195f72d1eSYinghai Lu /* 295f72d1eSYinghai Lu * Procedures for maintaining information about logical memory blocks. 395f72d1eSYinghai Lu * 495f72d1eSYinghai Lu * Peter Bergner, IBM Corp. June 2001. 595f72d1eSYinghai Lu * Copyright (C) 2001 Peter Bergner. 695f72d1eSYinghai Lu * 795f72d1eSYinghai Lu * This program is free software; you can redistribute it and/or 895f72d1eSYinghai Lu * modify it under the terms of the GNU General Public License 995f72d1eSYinghai Lu * as published by the Free Software Foundation; either version 1095f72d1eSYinghai Lu * 2 of the License, or (at your option) any later version. 1195f72d1eSYinghai Lu */ 1295f72d1eSYinghai Lu 1395f72d1eSYinghai Lu #include <linux/kernel.h> 14142b45a7SBenjamin Herrenschmidt #include <linux/slab.h> 1595f72d1eSYinghai Lu #include <linux/init.h> 1695f72d1eSYinghai Lu #include <linux/bitops.h> 17449e8df3SBenjamin Herrenschmidt #include <linux/poison.h> 18c196f76fSBenjamin Herrenschmidt #include <linux/pfn.h> 196d03b885SBenjamin Herrenschmidt #include <linux/debugfs.h> 20514c6032SRandy Dunlap #include <linux/kmemleak.h> 216d03b885SBenjamin Herrenschmidt #include <linux/seq_file.h> 2295f72d1eSYinghai Lu #include <linux/memblock.h> 2395f72d1eSYinghai Lu 24c4c5ad6bSChristoph Hellwig #include <asm/sections.h> 2526f09e9bSSantosh Shilimkar #include <linux/io.h> 2626f09e9bSSantosh Shilimkar 2726f09e9bSSantosh Shilimkar #include "internal.h" 2879442ed1STang Chen 293e039c5cSMike Rapoport /** 303e039c5cSMike Rapoport * DOC: memblock overview 313e039c5cSMike Rapoport * 323e039c5cSMike Rapoport * Memblock is a method of managing memory regions during the early 333e039c5cSMike Rapoport * boot period when the usual kernel memory allocators are not up and 343e039c5cSMike Rapoport * running. 353e039c5cSMike Rapoport * 363e039c5cSMike Rapoport * Memblock views the system memory as collections of contiguous 373e039c5cSMike Rapoport * regions. There are several types of these collections: 383e039c5cSMike Rapoport * 393e039c5cSMike Rapoport * * ``memory`` - describes the physical memory available to the 403e039c5cSMike Rapoport * kernel; this may differ from the actual physical memory installed 413e039c5cSMike Rapoport * in the system, for instance when the memory is restricted with 423e039c5cSMike Rapoport * ``mem=`` command line parameter 433e039c5cSMike Rapoport * * ``reserved`` - describes the regions that were allocated 443e039c5cSMike Rapoport * * ``physmap`` - describes the actual physical memory regardless of 453e039c5cSMike Rapoport * the possible restrictions; the ``physmap`` type is only available 463e039c5cSMike Rapoport * on some architectures. 473e039c5cSMike Rapoport * 483e039c5cSMike Rapoport * Each region is represented by :c:type:`struct memblock_region` that 493e039c5cSMike Rapoport * defines the region extents, its attributes and NUMA node id on NUMA 503e039c5cSMike Rapoport * systems. Every memory type is described by the :c:type:`struct 513e039c5cSMike Rapoport * memblock_type` which contains an array of memory regions along with 523e039c5cSMike Rapoport * the allocator metadata. The memory types are nicely wrapped with 533e039c5cSMike Rapoport * :c:type:`struct memblock`. This structure is statically initialzed 543e039c5cSMike Rapoport * at build time. The region arrays for the "memory" and "reserved" 553e039c5cSMike Rapoport * types are initially sized to %INIT_MEMBLOCK_REGIONS and for the 563e039c5cSMike Rapoport * "physmap" type to %INIT_PHYSMEM_REGIONS. 573e039c5cSMike Rapoport * The :c:func:`memblock_allow_resize` enables automatic resizing of 583e039c5cSMike Rapoport * the region arrays during addition of new regions. This feature 593e039c5cSMike Rapoport * should be used with care so that memory allocated for the region 603e039c5cSMike Rapoport * array will not overlap with areas that should be reserved, for 613e039c5cSMike Rapoport * example initrd. 623e039c5cSMike Rapoport * 633e039c5cSMike Rapoport * The early architecture setup should tell memblock what the physical 643e039c5cSMike Rapoport * memory layout is by using :c:func:`memblock_add` or 653e039c5cSMike Rapoport * :c:func:`memblock_add_node` functions. The first function does not 663e039c5cSMike Rapoport * assign the region to a NUMA node and it is appropriate for UMA 673e039c5cSMike Rapoport * systems. Yet, it is possible to use it on NUMA systems as well and 683e039c5cSMike Rapoport * assign the region to a NUMA node later in the setup process using 693e039c5cSMike Rapoport * :c:func:`memblock_set_node`. The :c:func:`memblock_add_node` 703e039c5cSMike Rapoport * performs such an assignment directly. 713e039c5cSMike Rapoport * 723e039c5cSMike Rapoport * Once memblock is setup the memory can be allocated using either 733e039c5cSMike Rapoport * memblock or bootmem APIs. 743e039c5cSMike Rapoport * 753e039c5cSMike Rapoport * As the system boot progresses, the architecture specific 763e039c5cSMike Rapoport * :c:func:`mem_init` function frees all the memory to the buddy page 773e039c5cSMike Rapoport * allocator. 783e039c5cSMike Rapoport * 793e039c5cSMike Rapoport * If an architecure enables %CONFIG_ARCH_DISCARD_MEMBLOCK, the 803e039c5cSMike Rapoport * memblock data structures will be discarded after the system 813e039c5cSMike Rapoport * initialization compltes. 823e039c5cSMike Rapoport */ 833e039c5cSMike Rapoport 84bda49a81SMike Rapoport #ifndef CONFIG_NEED_MULTIPLE_NODES 85bda49a81SMike Rapoport struct pglist_data __refdata contig_page_data; 86bda49a81SMike Rapoport EXPORT_SYMBOL(contig_page_data); 87bda49a81SMike Rapoport #endif 88bda49a81SMike Rapoport 89bda49a81SMike Rapoport unsigned long max_low_pfn; 90bda49a81SMike Rapoport unsigned long min_low_pfn; 91bda49a81SMike Rapoport unsigned long max_pfn; 92bda49a81SMike Rapoport unsigned long long max_possible_pfn; 93bda49a81SMike Rapoport 94fe091c20STejun Heo static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; 95fe091c20STejun Heo static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; 9670210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 9770210ed9SPhilipp Hachtmann static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock; 9870210ed9SPhilipp Hachtmann #endif 99fe091c20STejun Heo 100fe091c20STejun Heo struct memblock memblock __initdata_memblock = { 101fe091c20STejun Heo .memory.regions = memblock_memory_init_regions, 102fe091c20STejun Heo .memory.cnt = 1, /* empty dummy entry */ 103fe091c20STejun Heo .memory.max = INIT_MEMBLOCK_REGIONS, 1040262d9c8SHeiko Carstens .memory.name = "memory", 105fe091c20STejun Heo 106fe091c20STejun Heo .reserved.regions = memblock_reserved_init_regions, 107fe091c20STejun Heo .reserved.cnt = 1, /* empty dummy entry */ 108fe091c20STejun Heo .reserved.max = INIT_MEMBLOCK_REGIONS, 1090262d9c8SHeiko Carstens .reserved.name = "reserved", 110fe091c20STejun Heo 11170210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 11270210ed9SPhilipp Hachtmann .physmem.regions = memblock_physmem_init_regions, 11370210ed9SPhilipp Hachtmann .physmem.cnt = 1, /* empty dummy entry */ 11470210ed9SPhilipp Hachtmann .physmem.max = INIT_PHYSMEM_REGIONS, 1150262d9c8SHeiko Carstens .physmem.name = "physmem", 11670210ed9SPhilipp Hachtmann #endif 11770210ed9SPhilipp Hachtmann 11879442ed1STang Chen .bottom_up = false, 119fe091c20STejun Heo .current_limit = MEMBLOCK_ALLOC_ANYWHERE, 120fe091c20STejun Heo }; 12195f72d1eSYinghai Lu 12210d06439SYinghai Lu int memblock_debug __initdata_memblock; 123a3f5bafcSTony Luck static bool system_has_some_mirror __initdata_memblock = false; 1241aadc056STejun Heo static int memblock_can_resize __initdata_memblock; 125181eb394SGavin Shan static int memblock_memory_in_slab __initdata_memblock = 0; 126181eb394SGavin Shan static int memblock_reserved_in_slab __initdata_memblock = 0; 12795f72d1eSYinghai Lu 128e1720feeSMike Rapoport enum memblock_flags __init_memblock choose_memblock_flags(void) 129a3f5bafcSTony Luck { 130a3f5bafcSTony Luck return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE; 131a3f5bafcSTony Luck } 132a3f5bafcSTony Luck 133eb18f1b5STejun Heo /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ 134eb18f1b5STejun Heo static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) 135eb18f1b5STejun Heo { 1361c4bc43dSStefan Agner return *size = min(*size, PHYS_ADDR_MAX - base); 137eb18f1b5STejun Heo } 138eb18f1b5STejun Heo 1396ed311b2SBenjamin Herrenschmidt /* 1406ed311b2SBenjamin Herrenschmidt * Address comparison utilities 1416ed311b2SBenjamin Herrenschmidt */ 14210d06439SYinghai Lu static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, 1432898cc4cSBenjamin Herrenschmidt phys_addr_t base2, phys_addr_t size2) 14495f72d1eSYinghai Lu { 14595f72d1eSYinghai Lu return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); 14695f72d1eSYinghai Lu } 14795f72d1eSYinghai Lu 14895cf82ecSTang Chen bool __init_memblock memblock_overlaps_region(struct memblock_type *type, 1492d7d3eb2SH Hartley Sweeten phys_addr_t base, phys_addr_t size) 1506ed311b2SBenjamin Herrenschmidt { 1516ed311b2SBenjamin Herrenschmidt unsigned long i; 1526ed311b2SBenjamin Herrenschmidt 153f14516fbSAlexander Kuleshov for (i = 0; i < type->cnt; i++) 154f14516fbSAlexander Kuleshov if (memblock_addrs_overlap(base, size, type->regions[i].base, 155f14516fbSAlexander Kuleshov type->regions[i].size)) 1566ed311b2SBenjamin Herrenschmidt break; 157c5c5c9d1STang Chen return i < type->cnt; 1586ed311b2SBenjamin Herrenschmidt } 1596ed311b2SBenjamin Herrenschmidt 16047cec443SMike Rapoport /** 16179442ed1STang Chen * __memblock_find_range_bottom_up - find free area utility in bottom-up 16279442ed1STang Chen * @start: start of candidate range 16347cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 16447cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 16579442ed1STang Chen * @size: size of free area to find 16679442ed1STang Chen * @align: alignment of free area to find 167b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 168fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 16979442ed1STang Chen * 17079442ed1STang Chen * Utility called from memblock_find_in_range_node(), find free area bottom-up. 17179442ed1STang Chen * 17247cec443SMike Rapoport * Return: 17379442ed1STang Chen * Found address on success, 0 on failure. 17479442ed1STang Chen */ 17579442ed1STang Chen static phys_addr_t __init_memblock 17679442ed1STang Chen __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, 177fc6daaf9STony Luck phys_addr_t size, phys_addr_t align, int nid, 178e1720feeSMike Rapoport enum memblock_flags flags) 17979442ed1STang Chen { 18079442ed1STang Chen phys_addr_t this_start, this_end, cand; 18179442ed1STang Chen u64 i; 18279442ed1STang Chen 183fc6daaf9STony Luck for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) { 18479442ed1STang Chen this_start = clamp(this_start, start, end); 18579442ed1STang Chen this_end = clamp(this_end, start, end); 18679442ed1STang Chen 18779442ed1STang Chen cand = round_up(this_start, align); 18879442ed1STang Chen if (cand < this_end && this_end - cand >= size) 18979442ed1STang Chen return cand; 19079442ed1STang Chen } 19179442ed1STang Chen 19279442ed1STang Chen return 0; 19379442ed1STang Chen } 19479442ed1STang Chen 1957bd0b0f0STejun Heo /** 1961402899eSTang Chen * __memblock_find_range_top_down - find free area utility, in top-down 1971402899eSTang Chen * @start: start of candidate range 19847cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 19947cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 2001402899eSTang Chen * @size: size of free area to find 2011402899eSTang Chen * @align: alignment of free area to find 202b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 203fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 2041402899eSTang Chen * 2051402899eSTang Chen * Utility called from memblock_find_in_range_node(), find free area top-down. 2061402899eSTang Chen * 20747cec443SMike Rapoport * Return: 20879442ed1STang Chen * Found address on success, 0 on failure. 2091402899eSTang Chen */ 2101402899eSTang Chen static phys_addr_t __init_memblock 2111402899eSTang Chen __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, 212fc6daaf9STony Luck phys_addr_t size, phys_addr_t align, int nid, 213e1720feeSMike Rapoport enum memblock_flags flags) 2141402899eSTang Chen { 2151402899eSTang Chen phys_addr_t this_start, this_end, cand; 2161402899eSTang Chen u64 i; 2171402899eSTang Chen 218fc6daaf9STony Luck for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end, 219fc6daaf9STony Luck NULL) { 2201402899eSTang Chen this_start = clamp(this_start, start, end); 2211402899eSTang Chen this_end = clamp(this_end, start, end); 2221402899eSTang Chen 2231402899eSTang Chen if (this_end < size) 2241402899eSTang Chen continue; 2251402899eSTang Chen 2261402899eSTang Chen cand = round_down(this_end - size, align); 2271402899eSTang Chen if (cand >= this_start) 2281402899eSTang Chen return cand; 2291402899eSTang Chen } 2301402899eSTang Chen 2311402899eSTang Chen return 0; 2321402899eSTang Chen } 2331402899eSTang Chen 2341402899eSTang Chen /** 2357bd0b0f0STejun Heo * memblock_find_in_range_node - find free area in given range and node 2367bd0b0f0STejun Heo * @size: size of free area to find 2377bd0b0f0STejun Heo * @align: alignment of free area to find 23887029ee9SGrygorii Strashko * @start: start of candidate range 23947cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 24047cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 241b1154233SGrygorii Strashko * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 242fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 2437bd0b0f0STejun Heo * 2447bd0b0f0STejun Heo * Find @size free area aligned to @align in the specified range and node. 2457bd0b0f0STejun Heo * 24679442ed1STang Chen * When allocation direction is bottom-up, the @start should be greater 24779442ed1STang Chen * than the end of the kernel image. Otherwise, it will be trimmed. The 24879442ed1STang Chen * reason is that we want the bottom-up allocation just near the kernel 24979442ed1STang Chen * image so it is highly likely that the allocated memory and the kernel 25079442ed1STang Chen * will reside in the same node. 25179442ed1STang Chen * 25279442ed1STang Chen * If bottom-up allocation failed, will try to allocate memory top-down. 25379442ed1STang Chen * 25447cec443SMike Rapoport * Return: 25579442ed1STang Chen * Found address on success, 0 on failure. 2566ed311b2SBenjamin Herrenschmidt */ 25787029ee9SGrygorii Strashko phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, 25887029ee9SGrygorii Strashko phys_addr_t align, phys_addr_t start, 259e1720feeSMike Rapoport phys_addr_t end, int nid, 260e1720feeSMike Rapoport enum memblock_flags flags) 261f7210e6cSTang Chen { 2620cfb8f0cSTang Chen phys_addr_t kernel_end, ret; 26379442ed1STang Chen 264f7210e6cSTang Chen /* pump up @end */ 265f7210e6cSTang Chen if (end == MEMBLOCK_ALLOC_ACCESSIBLE) 266f7210e6cSTang Chen end = memblock.current_limit; 267f7210e6cSTang Chen 268f7210e6cSTang Chen /* avoid allocating the first page */ 269f7210e6cSTang Chen start = max_t(phys_addr_t, start, PAGE_SIZE); 270f7210e6cSTang Chen end = max(start, end); 27179442ed1STang Chen kernel_end = __pa_symbol(_end); 27279442ed1STang Chen 27379442ed1STang Chen /* 27479442ed1STang Chen * try bottom-up allocation only when bottom-up mode 27579442ed1STang Chen * is set and @end is above the kernel image. 27679442ed1STang Chen */ 27779442ed1STang Chen if (memblock_bottom_up() && end > kernel_end) { 27879442ed1STang Chen phys_addr_t bottom_up_start; 27979442ed1STang Chen 28079442ed1STang Chen /* make sure we will allocate above the kernel */ 28179442ed1STang Chen bottom_up_start = max(start, kernel_end); 28279442ed1STang Chen 28379442ed1STang Chen /* ok, try bottom-up allocation first */ 28479442ed1STang Chen ret = __memblock_find_range_bottom_up(bottom_up_start, end, 285fc6daaf9STony Luck size, align, nid, flags); 28679442ed1STang Chen if (ret) 28779442ed1STang Chen return ret; 28879442ed1STang Chen 28979442ed1STang Chen /* 29079442ed1STang Chen * we always limit bottom-up allocation above the kernel, 29179442ed1STang Chen * but top-down allocation doesn't have the limit, so 29279442ed1STang Chen * retrying top-down allocation may succeed when bottom-up 29379442ed1STang Chen * allocation failed. 29479442ed1STang Chen * 29579442ed1STang Chen * bottom-up allocation is expected to be fail very rarely, 29679442ed1STang Chen * so we use WARN_ONCE() here to see the stack trace if 29779442ed1STang Chen * fail happens. 29879442ed1STang Chen */ 299e3d301caSMichal Hocko WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE), 300e3d301caSMichal Hocko "memblock: bottom-up allocation failed, memory hotremove may be affected\n"); 30179442ed1STang Chen } 302f7210e6cSTang Chen 303fc6daaf9STony Luck return __memblock_find_range_top_down(start, end, size, align, nid, 304fc6daaf9STony Luck flags); 305f7210e6cSTang Chen } 3066ed311b2SBenjamin Herrenschmidt 3077bd0b0f0STejun Heo /** 3087bd0b0f0STejun Heo * memblock_find_in_range - find free area in given range 3097bd0b0f0STejun Heo * @start: start of candidate range 31047cec443SMike Rapoport * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 31147cec443SMike Rapoport * %MEMBLOCK_ALLOC_ACCESSIBLE 3127bd0b0f0STejun Heo * @size: size of free area to find 3137bd0b0f0STejun Heo * @align: alignment of free area to find 3147bd0b0f0STejun Heo * 3157bd0b0f0STejun Heo * Find @size free area aligned to @align in the specified range. 3167bd0b0f0STejun Heo * 31747cec443SMike Rapoport * Return: 31879442ed1STang Chen * Found address on success, 0 on failure. 3197bd0b0f0STejun Heo */ 3207bd0b0f0STejun Heo phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, 3217bd0b0f0STejun Heo phys_addr_t end, phys_addr_t size, 3227bd0b0f0STejun Heo phys_addr_t align) 3237bd0b0f0STejun Heo { 324a3f5bafcSTony Luck phys_addr_t ret; 325e1720feeSMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 326a3f5bafcSTony Luck 327a3f5bafcSTony Luck again: 328a3f5bafcSTony Luck ret = memblock_find_in_range_node(size, align, start, end, 329a3f5bafcSTony Luck NUMA_NO_NODE, flags); 330a3f5bafcSTony Luck 331a3f5bafcSTony Luck if (!ret && (flags & MEMBLOCK_MIRROR)) { 332a3f5bafcSTony Luck pr_warn("Could not allocate %pap bytes of mirrored memory\n", 333a3f5bafcSTony Luck &size); 334a3f5bafcSTony Luck flags &= ~MEMBLOCK_MIRROR; 335a3f5bafcSTony Luck goto again; 336a3f5bafcSTony Luck } 337a3f5bafcSTony Luck 338a3f5bafcSTony Luck return ret; 3397bd0b0f0STejun Heo } 3407bd0b0f0STejun Heo 34110d06439SYinghai Lu static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) 34295f72d1eSYinghai Lu { 3431440c4e2STejun Heo type->total_size -= type->regions[r].size; 3447c0caeb8STejun Heo memmove(&type->regions[r], &type->regions[r + 1], 3457c0caeb8STejun Heo (type->cnt - (r + 1)) * sizeof(type->regions[r])); 346e3239ff9SBenjamin Herrenschmidt type->cnt--; 34795f72d1eSYinghai Lu 3488f7a6605SBenjamin Herrenschmidt /* Special case for empty arrays */ 3498f7a6605SBenjamin Herrenschmidt if (type->cnt == 0) { 3501440c4e2STejun Heo WARN_ON(type->total_size != 0); 3518f7a6605SBenjamin Herrenschmidt type->cnt = 1; 3528f7a6605SBenjamin Herrenschmidt type->regions[0].base = 0; 3538f7a6605SBenjamin Herrenschmidt type->regions[0].size = 0; 35466a20757STang Chen type->regions[0].flags = 0; 3557c0caeb8STejun Heo memblock_set_region_node(&type->regions[0], MAX_NUMNODES); 3568f7a6605SBenjamin Herrenschmidt } 35795f72d1eSYinghai Lu } 35895f72d1eSYinghai Lu 359354f17e1SPhilipp Hachtmann #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK 3603010f876SPavel Tatashin /** 36147cec443SMike Rapoport * memblock_discard - discard memory and reserved arrays if they were allocated 3623010f876SPavel Tatashin */ 3633010f876SPavel Tatashin void __init memblock_discard(void) 36429f67386SYinghai Lu { 3653010f876SPavel Tatashin phys_addr_t addr, size; 36629f67386SYinghai Lu 3673010f876SPavel Tatashin if (memblock.reserved.regions != memblock_reserved_init_regions) { 3683010f876SPavel Tatashin addr = __pa(memblock.reserved.regions); 3693010f876SPavel Tatashin size = PAGE_ALIGN(sizeof(struct memblock_region) * 37029f67386SYinghai Lu memblock.reserved.max); 3713010f876SPavel Tatashin __memblock_free_late(addr, size); 37229f67386SYinghai Lu } 37329f67386SYinghai Lu 37491b540f9SPavel Tatashin if (memblock.memory.regions != memblock_memory_init_regions) { 3753010f876SPavel Tatashin addr = __pa(memblock.memory.regions); 3763010f876SPavel Tatashin size = PAGE_ALIGN(sizeof(struct memblock_region) * 3775e270e25SPhilipp Hachtmann memblock.memory.max); 3783010f876SPavel Tatashin __memblock_free_late(addr, size); 3795e270e25SPhilipp Hachtmann } 3803010f876SPavel Tatashin } 3815e270e25SPhilipp Hachtmann #endif 3825e270e25SPhilipp Hachtmann 38348c3b583SGreg Pearson /** 38448c3b583SGreg Pearson * memblock_double_array - double the size of the memblock regions array 38548c3b583SGreg Pearson * @type: memblock type of the regions array being doubled 38648c3b583SGreg Pearson * @new_area_start: starting address of memory range to avoid overlap with 38748c3b583SGreg Pearson * @new_area_size: size of memory range to avoid overlap with 38848c3b583SGreg Pearson * 38948c3b583SGreg Pearson * Double the size of the @type regions array. If memblock is being used to 39048c3b583SGreg Pearson * allocate memory for a new reserved regions array and there is a previously 39148c3b583SGreg Pearson * allocated memory range [@new_area_start, @new_area_start + @new_area_size] 39248c3b583SGreg Pearson * waiting to be reserved, ensure the memory used by the new array does 39348c3b583SGreg Pearson * not overlap. 39448c3b583SGreg Pearson * 39547cec443SMike Rapoport * Return: 39648c3b583SGreg Pearson * 0 on success, -1 on failure. 39748c3b583SGreg Pearson */ 39848c3b583SGreg Pearson static int __init_memblock memblock_double_array(struct memblock_type *type, 39948c3b583SGreg Pearson phys_addr_t new_area_start, 40048c3b583SGreg Pearson phys_addr_t new_area_size) 401142b45a7SBenjamin Herrenschmidt { 402142b45a7SBenjamin Herrenschmidt struct memblock_region *new_array, *old_array; 40329f67386SYinghai Lu phys_addr_t old_alloc_size, new_alloc_size; 404a36aab89SMike Rapoport phys_addr_t old_size, new_size, addr, new_end; 405142b45a7SBenjamin Herrenschmidt int use_slab = slab_is_available(); 406181eb394SGavin Shan int *in_slab; 407142b45a7SBenjamin Herrenschmidt 408142b45a7SBenjamin Herrenschmidt /* We don't allow resizing until we know about the reserved regions 409142b45a7SBenjamin Herrenschmidt * of memory that aren't suitable for allocation 410142b45a7SBenjamin Herrenschmidt */ 411142b45a7SBenjamin Herrenschmidt if (!memblock_can_resize) 412142b45a7SBenjamin Herrenschmidt return -1; 413142b45a7SBenjamin Herrenschmidt 414142b45a7SBenjamin Herrenschmidt /* Calculate new doubled size */ 415142b45a7SBenjamin Herrenschmidt old_size = type->max * sizeof(struct memblock_region); 416142b45a7SBenjamin Herrenschmidt new_size = old_size << 1; 41729f67386SYinghai Lu /* 41829f67386SYinghai Lu * We need to allocated new one align to PAGE_SIZE, 41929f67386SYinghai Lu * so we can free them completely later. 42029f67386SYinghai Lu */ 42129f67386SYinghai Lu old_alloc_size = PAGE_ALIGN(old_size); 42229f67386SYinghai Lu new_alloc_size = PAGE_ALIGN(new_size); 423142b45a7SBenjamin Herrenschmidt 424181eb394SGavin Shan /* Retrieve the slab flag */ 425181eb394SGavin Shan if (type == &memblock.memory) 426181eb394SGavin Shan in_slab = &memblock_memory_in_slab; 427181eb394SGavin Shan else 428181eb394SGavin Shan in_slab = &memblock_reserved_in_slab; 429181eb394SGavin Shan 430142b45a7SBenjamin Herrenschmidt /* Try to find some space for it. 431142b45a7SBenjamin Herrenschmidt * 432142b45a7SBenjamin Herrenschmidt * WARNING: We assume that either slab_is_available() and we use it or 433fd07383bSAndrew Morton * we use MEMBLOCK for allocations. That means that this is unsafe to 434fd07383bSAndrew Morton * use when bootmem is currently active (unless bootmem itself is 435fd07383bSAndrew Morton * implemented on top of MEMBLOCK which isn't the case yet) 436142b45a7SBenjamin Herrenschmidt * 437142b45a7SBenjamin Herrenschmidt * This should however not be an issue for now, as we currently only 438fd07383bSAndrew Morton * call into MEMBLOCK while it's still active, or much later when slab 439fd07383bSAndrew Morton * is active for memory hotplug operations 440142b45a7SBenjamin Herrenschmidt */ 441142b45a7SBenjamin Herrenschmidt if (use_slab) { 442142b45a7SBenjamin Herrenschmidt new_array = kmalloc(new_size, GFP_KERNEL); 4431f5026a7STejun Heo addr = new_array ? __pa(new_array) : 0; 4444e2f0775SGavin Shan } else { 44548c3b583SGreg Pearson /* only exclude range when trying to double reserved.regions */ 44648c3b583SGreg Pearson if (type != &memblock.reserved) 44748c3b583SGreg Pearson new_area_start = new_area_size = 0; 44848c3b583SGreg Pearson 44948c3b583SGreg Pearson addr = memblock_find_in_range(new_area_start + new_area_size, 45048c3b583SGreg Pearson memblock.current_limit, 45129f67386SYinghai Lu new_alloc_size, PAGE_SIZE); 45248c3b583SGreg Pearson if (!addr && new_area_size) 45348c3b583SGreg Pearson addr = memblock_find_in_range(0, 45448c3b583SGreg Pearson min(new_area_start, memblock.current_limit), 45529f67386SYinghai Lu new_alloc_size, PAGE_SIZE); 45648c3b583SGreg Pearson 45715674868SSachin Kamat new_array = addr ? __va(addr) : NULL; 4584e2f0775SGavin Shan } 4591f5026a7STejun Heo if (!addr) { 460142b45a7SBenjamin Herrenschmidt pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 4610262d9c8SHeiko Carstens type->name, type->max, type->max * 2); 462142b45a7SBenjamin Herrenschmidt return -1; 463142b45a7SBenjamin Herrenschmidt } 464142b45a7SBenjamin Herrenschmidt 465a36aab89SMike Rapoport new_end = addr + new_size - 1; 466a36aab89SMike Rapoport memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]", 467a36aab89SMike Rapoport type->name, type->max * 2, &addr, &new_end); 468ea9e4376SYinghai Lu 469fd07383bSAndrew Morton /* 470fd07383bSAndrew Morton * Found space, we now need to move the array over before we add the 471fd07383bSAndrew Morton * reserved region since it may be our reserved array itself that is 472fd07383bSAndrew Morton * full. 473142b45a7SBenjamin Herrenschmidt */ 474142b45a7SBenjamin Herrenschmidt memcpy(new_array, type->regions, old_size); 475142b45a7SBenjamin Herrenschmidt memset(new_array + type->max, 0, old_size); 476142b45a7SBenjamin Herrenschmidt old_array = type->regions; 477142b45a7SBenjamin Herrenschmidt type->regions = new_array; 478142b45a7SBenjamin Herrenschmidt type->max <<= 1; 479142b45a7SBenjamin Herrenschmidt 480fd07383bSAndrew Morton /* Free old array. We needn't free it if the array is the static one */ 481181eb394SGavin Shan if (*in_slab) 482181eb394SGavin Shan kfree(old_array); 483181eb394SGavin Shan else if (old_array != memblock_memory_init_regions && 484142b45a7SBenjamin Herrenschmidt old_array != memblock_reserved_init_regions) 48529f67386SYinghai Lu memblock_free(__pa(old_array), old_alloc_size); 486142b45a7SBenjamin Herrenschmidt 487fd07383bSAndrew Morton /* 488fd07383bSAndrew Morton * Reserve the new array if that comes from the memblock. Otherwise, we 489fd07383bSAndrew Morton * needn't do it 490181eb394SGavin Shan */ 491181eb394SGavin Shan if (!use_slab) 49229f67386SYinghai Lu BUG_ON(memblock_reserve(addr, new_alloc_size)); 493181eb394SGavin Shan 494181eb394SGavin Shan /* Update slab flag */ 495181eb394SGavin Shan *in_slab = use_slab; 496181eb394SGavin Shan 497142b45a7SBenjamin Herrenschmidt return 0; 498142b45a7SBenjamin Herrenschmidt } 499142b45a7SBenjamin Herrenschmidt 500784656f9STejun Heo /** 501784656f9STejun Heo * memblock_merge_regions - merge neighboring compatible regions 502784656f9STejun Heo * @type: memblock type to scan 503784656f9STejun Heo * 504784656f9STejun Heo * Scan @type and merge neighboring compatible regions. 505784656f9STejun Heo */ 506784656f9STejun Heo static void __init_memblock memblock_merge_regions(struct memblock_type *type) 507784656f9STejun Heo { 508784656f9STejun Heo int i = 0; 509784656f9STejun Heo 510784656f9STejun Heo /* cnt never goes below 1 */ 511784656f9STejun Heo while (i < type->cnt - 1) { 512784656f9STejun Heo struct memblock_region *this = &type->regions[i]; 513784656f9STejun Heo struct memblock_region *next = &type->regions[i + 1]; 514784656f9STejun Heo 5157c0caeb8STejun Heo if (this->base + this->size != next->base || 5167c0caeb8STejun Heo memblock_get_region_node(this) != 51766a20757STang Chen memblock_get_region_node(next) || 51866a20757STang Chen this->flags != next->flags) { 519784656f9STejun Heo BUG_ON(this->base + this->size > next->base); 520784656f9STejun Heo i++; 521784656f9STejun Heo continue; 522784656f9STejun Heo } 523784656f9STejun Heo 524784656f9STejun Heo this->size += next->size; 525c0232ae8SLin Feng /* move forward from next + 1, index of which is i + 2 */ 526c0232ae8SLin Feng memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 527784656f9STejun Heo type->cnt--; 528784656f9STejun Heo } 529784656f9STejun Heo } 530784656f9STejun Heo 531784656f9STejun Heo /** 532784656f9STejun Heo * memblock_insert_region - insert new memblock region 533784656f9STejun Heo * @type: memblock type to insert into 534784656f9STejun Heo * @idx: index for the insertion point 535784656f9STejun Heo * @base: base address of the new region 536784656f9STejun Heo * @size: size of the new region 537209ff86dSTang Chen * @nid: node id of the new region 53866a20757STang Chen * @flags: flags of the new region 539784656f9STejun Heo * 540784656f9STejun Heo * Insert new memblock region [@base, @base + @size) into @type at @idx. 541412d0008SAlexander Kuleshov * @type must already have extra room to accommodate the new region. 542784656f9STejun Heo */ 543784656f9STejun Heo static void __init_memblock memblock_insert_region(struct memblock_type *type, 544784656f9STejun Heo int idx, phys_addr_t base, 54566a20757STang Chen phys_addr_t size, 546e1720feeSMike Rapoport int nid, 547e1720feeSMike Rapoport enum memblock_flags flags) 548784656f9STejun Heo { 549784656f9STejun Heo struct memblock_region *rgn = &type->regions[idx]; 550784656f9STejun Heo 551784656f9STejun Heo BUG_ON(type->cnt >= type->max); 552784656f9STejun Heo memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 553784656f9STejun Heo rgn->base = base; 554784656f9STejun Heo rgn->size = size; 55566a20757STang Chen rgn->flags = flags; 5567c0caeb8STejun Heo memblock_set_region_node(rgn, nid); 557784656f9STejun Heo type->cnt++; 5581440c4e2STejun Heo type->total_size += size; 559784656f9STejun Heo } 560784656f9STejun Heo 561784656f9STejun Heo /** 562f1af9d3aSPhilipp Hachtmann * memblock_add_range - add new memblock region 563784656f9STejun Heo * @type: memblock type to add new region into 564784656f9STejun Heo * @base: base address of the new region 565784656f9STejun Heo * @size: size of the new region 5667fb0bc3fSTejun Heo * @nid: nid of the new region 56766a20757STang Chen * @flags: flags of the new region 568784656f9STejun Heo * 569784656f9STejun Heo * Add new memblock region [@base, @base + @size) into @type. The new region 570784656f9STejun Heo * is allowed to overlap with existing ones - overlaps don't affect already 571784656f9STejun Heo * existing regions. @type is guaranteed to be minimal (all neighbouring 572784656f9STejun Heo * compatible regions are merged) after the addition. 573784656f9STejun Heo * 57447cec443SMike Rapoport * Return: 575784656f9STejun Heo * 0 on success, -errno on failure. 576784656f9STejun Heo */ 577f1af9d3aSPhilipp Hachtmann int __init_memblock memblock_add_range(struct memblock_type *type, 57866a20757STang Chen phys_addr_t base, phys_addr_t size, 579e1720feeSMike Rapoport int nid, enum memblock_flags flags) 58095f72d1eSYinghai Lu { 581784656f9STejun Heo bool insert = false; 582eb18f1b5STejun Heo phys_addr_t obase = base; 583eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 5848c9c1701SAlexander Kuleshov int idx, nr_new; 5858c9c1701SAlexander Kuleshov struct memblock_region *rgn; 58695f72d1eSYinghai Lu 587b3dc627cSTejun Heo if (!size) 588b3dc627cSTejun Heo return 0; 589b3dc627cSTejun Heo 590784656f9STejun Heo /* special case for empty array */ 591784656f9STejun Heo if (type->regions[0].size == 0) { 5921440c4e2STejun Heo WARN_ON(type->cnt != 1 || type->total_size); 593784656f9STejun Heo type->regions[0].base = base; 594784656f9STejun Heo type->regions[0].size = size; 59566a20757STang Chen type->regions[0].flags = flags; 5967fb0bc3fSTejun Heo memblock_set_region_node(&type->regions[0], nid); 5971440c4e2STejun Heo type->total_size = size; 598784656f9STejun Heo return 0; 599784656f9STejun Heo } 600784656f9STejun Heo repeat: 601784656f9STejun Heo /* 602784656f9STejun Heo * The following is executed twice. Once with %false @insert and 603784656f9STejun Heo * then with %true. The first counts the number of regions needed 604412d0008SAlexander Kuleshov * to accommodate the new area. The second actually inserts them. 605784656f9STejun Heo */ 606784656f9STejun Heo base = obase; 607784656f9STejun Heo nr_new = 0; 608784656f9STejun Heo 60966e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 610784656f9STejun Heo phys_addr_t rbase = rgn->base; 611784656f9STejun Heo phys_addr_t rend = rbase + rgn->size; 6128f7a6605SBenjamin Herrenschmidt 613784656f9STejun Heo if (rbase >= end) 6148f7a6605SBenjamin Herrenschmidt break; 615784656f9STejun Heo if (rend <= base) 616784656f9STejun Heo continue; 617784656f9STejun Heo /* 618784656f9STejun Heo * @rgn overlaps. If it separates the lower part of new 619784656f9STejun Heo * area, insert that portion. 6208f7a6605SBenjamin Herrenschmidt */ 621784656f9STejun Heo if (rbase > base) { 622c0a29498SWei Yang #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 623c0a29498SWei Yang WARN_ON(nid != memblock_get_region_node(rgn)); 624c0a29498SWei Yang #endif 6254fcab5f4SWei Yang WARN_ON(flags != rgn->flags); 626784656f9STejun Heo nr_new++; 627784656f9STejun Heo if (insert) 6288c9c1701SAlexander Kuleshov memblock_insert_region(type, idx++, base, 62966a20757STang Chen rbase - base, nid, 63066a20757STang Chen flags); 631784656f9STejun Heo } 632784656f9STejun Heo /* area below @rend is dealt with, forget about it */ 633784656f9STejun Heo base = min(rend, end); 6348f7a6605SBenjamin Herrenschmidt } 6358f7a6605SBenjamin Herrenschmidt 636784656f9STejun Heo /* insert the remaining portion */ 637784656f9STejun Heo if (base < end) { 638784656f9STejun Heo nr_new++; 639784656f9STejun Heo if (insert) 6408c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, base, end - base, 64166a20757STang Chen nid, flags); 6428f7a6605SBenjamin Herrenschmidt } 6438f7a6605SBenjamin Herrenschmidt 644ef3cc4dbSnimisolo if (!nr_new) 645ef3cc4dbSnimisolo return 0; 646ef3cc4dbSnimisolo 647784656f9STejun Heo /* 648784656f9STejun Heo * If this was the first round, resize array and repeat for actual 649784656f9STejun Heo * insertions; otherwise, merge and return. 6508f7a6605SBenjamin Herrenschmidt */ 651784656f9STejun Heo if (!insert) { 652784656f9STejun Heo while (type->cnt + nr_new > type->max) 65348c3b583SGreg Pearson if (memblock_double_array(type, obase, size) < 0) 654784656f9STejun Heo return -ENOMEM; 655784656f9STejun Heo insert = true; 656784656f9STejun Heo goto repeat; 65795f72d1eSYinghai Lu } else { 658784656f9STejun Heo memblock_merge_regions(type); 65995f72d1eSYinghai Lu return 0; 66095f72d1eSYinghai Lu } 661784656f9STejun Heo } 66295f72d1eSYinghai Lu 66348a833ccSMike Rapoport /** 66448a833ccSMike Rapoport * memblock_add_node - add new memblock region within a NUMA node 66548a833ccSMike Rapoport * @base: base address of the new region 66648a833ccSMike Rapoport * @size: size of the new region 66748a833ccSMike Rapoport * @nid: nid of the new region 66848a833ccSMike Rapoport * 66948a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 67048a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 67148a833ccSMike Rapoport * 67248a833ccSMike Rapoport * Return: 67348a833ccSMike Rapoport * 0 on success, -errno on failure. 67448a833ccSMike Rapoport */ 6757fb0bc3fSTejun Heo int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 6767fb0bc3fSTejun Heo int nid) 6777fb0bc3fSTejun Heo { 678f1af9d3aSPhilipp Hachtmann return memblock_add_range(&memblock.memory, base, size, nid, 0); 6797fb0bc3fSTejun Heo } 6807fb0bc3fSTejun Heo 68148a833ccSMike Rapoport /** 68248a833ccSMike Rapoport * memblock_add - add new memblock region 68348a833ccSMike Rapoport * @base: base address of the new region 68448a833ccSMike Rapoport * @size: size of the new region 68548a833ccSMike Rapoport * 68648a833ccSMike Rapoport * Add new memblock region [@base, @base + @size) to the "memory" 68748a833ccSMike Rapoport * type. See memblock_add_range() description for mode details 68848a833ccSMike Rapoport * 68948a833ccSMike Rapoport * Return: 69048a833ccSMike Rapoport * 0 on success, -errno on failure. 69148a833ccSMike Rapoport */ 692f705ac4bSAlexander Kuleshov int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 6936a4055bcSAlexander Kuleshov { 6945d63f81cSMiles Chen phys_addr_t end = base + size - 1; 6955d63f81cSMiles Chen 6965d63f81cSMiles Chen memblock_dbg("memblock_add: [%pa-%pa] %pF\n", 6975d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 6986a4055bcSAlexander Kuleshov 699f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 70095f72d1eSYinghai Lu } 70195f72d1eSYinghai Lu 7026a9ceb31STejun Heo /** 7036a9ceb31STejun Heo * memblock_isolate_range - isolate given range into disjoint memblocks 7046a9ceb31STejun Heo * @type: memblock type to isolate range for 7056a9ceb31STejun Heo * @base: base of range to isolate 7066a9ceb31STejun Heo * @size: size of range to isolate 7076a9ceb31STejun Heo * @start_rgn: out parameter for the start of isolated region 7086a9ceb31STejun Heo * @end_rgn: out parameter for the end of isolated region 7096a9ceb31STejun Heo * 7106a9ceb31STejun Heo * Walk @type and ensure that regions don't cross the boundaries defined by 7116a9ceb31STejun Heo * [@base, @base + @size). Crossing regions are split at the boundaries, 7126a9ceb31STejun Heo * which may create at most two more regions. The index of the first 7136a9ceb31STejun Heo * region inside the range is returned in *@start_rgn and end in *@end_rgn. 7146a9ceb31STejun Heo * 71547cec443SMike Rapoport * Return: 7166a9ceb31STejun Heo * 0 on success, -errno on failure. 7176a9ceb31STejun Heo */ 7186a9ceb31STejun Heo static int __init_memblock memblock_isolate_range(struct memblock_type *type, 7196a9ceb31STejun Heo phys_addr_t base, phys_addr_t size, 7206a9ceb31STejun Heo int *start_rgn, int *end_rgn) 7216a9ceb31STejun Heo { 722eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 7238c9c1701SAlexander Kuleshov int idx; 7248c9c1701SAlexander Kuleshov struct memblock_region *rgn; 7256a9ceb31STejun Heo 7266a9ceb31STejun Heo *start_rgn = *end_rgn = 0; 7276a9ceb31STejun Heo 728b3dc627cSTejun Heo if (!size) 729b3dc627cSTejun Heo return 0; 730b3dc627cSTejun Heo 7316a9ceb31STejun Heo /* we'll create at most two more regions */ 7326a9ceb31STejun Heo while (type->cnt + 2 > type->max) 73348c3b583SGreg Pearson if (memblock_double_array(type, base, size) < 0) 7346a9ceb31STejun Heo return -ENOMEM; 7356a9ceb31STejun Heo 73666e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 7376a9ceb31STejun Heo phys_addr_t rbase = rgn->base; 7386a9ceb31STejun Heo phys_addr_t rend = rbase + rgn->size; 7396a9ceb31STejun Heo 7406a9ceb31STejun Heo if (rbase >= end) 7416a9ceb31STejun Heo break; 7426a9ceb31STejun Heo if (rend <= base) 7436a9ceb31STejun Heo continue; 7446a9ceb31STejun Heo 7456a9ceb31STejun Heo if (rbase < base) { 7466a9ceb31STejun Heo /* 7476a9ceb31STejun Heo * @rgn intersects from below. Split and continue 7486a9ceb31STejun Heo * to process the next region - the new top half. 7496a9ceb31STejun Heo */ 7506a9ceb31STejun Heo rgn->base = base; 7511440c4e2STejun Heo rgn->size -= base - rbase; 7521440c4e2STejun Heo type->total_size -= base - rbase; 7538c9c1701SAlexander Kuleshov memblock_insert_region(type, idx, rbase, base - rbase, 75466a20757STang Chen memblock_get_region_node(rgn), 75566a20757STang Chen rgn->flags); 7566a9ceb31STejun Heo } else if (rend > end) { 7576a9ceb31STejun Heo /* 7586a9ceb31STejun Heo * @rgn intersects from above. Split and redo the 7596a9ceb31STejun Heo * current region - the new bottom half. 7606a9ceb31STejun Heo */ 7616a9ceb31STejun Heo rgn->base = end; 7621440c4e2STejun Heo rgn->size -= end - rbase; 7631440c4e2STejun Heo type->total_size -= end - rbase; 7648c9c1701SAlexander Kuleshov memblock_insert_region(type, idx--, rbase, end - rbase, 76566a20757STang Chen memblock_get_region_node(rgn), 76666a20757STang Chen rgn->flags); 7676a9ceb31STejun Heo } else { 7686a9ceb31STejun Heo /* @rgn is fully contained, record it */ 7696a9ceb31STejun Heo if (!*end_rgn) 7708c9c1701SAlexander Kuleshov *start_rgn = idx; 7718c9c1701SAlexander Kuleshov *end_rgn = idx + 1; 7726a9ceb31STejun Heo } 7736a9ceb31STejun Heo } 7746a9ceb31STejun Heo 7756a9ceb31STejun Heo return 0; 7766a9ceb31STejun Heo } 7776a9ceb31STejun Heo 77835bd16a2SAlexander Kuleshov static int __init_memblock memblock_remove_range(struct memblock_type *type, 7798f7a6605SBenjamin Herrenschmidt phys_addr_t base, phys_addr_t size) 78095f72d1eSYinghai Lu { 78171936180STejun Heo int start_rgn, end_rgn; 78271936180STejun Heo int i, ret; 78395f72d1eSYinghai Lu 78471936180STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 78571936180STejun Heo if (ret) 78671936180STejun Heo return ret; 78795f72d1eSYinghai Lu 78871936180STejun Heo for (i = end_rgn - 1; i >= start_rgn; i--) 78971936180STejun Heo memblock_remove_region(type, i); 79095f72d1eSYinghai Lu return 0; 79195f72d1eSYinghai Lu } 79295f72d1eSYinghai Lu 793581adcbeSTejun Heo int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 79495f72d1eSYinghai Lu { 79525cf23d7SMinchan Kim phys_addr_t end = base + size - 1; 79625cf23d7SMinchan Kim 79725cf23d7SMinchan Kim memblock_dbg("memblock_remove: [%pa-%pa] %pS\n", 79825cf23d7SMinchan Kim &base, &end, (void *)_RET_IP_); 79925cf23d7SMinchan Kim 800f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.memory, base, size); 80195f72d1eSYinghai Lu } 80295f72d1eSYinghai Lu 803f1af9d3aSPhilipp Hachtmann 804581adcbeSTejun Heo int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) 80595f72d1eSYinghai Lu { 8065d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8075d63f81cSMiles Chen 8085d63f81cSMiles Chen memblock_dbg(" memblock_free: [%pa-%pa] %pF\n", 8095d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 81024aa0788STejun Heo 8119099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 812f1af9d3aSPhilipp Hachtmann return memblock_remove_range(&memblock.reserved, base, size); 81395f72d1eSYinghai Lu } 81495f72d1eSYinghai Lu 815f705ac4bSAlexander Kuleshov int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 81695f72d1eSYinghai Lu { 8175d63f81cSMiles Chen phys_addr_t end = base + size - 1; 8185d63f81cSMiles Chen 8195d63f81cSMiles Chen memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n", 8205d63f81cSMiles Chen &base, &end, (void *)_RET_IP_); 82195f72d1eSYinghai Lu 822f705ac4bSAlexander Kuleshov return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 82395f72d1eSYinghai Lu } 82495f72d1eSYinghai Lu 82535fd0808STejun Heo /** 82647cec443SMike Rapoport * memblock_setclr_flag - set or clear flag for a memory region 82747cec443SMike Rapoport * @base: base address of the region 82847cec443SMike Rapoport * @size: size of the region 82947cec443SMike Rapoport * @set: set or clear the flag 83047cec443SMike Rapoport * @flag: the flag to udpate 83166b16edfSTang Chen * 8324308ce17STony Luck * This function isolates region [@base, @base + @size), and sets/clears flag 83366b16edfSTang Chen * 83447cec443SMike Rapoport * Return: 0 on success, -errno on failure. 83566b16edfSTang Chen */ 8364308ce17STony Luck static int __init_memblock memblock_setclr_flag(phys_addr_t base, 8374308ce17STony Luck phys_addr_t size, int set, int flag) 83866b16edfSTang Chen { 83966b16edfSTang Chen struct memblock_type *type = &memblock.memory; 84066b16edfSTang Chen int i, ret, start_rgn, end_rgn; 84166b16edfSTang Chen 84266b16edfSTang Chen ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 84366b16edfSTang Chen if (ret) 84466b16edfSTang Chen return ret; 84566b16edfSTang Chen 84666b16edfSTang Chen for (i = start_rgn; i < end_rgn; i++) 8474308ce17STony Luck if (set) 8484308ce17STony Luck memblock_set_region_flags(&type->regions[i], flag); 8494308ce17STony Luck else 8504308ce17STony Luck memblock_clear_region_flags(&type->regions[i], flag); 85166b16edfSTang Chen 85266b16edfSTang Chen memblock_merge_regions(type); 85366b16edfSTang Chen return 0; 85466b16edfSTang Chen } 85566b16edfSTang Chen 85666b16edfSTang Chen /** 8574308ce17STony Luck * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 8584308ce17STony Luck * @base: the base phys addr of the region 8594308ce17STony Luck * @size: the size of the region 8604308ce17STony Luck * 86147cec443SMike Rapoport * Return: 0 on success, -errno on failure. 8624308ce17STony Luck */ 8634308ce17STony Luck int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 8644308ce17STony Luck { 8654308ce17STony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 8664308ce17STony Luck } 8674308ce17STony Luck 8684308ce17STony Luck /** 86966b16edfSTang Chen * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 87066b16edfSTang Chen * @base: the base phys addr of the region 87166b16edfSTang Chen * @size: the size of the region 87266b16edfSTang Chen * 87347cec443SMike Rapoport * Return: 0 on success, -errno on failure. 87466b16edfSTang Chen */ 87566b16edfSTang Chen int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 87666b16edfSTang Chen { 8774308ce17STony Luck return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 87866b16edfSTang Chen } 87966b16edfSTang Chen 88066b16edfSTang Chen /** 881a3f5bafcSTony Luck * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 882a3f5bafcSTony Luck * @base: the base phys addr of the region 883a3f5bafcSTony Luck * @size: the size of the region 884a3f5bafcSTony Luck * 88547cec443SMike Rapoport * Return: 0 on success, -errno on failure. 886a3f5bafcSTony Luck */ 887a3f5bafcSTony Luck int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 888a3f5bafcSTony Luck { 889a3f5bafcSTony Luck system_has_some_mirror = true; 890a3f5bafcSTony Luck 891a3f5bafcSTony Luck return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 892a3f5bafcSTony Luck } 893a3f5bafcSTony Luck 894bf3d3cc5SArd Biesheuvel /** 895bf3d3cc5SArd Biesheuvel * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 896bf3d3cc5SArd Biesheuvel * @base: the base phys addr of the region 897bf3d3cc5SArd Biesheuvel * @size: the size of the region 898bf3d3cc5SArd Biesheuvel * 89947cec443SMike Rapoport * Return: 0 on success, -errno on failure. 900bf3d3cc5SArd Biesheuvel */ 901bf3d3cc5SArd Biesheuvel int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 902bf3d3cc5SArd Biesheuvel { 903bf3d3cc5SArd Biesheuvel return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 904bf3d3cc5SArd Biesheuvel } 905a3f5bafcSTony Luck 906a3f5bafcSTony Luck /** 9074c546b8aSAKASHI Takahiro * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 9084c546b8aSAKASHI Takahiro * @base: the base phys addr of the region 9094c546b8aSAKASHI Takahiro * @size: the size of the region 9104c546b8aSAKASHI Takahiro * 91147cec443SMike Rapoport * Return: 0 on success, -errno on failure. 9124c546b8aSAKASHI Takahiro */ 9134c546b8aSAKASHI Takahiro int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 9144c546b8aSAKASHI Takahiro { 9154c546b8aSAKASHI Takahiro return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 9164c546b8aSAKASHI Takahiro } 9174c546b8aSAKASHI Takahiro 9184c546b8aSAKASHI Takahiro /** 9198e7a7f86SRobin Holt * __next_reserved_mem_region - next function for for_each_reserved_region() 9208e7a7f86SRobin Holt * @idx: pointer to u64 loop variable 9218e7a7f86SRobin Holt * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL 9228e7a7f86SRobin Holt * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL 9238e7a7f86SRobin Holt * 9248e7a7f86SRobin Holt * Iterate over all reserved memory regions. 9258e7a7f86SRobin Holt */ 9268e7a7f86SRobin Holt void __init_memblock __next_reserved_mem_region(u64 *idx, 9278e7a7f86SRobin Holt phys_addr_t *out_start, 9288e7a7f86SRobin Holt phys_addr_t *out_end) 9298e7a7f86SRobin Holt { 930567d117bSAlexander Kuleshov struct memblock_type *type = &memblock.reserved; 9318e7a7f86SRobin Holt 932cd33a76bSRichard Leitner if (*idx < type->cnt) { 933567d117bSAlexander Kuleshov struct memblock_region *r = &type->regions[*idx]; 9348e7a7f86SRobin Holt phys_addr_t base = r->base; 9358e7a7f86SRobin Holt phys_addr_t size = r->size; 9368e7a7f86SRobin Holt 9378e7a7f86SRobin Holt if (out_start) 9388e7a7f86SRobin Holt *out_start = base; 9398e7a7f86SRobin Holt if (out_end) 9408e7a7f86SRobin Holt *out_end = base + size - 1; 9418e7a7f86SRobin Holt 9428e7a7f86SRobin Holt *idx += 1; 9438e7a7f86SRobin Holt return; 9448e7a7f86SRobin Holt } 9458e7a7f86SRobin Holt 9468e7a7f86SRobin Holt /* signal end of iteration */ 9478e7a7f86SRobin Holt *idx = ULLONG_MAX; 9488e7a7f86SRobin Holt } 9498e7a7f86SRobin Holt 9508e7a7f86SRobin Holt /** 951f1af9d3aSPhilipp Hachtmann * __next__mem_range - next function for for_each_free_mem_range() etc. 95235fd0808STejun Heo * @idx: pointer to u64 loop variable 953b1154233SGrygorii Strashko * @nid: node selector, %NUMA_NO_NODE for all nodes 954fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 955f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 956f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 957dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 958dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 959dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 96035fd0808STejun Heo * 961f1af9d3aSPhilipp Hachtmann * Find the first area from *@idx which matches @nid, fill the out 96235fd0808STejun Heo * parameters, and update *@idx for the next iteration. The lower 32bit of 963f1af9d3aSPhilipp Hachtmann * *@idx contains index into type_a and the upper 32bit indexes the 964f1af9d3aSPhilipp Hachtmann * areas before each region in type_b. For example, if type_b regions 96535fd0808STejun Heo * look like the following, 96635fd0808STejun Heo * 96735fd0808STejun Heo * 0:[0-16), 1:[32-48), 2:[128-130) 96835fd0808STejun Heo * 96935fd0808STejun Heo * The upper 32bit indexes the following regions. 97035fd0808STejun Heo * 97135fd0808STejun Heo * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 97235fd0808STejun Heo * 97335fd0808STejun Heo * As both region arrays are sorted, the function advances the two indices 97435fd0808STejun Heo * in lockstep and returns each intersection. 97535fd0808STejun Heo */ 976e1720feeSMike Rapoport void __init_memblock __next_mem_range(u64 *idx, int nid, 977e1720feeSMike Rapoport enum memblock_flags flags, 978f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 979f1af9d3aSPhilipp Hachtmann struct memblock_type *type_b, 98035fd0808STejun Heo phys_addr_t *out_start, 98135fd0808STejun Heo phys_addr_t *out_end, int *out_nid) 98235fd0808STejun Heo { 983f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 984f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 985b1154233SGrygorii Strashko 986f1af9d3aSPhilipp Hachtmann if (WARN_ONCE(nid == MAX_NUMNODES, 987f1af9d3aSPhilipp Hachtmann "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 988560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 98935fd0808STejun Heo 990f1af9d3aSPhilipp Hachtmann for (; idx_a < type_a->cnt; idx_a++) { 991f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 992f1af9d3aSPhilipp Hachtmann 99335fd0808STejun Heo phys_addr_t m_start = m->base; 99435fd0808STejun Heo phys_addr_t m_end = m->base + m->size; 995f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 99635fd0808STejun Heo 99735fd0808STejun Heo /* only memory regions are associated with nodes, check it */ 998f1af9d3aSPhilipp Hachtmann if (nid != NUMA_NO_NODE && nid != m_nid) 99935fd0808STejun Heo continue; 100035fd0808STejun Heo 10010a313a99SXishi Qiu /* skip hotpluggable memory regions if needed */ 10020a313a99SXishi Qiu if (movable_node_is_enabled() && memblock_is_hotpluggable(m)) 10030a313a99SXishi Qiu continue; 10040a313a99SXishi Qiu 1005a3f5bafcSTony Luck /* if we want mirror memory skip non-mirror memory regions */ 1006a3f5bafcSTony Luck if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1007a3f5bafcSTony Luck continue; 1008a3f5bafcSTony Luck 1009bf3d3cc5SArd Biesheuvel /* skip nomap memory unless we were asked for it explicitly */ 1010bf3d3cc5SArd Biesheuvel if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1011bf3d3cc5SArd Biesheuvel continue; 1012bf3d3cc5SArd Biesheuvel 1013f1af9d3aSPhilipp Hachtmann if (!type_b) { 1014f1af9d3aSPhilipp Hachtmann if (out_start) 1015f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1016f1af9d3aSPhilipp Hachtmann if (out_end) 1017f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1018f1af9d3aSPhilipp Hachtmann if (out_nid) 1019f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1020f1af9d3aSPhilipp Hachtmann idx_a++; 1021f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1022f1af9d3aSPhilipp Hachtmann return; 1023f1af9d3aSPhilipp Hachtmann } 102435fd0808STejun Heo 1025f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1026f1af9d3aSPhilipp Hachtmann for (; idx_b < type_b->cnt + 1; idx_b++) { 1027f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1028f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1029f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1030f1af9d3aSPhilipp Hachtmann 1031f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1032f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1033f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 10341c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1035f1af9d3aSPhilipp Hachtmann 1036f1af9d3aSPhilipp Hachtmann /* 1037f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1038f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1039f1af9d3aSPhilipp Hachtmann */ 104035fd0808STejun Heo if (r_start >= m_end) 104135fd0808STejun Heo break; 104235fd0808STejun Heo /* if the two regions intersect, we're done */ 104335fd0808STejun Heo if (m_start < r_end) { 104435fd0808STejun Heo if (out_start) 1045f1af9d3aSPhilipp Hachtmann *out_start = 1046f1af9d3aSPhilipp Hachtmann max(m_start, r_start); 104735fd0808STejun Heo if (out_end) 104835fd0808STejun Heo *out_end = min(m_end, r_end); 104935fd0808STejun Heo if (out_nid) 1050f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 105135fd0808STejun Heo /* 1052f1af9d3aSPhilipp Hachtmann * The region which ends first is 1053f1af9d3aSPhilipp Hachtmann * advanced for the next iteration. 105435fd0808STejun Heo */ 105535fd0808STejun Heo if (m_end <= r_end) 1056f1af9d3aSPhilipp Hachtmann idx_a++; 105735fd0808STejun Heo else 1058f1af9d3aSPhilipp Hachtmann idx_b++; 1059f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 106035fd0808STejun Heo return; 106135fd0808STejun Heo } 106235fd0808STejun Heo } 106335fd0808STejun Heo } 106435fd0808STejun Heo 106535fd0808STejun Heo /* signal end of iteration */ 106635fd0808STejun Heo *idx = ULLONG_MAX; 106735fd0808STejun Heo } 106835fd0808STejun Heo 10697bd0b0f0STejun Heo /** 1070f1af9d3aSPhilipp Hachtmann * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1071f1af9d3aSPhilipp Hachtmann * 10727bd0b0f0STejun Heo * @idx: pointer to u64 loop variable 1073ad5ea8cdSAlexander Kuleshov * @nid: node selector, %NUMA_NO_NODE for all nodes 1074fc6daaf9STony Luck * @flags: pick from blocks based on memory attributes 1075f1af9d3aSPhilipp Hachtmann * @type_a: pointer to memblock_type from where the range is taken 1076f1af9d3aSPhilipp Hachtmann * @type_b: pointer to memblock_type which excludes memory from being taken 1077dad7557eSWanpeng Li * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1078dad7557eSWanpeng Li * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1079dad7557eSWanpeng Li * @out_nid: ptr to int for nid of the range, can be %NULL 10807bd0b0f0STejun Heo * 108147cec443SMike Rapoport * Finds the next range from type_a which is not marked as unsuitable 108247cec443SMike Rapoport * in type_b. 108347cec443SMike Rapoport * 1084f1af9d3aSPhilipp Hachtmann * Reverse of __next_mem_range(). 10857bd0b0f0STejun Heo */ 1086e1720feeSMike Rapoport void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1087e1720feeSMike Rapoport enum memblock_flags flags, 1088f1af9d3aSPhilipp Hachtmann struct memblock_type *type_a, 1089f1af9d3aSPhilipp Hachtmann struct memblock_type *type_b, 10907bd0b0f0STejun Heo phys_addr_t *out_start, 10917bd0b0f0STejun Heo phys_addr_t *out_end, int *out_nid) 10927bd0b0f0STejun Heo { 1093f1af9d3aSPhilipp Hachtmann int idx_a = *idx & 0xffffffff; 1094f1af9d3aSPhilipp Hachtmann int idx_b = *idx >> 32; 1095b1154233SGrygorii Strashko 1096560dca27SGrygorii Strashko if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1097560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 10987bd0b0f0STejun Heo 10997bd0b0f0STejun Heo if (*idx == (u64)ULLONG_MAX) { 1100f1af9d3aSPhilipp Hachtmann idx_a = type_a->cnt - 1; 1101e47608abSzijun_hu if (type_b != NULL) 1102f1af9d3aSPhilipp Hachtmann idx_b = type_b->cnt; 1103e47608abSzijun_hu else 1104e47608abSzijun_hu idx_b = 0; 11057bd0b0f0STejun Heo } 11067bd0b0f0STejun Heo 1107f1af9d3aSPhilipp Hachtmann for (; idx_a >= 0; idx_a--) { 1108f1af9d3aSPhilipp Hachtmann struct memblock_region *m = &type_a->regions[idx_a]; 1109f1af9d3aSPhilipp Hachtmann 11107bd0b0f0STejun Heo phys_addr_t m_start = m->base; 11117bd0b0f0STejun Heo phys_addr_t m_end = m->base + m->size; 1112f1af9d3aSPhilipp Hachtmann int m_nid = memblock_get_region_node(m); 11137bd0b0f0STejun Heo 11147bd0b0f0STejun Heo /* only memory regions are associated with nodes, check it */ 1115f1af9d3aSPhilipp Hachtmann if (nid != NUMA_NO_NODE && nid != m_nid) 11167bd0b0f0STejun Heo continue; 11177bd0b0f0STejun Heo 111855ac590cSTang Chen /* skip hotpluggable memory regions if needed */ 111955ac590cSTang Chen if (movable_node_is_enabled() && memblock_is_hotpluggable(m)) 112055ac590cSTang Chen continue; 112155ac590cSTang Chen 1122a3f5bafcSTony Luck /* if we want mirror memory skip non-mirror memory regions */ 1123a3f5bafcSTony Luck if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1124a3f5bafcSTony Luck continue; 1125a3f5bafcSTony Luck 1126bf3d3cc5SArd Biesheuvel /* skip nomap memory unless we were asked for it explicitly */ 1127bf3d3cc5SArd Biesheuvel if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1128bf3d3cc5SArd Biesheuvel continue; 1129bf3d3cc5SArd Biesheuvel 1130f1af9d3aSPhilipp Hachtmann if (!type_b) { 1131f1af9d3aSPhilipp Hachtmann if (out_start) 1132f1af9d3aSPhilipp Hachtmann *out_start = m_start; 1133f1af9d3aSPhilipp Hachtmann if (out_end) 1134f1af9d3aSPhilipp Hachtmann *out_end = m_end; 1135f1af9d3aSPhilipp Hachtmann if (out_nid) 1136f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 1137fb399b48Szijun_hu idx_a--; 1138f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 1139f1af9d3aSPhilipp Hachtmann return; 1140f1af9d3aSPhilipp Hachtmann } 11417bd0b0f0STejun Heo 1142f1af9d3aSPhilipp Hachtmann /* scan areas before each reservation */ 1143f1af9d3aSPhilipp Hachtmann for (; idx_b >= 0; idx_b--) { 1144f1af9d3aSPhilipp Hachtmann struct memblock_region *r; 1145f1af9d3aSPhilipp Hachtmann phys_addr_t r_start; 1146f1af9d3aSPhilipp Hachtmann phys_addr_t r_end; 1147f1af9d3aSPhilipp Hachtmann 1148f1af9d3aSPhilipp Hachtmann r = &type_b->regions[idx_b]; 1149f1af9d3aSPhilipp Hachtmann r_start = idx_b ? r[-1].base + r[-1].size : 0; 1150f1af9d3aSPhilipp Hachtmann r_end = idx_b < type_b->cnt ? 11511c4bc43dSStefan Agner r->base : PHYS_ADDR_MAX; 1152f1af9d3aSPhilipp Hachtmann /* 1153f1af9d3aSPhilipp Hachtmann * if idx_b advanced past idx_a, 1154f1af9d3aSPhilipp Hachtmann * break out to advance idx_a 1155f1af9d3aSPhilipp Hachtmann */ 1156f1af9d3aSPhilipp Hachtmann 11577bd0b0f0STejun Heo if (r_end <= m_start) 11587bd0b0f0STejun Heo break; 11597bd0b0f0STejun Heo /* if the two regions intersect, we're done */ 11607bd0b0f0STejun Heo if (m_end > r_start) { 11617bd0b0f0STejun Heo if (out_start) 11627bd0b0f0STejun Heo *out_start = max(m_start, r_start); 11637bd0b0f0STejun Heo if (out_end) 11647bd0b0f0STejun Heo *out_end = min(m_end, r_end); 11657bd0b0f0STejun Heo if (out_nid) 1166f1af9d3aSPhilipp Hachtmann *out_nid = m_nid; 11677bd0b0f0STejun Heo if (m_start >= r_start) 1168f1af9d3aSPhilipp Hachtmann idx_a--; 11697bd0b0f0STejun Heo else 1170f1af9d3aSPhilipp Hachtmann idx_b--; 1171f1af9d3aSPhilipp Hachtmann *idx = (u32)idx_a | (u64)idx_b << 32; 11727bd0b0f0STejun Heo return; 11737bd0b0f0STejun Heo } 11747bd0b0f0STejun Heo } 11757bd0b0f0STejun Heo } 1176f1af9d3aSPhilipp Hachtmann /* signal end of iteration */ 11777bd0b0f0STejun Heo *idx = ULLONG_MAX; 11787bd0b0f0STejun Heo } 11797bd0b0f0STejun Heo 11807c0caeb8STejun Heo #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 11817c0caeb8STejun Heo /* 11827c0caeb8STejun Heo * Common iterator interface used to define for_each_mem_range(). 11837c0caeb8STejun Heo */ 11847c0caeb8STejun Heo void __init_memblock __next_mem_pfn_range(int *idx, int nid, 11857c0caeb8STejun Heo unsigned long *out_start_pfn, 11867c0caeb8STejun Heo unsigned long *out_end_pfn, int *out_nid) 11877c0caeb8STejun Heo { 11887c0caeb8STejun Heo struct memblock_type *type = &memblock.memory; 11897c0caeb8STejun Heo struct memblock_region *r; 11907c0caeb8STejun Heo 11917c0caeb8STejun Heo while (++*idx < type->cnt) { 11927c0caeb8STejun Heo r = &type->regions[*idx]; 11937c0caeb8STejun Heo 11947c0caeb8STejun Heo if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 11957c0caeb8STejun Heo continue; 11967c0caeb8STejun Heo if (nid == MAX_NUMNODES || nid == r->nid) 11977c0caeb8STejun Heo break; 11987c0caeb8STejun Heo } 11997c0caeb8STejun Heo if (*idx >= type->cnt) { 12007c0caeb8STejun Heo *idx = -1; 12017c0caeb8STejun Heo return; 12027c0caeb8STejun Heo } 12037c0caeb8STejun Heo 12047c0caeb8STejun Heo if (out_start_pfn) 12057c0caeb8STejun Heo *out_start_pfn = PFN_UP(r->base); 12067c0caeb8STejun Heo if (out_end_pfn) 12077c0caeb8STejun Heo *out_end_pfn = PFN_DOWN(r->base + r->size); 12087c0caeb8STejun Heo if (out_nid) 12097c0caeb8STejun Heo *out_nid = r->nid; 12107c0caeb8STejun Heo } 12117c0caeb8STejun Heo 12127c0caeb8STejun Heo /** 12137c0caeb8STejun Heo * memblock_set_node - set node ID on memblock regions 12147c0caeb8STejun Heo * @base: base of area to set node ID for 12157c0caeb8STejun Heo * @size: size of area to set node ID for 1216e7e8de59STang Chen * @type: memblock type to set node ID for 12177c0caeb8STejun Heo * @nid: node ID to set 12187c0caeb8STejun Heo * 1219e7e8de59STang Chen * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 12207c0caeb8STejun Heo * Regions which cross the area boundaries are split as necessary. 12217c0caeb8STejun Heo * 122247cec443SMike Rapoport * Return: 12237c0caeb8STejun Heo * 0 on success, -errno on failure. 12247c0caeb8STejun Heo */ 12257c0caeb8STejun Heo int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1226e7e8de59STang Chen struct memblock_type *type, int nid) 12277c0caeb8STejun Heo { 12286a9ceb31STejun Heo int start_rgn, end_rgn; 12296a9ceb31STejun Heo int i, ret; 12307c0caeb8STejun Heo 12316a9ceb31STejun Heo ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 12326a9ceb31STejun Heo if (ret) 12336a9ceb31STejun Heo return ret; 12347c0caeb8STejun Heo 12356a9ceb31STejun Heo for (i = start_rgn; i < end_rgn; i++) 1236e9d24ad3SWanpeng Li memblock_set_region_node(&type->regions[i], nid); 12377c0caeb8STejun Heo 12387c0caeb8STejun Heo memblock_merge_regions(type); 12397c0caeb8STejun Heo return 0; 12407c0caeb8STejun Heo } 12417c0caeb8STejun Heo #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ 12427c0caeb8STejun Heo 12432bfc2862SAkinobu Mita static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 12442bfc2862SAkinobu Mita phys_addr_t align, phys_addr_t start, 1245e1720feeSMike Rapoport phys_addr_t end, int nid, 1246e1720feeSMike Rapoport enum memblock_flags flags) 124795f72d1eSYinghai Lu { 12486ed311b2SBenjamin Herrenschmidt phys_addr_t found; 124995f72d1eSYinghai Lu 1250*2f770806SMike Rapoport if (!align) { 1251*2f770806SMike Rapoport /* Can't use WARNs this early in boot on powerpc */ 1252*2f770806SMike Rapoport dump_stack(); 1253*2f770806SMike Rapoport align = SMP_CACHE_BYTES; 1254*2f770806SMike Rapoport } 1255*2f770806SMike Rapoport 1256fc6daaf9STony Luck found = memblock_find_in_range_node(size, align, start, end, nid, 1257fc6daaf9STony Luck flags); 1258aedf95eaSCatalin Marinas if (found && !memblock_reserve(found, size)) { 1259aedf95eaSCatalin Marinas /* 1260aedf95eaSCatalin Marinas * The min_count is set to 0 so that memblock allocations are 1261aedf95eaSCatalin Marinas * never reported as leaks. 1262aedf95eaSCatalin Marinas */ 12639099daedSCatalin Marinas kmemleak_alloc_phys(found, size, 0, 0); 12646ed311b2SBenjamin Herrenschmidt return found; 1265aedf95eaSCatalin Marinas } 12666ed311b2SBenjamin Herrenschmidt return 0; 126795f72d1eSYinghai Lu } 126895f72d1eSYinghai Lu 12692bfc2862SAkinobu Mita phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align, 1270fc6daaf9STony Luck phys_addr_t start, phys_addr_t end, 1271e1720feeSMike Rapoport enum memblock_flags flags) 12722bfc2862SAkinobu Mita { 1273fc6daaf9STony Luck return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 1274fc6daaf9STony Luck flags); 12752bfc2862SAkinobu Mita } 12762bfc2862SAkinobu Mita 1277b575454fSNicholas Piggin phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size, 12782bfc2862SAkinobu Mita phys_addr_t align, phys_addr_t max_addr, 1279e1720feeSMike Rapoport int nid, enum memblock_flags flags) 12802bfc2862SAkinobu Mita { 1281fc6daaf9STony Luck return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags); 12822bfc2862SAkinobu Mita } 12832bfc2862SAkinobu Mita 12849a8dd708SMike Rapoport phys_addr_t __init memblock_phys_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) 12857bd0b0f0STejun Heo { 1286e1720feeSMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 1287a3f5bafcSTony Luck phys_addr_t ret; 1288a3f5bafcSTony Luck 1289a3f5bafcSTony Luck again: 1290a3f5bafcSTony Luck ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, 1291a3f5bafcSTony Luck nid, flags); 1292a3f5bafcSTony Luck 1293a3f5bafcSTony Luck if (!ret && (flags & MEMBLOCK_MIRROR)) { 1294a3f5bafcSTony Luck flags &= ~MEMBLOCK_MIRROR; 1295a3f5bafcSTony Luck goto again; 1296a3f5bafcSTony Luck } 1297a3f5bafcSTony Luck return ret; 12987bd0b0f0STejun Heo } 12997bd0b0f0STejun Heo 13007bd0b0f0STejun Heo phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) 13017bd0b0f0STejun Heo { 1302fc6daaf9STony Luck return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE, 1303fc6daaf9STony Luck MEMBLOCK_NONE); 13047bd0b0f0STejun Heo } 13057bd0b0f0STejun Heo 13066ed311b2SBenjamin Herrenschmidt phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) 130795f72d1eSYinghai Lu { 13086ed311b2SBenjamin Herrenschmidt phys_addr_t alloc; 13096ed311b2SBenjamin Herrenschmidt 13106ed311b2SBenjamin Herrenschmidt alloc = __memblock_alloc_base(size, align, max_addr); 13116ed311b2SBenjamin Herrenschmidt 13126ed311b2SBenjamin Herrenschmidt if (alloc == 0) 13135d63f81cSMiles Chen panic("ERROR: Failed to allocate %pa bytes below %pa.\n", 13145d63f81cSMiles Chen &size, &max_addr); 13156ed311b2SBenjamin Herrenschmidt 13166ed311b2SBenjamin Herrenschmidt return alloc; 131795f72d1eSYinghai Lu } 131895f72d1eSYinghai Lu 13199a8dd708SMike Rapoport phys_addr_t __init memblock_phys_alloc(phys_addr_t size, phys_addr_t align) 132095f72d1eSYinghai Lu { 13216ed311b2SBenjamin Herrenschmidt return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); 132295f72d1eSYinghai Lu } 132395f72d1eSYinghai Lu 13249a8dd708SMike Rapoport phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 13259d1e2492SBenjamin Herrenschmidt { 13269a8dd708SMike Rapoport phys_addr_t res = memblock_phys_alloc_nid(size, align, nid); 13279d1e2492SBenjamin Herrenschmidt 13289d1e2492SBenjamin Herrenschmidt if (res) 13299d1e2492SBenjamin Herrenschmidt return res; 133015fb0972STejun Heo return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); 133195f72d1eSYinghai Lu } 133295f72d1eSYinghai Lu 133326f09e9bSSantosh Shilimkar /** 1334eb31d559SMike Rapoport * memblock_alloc_internal - allocate boot memory block 133526f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 133626f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 133726f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region to allocate (phys address) 133826f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region to allocate (phys address) 133926f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 134026f09e9bSSantosh Shilimkar * 134126f09e9bSSantosh Shilimkar * The @min_addr limit is dropped if it can not be satisfied and the allocation 134226f09e9bSSantosh Shilimkar * will fall back to memory below @min_addr. Also, allocation may fall back 134326f09e9bSSantosh Shilimkar * to any node in the system if the specified node can not 134426f09e9bSSantosh Shilimkar * hold the requested memory. 134526f09e9bSSantosh Shilimkar * 134626f09e9bSSantosh Shilimkar * The allocation is performed from memory region limited by 134797ad1087SMike Rapoport * memblock.current_limit if @max_addr == %MEMBLOCK_ALLOC_ACCESSIBLE. 134826f09e9bSSantosh Shilimkar * 134926f09e9bSSantosh Shilimkar * The phys address of allocated boot memory block is converted to virtual and 135026f09e9bSSantosh Shilimkar * allocated memory is reset to 0. 135126f09e9bSSantosh Shilimkar * 135226f09e9bSSantosh Shilimkar * In addition, function sets the min_count to 0 using kmemleak_alloc for 135326f09e9bSSantosh Shilimkar * allocated boot memory block, so that it is never reported as leaks. 135426f09e9bSSantosh Shilimkar * 135547cec443SMike Rapoport * Return: 135626f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 135726f09e9bSSantosh Shilimkar */ 1358eb31d559SMike Rapoport static void * __init memblock_alloc_internal( 135926f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 136026f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 136126f09e9bSSantosh Shilimkar int nid) 136226f09e9bSSantosh Shilimkar { 136326f09e9bSSantosh Shilimkar phys_addr_t alloc; 136426f09e9bSSantosh Shilimkar void *ptr; 1365e1720feeSMike Rapoport enum memblock_flags flags = choose_memblock_flags(); 136626f09e9bSSantosh Shilimkar 1367560dca27SGrygorii Strashko if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1368560dca27SGrygorii Strashko nid = NUMA_NO_NODE; 136926f09e9bSSantosh Shilimkar 137026f09e9bSSantosh Shilimkar /* 137126f09e9bSSantosh Shilimkar * Detect any accidental use of these APIs after slab is ready, as at 137226f09e9bSSantosh Shilimkar * this moment memblock may be deinitialized already and its 1373c6ffc5caSMike Rapoport * internal data may be destroyed (after execution of memblock_free_all) 137426f09e9bSSantosh Shilimkar */ 137526f09e9bSSantosh Shilimkar if (WARN_ON_ONCE(slab_is_available())) 137626f09e9bSSantosh Shilimkar return kzalloc_node(size, GFP_NOWAIT, nid); 137726f09e9bSSantosh Shilimkar 1378*2f770806SMike Rapoport if (!align) { 1379*2f770806SMike Rapoport dump_stack(); 1380*2f770806SMike Rapoport align = SMP_CACHE_BYTES; 1381*2f770806SMike Rapoport } 1382*2f770806SMike Rapoport 1383f544e14fSYinghai Lu if (max_addr > memblock.current_limit) 1384f544e14fSYinghai Lu max_addr = memblock.current_limit; 138526f09e9bSSantosh Shilimkar again: 138626f09e9bSSantosh Shilimkar alloc = memblock_find_in_range_node(size, align, min_addr, max_addr, 1387a3f5bafcSTony Luck nid, flags); 13887d41c03eSWei Yang if (alloc && !memblock_reserve(alloc, size)) 138926f09e9bSSantosh Shilimkar goto done; 139026f09e9bSSantosh Shilimkar 139126f09e9bSSantosh Shilimkar if (nid != NUMA_NO_NODE) { 139226f09e9bSSantosh Shilimkar alloc = memblock_find_in_range_node(size, align, min_addr, 1393fc6daaf9STony Luck max_addr, NUMA_NO_NODE, 1394a3f5bafcSTony Luck flags); 13957d41c03eSWei Yang if (alloc && !memblock_reserve(alloc, size)) 139626f09e9bSSantosh Shilimkar goto done; 139726f09e9bSSantosh Shilimkar } 139826f09e9bSSantosh Shilimkar 139926f09e9bSSantosh Shilimkar if (min_addr) { 140026f09e9bSSantosh Shilimkar min_addr = 0; 140126f09e9bSSantosh Shilimkar goto again; 140226f09e9bSSantosh Shilimkar } 140326f09e9bSSantosh Shilimkar 1404a3f5bafcSTony Luck if (flags & MEMBLOCK_MIRROR) { 1405a3f5bafcSTony Luck flags &= ~MEMBLOCK_MIRROR; 1406a3f5bafcSTony Luck pr_warn("Could not allocate %pap bytes of mirrored memory\n", 1407a3f5bafcSTony Luck &size); 1408a3f5bafcSTony Luck goto again; 1409a3f5bafcSTony Luck } 1410a3f5bafcSTony Luck 1411a3f5bafcSTony Luck return NULL; 141226f09e9bSSantosh Shilimkar done: 141326f09e9bSSantosh Shilimkar ptr = phys_to_virt(alloc); 141426f09e9bSSantosh Shilimkar 141526f09e9bSSantosh Shilimkar /* 141626f09e9bSSantosh Shilimkar * The min_count is set to 0 so that bootmem allocated blocks 141726f09e9bSSantosh Shilimkar * are never reported as leaks. This is because many of these blocks 141826f09e9bSSantosh Shilimkar * are only referred via the physical address which is not 141926f09e9bSSantosh Shilimkar * looked up by kmemleak. 142026f09e9bSSantosh Shilimkar */ 142126f09e9bSSantosh Shilimkar kmemleak_alloc(ptr, size, 0, 0); 142226f09e9bSSantosh Shilimkar 142326f09e9bSSantosh Shilimkar return ptr; 142426f09e9bSSantosh Shilimkar } 142526f09e9bSSantosh Shilimkar 142626f09e9bSSantosh Shilimkar /** 1427eb31d559SMike Rapoport * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1428ea1f5f37SPavel Tatashin * memory and without panicking 1429ea1f5f37SPavel Tatashin * @size: size of memory block to be allocated in bytes 1430ea1f5f37SPavel Tatashin * @align: alignment of the region and block's size 1431ea1f5f37SPavel Tatashin * @min_addr: the lower bound of the memory region from where the allocation 1432ea1f5f37SPavel Tatashin * is preferred (phys address) 1433ea1f5f37SPavel Tatashin * @max_addr: the upper bound of the memory region from where the allocation 143497ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1435ea1f5f37SPavel Tatashin * allocate only from memory limited by memblock.current_limit value 1436ea1f5f37SPavel Tatashin * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1437ea1f5f37SPavel Tatashin * 1438ea1f5f37SPavel Tatashin * Public function, provides additional debug information (including caller 1439ea1f5f37SPavel Tatashin * info), if enabled. Does not zero allocated memory, does not panic if request 1440ea1f5f37SPavel Tatashin * cannot be satisfied. 1441ea1f5f37SPavel Tatashin * 144247cec443SMike Rapoport * Return: 1443ea1f5f37SPavel Tatashin * Virtual address of allocated memory block on success, NULL on failure. 1444ea1f5f37SPavel Tatashin */ 1445eb31d559SMike Rapoport void * __init memblock_alloc_try_nid_raw( 1446ea1f5f37SPavel Tatashin phys_addr_t size, phys_addr_t align, 1447ea1f5f37SPavel Tatashin phys_addr_t min_addr, phys_addr_t max_addr, 1448ea1f5f37SPavel Tatashin int nid) 1449ea1f5f37SPavel Tatashin { 1450ea1f5f37SPavel Tatashin void *ptr; 1451ea1f5f37SPavel Tatashin 1452a36aab89SMike Rapoport memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n", 1453a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1454a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1455ea1f5f37SPavel Tatashin 1456eb31d559SMike Rapoport ptr = memblock_alloc_internal(size, align, 1457ea1f5f37SPavel Tatashin min_addr, max_addr, nid); 1458ea1f5f37SPavel Tatashin if (ptr && size > 0) 1459f682a97aSAlexander Duyck page_init_poison(ptr, size); 1460f682a97aSAlexander Duyck 1461ea1f5f37SPavel Tatashin return ptr; 1462ea1f5f37SPavel Tatashin } 1463ea1f5f37SPavel Tatashin 1464ea1f5f37SPavel Tatashin /** 1465eb31d559SMike Rapoport * memblock_alloc_try_nid_nopanic - allocate boot memory block 146626f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 146726f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 146826f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region from where the allocation 146926f09e9bSSantosh Shilimkar * is preferred (phys address) 147026f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region from where the allocation 147197ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 147226f09e9bSSantosh Shilimkar * allocate only from memory limited by memblock.current_limit value 147326f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 147426f09e9bSSantosh Shilimkar * 1475ea1f5f37SPavel Tatashin * Public function, provides additional debug information (including caller 1476ea1f5f37SPavel Tatashin * info), if enabled. This function zeroes the allocated memory. 147726f09e9bSSantosh Shilimkar * 147847cec443SMike Rapoport * Return: 147926f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 148026f09e9bSSantosh Shilimkar */ 1481eb31d559SMike Rapoport void * __init memblock_alloc_try_nid_nopanic( 148226f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 148326f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 148426f09e9bSSantosh Shilimkar int nid) 148526f09e9bSSantosh Shilimkar { 1486ea1f5f37SPavel Tatashin void *ptr; 1487ea1f5f37SPavel Tatashin 1488a36aab89SMike Rapoport memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n", 1489a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1490a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1491ea1f5f37SPavel Tatashin 1492eb31d559SMike Rapoport ptr = memblock_alloc_internal(size, align, 1493ea1f5f37SPavel Tatashin min_addr, max_addr, nid); 1494ea1f5f37SPavel Tatashin if (ptr) 1495ea1f5f37SPavel Tatashin memset(ptr, 0, size); 1496ea1f5f37SPavel Tatashin return ptr; 149726f09e9bSSantosh Shilimkar } 149826f09e9bSSantosh Shilimkar 149926f09e9bSSantosh Shilimkar /** 1500eb31d559SMike Rapoport * memblock_alloc_try_nid - allocate boot memory block with panicking 150126f09e9bSSantosh Shilimkar * @size: size of memory block to be allocated in bytes 150226f09e9bSSantosh Shilimkar * @align: alignment of the region and block's size 150326f09e9bSSantosh Shilimkar * @min_addr: the lower bound of the memory region from where the allocation 150426f09e9bSSantosh Shilimkar * is preferred (phys address) 150526f09e9bSSantosh Shilimkar * @max_addr: the upper bound of the memory region from where the allocation 150697ad1087SMike Rapoport * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 150726f09e9bSSantosh Shilimkar * allocate only from memory limited by memblock.current_limit value 150826f09e9bSSantosh Shilimkar * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 150926f09e9bSSantosh Shilimkar * 1510eb31d559SMike Rapoport * Public panicking version of memblock_alloc_try_nid_nopanic() 151126f09e9bSSantosh Shilimkar * which provides debug information (including caller info), if enabled, 151226f09e9bSSantosh Shilimkar * and panics if the request can not be satisfied. 151326f09e9bSSantosh Shilimkar * 151447cec443SMike Rapoport * Return: 151526f09e9bSSantosh Shilimkar * Virtual address of allocated memory block on success, NULL on failure. 151626f09e9bSSantosh Shilimkar */ 1517eb31d559SMike Rapoport void * __init memblock_alloc_try_nid( 151826f09e9bSSantosh Shilimkar phys_addr_t size, phys_addr_t align, 151926f09e9bSSantosh Shilimkar phys_addr_t min_addr, phys_addr_t max_addr, 152026f09e9bSSantosh Shilimkar int nid) 152126f09e9bSSantosh Shilimkar { 152226f09e9bSSantosh Shilimkar void *ptr; 152326f09e9bSSantosh Shilimkar 1524a36aab89SMike Rapoport memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pF\n", 1525a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, 1526a36aab89SMike Rapoport &max_addr, (void *)_RET_IP_); 1527eb31d559SMike Rapoport ptr = memblock_alloc_internal(size, align, 152826f09e9bSSantosh Shilimkar min_addr, max_addr, nid); 1529ea1f5f37SPavel Tatashin if (ptr) { 1530ea1f5f37SPavel Tatashin memset(ptr, 0, size); 153126f09e9bSSantosh Shilimkar return ptr; 1532ea1f5f37SPavel Tatashin } 153326f09e9bSSantosh Shilimkar 1534a36aab89SMike Rapoport panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa\n", 1535a36aab89SMike Rapoport __func__, (u64)size, (u64)align, nid, &min_addr, &max_addr); 153626f09e9bSSantosh Shilimkar return NULL; 153726f09e9bSSantosh Shilimkar } 153826f09e9bSSantosh Shilimkar 153926f09e9bSSantosh Shilimkar /** 154026f09e9bSSantosh Shilimkar * __memblock_free_early - free boot memory block 154126f09e9bSSantosh Shilimkar * @base: phys starting address of the boot memory block 154226f09e9bSSantosh Shilimkar * @size: size of the boot memory block in bytes 154326f09e9bSSantosh Shilimkar * 1544eb31d559SMike Rapoport * Free boot memory block previously allocated by memblock_alloc_xx() API. 154526f09e9bSSantosh Shilimkar * The freeing memory will not be released to the buddy allocator. 154626f09e9bSSantosh Shilimkar */ 154726f09e9bSSantosh Shilimkar void __init __memblock_free_early(phys_addr_t base, phys_addr_t size) 154826f09e9bSSantosh Shilimkar { 1549a36aab89SMike Rapoport phys_addr_t end = base + size - 1; 1550a36aab89SMike Rapoport 1551a36aab89SMike Rapoport memblock_dbg("%s: [%pa-%pa] %pF\n", 1552a36aab89SMike Rapoport __func__, &base, &end, (void *)_RET_IP_); 15539099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 1554f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.reserved, base, size); 155526f09e9bSSantosh Shilimkar } 155626f09e9bSSantosh Shilimkar 155748a833ccSMike Rapoport /** 155826f09e9bSSantosh Shilimkar * __memblock_free_late - free bootmem block pages directly to buddy allocator 155948a833ccSMike Rapoport * @base: phys starting address of the boot memory block 156026f09e9bSSantosh Shilimkar * @size: size of the boot memory block in bytes 156126f09e9bSSantosh Shilimkar * 156226f09e9bSSantosh Shilimkar * This is only useful when the bootmem allocator has already been torn 156326f09e9bSSantosh Shilimkar * down, but we are still initializing the system. Pages are released directly 156426f09e9bSSantosh Shilimkar * to the buddy allocator, no bootmem metadata is updated because it is gone. 156526f09e9bSSantosh Shilimkar */ 156626f09e9bSSantosh Shilimkar void __init __memblock_free_late(phys_addr_t base, phys_addr_t size) 156726f09e9bSSantosh Shilimkar { 1568a36aab89SMike Rapoport phys_addr_t cursor, end; 156926f09e9bSSantosh Shilimkar 1570a36aab89SMike Rapoport end = base + size - 1; 1571a36aab89SMike Rapoport memblock_dbg("%s: [%pa-%pa] %pF\n", 1572a36aab89SMike Rapoport __func__, &base, &end, (void *)_RET_IP_); 15739099daedSCatalin Marinas kmemleak_free_part_phys(base, size); 157426f09e9bSSantosh Shilimkar cursor = PFN_UP(base); 157526f09e9bSSantosh Shilimkar end = PFN_DOWN(base + size); 157626f09e9bSSantosh Shilimkar 157726f09e9bSSantosh Shilimkar for (; cursor < end; cursor++) { 15787c2ee349SMike Rapoport memblock_free_pages(pfn_to_page(cursor), cursor, 0); 157926f09e9bSSantosh Shilimkar totalram_pages++; 158026f09e9bSSantosh Shilimkar } 158126f09e9bSSantosh Shilimkar } 15829d1e2492SBenjamin Herrenschmidt 15839d1e2492SBenjamin Herrenschmidt /* 15849d1e2492SBenjamin Herrenschmidt * Remaining API functions 15859d1e2492SBenjamin Herrenschmidt */ 15869d1e2492SBenjamin Herrenschmidt 15871f1ffb8aSDavid Gibson phys_addr_t __init_memblock memblock_phys_mem_size(void) 158895f72d1eSYinghai Lu { 15891440c4e2STejun Heo return memblock.memory.total_size; 159095f72d1eSYinghai Lu } 159195f72d1eSYinghai Lu 15928907de5dSSrikar Dronamraju phys_addr_t __init_memblock memblock_reserved_size(void) 15938907de5dSSrikar Dronamraju { 15948907de5dSSrikar Dronamraju return memblock.reserved.total_size; 15958907de5dSSrikar Dronamraju } 15968907de5dSSrikar Dronamraju 1597595ad9afSYinghai Lu phys_addr_t __init memblock_mem_size(unsigned long limit_pfn) 1598595ad9afSYinghai Lu { 1599595ad9afSYinghai Lu unsigned long pages = 0; 1600595ad9afSYinghai Lu struct memblock_region *r; 1601595ad9afSYinghai Lu unsigned long start_pfn, end_pfn; 1602595ad9afSYinghai Lu 1603595ad9afSYinghai Lu for_each_memblock(memory, r) { 1604595ad9afSYinghai Lu start_pfn = memblock_region_memory_base_pfn(r); 1605595ad9afSYinghai Lu end_pfn = memblock_region_memory_end_pfn(r); 1606595ad9afSYinghai Lu start_pfn = min_t(unsigned long, start_pfn, limit_pfn); 1607595ad9afSYinghai Lu end_pfn = min_t(unsigned long, end_pfn, limit_pfn); 1608595ad9afSYinghai Lu pages += end_pfn - start_pfn; 1609595ad9afSYinghai Lu } 1610595ad9afSYinghai Lu 161116763230SFabian Frederick return PFN_PHYS(pages); 1612595ad9afSYinghai Lu } 1613595ad9afSYinghai Lu 16140a93ebefSSam Ravnborg /* lowest address */ 16150a93ebefSSam Ravnborg phys_addr_t __init_memblock memblock_start_of_DRAM(void) 16160a93ebefSSam Ravnborg { 16170a93ebefSSam Ravnborg return memblock.memory.regions[0].base; 16180a93ebefSSam Ravnborg } 16190a93ebefSSam Ravnborg 162010d06439SYinghai Lu phys_addr_t __init_memblock memblock_end_of_DRAM(void) 162195f72d1eSYinghai Lu { 162295f72d1eSYinghai Lu int idx = memblock.memory.cnt - 1; 162395f72d1eSYinghai Lu 1624e3239ff9SBenjamin Herrenschmidt return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 162595f72d1eSYinghai Lu } 162695f72d1eSYinghai Lu 1627a571d4ebSDennis Chen static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 162895f72d1eSYinghai Lu { 16291c4bc43dSStefan Agner phys_addr_t max_addr = PHYS_ADDR_MAX; 1630136199f0SEmil Medve struct memblock_region *r; 163195f72d1eSYinghai Lu 1632a571d4ebSDennis Chen /* 1633a571d4ebSDennis Chen * translate the memory @limit size into the max address within one of 1634a571d4ebSDennis Chen * the memory memblock regions, if the @limit exceeds the total size 16351c4bc43dSStefan Agner * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1636a571d4ebSDennis Chen */ 1637136199f0SEmil Medve for_each_memblock(memory, r) { 1638c0ce8fefSTejun Heo if (limit <= r->size) { 1639c0ce8fefSTejun Heo max_addr = r->base + limit; 164095f72d1eSYinghai Lu break; 164195f72d1eSYinghai Lu } 1642c0ce8fefSTejun Heo limit -= r->size; 164395f72d1eSYinghai Lu } 1644c0ce8fefSTejun Heo 1645a571d4ebSDennis Chen return max_addr; 1646a571d4ebSDennis Chen } 1647a571d4ebSDennis Chen 1648a571d4ebSDennis Chen void __init memblock_enforce_memory_limit(phys_addr_t limit) 1649a571d4ebSDennis Chen { 16501c4bc43dSStefan Agner phys_addr_t max_addr = PHYS_ADDR_MAX; 1651a571d4ebSDennis Chen 1652a571d4ebSDennis Chen if (!limit) 1653a571d4ebSDennis Chen return; 1654a571d4ebSDennis Chen 1655a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1656a571d4ebSDennis Chen 1657a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 16581c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1659a571d4ebSDennis Chen return; 1660a571d4ebSDennis Chen 1661c0ce8fefSTejun Heo /* truncate both memory and reserved regions */ 1662f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.memory, max_addr, 16631c4bc43dSStefan Agner PHYS_ADDR_MAX); 1664f1af9d3aSPhilipp Hachtmann memblock_remove_range(&memblock.reserved, max_addr, 16651c4bc43dSStefan Agner PHYS_ADDR_MAX); 166695f72d1eSYinghai Lu } 166795f72d1eSYinghai Lu 1668c9ca9b4eSAKASHI Takahiro void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1669c9ca9b4eSAKASHI Takahiro { 1670c9ca9b4eSAKASHI Takahiro int start_rgn, end_rgn; 1671c9ca9b4eSAKASHI Takahiro int i, ret; 1672c9ca9b4eSAKASHI Takahiro 1673c9ca9b4eSAKASHI Takahiro if (!size) 1674c9ca9b4eSAKASHI Takahiro return; 1675c9ca9b4eSAKASHI Takahiro 1676c9ca9b4eSAKASHI Takahiro ret = memblock_isolate_range(&memblock.memory, base, size, 1677c9ca9b4eSAKASHI Takahiro &start_rgn, &end_rgn); 1678c9ca9b4eSAKASHI Takahiro if (ret) 1679c9ca9b4eSAKASHI Takahiro return; 1680c9ca9b4eSAKASHI Takahiro 1681c9ca9b4eSAKASHI Takahiro /* remove all the MAP regions */ 1682c9ca9b4eSAKASHI Takahiro for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1683c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1684c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1685c9ca9b4eSAKASHI Takahiro 1686c9ca9b4eSAKASHI Takahiro for (i = start_rgn - 1; i >= 0; i--) 1687c9ca9b4eSAKASHI Takahiro if (!memblock_is_nomap(&memblock.memory.regions[i])) 1688c9ca9b4eSAKASHI Takahiro memblock_remove_region(&memblock.memory, i); 1689c9ca9b4eSAKASHI Takahiro 1690c9ca9b4eSAKASHI Takahiro /* truncate the reserved regions */ 1691c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 0, base); 1692c9ca9b4eSAKASHI Takahiro memblock_remove_range(&memblock.reserved, 16931c4bc43dSStefan Agner base + size, PHYS_ADDR_MAX); 1694c9ca9b4eSAKASHI Takahiro } 1695c9ca9b4eSAKASHI Takahiro 1696a571d4ebSDennis Chen void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1697a571d4ebSDennis Chen { 1698a571d4ebSDennis Chen phys_addr_t max_addr; 1699a571d4ebSDennis Chen 1700a571d4ebSDennis Chen if (!limit) 1701a571d4ebSDennis Chen return; 1702a571d4ebSDennis Chen 1703a571d4ebSDennis Chen max_addr = __find_max_addr(limit); 1704a571d4ebSDennis Chen 1705a571d4ebSDennis Chen /* @limit exceeds the total size of the memory, do nothing */ 17061c4bc43dSStefan Agner if (max_addr == PHYS_ADDR_MAX) 1707a571d4ebSDennis Chen return; 1708a571d4ebSDennis Chen 1709c9ca9b4eSAKASHI Takahiro memblock_cap_memory_range(0, max_addr); 1710a571d4ebSDennis Chen } 1711a571d4ebSDennis Chen 1712cd79481dSYinghai Lu static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 171372d4b0b4SBenjamin Herrenschmidt { 171472d4b0b4SBenjamin Herrenschmidt unsigned int left = 0, right = type->cnt; 171572d4b0b4SBenjamin Herrenschmidt 171672d4b0b4SBenjamin Herrenschmidt do { 171772d4b0b4SBenjamin Herrenschmidt unsigned int mid = (right + left) / 2; 171872d4b0b4SBenjamin Herrenschmidt 171972d4b0b4SBenjamin Herrenschmidt if (addr < type->regions[mid].base) 172072d4b0b4SBenjamin Herrenschmidt right = mid; 172172d4b0b4SBenjamin Herrenschmidt else if (addr >= (type->regions[mid].base + 172272d4b0b4SBenjamin Herrenschmidt type->regions[mid].size)) 172372d4b0b4SBenjamin Herrenschmidt left = mid + 1; 172472d4b0b4SBenjamin Herrenschmidt else 172572d4b0b4SBenjamin Herrenschmidt return mid; 172672d4b0b4SBenjamin Herrenschmidt } while (left < right); 172772d4b0b4SBenjamin Herrenschmidt return -1; 172872d4b0b4SBenjamin Herrenschmidt } 172972d4b0b4SBenjamin Herrenschmidt 1730b4ad0c7eSYaowei Bai bool __init memblock_is_reserved(phys_addr_t addr) 173195f72d1eSYinghai Lu { 173272d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.reserved, addr) != -1; 173395f72d1eSYinghai Lu } 173472d4b0b4SBenjamin Herrenschmidt 1735b4ad0c7eSYaowei Bai bool __init_memblock memblock_is_memory(phys_addr_t addr) 173672d4b0b4SBenjamin Herrenschmidt { 173772d4b0b4SBenjamin Herrenschmidt return memblock_search(&memblock.memory, addr) != -1; 173872d4b0b4SBenjamin Herrenschmidt } 173972d4b0b4SBenjamin Herrenschmidt 1740937f0c26SYaowei Bai bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1741bf3d3cc5SArd Biesheuvel { 1742bf3d3cc5SArd Biesheuvel int i = memblock_search(&memblock.memory, addr); 1743bf3d3cc5SArd Biesheuvel 1744bf3d3cc5SArd Biesheuvel if (i == -1) 1745bf3d3cc5SArd Biesheuvel return false; 1746bf3d3cc5SArd Biesheuvel return !memblock_is_nomap(&memblock.memory.regions[i]); 1747bf3d3cc5SArd Biesheuvel } 1748bf3d3cc5SArd Biesheuvel 1749e76b63f8SYinghai Lu #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 1750e76b63f8SYinghai Lu int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1751e76b63f8SYinghai Lu unsigned long *start_pfn, unsigned long *end_pfn) 1752e76b63f8SYinghai Lu { 1753e76b63f8SYinghai Lu struct memblock_type *type = &memblock.memory; 175416763230SFabian Frederick int mid = memblock_search(type, PFN_PHYS(pfn)); 1755e76b63f8SYinghai Lu 1756e76b63f8SYinghai Lu if (mid == -1) 1757e76b63f8SYinghai Lu return -1; 1758e76b63f8SYinghai Lu 1759f7e2f7e8SFabian Frederick *start_pfn = PFN_DOWN(type->regions[mid].base); 1760f7e2f7e8SFabian Frederick *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1761e76b63f8SYinghai Lu 1762e76b63f8SYinghai Lu return type->regions[mid].nid; 1763e76b63f8SYinghai Lu } 1764e76b63f8SYinghai Lu #endif 1765e76b63f8SYinghai Lu 1766eab30949SStephen Boyd /** 1767eab30949SStephen Boyd * memblock_is_region_memory - check if a region is a subset of memory 1768eab30949SStephen Boyd * @base: base of region to check 1769eab30949SStephen Boyd * @size: size of region to check 1770eab30949SStephen Boyd * 1771eab30949SStephen Boyd * Check if the region [@base, @base + @size) is a subset of a memory block. 1772eab30949SStephen Boyd * 177347cec443SMike Rapoport * Return: 1774eab30949SStephen Boyd * 0 if false, non-zero if true 1775eab30949SStephen Boyd */ 1776937f0c26SYaowei Bai bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 177772d4b0b4SBenjamin Herrenschmidt { 1778abb65272STomi Valkeinen int idx = memblock_search(&memblock.memory, base); 1779eb18f1b5STejun Heo phys_addr_t end = base + memblock_cap_size(base, &size); 178072d4b0b4SBenjamin Herrenschmidt 178172d4b0b4SBenjamin Herrenschmidt if (idx == -1) 1782937f0c26SYaowei Bai return false; 1783ef415ef4SWei Yang return (memblock.memory.regions[idx].base + 1784eb18f1b5STejun Heo memblock.memory.regions[idx].size) >= end; 178595f72d1eSYinghai Lu } 178695f72d1eSYinghai Lu 1787eab30949SStephen Boyd /** 1788eab30949SStephen Boyd * memblock_is_region_reserved - check if a region intersects reserved memory 1789eab30949SStephen Boyd * @base: base of region to check 1790eab30949SStephen Boyd * @size: size of region to check 1791eab30949SStephen Boyd * 179247cec443SMike Rapoport * Check if the region [@base, @base + @size) intersects a reserved 179347cec443SMike Rapoport * memory block. 1794eab30949SStephen Boyd * 179547cec443SMike Rapoport * Return: 1796c5c5c9d1STang Chen * True if they intersect, false if not. 1797eab30949SStephen Boyd */ 1798c5c5c9d1STang Chen bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 179995f72d1eSYinghai Lu { 1800eb18f1b5STejun Heo memblock_cap_size(base, &size); 1801c5c5c9d1STang Chen return memblock_overlaps_region(&memblock.reserved, base, size); 180295f72d1eSYinghai Lu } 180395f72d1eSYinghai Lu 18046ede1fd3SYinghai Lu void __init_memblock memblock_trim_memory(phys_addr_t align) 18056ede1fd3SYinghai Lu { 18066ede1fd3SYinghai Lu phys_addr_t start, end, orig_start, orig_end; 1807136199f0SEmil Medve struct memblock_region *r; 18086ede1fd3SYinghai Lu 1809136199f0SEmil Medve for_each_memblock(memory, r) { 1810136199f0SEmil Medve orig_start = r->base; 1811136199f0SEmil Medve orig_end = r->base + r->size; 18126ede1fd3SYinghai Lu start = round_up(orig_start, align); 18136ede1fd3SYinghai Lu end = round_down(orig_end, align); 18146ede1fd3SYinghai Lu 18156ede1fd3SYinghai Lu if (start == orig_start && end == orig_end) 18166ede1fd3SYinghai Lu continue; 18176ede1fd3SYinghai Lu 18186ede1fd3SYinghai Lu if (start < end) { 1819136199f0SEmil Medve r->base = start; 1820136199f0SEmil Medve r->size = end - start; 18216ede1fd3SYinghai Lu } else { 1822136199f0SEmil Medve memblock_remove_region(&memblock.memory, 1823136199f0SEmil Medve r - memblock.memory.regions); 1824136199f0SEmil Medve r--; 18256ede1fd3SYinghai Lu } 18266ede1fd3SYinghai Lu } 18276ede1fd3SYinghai Lu } 1828e63075a3SBenjamin Herrenschmidt 18293661ca66SYinghai Lu void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1830e63075a3SBenjamin Herrenschmidt { 1831e63075a3SBenjamin Herrenschmidt memblock.current_limit = limit; 1832e63075a3SBenjamin Herrenschmidt } 1833e63075a3SBenjamin Herrenschmidt 1834fec51014SLaura Abbott phys_addr_t __init_memblock memblock_get_current_limit(void) 1835fec51014SLaura Abbott { 1836fec51014SLaura Abbott return memblock.current_limit; 1837fec51014SLaura Abbott } 1838fec51014SLaura Abbott 18390262d9c8SHeiko Carstens static void __init_memblock memblock_dump(struct memblock_type *type) 18406ed311b2SBenjamin Herrenschmidt { 18415d63f81cSMiles Chen phys_addr_t base, end, size; 1842e1720feeSMike Rapoport enum memblock_flags flags; 18438c9c1701SAlexander Kuleshov int idx; 18448c9c1701SAlexander Kuleshov struct memblock_region *rgn; 18456ed311b2SBenjamin Herrenschmidt 18460262d9c8SHeiko Carstens pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 18476ed311b2SBenjamin Herrenschmidt 184866e8b438SGioh Kim for_each_memblock_type(idx, type, rgn) { 18497c0caeb8STejun Heo char nid_buf[32] = ""; 18506ed311b2SBenjamin Herrenschmidt 18517c0caeb8STejun Heo base = rgn->base; 18527c0caeb8STejun Heo size = rgn->size; 18535d63f81cSMiles Chen end = base + size - 1; 185466a20757STang Chen flags = rgn->flags; 18557c0caeb8STejun Heo #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 18567c0caeb8STejun Heo if (memblock_get_region_node(rgn) != MAX_NUMNODES) 18577c0caeb8STejun Heo snprintf(nid_buf, sizeof(nid_buf), " on node %d", 18587c0caeb8STejun Heo memblock_get_region_node(rgn)); 18597c0caeb8STejun Heo #endif 1860e1720feeSMike Rapoport pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 18610262d9c8SHeiko Carstens type->name, idx, &base, &end, &size, nid_buf, flags); 18626ed311b2SBenjamin Herrenschmidt } 18636ed311b2SBenjamin Herrenschmidt } 18646ed311b2SBenjamin Herrenschmidt 18654ff7b82fSTejun Heo void __init_memblock __memblock_dump_all(void) 18666ed311b2SBenjamin Herrenschmidt { 18676ed311b2SBenjamin Herrenschmidt pr_info("MEMBLOCK configuration:\n"); 18685d63f81cSMiles Chen pr_info(" memory size = %pa reserved size = %pa\n", 18695d63f81cSMiles Chen &memblock.memory.total_size, 18705d63f81cSMiles Chen &memblock.reserved.total_size); 18716ed311b2SBenjamin Herrenschmidt 18720262d9c8SHeiko Carstens memblock_dump(&memblock.memory); 18730262d9c8SHeiko Carstens memblock_dump(&memblock.reserved); 1874409efd4cSHeiko Carstens #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 18750262d9c8SHeiko Carstens memblock_dump(&memblock.physmem); 1876409efd4cSHeiko Carstens #endif 18776ed311b2SBenjamin Herrenschmidt } 18786ed311b2SBenjamin Herrenschmidt 18791aadc056STejun Heo void __init memblock_allow_resize(void) 18806ed311b2SBenjamin Herrenschmidt { 1881142b45a7SBenjamin Herrenschmidt memblock_can_resize = 1; 18826ed311b2SBenjamin Herrenschmidt } 18836ed311b2SBenjamin Herrenschmidt 18846ed311b2SBenjamin Herrenschmidt static int __init early_memblock(char *p) 18856ed311b2SBenjamin Herrenschmidt { 18866ed311b2SBenjamin Herrenschmidt if (p && strstr(p, "debug")) 18876ed311b2SBenjamin Herrenschmidt memblock_debug = 1; 18886ed311b2SBenjamin Herrenschmidt return 0; 18896ed311b2SBenjamin Herrenschmidt } 18906ed311b2SBenjamin Herrenschmidt early_param("memblock", early_memblock); 18916ed311b2SBenjamin Herrenschmidt 1892bda49a81SMike Rapoport static void __init __free_pages_memory(unsigned long start, unsigned long end) 1893bda49a81SMike Rapoport { 1894bda49a81SMike Rapoport int order; 1895bda49a81SMike Rapoport 1896bda49a81SMike Rapoport while (start < end) { 1897bda49a81SMike Rapoport order = min(MAX_ORDER - 1UL, __ffs(start)); 1898bda49a81SMike Rapoport 1899bda49a81SMike Rapoport while (start + (1UL << order) > end) 1900bda49a81SMike Rapoport order--; 1901bda49a81SMike Rapoport 1902bda49a81SMike Rapoport memblock_free_pages(pfn_to_page(start), start, order); 1903bda49a81SMike Rapoport 1904bda49a81SMike Rapoport start += (1UL << order); 1905bda49a81SMike Rapoport } 1906bda49a81SMike Rapoport } 1907bda49a81SMike Rapoport 1908bda49a81SMike Rapoport static unsigned long __init __free_memory_core(phys_addr_t start, 1909bda49a81SMike Rapoport phys_addr_t end) 1910bda49a81SMike Rapoport { 1911bda49a81SMike Rapoport unsigned long start_pfn = PFN_UP(start); 1912bda49a81SMike Rapoport unsigned long end_pfn = min_t(unsigned long, 1913bda49a81SMike Rapoport PFN_DOWN(end), max_low_pfn); 1914bda49a81SMike Rapoport 1915bda49a81SMike Rapoport if (start_pfn >= end_pfn) 1916bda49a81SMike Rapoport return 0; 1917bda49a81SMike Rapoport 1918bda49a81SMike Rapoport __free_pages_memory(start_pfn, end_pfn); 1919bda49a81SMike Rapoport 1920bda49a81SMike Rapoport return end_pfn - start_pfn; 1921bda49a81SMike Rapoport } 1922bda49a81SMike Rapoport 1923bda49a81SMike Rapoport static unsigned long __init free_low_memory_core_early(void) 1924bda49a81SMike Rapoport { 1925bda49a81SMike Rapoport unsigned long count = 0; 1926bda49a81SMike Rapoport phys_addr_t start, end; 1927bda49a81SMike Rapoport u64 i; 1928bda49a81SMike Rapoport 1929bda49a81SMike Rapoport memblock_clear_hotplug(0, -1); 1930bda49a81SMike Rapoport 1931bda49a81SMike Rapoport for_each_reserved_mem_region(i, &start, &end) 1932bda49a81SMike Rapoport reserve_bootmem_region(start, end); 1933bda49a81SMike Rapoport 1934bda49a81SMike Rapoport /* 1935bda49a81SMike Rapoport * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 1936bda49a81SMike Rapoport * because in some case like Node0 doesn't have RAM installed 1937bda49a81SMike Rapoport * low ram will be on Node1 1938bda49a81SMike Rapoport */ 1939bda49a81SMike Rapoport for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 1940bda49a81SMike Rapoport NULL) 1941bda49a81SMike Rapoport count += __free_memory_core(start, end); 1942bda49a81SMike Rapoport 1943bda49a81SMike Rapoport return count; 1944bda49a81SMike Rapoport } 1945bda49a81SMike Rapoport 1946bda49a81SMike Rapoport static int reset_managed_pages_done __initdata; 1947bda49a81SMike Rapoport 1948bda49a81SMike Rapoport void reset_node_managed_pages(pg_data_t *pgdat) 1949bda49a81SMike Rapoport { 1950bda49a81SMike Rapoport struct zone *z; 1951bda49a81SMike Rapoport 1952bda49a81SMike Rapoport for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 1953bda49a81SMike Rapoport z->managed_pages = 0; 1954bda49a81SMike Rapoport } 1955bda49a81SMike Rapoport 1956bda49a81SMike Rapoport void __init reset_all_zones_managed_pages(void) 1957bda49a81SMike Rapoport { 1958bda49a81SMike Rapoport struct pglist_data *pgdat; 1959bda49a81SMike Rapoport 1960bda49a81SMike Rapoport if (reset_managed_pages_done) 1961bda49a81SMike Rapoport return; 1962bda49a81SMike Rapoport 1963bda49a81SMike Rapoport for_each_online_pgdat(pgdat) 1964bda49a81SMike Rapoport reset_node_managed_pages(pgdat); 1965bda49a81SMike Rapoport 1966bda49a81SMike Rapoport reset_managed_pages_done = 1; 1967bda49a81SMike Rapoport } 1968bda49a81SMike Rapoport 1969bda49a81SMike Rapoport /** 1970bda49a81SMike Rapoport * memblock_free_all - release free pages to the buddy allocator 1971bda49a81SMike Rapoport * 1972bda49a81SMike Rapoport * Return: the number of pages actually released. 1973bda49a81SMike Rapoport */ 1974bda49a81SMike Rapoport unsigned long __init memblock_free_all(void) 1975bda49a81SMike Rapoport { 1976bda49a81SMike Rapoport unsigned long pages; 1977bda49a81SMike Rapoport 1978bda49a81SMike Rapoport reset_all_zones_managed_pages(); 1979bda49a81SMike Rapoport 1980bda49a81SMike Rapoport pages = free_low_memory_core_early(); 1981bda49a81SMike Rapoport totalram_pages += pages; 1982bda49a81SMike Rapoport 1983bda49a81SMike Rapoport return pages; 1984bda49a81SMike Rapoport } 1985bda49a81SMike Rapoport 1986c378ddd5STejun Heo #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) 19876d03b885SBenjamin Herrenschmidt 19886d03b885SBenjamin Herrenschmidt static int memblock_debug_show(struct seq_file *m, void *private) 19896d03b885SBenjamin Herrenschmidt { 19906d03b885SBenjamin Herrenschmidt struct memblock_type *type = m->private; 19916d03b885SBenjamin Herrenschmidt struct memblock_region *reg; 19926d03b885SBenjamin Herrenschmidt int i; 19935d63f81cSMiles Chen phys_addr_t end; 19946d03b885SBenjamin Herrenschmidt 19956d03b885SBenjamin Herrenschmidt for (i = 0; i < type->cnt; i++) { 19966d03b885SBenjamin Herrenschmidt reg = &type->regions[i]; 19975d63f81cSMiles Chen end = reg->base + reg->size - 1; 19986d03b885SBenjamin Herrenschmidt 19995d63f81cSMiles Chen seq_printf(m, "%4d: ", i); 20005d63f81cSMiles Chen seq_printf(m, "%pa..%pa\n", ®->base, &end); 20016d03b885SBenjamin Herrenschmidt } 20026d03b885SBenjamin Herrenschmidt return 0; 20036d03b885SBenjamin Herrenschmidt } 20045ad35093SAndy Shevchenko DEFINE_SHOW_ATTRIBUTE(memblock_debug); 20056d03b885SBenjamin Herrenschmidt 20066d03b885SBenjamin Herrenschmidt static int __init memblock_init_debugfs(void) 20076d03b885SBenjamin Herrenschmidt { 20086d03b885SBenjamin Herrenschmidt struct dentry *root = debugfs_create_dir("memblock", NULL); 20096d03b885SBenjamin Herrenschmidt if (!root) 20106d03b885SBenjamin Herrenschmidt return -ENXIO; 20110825a6f9SJoe Perches debugfs_create_file("memory", 0444, root, 20120825a6f9SJoe Perches &memblock.memory, &memblock_debug_fops); 20130825a6f9SJoe Perches debugfs_create_file("reserved", 0444, root, 20140825a6f9SJoe Perches &memblock.reserved, &memblock_debug_fops); 201570210ed9SPhilipp Hachtmann #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 20160825a6f9SJoe Perches debugfs_create_file("physmem", 0444, root, 20170825a6f9SJoe Perches &memblock.physmem, &memblock_debug_fops); 201870210ed9SPhilipp Hachtmann #endif 20196d03b885SBenjamin Herrenschmidt 20206d03b885SBenjamin Herrenschmidt return 0; 20216d03b885SBenjamin Herrenschmidt } 20226d03b885SBenjamin Herrenschmidt __initcall(memblock_init_debugfs); 20236d03b885SBenjamin Herrenschmidt 20246d03b885SBenjamin Herrenschmidt #endif /* CONFIG_DEBUG_FS */ 2025