of 1011f1af9d3aSPhilipp Hachtmann * *@idx contains index into type_a and the upper 32bit indexes the 1012f1af9d3aSPhilipp Hachtmann * areas before each region in type_b. For example, if type_b regions 101335fd0808STejun Heo * look like the following, 101435fd0808STejun Heo * 101535fd0808STejun Heo * 0:[0-16), 1:[32-48), 2:[128-130) 101635fd0808STejun Heo * 101735fd0808STejun Heo * The upper 32bit indexes the following regions. 101835fd0808STejun Heo * 101935fd0808STejun Heo * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 102035fd0808STejun Heo * 102135fd0808STejun Heo * As both region arrays are sorted, the function advances the two indices 102235fd0808STejun Heo * in lockstep and returns each intersection. 102335fd0808STejun Heo */ 102477649905SDavid Hildenbrand void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags, 1025f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 102677649905SDavid Hildenbrand struct memblock_type *type_b, phys_addr_t *out_start, 102735fd0808STejun Heo phys_addr_t *out_end, int *out_nid) 102835fd0808STejun Heo { 1029f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 1030f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 1031b1154233SGrygorii Strashko 1032f1af9d3aSPhilipp Hachtmann if (WARN_ONCE(nid == MAX_NUMNODES, 1033f1af9d3aSPhilipp Hachtmann "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1034560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 103535fd0808STejun Heo 1036f1af9d3aSPhilipp Hachtmann for (; idx_a < type_a->cnt; idx_a++) { 1037f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1038f1af9d3aSPhilipp Hachtmann 103935fd0808STejun Heo phys_addr_t m_start = m->base; 104035fd0808STejun Heo phys_addr_t m_end = m->base + m->size; 1041f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 104235fd0808STejun Heo 10439f3d5eaaSMike Rapoport if (should_skip_region(type_a, m, nid, flags)) 1044bf3d3cc5SArd Biesheuvel continue; 1045bf3d3cc5SArd Biesheuvel 1046f1af9d3aSPhilipp Hachtmann if (!type_b) { 1047f1af9d3aSPhilipp Hachtmann if (out_start) 1048f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1049f1af9d3aSPhilipp Hachtmann if (out_end) 1050f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1051f1af9d3aSPhilipp Hachtmann if (out_nid) 1052f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1053f1af9d3aSPhilipp Hachtmann idx_a++; 1054f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1055f1af9d3aSPhilipp Hachtmann return; 1056f1af9d3aSPhilipp Hachtmann } 105735fd0808STejun Heo 1058f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1059f1af9d3aSPhilipp Hachtmann for (; idx_b < type_b->cnt + 1; idx_b++) { 1060f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1061f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1062f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1063f1af9d3aSPhilipp Hachtmann 1064f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1065f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1066f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 10671c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1068f1af9d3aSPhilipp Hachtmann 1069f1af9d3aSPhilipp Hachtmann /* 1070f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1071f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1072f1af9d3aSPhilipp Hachtmann */ 107335fd0808STejun Heo if (r_start >= m_end) 107435fd0808STejun Heo break; 107535fd0808STejun Heo /* if the two regions intersect, we're done */ 107635fd0808STejun Heo if (m_start < r_end) { 107735fd0808STejun Heo if (out_start) 1078f1af9d3aSPhilipp Hachtmann *out_start = 1079f1af9d3aSPhilipp Hachtmann max(m_start, r_start); 108035fd0808STejun Heo if (out_end) 108135fd0808STejun Heo *out_end = min(m_end, r_end); 108235fd0808STejun Heo if (out_nid) 1083f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 108435fd0808STejun Heo /* 1085f1af9d3aSPhilipp Hachtmann * The region which ends first is 1086f1af9d3aSPhilipp Hachtmann * advanced for the next iteration. 108735fd0808STejun Heo */ 108835fd0808STejun Heo if (m_end <= r_end) 1089f1af9d3aSPhilipp Hachtmann idx_a++; 109035fd0808STejun Heo else 1091f1af9d3aSPhilipp Hachtmann idx_b++; 1092f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 109335fd0808STejun Heo return; 109435fd0808STejun Heo } 109535fd0808STejun Heo } 109635fd0808STejun Heo } 109735fd0808STejun Heo 109835fd0808STejun Heo /* signal end of iteration */ 109935fd0808STejun Heo *idx = ULLONG_MAX; 110035fd0808STejun Heo } 110135fd0808STejun Heo 11027bd0b0f0STejun Heo /** 1103f1af9d3aSPhilipp Hachtmann * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1104f1af9d3aSPhilipp Hachtmann * 11057bd0b0f0STejun Heo * @idx: pointer to u64 loop variable 1106ad5ea8cdSAlexander Kuleshov * @nid: node selector, %NUMA_NO_NODE for all nodes 1107fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 1108f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 1109f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 1110dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1111dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1112dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 11137bd0b0f0STejun Heo * 111447cec443SMike Rapoport * Finds the next range from type_a which is not marked as unsuitable 111547cec443SMike Rapoport * in type_b. 111647cec443SMike Rapoport * 1117f1af9d3aSPhilipp Hachtmann * Reverse of __next_mem_range(). 11187bd0b0f0STejun Heo */ 1119e1720feeSMike Rapoport void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1120e1720feeSMike Rapoport enum memblock_flags flags, 1121f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 1122f1af9d3aSPhilipp Hachtmann struct memblock_type *type_b, 11237bd0b0f0STejun Heo phys_addr_t *out_start, 11247bd0b0f0STejun Heo phys_addr_t *out_end, int *out_nid) 11257bd0b0f0STejun Heo { 1126f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 1127f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 1128b1154233SGrygorii Strashko 1129560dca27SGrygorii Strashko if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1130560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 11317bd0b0f0STejun Heo 11327bd0b0f0STejun Heo if (*idx == (u64)ULLONG_MAX) { 1133f1af9d3aSPhilipp Hachtmann idx_a = type_a->cnt - 1; 1134e47608abSzijun_hu if (type_b != NULL) 1135f1af9d3aSPhilipp Hachtmann idx_b = type_b->cnt; 1136e47608abSzijun_hu else 1137e47608abSzijun_hu idx_b = 0; 11387bd0b0f0STejun Heo } 11397bd0b0f0STejun Heo 1140f1af9d3aSPhilipp Hachtmann for (; idx_a >= 0; idx_a--) { 1141f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1142f1af9d3aSPhilipp Hachtmann 11437bd0b0f0STejun Heo phys_addr_t m_start = m->base; 11447bd0b0f0STejun Heo phys_addr_t m_end = m->base + m->size; 1145f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 11467bd0b0f0STejun Heo 11479f3d5eaaSMike Rapoport if (should_skip_region(type_a, m, nid, flags)) 1148bf3d3cc5SArd Biesheuvel continue; 1149bf3d3cc5SArd Biesheuvel 1150f1af9d3aSPhilipp Hachtmann if (!type_b) { 1151f1af9d3aSPhilipp Hachtmann if (out_start) 1152f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1153f1af9d3aSPhilipp Hachtmann if (out_end) 1154f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1155f1af9d3aSPhilipp Hachtmann if (out_nid) 1156f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1157fb399b48Szijun_hu idx_a--; 1158f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1159f1af9d3aSPhilipp Hachtmann return; 1160f1af9d3aSPhilipp Hachtmann } 11617bd0b0f0STejun Heo 1162f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1163f1af9d3aSPhilipp Hachtmann for (; idx_b >= 0; idx_b--) { 1164f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1165f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1166f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1167f1af9d3aSPhilipp Hachtmann 1168f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1169f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1170f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 11711c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1172f1af9d3aSPhilipp Hachtmann /* 1173f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1174f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1175f1af9d3aSPhilipp Hachtmann */ 1176f1af9d3aSPhilipp Hachtmann 11777bd0b0f0STejun Heo if (r_end <= m_start) 11787bd0b0f0STejun Heo break; 11797bd0b0f0STejun Heo /* if the two regions intersect, we're done */ 11807bd0b0f0STejun Heo if (m_end > r_start) { 11817bd0b0f0STejun Heo if (out_start) 11827bd0b0f0STejun Heo *out_start = max(m_start, r_start); 11837bd0b0f0STejun Heo if (out_end) 11847bd0b0f0STejun Heo *out_end = min(m_end, r_end); 11857bd0b0f0STejun Heo if (out_nid) 1186f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 11877bd0b0f0STejun Heo if (m_start >= r_start) 1188f1af9d3aSPhilipp Hachtmann idx_a--; 11897bd0b0f0STejun Heo else 1190f1af9d3aSPhilipp Hachtmann idx_b--; 1191f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 11927bd0b0f0STejun Heo return; 11937bd0b0f0STejun Heo } 11947bd0b0f0STejun Heo } 11957bd0b0f0STejun Heo } 1196f1af9d3aSPhilipp Hachtmann /* signal end of iteration */ 11977bd0b0f0STejun Heo *idx = ULLONG_MAX; 11987bd0b0f0STejun Heo } 11997bd0b0f0STejun Heo 12007c0caeb8STejun Heo /* 120145e79815SChen Chang * Common iterator interface used to define for_each_mem_pfn_range(). 12027c0caeb8STejun Heo */ 12037c0caeb8STejun Heo void __init_memblock __next_mem_pfn_range(int *idx, int nid, 12047c0caeb8STejun Heo unsigned long *out_start_pfn, 12057c0caeb8STejun Heo unsigned long *out_end_pfn, int *out_nid) 12067c0caeb8STejun Heo { 12077c0caeb8STejun Heo struct memblock_type *type = &memblock.memory; 12087c0caeb8STejun Heo struct memblock_region *r; 1209d622abf7SMike Rapoport int r_nid; 12107c0caeb8STejun Heo 12117c0caeb8STejun Heo while (++*idx < type->cnt) { 12127c0caeb8STejun Heo r = &type->regions[*idx]; 1213d622abf7SMike Rapoport r_nid = memblock_get_region_node(r); 12147c0caeb8STejun Heo 12157c0caeb8STejun Heo if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 12167c0caeb8STejun Heo continue; 1217d622abf7SMike Rapoport if (nid == MAX_NUMNODES || nid == r_nid) 12187c0caeb8STejun Heo break; 12197c0caeb8STejun Heo } 12207c0caeb8STejun Heo if (*idx >= type->cnt) { 12217c0caeb8STejun Heo *idx = -1; 12227c0caeb8STejun Heo return; 12237c0caeb8STejun Heo } 12247c0caeb8STejun Heo 12257c0caeb8STejun Heo if (out_start_pfn) 12267c0caeb8STejun Heo *out_start_pfn = PFN_UP(r->base); 12277c0caeb8STejun Heo if (out_end_pfn) 12287c0caeb8STejun Heo *out_end_pfn = PFN_DOWN(r->base + r->size); 12297c0caeb8STejun Heo if (out_nid) 1230d622abf7SMike Rapoport *out_nid = r_nid; 12317c0caeb8STejun Heo } 12327c0caeb8STejun Heo 12337c0caeb8STejun Heo /** 12347c0caeb8STejun Heo * memblock_set_node - set node ID on memblock regions 12357c0caeb8STejun Heo * @base: base of area to set node ID for 12367c0caeb8STejun Heo * @size: size of area to set node ID for 1237e7e8de59STang Chen * @type: memblock type to set node ID for 12387c0caeb8STejun Heo * @nid: node ID to set 12397c0caeb8STejun Heo * 1240e7e8de59STang Chen * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 12417c0caeb8STejun Heo * Regions which cross the area boundaries are split as necessary. 12427c0caeb8STejun Heo * 124347cec443SMike Rapoport * Return: 12447c0caeb8STejun Heo * 0 on success, -errno on failure. 12457c0caeb8STejun Heo */ 12467c0caeb8STejun Heo int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1247e7e8de59STang Chen struct memblock_type *type, int nid) 12487c0caeb8STejun Heo { 1249a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 12506a9ceb31STejun Heo int start_rgn, end_rgn; 12516a9ceb31STejun Heo int i, ret; 12527c0caeb8STejun Heo 12536a9ceb31STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 12546a9ceb31STejun Heo if (ret) 12556a9ceb31STejun Heo return ret; 12567c0caeb8STejun Heo 12576a9ceb31STejun Heo for (i = start_rgn; i < end_rgn; i++) 1258e9d24ad3SWanpeng Li memblock_set_region_node(&type->regions[i], nid); 12597c0caeb8STejun Heo 12607c0caeb8STejun Heo memblock_merge_regions(type); 12613f08a302SMike Rapoport #endif 12627c0caeb8STejun Heo return 0; 12637c0caeb8STejun Heo } 12643f08a302SMike Rapoport 1265837566e7SAlexander Duyck #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1266837566e7SAlexander Duyck /** 1267837566e7SAlexander Duyck * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1268837566e7SAlexander Duyck * 1269837566e7SAlexander Duyck * @idx: pointer to u64 loop variable 1270837566e7SAlexander Duyck * @zone: zone in which all of the memory blocks reside 1271837566e7SAlexander Duyck * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1272837566e7SAlexander Duyck * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1273837566e7SAlexander Duyck * 1274837566e7SAlexander Duyck * This function is meant to be a zone/pfn specific wrapper for the 1275837566e7SAlexander Duyck * for_each_mem_range type iterators. Specifically they are used in the 1276837566e7SAlexander Duyck * deferred memory init routines and as such we were duplicating much of 1277837566e7SAlexander Duyck * this logic throughout the code. So instead of having it in multiple 1278837566e7SAlexander Duyck * locations it seemed like it would make more sense to centralize this to 1279837566e7SAlexander Duyck * one new iterator that does everything they need. 1280837566e7SAlexander Duyck */ 1281837566e7SAlexander Duyck void __init_memblock 1282837566e7SAlexander Duyck __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1283837566e7SAlexander Duyck unsigned long *out_spfn, unsigned long *out_epfn) 1284837566e7SAlexander Duyck { 1285837566e7SAlexander Duyck int zone_nid = zone_to_nid(zone); 1286837566e7SAlexander Duyck phys_addr_t spa, epa; 1287837566e7SAlexander Duyck 1288837566e7SAlexander Duyck __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1289837566e7SAlexander Duyck &memblock.memory, &memblock.reserved, 1290f30b002cSMiaohe Lin &spa, &epa, NULL); 1291837566e7SAlexander Duyck 1292837566e7SAlexander Duyck while (*idx != U64_MAX) { 1293837566e7SAlexander Duyck unsigned long epfn = PFN_DOWN(epa); 1294837566e7SAlexander Duyck unsigned long spfn = PFN_UP(spa); 1295837566e7SAlexander Duyck 1296837566e7SAlexander Duyck /* 1297837566e7SAlexander Duyck * Verify the end is at least past the start of the zone and 1298837566e7SAlexander Duyck * that we have at least one PFN to initialize. 1299837566e7SAlexander Duyck */ 1300837566e7SAlexander Duyck if (zone->zone_start_pfn < epfn && spfn < epfn) { 1301837566e7SAlexander Duyck /* if we went too far just stop searching */ 1302837566e7SAlexander Duyck if (zone_end_pfn(zone) <= spfn) { 1303837566e7SAlexander Duyck *idx = U64_MAX; 1304837566e7SAlexander Duyck break; 1305837566e7SAlexander Duyck } 1306837566e7SAlexander Duyck 1307837566e7SAlexander Duyck if (out_spfn) 1308837566e7SAlexander Duyck *out_spfn = max(zone->zone_start_pfn, spfn); 1309837566e7SAlexander Duyck if (out_epfn) 1310837566e7SAlexander Duyck *out_epfn = min(zone_end_pfn(zone), epfn); 1311837566e7SAlexander Duyck 1312837566e7SAlexander Duyck return; 1313837566e7SAlexander Duyck } 1314837566e7SAlexander Duyck 1315837566e7SAlexander Duyck __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1316837566e7SAlexander Duyck &memblock.memory, &memblock.reserved, 1317f30b002cSMiaohe Lin &spa, &epa, NULL); 1318837566e7SAlexander Duyck } 1319837566e7SAlexander Duyck 1320837566e7SAlexander Duyck /* signal end of iteration */ 1321837566e7SAlexander Duyck if (out_spfn) 1322837566e7SAlexander Duyck *out_spfn = ULONG_MAX; 1323837566e7SAlexander Duyck if (out_epfn) 1324837566e7SAlexander Duyck *out_epfn = 0; 1325837566e7SAlexander Duyck } 1326837566e7SAlexander Duyck 1327837566e7SAlexander Duyck #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 13287c0caeb8STejun Heo 132992d12f95SMike Rapoport /** 133092d12f95SMike Rapoport * memblock_alloc_range_nid - allocate boot memory block 133192d12f95SMike Rapoport * @size: size of memory block to be allocated in bytes 133292d12f95SMike Rapoport * @align: alignment of the region and block's size 133392d12f95SMike Rapoport * @start: the lower bound of the memory region to allocate (phys address) 133492d12f95SMike Rapoport * @end: the upper bound of the memory region to allocate (phys address) 133592d12f95SMike Rapoport * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 13360ac398b1SYunfeng Ye * @exact_nid: control the allocation fall back to other nodes 133792d12f95SMike Rapoport * 133892d12f95SMike Rapoport * The allocation is performed from memory region limited by 133995830666SCao jin * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 134092d12f95SMike Rapoport * 13410ac398b1SYunfeng Ye * If the specified node can not hold the requested memory and @exact_nid 13420ac398b1SYunfeng Ye * is false, the allocation falls back to any node in the system. 134392d12f95SMike Rapoport * 134492d12f95SMike Rapoport * For systems with memory mirroring, the allocation is attempted first 134592d12f95SMike Rapoport * from the regions with mirroring enabled and then retried from any 134692d12f95SMike Rapoport * memory region. 134792d12f95SMike Rapoport * 1348*c200d900SPatrick Wang * In addition, function using kmemleak_alloc_phys for allocated boot 1349*c200d900SPatrick Wang * memory block, it is never reported as leaks. 135092d12f95SMike Rapoport * 135192d12f95SMike Rapoport * Return: 135292d12f95SMike Rapoport * Physical address of allocated memory block on success, %0 on failure. 135392d12f95SMike Rapoport */ 13548676af1fSAslan Bakirov phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 13552bfc2862SAkinobu Mita phys_addr_t align, phys_addr_t start, 13560ac398b1SYunfeng Ye phys_addr_t end, int nid, 13570ac398b1SYunfeng Ye bool exact_nid) 135895f72d1eSYinghai Lu { 135992d12f95SMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 13606ed311b2SBenjamin Herrenschmidt phys_addr_t found; 136195f72d1eSYinghai Lu 136292d12f95SMike Rapoport if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 136392d12f95SMike Rapoport nid = NUMA_NO_NODE; 136492d12f95SMike Rapoport 13652f770806SMike Rapoport if (!align) { 13662f770806SMike Rapoport /* Can't use WARNs this early in boot on powerpc */ 13672f770806SMike Rapoport dump_stack(); 13682f770806SMike Rapoport align = SMP_CACHE_BYTES; 13692f770806SMike Rapoport } 13702f770806SMike Rapoport 137192d12f95SMike Rapoport again: 1372fc6daaf9STony Luck found = memblock_find_in_range_node(size, align, start, end, nid, 1373fc6daaf9STony Luck flags); 137492d12f95SMike Rapoport if (found && !memblock_reserve(found, size)) 137592d12f95SMike Rapoport goto done; 137692d12f95SMike Rapoport 13770ac398b1SYunfeng Ye if (nid != NUMA_NO_NODE && !exact_nid) { 137892d12f95SMike Rapoport found = memblock_find_in_range_node(size, align, start, 137992d12f95SMike Rapoport end, NUMA_NO_NODE, 138092d12f95SMike Rapoport flags); 138192d12f95SMike Rapoport if (found && !memblock_reserve(found, size)) 138292d12f95SMike Rapoport goto done; 138392d12f95SMike Rapoport } 138492d12f95SMike Rapoport 138592d12f95SMike Rapoport if (flags & MEMBLOCK_MIRROR) { 138692d12f95SMike Rapoport flags &= ~MEMBLOCK_MIRROR; 138792d12f95SMike Rapoport pr_warn("Could not allocate %pap bytes of mirrored memory\n", 138892d12f95SMike Rapoport &size); 138992d12f95SMike Rapoport goto again; 139092d12f95SMike Rapoport } 139192d12f95SMike Rapoport 139292d12f95SMike Rapoport return 0; 139392d12f95SMike Rapoport 139492d12f95SMike Rapoport done: 1395c6975d7cSQian Cai /* 1396c6975d7cSQian Cai * Skip kmemleak for those places like kasan_init() and 1397c6975d7cSQian Cai * early_pgtable_alloc() due to high volume. 1398c6975d7cSQian Cai */ 1399c6975d7cSQian Cai if (end != MEMBLOCK_ALLOC_NOLEAKTRACE) 1400aedf95eaSCatalin Marinas /* 1401*c200d900SPatrick Wang * Memblock allocated blocks are never reported as 1402*c200d900SPatrick Wang * leaks. This is because many of these blocks are 1403*c200d900SPatrick Wang * only referred via the physical address which is 1404*c200d900SPatrick Wang * not looked up by kmemleak. 1405aedf95eaSCatalin Marinas */ 1406*c200d900SPatrick Wang kmemleak_alloc_phys(found, size, 0); 140792d12f95SMike Rapoport 14086ed311b2SBenjamin Herrenschmidt return found; 1409aedf95eaSCatalin Marinas } 141095f72d1eSYinghai Lu 1411a2974133SMike Rapoport /** 1412a2974133SMike Rapoport * memblock_phys_alloc_range - allocate a memory block inside specified range 1413a2974133SMike Rapoport * @size: size of memory block to be allocated in bytes 1414a2974133SMike Rapoport * @align: alignment of the region and block's size 1415a2974133SMike Rapoport * @start: the lower bound of the memory region to allocate (physical address) 1416a2974133SMike Rapoport * @end: the upper bound of the memory region to allocate (physical address) 1417a2974133SMike Rapoport * 1418a2974133SMike Rapoport * Allocate @size bytes in the between @start and @end. 1419a2974133SMike Rapoport * 1420a2974133SMike Rapoport * Return: physical address of the allocated memory block on success, 1421a2974133SMike Rapoport * %0 on failure. 1422a2974133SMike Rapoport */ 14238a770c2aSMike Rapoport phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 14248a770c2aSMike Rapoport phys_addr_t align, 14258a770c2aSMike Rapoport phys_addr_t start, 14268a770c2aSMike Rapoport phys_addr_t end) 14272bfc2862SAkinobu Mita { 1428b5cf2d6cSFaiyaz Mohammed memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1429b5cf2d6cSFaiyaz Mohammed __func__, (u64)size, (u64)align, &start, &end, 1430b5cf2d6cSFaiyaz Mohammed (void *)_RET_IP_); 14310ac398b1SYunfeng Ye return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 14320ac398b1SYunfeng Ye false); 14337bd0b0f0STejun Heo } 14347bd0b0f0STejun Heo 1435a2974133SMike Rapoport /** 143617cbe038SLevi Yun * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1437a2974133SMike Rapoport * @size: size of memory block to be allocated in bytes 1438a2974133SMike Rapoport * @align: alignment of the region and block's size 1439a2974133SMike Rapoport * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1440a2974133SMike Rapoport * 1441a2974133SMike Rapoport * Allocates memory block from the specified NUMA node. If the node 1442a2974133SMike Rapoport * has no available memory, attempts to allocated from any node in the 1443a2974133SMike Rapoport * system. 1444a2974133SMike Rapoport * 1445a2974133SMike Rapoport * Return: physical address of the allocated memory block on success, 1446a2974133SMike Rapoport * %0 on failure. 1447a2974133SMike Rapoport */ 14489a8dd708SMike Rapoport phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 14499d1e2492SBenjamin Herrenschmidt { 145033755574SMike Rapoport return memblock_alloc_range_nid(size, align, 0, 14510ac398b1SYunfeng Ye MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 145295f72d1eSYinghai Lu } 145395f72d1eSYinghai Lu 145426f09e9bSSantosh Shilimkar /** 1455eb31d559SMike Rapoport * memblock_alloc_internal - allocate boot memory block 145626f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 145726f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 145826f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region to allocate (phys address) 145926f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region to allocate (phys address) 146026f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 14610ac398b1SYunfeng Ye * @exact_nid: control the allocation fall back to other nodes 146226f09e9bSSantosh Shilimkar * 146392d12f95SMike Rapoport * Allocates memory block using memblock_alloc_range_nid() and 146492d12f95SMike Rapoport * converts the returned physical address to virtual. 146592d12f95SMike Rapoport * 146626f09e9bSSantosh Shilimkar * The @min_addr limit is dropped if it can not be satisfied and the allocation 146792d12f95SMike Rapoport * will fall back to memory below @min_addr. Other constraints, such 146892d12f95SMike Rapoport * as node and mirrored memory will be handled again in 146992d12f95SMike Rapoport * memblock_alloc_range_nid(). 147026f09e9bSSantosh Shilimkar * 147147cec443SMike Rapoport * Return: 147226f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 147326f09e9bSSantosh Shilimkar */ 1474eb31d559SMike Rapoport static void * __init memblock_alloc_internal( 147526f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 147626f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 14770ac398b1SYunfeng Ye int nid, bool exact_nid) 147826f09e9bSSantosh Shilimkar { 147926f09e9bSSantosh Shilimkar phys_addr_t alloc; 148026f09e9bSSantosh Shilimkar 148126f09e9bSSantosh Shilimkar /* 148226f09e9bSSantosh Shilimkar * Detect any accidental use of these APIs after slab is ready, as at 148326f09e9bSSantosh Shilimkar * this moment memblock may be deinitialized already and its 1484c6ffc5caSMike Rapoport * internal data may be destroyed (after execution of memblock_free_all) 148526f09e9bSSantosh Shilimkar */ 148626f09e9bSSantosh Shilimkar if (WARN_ON_ONCE(slab_is_available())) 148726f09e9bSSantosh Shilimkar return kzalloc_node(size, GFP_NOWAIT, nid); 148826f09e9bSSantosh Shilimkar 1489f3057ad7SMike Rapoport if (max_addr > memblock.current_limit) 1490f3057ad7SMike Rapoport max_addr = memblock.current_limit; 1491f3057ad7SMike Rapoport 14920ac398b1SYunfeng Ye alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 14930ac398b1SYunfeng Ye exact_nid); 14942f770806SMike Rapoport 149592d12f95SMike Rapoport /* retry allocation without lower limit */ 149692d12f95SMike Rapoport if (!alloc && min_addr) 14970ac398b1SYunfeng Ye alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 14980ac398b1SYunfeng Ye exact_nid); 149926f09e9bSSantosh Shilimkar 150092d12f95SMike Rapoport if (!alloc) 1501a3f5bafcSTony Luck return NULL; 150226f09e9bSSantosh Shilimkar 150392d12f95SMike Rapoport return phys_to_virt(alloc); 150426f09e9bSSantosh Shilimkar } 150526f09e9bSSantosh Shilimkar 150626f09e9bSSantosh Shilimkar /** 15070ac398b1SYunfeng Ye * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 15080ac398b1SYunfeng Ye * without zeroing memory 15090ac398b1SYunfeng Ye * @size: size of memory block to be allocated in bytes 15100ac398b1SYunfeng Ye * @align: alignment of the region and block's size 15110ac398b1SYunfeng Ye * @min_addr: the lower bound of the memory region from where the allocation 15120ac398b1SYunfeng Ye * is preferred (phys address) 15130ac398b1SYunfeng Ye * @max_addr: the upper bound of the memory region from where the allocation 15140ac398b1SYunfeng Ye * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 15150ac398b1SYunfeng Ye * allocate only from memory limited by memblock.current_limit value 15160ac398b1SYunfeng Ye * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 15170ac398b1SYunfeng Ye * 15180ac398b1SYunfeng Ye * Public function, provides additional debug information (including caller 15190ac398b1SYunfeng Ye * info), if enabled. Does not zero allocated memory. 15200ac398b1SYunfeng Ye * 15210ac398b1SYunfeng Ye * Return: 15220ac398b1SYunfeng Ye * Virtual address of allocated memory block on success, NULL on failure. 15230ac398b1SYunfeng Ye */ 15240ac398b1SYunfeng Ye void * __init memblock_alloc_exact_nid_raw( 15250ac398b1SYunfeng Ye phys_addr_t size, phys_addr_t align, 15260ac398b1SYunfeng Ye phys_addr_t min_addr, phys_addr_t max_addr, 15270ac398b1SYunfeng Ye int nid) 15280ac398b1SYunfeng Ye { 15290ac398b1SYunfeng Ye memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 15300ac398b1SYunfeng Ye __func__, (u64)size, (u64)align, nid, &min_addr, 15310ac398b1SYunfeng Ye &max_addr, (void *)_RET_IP_); 15320ac398b1SYunfeng Ye 153308678804SMike Rapoport return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 153408678804SMike Rapoport true); 15350ac398b1SYunfeng Ye } 15360ac398b1SYunfeng Ye 15370ac398b1SYunfeng Ye /** 1538eb31d559SMike Rapoport * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1539ea1f5f37SPavel Tatashin * memory and without panicking 1540ea1f5f37SPavel Tatashin * @size: size of memory block to be allocated in bytes 1541ea1f5f37SPavel Tatashin * @align: alignment of the region and block's size 1542ea1f5f37SPavel Tatashin * @min_addr: the lower bound of the memory region from where the allocation 1543ea1f5f37SPavel Tatashin * is preferred (phys address) 1544ea1f5f37SPavel Tatashin * @max_addr: the upper bound of the memory region from where the allocation 154597ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1546ea1f5f37SPavel Tatashin * allocate only from memory limited by memblock.current_limit value 1547ea1f5f37SPavel Tatashin * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1548ea1f5f37SPavel Tatashin * 1549ea1f5f37SPavel Tatashin * Public function, provides additional debug information (including caller 1550ea1f5f37SPavel Tatashin * info), if enabled. Does not zero allocated memory, does not panic if request 1551ea1f5f37SPavel Tatashin * cannot be satisfied. 1552ea1f5f37SPavel Tatashin * 155347cec443SMike Rapoport * Return: 1554ea1f5f37SPavel Tatashin * Virtual address of allocated memory block on success, NULL on failure. 1555ea1f5f37SPavel Tatashin */ 1556eb31d559SMike Rapoport void * __init memblock_alloc_try_nid_raw( 1557ea1f5f37SPavel Tatashin phys_addr_t size, phys_addr_t align, 1558ea1f5f37SPavel Tatashin phys_addr_t min_addr, phys_addr_t max_addr, 1559ea1f5f37SPavel Tatashin int nid) 1560ea1f5f37SPavel Tatashin { 1561d75f773cSSakari Ailus memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1562a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1563a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1564ea1f5f37SPavel Tatashin 156508678804SMike Rapoport return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 156608678804SMike Rapoport false); 1567ea1f5f37SPavel Tatashin } 1568ea1f5f37SPavel Tatashin 1569ea1f5f37SPavel Tatashin /** 1570c0dbe825SMike Rapoport * memblock_alloc_try_nid - allocate boot memory block 157126f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 157226f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 157326f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region from where the allocation 157426f09e9bSSantosh Shilimkar * is preferred (phys address) 157526f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region from where the allocation 157697ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 157726f09e9bSSantosh Shilimkar * allocate only from memory limited by memblock.current_limit value 157826f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 157926f09e9bSSantosh Shilimkar * 1580c0dbe825SMike Rapoport * Public function, provides additional debug information (including caller 1581c0dbe825SMike Rapoport * info), if enabled. This function zeroes the allocated memory. 158226f09e9bSSantosh Shilimkar * 158347cec443SMike Rapoport * Return: 158426f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 158526f09e9bSSantosh Shilimkar */ 1586eb31d559SMike Rapoport void * __init memblock_alloc_try_nid( 158726f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 158826f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 158926f09e9bSSantosh Shilimkar int nid) 159026f09e9bSSantosh Shilimkar { 159126f09e9bSSantosh Shilimkar void *ptr; 159226f09e9bSSantosh Shilimkar 1593d75f773cSSakari Ailus memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1594a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1595a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1596eb31d559SMike Rapoport ptr = memblock_alloc_internal(size, align, 15970ac398b1SYunfeng Ye min_addr, max_addr, nid, false); 1598c0dbe825SMike Rapoport if (ptr) 1599ea1f5f37SPavel Tatashin memset(ptr, 0, size); 160026f09e9bSSantosh Shilimkar 1601c0dbe825SMike Rapoport return ptr; 160226f09e9bSSantosh Shilimkar } 160326f09e9bSSantosh Shilimkar 160426f09e9bSSantosh Shilimkar /** 1605621d9739SMike Rapoport * memblock_free_late - free pages directly to buddy allocator 160648a833ccSMike Rapoport * @base: phys starting address of the boot memory block 160726f09e9bSSantosh Shilimkar * @size: size of the boot memory block in bytes 160826f09e9bSSantosh Shilimkar * 1609a2974133SMike Rapoport * This is only useful when the memblock allocator has already been torn 161026f09e9bSSantosh Shilimkar * down, but we are still initializing the system. Pages are released directly 1611a2974133SMike Rapoport * to the buddy allocator. 161226f09e9bSSantosh Shilimkar */ 1613621d9739SMike Rapoport void __init memblock_free_late(phys_addr_t base, phys_addr_t size) 161426f09e9bSSantosh Shilimkar { 1615a36aab89SMike Rapoport phys_addr_t cursor, end; 161626f09e9bSSantosh Shilimkar 1617a36aab89SMike Rapoport end = base + size - 1; 1618d75f773cSSakari Ailus memblock_dbg("%s: [%pa-%pa] %pS\n", 1619a36aab89SMike Rapoport __func__, &base, &end, (void *)_RET_IP_); 16209099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 162126f09e9bSSantosh Shilimkar cursor = PFN_UP(base); 162226f09e9bSSantosh Shilimkar end = PFN_DOWN(base + size); 162326f09e9bSSantosh Shilimkar 162426f09e9bSSantosh Shilimkar for (; cursor < end; cursor++) { 16257c2ee349SMike Rapoport memblock_free_pages(pfn_to_page(cursor), cursor, 0); 1626ca79b0c2SArun KS totalram_pages_inc(); 162726f09e9bSSantosh Shilimkar } 162826f09e9bSSantosh Shilimkar } 16299d1e2492SBenjamin Herrenschmidt 16309d1e2492SBenjamin Herrenschmidt /* 16319d1e2492SBenjamin Herrenschmidt * Remaining API functions 16329d1e2492SBenjamin Herrenschmidt */ 16339d1e2492SBenjamin Herrenschmidt 16341f1ffb8aSDavid Gibson phys_addr_t __init_memblock memblock_phys_mem_size(void) 163595f72d1eSYinghai Lu { 16361440c4e2STejun Heo return memblock.memory.total_size; 163795f72d1eSYinghai Lu } 163895f72d1eSYinghai Lu 16398907de5dSSrikar Dronamraju phys_addr_t __init_memblock memblock_reserved_size(void) 16408907de5dSSrikar Dronamraju { 16418907de5dSSrikar Dronamraju return memblock.reserved.total_size; 16428907de5dSSrikar Dronamraju } 16438907de5dSSrikar Dronamraju 16440a93ebefSSam Ravnborg /* lowest address */ 16450a93ebefSSam Ravnborg phys_addr_t __init_memblock memblock_start_of_DRAM(void) 16460a93ebefSSam Ravnborg { 16470a93ebefSSam Ravnborg return memblock.memory.regions[0].base; 16480a93ebefSSam Ravnborg } 16490a93ebefSSam Ravnborg 165010d06439SYinghai Lu phys_addr_t __init_memblock memblock_end_of_DRAM(void) 165195f72d1eSYinghai Lu { 165295f72d1eSYinghai Lu int idx = memblock.memory.cnt - 1; 165395f72d1eSYinghai Lu 1654e3239ff9SBenjamin Herrenschmidt return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 165595f72d1eSYinghai Lu } 165695f72d1eSYinghai Lu 1657a571d4ebSDennis Chen static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 165895f72d1eSYinghai Lu { 16591c4bc43dSStefan Agner phys_addr_t max_addr = PHYS_ADDR_MAX; 1660136199f0SEmil Medve struct memblock_region *r; 166195f72d1eSYinghai Lu 1662a571d4ebSDennis Chen /* 1663a571d4ebSDennis Chen * translate the memory @limit size into the max address within one of 1664a571d4ebSDennis Chen * the memory memblock regions, if the @limit exceeds the total size 16651c4bc43dSStefan Agner * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1666a571d4ebSDennis Chen */ 1667cc6de168SMike Rapoport for_each_mem_region(r) { 1668c0ce8fefSTejun Heo if (limit <= r->size) { 1669c0ce8fefSTejun Heo max_addr = r->base + limit; 167095f72d1eSYinghai Lu break; 167195f72d1eSYinghai Lu } 1672c0ce8fefSTejun Heo limit -= r->size; 167395f72d1eSYinghai Lu } 1674c0ce8fefSTejun Heo 1675a571d4ebSDennis Chen return max_addr; 1676a571d4ebSDennis Chen } 1677a571d4ebSDennis Chen 1678a571d4ebSDennis Chen void __init memblock_enforce_memory_limit(phys_addr_t limit) 1679a571d4ebSDennis Chen { 168049aef717SColin Ian King phys_addr_t max_addr; 1681a571d4ebSDennis Chen 1682a571d4ebSDennis Chen if (!limit) 1683a571d4ebSDennis Chen return; 1684a571d4ebSDennis Chen 1685a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1686a571d4ebSDennis Chen 1687a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 16881c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1689a571d4ebSDennis Chen return; 1690a571d4ebSDennis Chen 1691c0ce8fefSTejun Heo /* truncate both memory and reserved regions */ 1692f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.memory, max_addr, 16931c4bc43dSStefan Agner PHYS_ADDR_MAX); 1694f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.reserved, max_addr, 16951c4bc43dSStefan Agner PHYS_ADDR_MAX); 169695f72d1eSYinghai Lu } 169795f72d1eSYinghai Lu 1698c9ca9b4eSAKASHI Takahiro void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1699c9ca9b4eSAKASHI Takahiro { 1700c9ca9b4eSAKASHI Takahiro int start_rgn, end_rgn; 1701c9ca9b4eSAKASHI Takahiro int i, ret; 1702c9ca9b4eSAKASHI Takahiro 1703c9ca9b4eSAKASHI Takahiro if (!size) 1704c9ca9b4eSAKASHI Takahiro return; 1705c9ca9b4eSAKASHI Takahiro 17065173ed72SPeng Fan if (!memblock_memory->total_size) { 1707e888fa7bSGeert Uytterhoeven pr_warn("%s: No memory registered yet\n", __func__); 1708e888fa7bSGeert Uytterhoeven return; 1709e888fa7bSGeert Uytterhoeven } 1710e888fa7bSGeert Uytterhoeven 1711c9ca9b4eSAKASHI Takahiro ret = memblock_isolate_range(&memblock.memory, base, size, 1712c9ca9b4eSAKASHI Takahiro &start_rgn, &end_rgn); 1713c9ca9b4eSAKASHI Takahiro if (ret) 1714c9ca9b4eSAKASHI Takahiro return; 1715c9ca9b4eSAKASHI Takahiro 1716c9ca9b4eSAKASHI Takahiro /* remove all the MAP regions */ 1717c9ca9b4eSAKASHI Takahiro for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1718c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1719c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1720c9ca9b4eSAKASHI Takahiro 1721c9ca9b4eSAKASHI Takahiro for (i = start_rgn - 1; i >= 0; i--) 1722c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1723c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1724c9ca9b4eSAKASHI Takahiro 1725c9ca9b4eSAKASHI Takahiro /* truncate the reserved regions */ 1726c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 0, base); 1727c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 17281c4bc43dSStefan Agner base + size, PHYS_ADDR_MAX); 1729c9ca9b4eSAKASHI Takahiro } 1730c9ca9b4eSAKASHI Takahiro 1731a571d4ebSDennis Chen void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1732a571d4ebSDennis Chen { 1733a571d4ebSDennis Chen phys_addr_t max_addr; 1734a571d4ebSDennis Chen 1735a571d4ebSDennis Chen if (!limit) 1736a571d4ebSDennis Chen return; 1737a571d4ebSDennis Chen 1738a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1739a571d4ebSDennis Chen 1740a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 17411c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1742a571d4ebSDennis Chen return; 1743a571d4ebSDennis Chen 1744c9ca9b4eSAKASHI Takahiro memblock_cap_memory_range(0, max_addr); 1745a571d4ebSDennis Chen } 1746a571d4ebSDennis Chen 1747cd79481dSYinghai Lu static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 174872d4b0b4SBenjamin Herrenschmidt { 174972d4b0b4SBenjamin Herrenschmidt unsigned int left = 0, right = type->cnt; 175072d4b0b4SBenjamin Herrenschmidt 175172d4b0b4SBenjamin Herrenschmidt do { 175272d4b0b4SBenjamin Herrenschmidt unsigned int mid = (right + left) / 2; 175372d4b0b4SBenjamin Herrenschmidt 175472d4b0b4SBenjamin Herrenschmidt if (addr < type->regions[mid].base) 175572d4b0b4SBenjamin Herrenschmidt right = mid; 175672d4b0b4SBenjamin Herrenschmidt else if (addr >= (type->regions[mid].base + 175772d4b0b4SBenjamin Herrenschmidt type->regions[mid].size)) 175872d4b0b4SBenjamin Herrenschmidt left = mid + 1; 175972d4b0b4SBenjamin Herrenschmidt else 176072d4b0b4SBenjamin Herrenschmidt return mid; 176172d4b0b4SBenjamin Herrenschmidt } while (left < right); 176272d4b0b4SBenjamin Herrenschmidt return -1; 176372d4b0b4SBenjamin Herrenschmidt } 176472d4b0b4SBenjamin Herrenschmidt 1765f5a222dcSYueyi Li bool __init_memblock memblock_is_reserved(phys_addr_t addr) 176695f72d1eSYinghai Lu { 176772d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.reserved, addr) != -1; 176895f72d1eSYinghai Lu } 176972d4b0b4SBenjamin Herrenschmidt 1770b4ad0c7eSYaowei Bai bool __init_memblock memblock_is_memory(phys_addr_t addr) 177172d4b0b4SBenjamin Herrenschmidt { 177272d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.memory, addr) != -1; 177372d4b0b4SBenjamin Herrenschmidt } 177472d4b0b4SBenjamin Herrenschmidt 1775937f0c26SYaowei Bai bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1776bf3d3cc5SArd Biesheuvel { 1777bf3d3cc5SArd Biesheuvel int i = memblock_search(&memblock.memory, addr); 1778bf3d3cc5SArd Biesheuvel 1779bf3d3cc5SArd Biesheuvel if (i == -1) 1780bf3d3cc5SArd Biesheuvel return false; 1781bf3d3cc5SArd Biesheuvel return !memblock_is_nomap(&memblock.memory.regions[i]); 1782bf3d3cc5SArd Biesheuvel } 1783bf3d3cc5SArd Biesheuvel 1784e76b63f8SYinghai Lu int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1785e76b63f8SYinghai Lu unsigned long *start_pfn, unsigned long *end_pfn) 1786e76b63f8SYinghai Lu { 1787e76b63f8SYinghai Lu struct memblock_type *type = &memblock.memory; 178816763230SFabian Frederick int mid = memblock_search(type, PFN_PHYS(pfn)); 1789e76b63f8SYinghai Lu 1790e76b63f8SYinghai Lu if (mid == -1) 1791e76b63f8SYinghai Lu return -1; 1792e76b63f8SYinghai Lu 1793f7e2f7e8SFabian Frederick *start_pfn = PFN_DOWN(type->regions[mid].base); 1794f7e2f7e8SFabian Frederick *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1795e76b63f8SYinghai Lu 1796d622abf7SMike Rapoport return memblock_get_region_node(&type->regions[mid]); 1797e76b63f8SYinghai Lu } 1798e76b63f8SYinghai Lu 1799eab30949SStephen Boyd /** 1800eab30949SStephen Boyd * memblock_is_region_memory - check if a region is a subset of memory 1801eab30949SStephen Boyd * @base: base of region to check 1802eab30949SStephen Boyd * @size: size of region to check 1803eab30949SStephen Boyd * 1804eab30949SStephen Boyd * Check if the region [@base, @base + @size) is a subset of a memory block. 1805eab30949SStephen Boyd * 180647cec443SMike Rapoport * Return: 1807eab30949SStephen Boyd * 0 if false, non-zero if true 1808eab30949SStephen Boyd */ 1809937f0c26SYaowei Bai bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 181072d4b0b4SBenjamin Herrenschmidt { 1811abb65272STomi Valkeinen int idx = memblock_search(&memblock.memory, base); 1812eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 181372d4b0b4SBenjamin Herrenschmidt 181472d4b0b4SBenjamin Herrenschmidt if (idx == -1) 1815937f0c26SYaowei Bai return false; 1816ef415ef4SWei Yang return (memblock.memory.regions[idx].base + 1817eb18f1b5STejun Heo memblock.memory.regions[idx].size) >= end; 181895f72d1eSYinghai Lu } 181995f72d1eSYinghai Lu 1820eab30949SStephen Boyd /** 1821eab30949SStephen Boyd * memblock_is_region_reserved - check if a region intersects reserved memory 1822eab30949SStephen Boyd * @base: base of region to check 1823eab30949SStephen Boyd * @size: size of region to check 1824eab30949SStephen Boyd * 182547cec443SMike Rapoport * Check if the region [@base, @base + @size) intersects a reserved 182647cec443SMike Rapoport * memory block. 1827eab30949SStephen Boyd * 182847cec443SMike Rapoport * Return: 1829c5c5c9d1STang Chen * True if they intersect, false if not. 1830eab30949SStephen Boyd */ 1831c5c5c9d1STang Chen bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 183295f72d1eSYinghai Lu { 1833c5c5c9d1STang Chen return memblock_overlaps_region(&memblock.reserved, base, size); 183495f72d1eSYinghai Lu } 183595f72d1eSYinghai Lu 18366ede1fd3SYinghai Lu void __init_memblock memblock_trim_memory(phys_addr_t align) 18376ede1fd3SYinghai Lu { 18386ede1fd3SYinghai Lu phys_addr_t start, end, orig_start, orig_end; 1839136199f0SEmil Medve struct memblock_region *r; 18406ede1fd3SYinghai Lu 1841cc6de168SMike Rapoport for_each_mem_region(r) { 1842136199f0SEmil Medve orig_start = r->base; 1843136199f0SEmil Medve orig_end = r->base + r->size; 18446ede1fd3SYinghai Lu start = round_up(orig_start, align); 18456ede1fd3SYinghai Lu end = round_down(orig_end, align); 18466ede1fd3SYinghai Lu 18476ede1fd3SYinghai Lu if (start == orig_start && end == orig_end) 18486ede1fd3SYinghai Lu continue; 18496ede1fd3SYinghai Lu 18506ede1fd3SYinghai Lu if (start < end) { 1851136199f0SEmil Medve r->base = start; 1852136199f0SEmil Medve r->size = end - start; 18536ede1fd3SYinghai Lu } else { 1854136199f0SEmil Medve memblock_remove_region(&memblock.memory, 1855136199f0SEmil Medve r - memblock.memory.regions); 1856136199f0SEmil Medve r--; 18576ede1fd3SYinghai Lu } 18586ede1fd3SYinghai Lu } 18596ede1fd3SYinghai Lu } 1860e63075a3SBenjamin Herrenschmidt 18613661ca66SYinghai Lu void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1862e63075a3SBenjamin Herrenschmidt { 1863e63075a3SBenjamin Herrenschmidt memblock.current_limit = limit; 1864e63075a3SBenjamin Herrenschmidt } 1865e63075a3SBenjamin Herrenschmidt 1866fec51014SLaura Abbott phys_addr_t __init_memblock memblock_get_current_limit(void) 1867fec51014SLaura Abbott { 1868fec51014SLaura Abbott return memblock.current_limit; 1869fec51014SLaura Abbott } 1870fec51014SLaura Abbott 18710262d9c8SHeiko Carstens static void __init_memblock memblock_dump(struct memblock_type *type) 18726ed311b2SBenjamin Herrenschmidt { 18735d63f81cSMiles Chen phys_addr_t base, end, size; 1874e1720feeSMike Rapoport enum memblock_flags flags; 18758c9c1701SAlexander Kuleshov int idx; 18768c9c1701SAlexander Kuleshov struct memblock_region *rgn; 18776ed311b2SBenjamin Herrenschmidt 18780262d9c8SHeiko Carstens pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 18796ed311b2SBenjamin Herrenschmidt 188066e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 18817c0caeb8STejun Heo char nid_buf[32] = ""; 18826ed311b2SBenjamin Herrenschmidt 18837c0caeb8STejun Heo base = rgn->base; 18847c0caeb8STejun Heo size = rgn->size; 18855d63f81cSMiles Chen end = base + size - 1; 188666a20757STang Chen flags = rgn->flags; 1887a9ee6cf5SMike Rapoport #ifdef CONFIG_NUMA 18887c0caeb8STejun Heo if (memblock_get_region_node(rgn) != MAX_NUMNODES) 18897c0caeb8STejun Heo snprintf(nid_buf, sizeof(nid_buf), " on node %d", 18907c0caeb8STejun Heo memblock_get_region_node(rgn)); 18917c0caeb8STejun Heo #endif 1892e1720feeSMike Rapoport pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 18930262d9c8SHeiko Carstens type->name, idx, &base, &end, &size, nid_buf, flags); 18946ed311b2SBenjamin Herrenschmidt } 18956ed311b2SBenjamin Herrenschmidt } 18966ed311b2SBenjamin Herrenschmidt 189787c55870SMike Rapoport static void __init_memblock __memblock_dump_all(void) 18986ed311b2SBenjamin Herrenschmidt { 18996ed311b2SBenjamin Herrenschmidt pr_info("MEMBLOCK configuration:\n"); 19005d63f81cSMiles Chen pr_info(" memory size = %pa reserved size = %pa\n", 19015d63f81cSMiles Chen &memblock.memory.total_size, 19025d63f81cSMiles Chen &memblock.reserved.total_size); 19036ed311b2SBenjamin Herrenschmidt 19040262d9c8SHeiko Carstens memblock_dump(&memblock.memory); 19050262d9c8SHeiko Carstens memblock_dump(&memblock.reserved); 1906409efd4cSHeiko Carstens #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 190777649905SDavid Hildenbrand memblock_dump(&physmem); 1908409efd4cSHeiko Carstens #endif 19096ed311b2SBenjamin Herrenschmidt } 19106ed311b2SBenjamin Herrenschmidt 191187c55870SMike Rapoport void __init_memblock memblock_dump_all(void) 191287c55870SMike Rapoport { 191387c55870SMike Rapoport if (memblock_debug) 191487c55870SMike Rapoport __memblock_dump_all(); 191587c55870SMike Rapoport } 191687c55870SMike Rapoport 19171aadc056STejun Heo void __init memblock_allow_resize(void) 19186ed311b2SBenjamin Herrenschmidt { 1919142b45a7SBenjamin Herrenschmidt memblock_can_resize = 1; 19206ed311b2SBenjamin Herrenschmidt } 19216ed311b2SBenjamin Herrenschmidt 19226ed311b2SBenjamin Herrenschmidt static int __init early_memblock(char *p) 19236ed311b2SBenjamin Herrenschmidt { 19246ed311b2SBenjamin Herrenschmidt if (p && strstr(p, "debug")) 19256ed311b2SBenjamin Herrenschmidt memblock_debug = 1; 19266ed311b2SBenjamin Herrenschmidt return 0; 19276ed311b2SBenjamin Herrenschmidt } 19286ed311b2SBenjamin Herrenschmidt early_param("memblock", early_memblock); 19296ed311b2SBenjamin Herrenschmidt 19304f5b0c17SMike Rapoport static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 19314f5b0c17SMike Rapoport { 19324f5b0c17SMike Rapoport struct page *start_pg, *end_pg; 19334f5b0c17SMike Rapoport phys_addr_t pg, pgend; 19344f5b0c17SMike Rapoport 19354f5b0c17SMike Rapoport /* 19364f5b0c17SMike Rapoport * Convert start_pfn/end_pfn to a struct page pointer. 19374f5b0c17SMike Rapoport */ 19384f5b0c17SMike Rapoport start_pg = pfn_to_page(start_pfn - 1) + 1; 19394f5b0c17SMike Rapoport end_pg = pfn_to_page(end_pfn - 1) + 1; 19404f5b0c17SMike Rapoport 19414f5b0c17SMike Rapoport /* 19424f5b0c17SMike Rapoport * Convert to physical addresses, and round start upwards and end 19434f5b0c17SMike Rapoport * downwards. 19444f5b0c17SMike Rapoport */ 19454f5b0c17SMike Rapoport pg = PAGE_ALIGN(__pa(start_pg)); 19464f5b0c17SMike Rapoport pgend = __pa(end_pg) & PAGE_MASK; 19474f5b0c17SMike Rapoport 19484f5b0c17SMike Rapoport /* 19494f5b0c17SMike Rapoport * If there are free pages between these, free the section of the 19504f5b0c17SMike Rapoport * memmap array. 19514f5b0c17SMike Rapoport */ 19524f5b0c17SMike Rapoport if (pg < pgend) 19533ecc6834SMike Rapoport memblock_phys_free(pg, pgend - pg); 19544f5b0c17SMike Rapoport } 19554f5b0c17SMike Rapoport 19564f5b0c17SMike Rapoport /* 19574f5b0c17SMike Rapoport * The mem_map array can get very big. Free the unused area of the memory map. 19584f5b0c17SMike Rapoport */ 19594f5b0c17SMike Rapoport static void __init free_unused_memmap(void) 19604f5b0c17SMike Rapoport { 19614f5b0c17SMike Rapoport unsigned long start, end, prev_end = 0; 19624f5b0c17SMike Rapoport int i; 19634f5b0c17SMike Rapoport 19644f5b0c17SMike Rapoport if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 19654f5b0c17SMike Rapoport IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 19664f5b0c17SMike Rapoport return; 19674f5b0c17SMike Rapoport 19684f5b0c17SMike Rapoport /* 19694f5b0c17SMike Rapoport * This relies on each bank being in address order. 19704f5b0c17SMike Rapoport * The banks are sorted previously in bootmem_init(). 19714f5b0c17SMike Rapoport */ 19724f5b0c17SMike Rapoport for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 19734f5b0c17SMike Rapoport #ifdef CONFIG_SPARSEMEM 19744f5b0c17SMike Rapoport /* 19754f5b0c17SMike Rapoport * Take care not to free memmap entries that don't exist 19764f5b0c17SMike Rapoport * due to SPARSEMEM sections which aren't present. 19774f5b0c17SMike Rapoport */ 19784f5b0c17SMike Rapoport start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 19794f5b0c17SMike Rapoport #endif 19804f5b0c17SMike Rapoport /* 1981e2a86800SMike Rapoport * Align down here since many operations in VM subsystem 1982e2a86800SMike Rapoport * presume that there are no holes in the memory map inside 1983e2a86800SMike Rapoport * a pageblock 19844f5b0c17SMike Rapoport */ 1985e2a86800SMike Rapoport start = round_down(start, pageblock_nr_pages); 19864f5b0c17SMike Rapoport 19874f5b0c17SMike Rapoport /* 19884f5b0c17SMike Rapoport * If we had a previous bank, and there is a space 19894f5b0c17SMike Rapoport * between the current bank and the previous, free it. 19904f5b0c17SMike Rapoport */ 19914f5b0c17SMike Rapoport if (prev_end && prev_end < start) 19924f5b0c17SMike Rapoport free_memmap(prev_end, start); 19934f5b0c17SMike Rapoport 19944f5b0c17SMike Rapoport /* 1995e2a86800SMike Rapoport * Align up here since many operations in VM subsystem 1996e2a86800SMike Rapoport * presume that there are no holes in the memory map inside 1997e2a86800SMike Rapoport * a pageblock 19984f5b0c17SMike Rapoport */ 1999e2a86800SMike Rapoport prev_end = ALIGN(end, pageblock_nr_pages); 20004f5b0c17SMike Rapoport } 20014f5b0c17SMike Rapoport 20024f5b0c17SMike Rapoport #ifdef CONFIG_SPARSEMEM 2003f921f53eSMike Rapoport if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 2004f921f53eSMike Rapoport prev_end = ALIGN(end, pageblock_nr_pages); 20054f5b0c17SMike Rapoport free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 2006f921f53eSMike Rapoport } 20074f5b0c17SMike Rapoport #endif 20084f5b0c17SMike Rapoport } 20094f5b0c17SMike Rapoport 2010bda49a81SMike Rapoport static void __init __free_pages_memory(unsigned long start, unsigned long end) 2011bda49a81SMike Rapoport { 2012bda49a81SMike Rapoport int order; 2013bda49a81SMike Rapoport 2014bda49a81SMike Rapoport while (start < end) { 2015bda49a81SMike Rapoport order = min(MAX_ORDER - 1UL, __ffs(start)); 2016bda49a81SMike Rapoport 2017bda49a81SMike Rapoport while (start + (1UL << order) > end) 2018bda49a81SMike Rapoport order--; 2019bda49a81SMike Rapoport 2020bda49a81SMike Rapoport memblock_free_pages(pfn_to_page(start), start, order); 2021bda49a81SMike Rapoport 2022bda49a81SMike Rapoport start += (1UL << order); 2023bda49a81SMike Rapoport } 2024bda49a81SMike Rapoport } 2025bda49a81SMike Rapoport 2026bda49a81SMike Rapoport static unsigned long __init __free_memory_core(phys_addr_t start, 2027bda49a81SMike Rapoport phys_addr_t end) 2028bda49a81SMike Rapoport { 2029bda49a81SMike Rapoport unsigned long start_pfn = PFN_UP(start); 2030bda49a81SMike Rapoport unsigned long end_pfn = min_t(unsigned long, 2031bda49a81SMike Rapoport PFN_DOWN(end), max_low_pfn); 2032bda49a81SMike Rapoport 2033bda49a81SMike Rapoport if (start_pfn >= end_pfn) 2034bda49a81SMike Rapoport return 0; 2035bda49a81SMike Rapoport 2036bda49a81SMike Rapoport __free_pages_memory(start_pfn, end_pfn); 2037bda49a81SMike Rapoport 2038bda49a81SMike Rapoport return end_pfn - start_pfn; 2039bda49a81SMike Rapoport } 2040bda49a81SMike Rapoport 20419092d4f7SMike Rapoport static void __init memmap_init_reserved_pages(void) 20429092d4f7SMike Rapoport { 20439092d4f7SMike Rapoport struct memblock_region *region; 20449092d4f7SMike Rapoport phys_addr_t start, end; 20459092d4f7SMike Rapoport u64 i; 20469092d4f7SMike Rapoport 20479092d4f7SMike Rapoport /* initialize struct pages for the reserved regions */ 20489092d4f7SMike Rapoport for_each_reserved_mem_range(i, &start, &end) 20499092d4f7SMike Rapoport reserve_bootmem_region(start, end); 20509092d4f7SMike Rapoport 20519092d4f7SMike Rapoport /* and also treat struct pages for the NOMAP regions as PageReserved */ 20529092d4f7SMike Rapoport for_each_mem_region(region) { 20539092d4f7SMike Rapoport if (memblock_is_nomap(region)) { 20549092d4f7SMike Rapoport start = region->base; 20559092d4f7SMike Rapoport end = start + region->size; 20569092d4f7SMike Rapoport reserve_bootmem_region(start, end); 20579092d4f7SMike Rapoport } 20589092d4f7SMike Rapoport } 20599092d4f7SMike Rapoport } 20609092d4f7SMike Rapoport 2061bda49a81SMike Rapoport static unsigned long __init free_low_memory_core_early(void) 2062bda49a81SMike Rapoport { 2063bda49a81SMike Rapoport unsigned long count = 0; 2064bda49a81SMike Rapoport phys_addr_t start, end; 2065bda49a81SMike Rapoport u64 i; 2066bda49a81SMike Rapoport 2067bda49a81SMike Rapoport memblock_clear_hotplug(0, -1); 2068bda49a81SMike Rapoport 20699092d4f7SMike Rapoport memmap_init_reserved_pages(); 2070bda49a81SMike Rapoport 2071bda49a81SMike Rapoport /* 2072bda49a81SMike Rapoport * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2073bda49a81SMike Rapoport * because in some case like Node0 doesn't have RAM installed 2074bda49a81SMike Rapoport * low ram will be on Node1 2075bda49a81SMike Rapoport */ 2076bda49a81SMike Rapoport for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2077bda49a81SMike Rapoport NULL) 2078bda49a81SMike Rapoport count += __free_memory_core(start, end); 2079bda49a81SMike Rapoport 2080bda49a81SMike Rapoport return count; 2081bda49a81SMike Rapoport } 2082bda49a81SMike Rapoport 2083bda49a81SMike Rapoport static int reset_managed_pages_done __initdata; 2084bda49a81SMike Rapoport 2085bda49a81SMike Rapoport void reset_node_managed_pages(pg_data_t *pgdat) 2086bda49a81SMike Rapoport { 2087bda49a81SMike Rapoport struct zone *z; 2088bda49a81SMike Rapoport 2089bda49a81SMike Rapoport for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 20909705bea5SArun KS atomic_long_set(&z->managed_pages, 0); 2091bda49a81SMike Rapoport } 2092bda49a81SMike Rapoport 2093bda49a81SMike Rapoport void __init reset_all_zones_managed_pages(void) 2094bda49a81SMike Rapoport { 2095bda49a81SMike Rapoport struct pglist_data *pgdat; 2096bda49a81SMike Rapoport 2097bda49a81SMike Rapoport if (reset_managed_pages_done) 2098bda49a81SMike Rapoport return; 2099bda49a81SMike Rapoport 2100bda49a81SMike Rapoport for_each_online_pgdat(pgdat) 2101bda49a81SMike Rapoport reset_node_managed_pages(pgdat); 2102bda49a81SMike Rapoport 2103bda49a81SMike Rapoport reset_managed_pages_done = 1; 2104bda49a81SMike Rapoport } 2105bda49a81SMike Rapoport 2106bda49a81SMike Rapoport /** 2107bda49a81SMike Rapoport * memblock_free_all - release free pages to the buddy allocator 2108bda49a81SMike Rapoport */ 2109097d43d8SDaeseok Youn void __init memblock_free_all(void) 2110bda49a81SMike Rapoport { 2111bda49a81SMike Rapoport unsigned long pages; 2112bda49a81SMike Rapoport 21134f5b0c17SMike Rapoport free_unused_memmap(); 2114bda49a81SMike Rapoport reset_all_zones_managed_pages(); 2115bda49a81SMike Rapoport 2116bda49a81SMike Rapoport pages = free_low_memory_core_early(); 2117ca79b0c2SArun KS totalram_pages_add(pages); 2118bda49a81SMike Rapoport } 2119bda49a81SMike Rapoport 2120350e88baSMike Rapoport #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 21216d03b885SBenjamin Herrenschmidt 21226d03b885SBenjamin Herrenschmidt static int memblock_debug_show(struct seq_file *m, void *private) 21236d03b885SBenjamin Herrenschmidt { 21246d03b885SBenjamin Herrenschmidt struct memblock_type *type = m->private; 21256d03b885SBenjamin Herrenschmidt struct memblock_region *reg; 21266d03b885SBenjamin Herrenschmidt int i; 21275d63f81cSMiles Chen phys_addr_t end; 21286d03b885SBenjamin Herrenschmidt 21296d03b885SBenjamin Herrenschmidt for (i = 0; i < type->cnt; i++) { 21306d03b885SBenjamin Herrenschmidt reg = &type->regions[i]; 21315d63f81cSMiles Chen end = reg->base + reg->size - 1; 21326d03b885SBenjamin Herrenschmidt 21335d63f81cSMiles Chen seq_printf(m, "%4d: ", i); 21345d63f81cSMiles Chen seq_printf(m, "%pa..%pa\n", ®->base, &end); 21356d03b885SBenjamin Herrenschmidt } 21366d03b885SBenjamin Herrenschmidt return 0; 21376d03b885SBenjamin Herrenschmidt } 21385ad35093SAndy Shevchenko DEFINE_SHOW_ATTRIBUTE(memblock_debug); 21396d03b885SBenjamin Herrenschmidt 21406d03b885SBenjamin Herrenschmidt static int __init memblock_init_debugfs(void) 21416d03b885SBenjamin Herrenschmidt { 21426d03b885SBenjamin Herrenschmidt struct dentry *root = debugfs_create_dir("memblock", NULL); 2143d9f7979cSGreg Kroah-Hartman 21440825a6f9SJoe Perches debugfs_create_file("memory", 0444, root, 21450825a6f9SJoe Perches &memblock.memory, &memblock_debug_fops); 21460825a6f9SJoe Perches debugfs_create_file("reserved", 0444, root, 21470825a6f9SJoe Perches &memblock.reserved, &memblock_debug_fops); 214870210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 214977649905SDavid Hildenbrand debugfs_create_file("physmem", 0444, root, &physmem, 215077649905SDavid Hildenbrand &memblock_debug_fops); 215170210ed9SPhilipp Hachtmann #endif 21526d03b885SBenjamin Herrenschmidt 21536d03b885SBenjamin Herrenschmidt return 0; 21546d03b885SBenjamin Herrenschmidt } 21556d03b885SBenjamin Herrenschmidt __initcall(memblock_init_debugfs); 21566d03b885SBenjamin Herrenschmidt 21576d03b885SBenjamin Herrenschmidt #endif /* CONFIG_DEBUG_FS */ 2158