Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1 | /* |
| 2 | * include/linux/topology.h |
| 3 | * |
| 4 | * Written by: Matthew Dobson, IBM Corporation |
| 5 | * |
| 6 | * Copyright (C) 2002, IBM Corp. |
| 7 | * |
| 8 | * All rights reserved. |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or modify |
| 11 | * it under the terms of the GNU General Public License as published by |
| 12 | * the Free Software Foundation; either version 2 of the License, or |
| 13 | * (at your option) any later version. |
| 14 | * |
| 15 | * This program is distributed in the hope that it will be useful, but |
| 16 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or |
| 18 | * NON INFRINGEMENT. See the GNU General Public License for more |
| 19 | * details. |
| 20 | * |
| 21 | * You should have received a copy of the GNU General Public License |
| 22 | * along with this program; if not, write to the Free Software |
| 23 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
| 24 | * |
| 25 | * Send feedback to <colpatch@us.ibm.com> |
| 26 | */ |
| 27 | #ifndef _LINUX_TOPOLOGY_H |
| 28 | #define _LINUX_TOPOLOGY_H |
| 29 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 30 | #include <linux/arch_topology.h> |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 31 | #include <linux/cpumask.h> |
| 32 | #include <linux/bitops.h> |
| 33 | #include <linux/mmzone.h> |
| 34 | #include <linux/smp.h> |
| 35 | #include <linux/percpu.h> |
| 36 | #include <asm/topology.h> |
| 37 | |
| 38 | #ifndef nr_cpus_node |
| 39 | #define nr_cpus_node(node) cpumask_weight(cpumask_of_node(node)) |
| 40 | #endif |
| 41 | |
| 42 | #define for_each_node_with_cpus(node) \ |
| 43 | for_each_online_node(node) \ |
| 44 | if (nr_cpus_node(node)) |
| 45 | |
| 46 | int arch_update_cpu_topology(void); |
| 47 | |
| 48 | /* Conform to ACPI 2.0 SLIT distance definitions */ |
| 49 | #define LOCAL_DISTANCE 10 |
| 50 | #define REMOTE_DISTANCE 20 |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame] | 51 | #define DISTANCE_BITS 8 |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 52 | #ifndef node_distance |
| 53 | #define node_distance(from,to) ((from) == (to) ? LOCAL_DISTANCE : REMOTE_DISTANCE) |
| 54 | #endif |
| 55 | #ifndef RECLAIM_DISTANCE |
| 56 | /* |
| 57 | * If the distance between nodes in a system is larger than RECLAIM_DISTANCE |
| 58 | * (in whatever arch specific measurement units returned by node_distance()) |
| 59 | * and node_reclaim_mode is enabled then the VM will only call node_reclaim() |
| 60 | * on nodes within this distance. |
| 61 | */ |
| 62 | #define RECLAIM_DISTANCE 30 |
| 63 | #endif |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 64 | |
| 65 | /* |
| 66 | * The following tunable allows platforms to override the default node |
| 67 | * reclaim distance (RECLAIM_DISTANCE) if remote memory accesses are |
| 68 | * sufficiently fast that the default value actually hurts |
| 69 | * performance. |
| 70 | * |
| 71 | * AMD EPYC machines use this because even though the 2-hop distance |
| 72 | * is 32 (3.2x slower than a local memory access) performance actually |
| 73 | * *improves* if allowed to reclaim memory and load balance tasks |
| 74 | * between NUMA nodes 2-hops apart. |
| 75 | */ |
| 76 | extern int __read_mostly node_reclaim_distance; |
| 77 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 78 | #ifndef PENALTY_FOR_NODE_WITH_CPUS |
| 79 | #define PENALTY_FOR_NODE_WITH_CPUS (1) |
| 80 | #endif |
| 81 | |
| 82 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
| 83 | DECLARE_PER_CPU(int, numa_node); |
| 84 | |
| 85 | #ifndef numa_node_id |
| 86 | /* Returns the number of the current Node. */ |
| 87 | static inline int numa_node_id(void) |
| 88 | { |
| 89 | return raw_cpu_read(numa_node); |
| 90 | } |
| 91 | #endif |
| 92 | |
| 93 | #ifndef cpu_to_node |
| 94 | static inline int cpu_to_node(int cpu) |
| 95 | { |
| 96 | return per_cpu(numa_node, cpu); |
| 97 | } |
| 98 | #endif |
| 99 | |
| 100 | #ifndef set_numa_node |
| 101 | static inline void set_numa_node(int node) |
| 102 | { |
| 103 | this_cpu_write(numa_node, node); |
| 104 | } |
| 105 | #endif |
| 106 | |
| 107 | #ifndef set_cpu_numa_node |
| 108 | static inline void set_cpu_numa_node(int cpu, int node) |
| 109 | { |
| 110 | per_cpu(numa_node, cpu) = node; |
| 111 | } |
| 112 | #endif |
| 113 | |
| 114 | #else /* !CONFIG_USE_PERCPU_NUMA_NODE_ID */ |
| 115 | |
| 116 | /* Returns the number of the current Node. */ |
| 117 | #ifndef numa_node_id |
| 118 | static inline int numa_node_id(void) |
| 119 | { |
| 120 | return cpu_to_node(raw_smp_processor_id()); |
| 121 | } |
| 122 | #endif |
| 123 | |
| 124 | #endif /* [!]CONFIG_USE_PERCPU_NUMA_NODE_ID */ |
| 125 | |
| 126 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 127 | |
| 128 | /* |
| 129 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. |
| 130 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. |
| 131 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem(). |
| 132 | */ |
| 133 | DECLARE_PER_CPU(int, _numa_mem_); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 134 | |
| 135 | #ifndef set_numa_mem |
| 136 | static inline void set_numa_mem(int node) |
| 137 | { |
| 138 | this_cpu_write(_numa_mem_, node); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 139 | } |
| 140 | #endif |
| 141 | |
| 142 | #ifndef numa_mem_id |
| 143 | /* Returns the number of the nearest Node with memory */ |
| 144 | static inline int numa_mem_id(void) |
| 145 | { |
| 146 | return raw_cpu_read(_numa_mem_); |
| 147 | } |
| 148 | #endif |
| 149 | |
| 150 | #ifndef cpu_to_mem |
| 151 | static inline int cpu_to_mem(int cpu) |
| 152 | { |
| 153 | return per_cpu(_numa_mem_, cpu); |
| 154 | } |
| 155 | #endif |
| 156 | |
| 157 | #ifndef set_cpu_numa_mem |
| 158 | static inline void set_cpu_numa_mem(int cpu, int node) |
| 159 | { |
| 160 | per_cpu(_numa_mem_, cpu) = node; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 161 | } |
| 162 | #endif |
| 163 | |
| 164 | #else /* !CONFIG_HAVE_MEMORYLESS_NODES */ |
| 165 | |
| 166 | #ifndef numa_mem_id |
| 167 | /* Returns the number of the nearest Node with memory */ |
| 168 | static inline int numa_mem_id(void) |
| 169 | { |
| 170 | return numa_node_id(); |
| 171 | } |
| 172 | #endif |
| 173 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 174 | #ifndef cpu_to_mem |
| 175 | static inline int cpu_to_mem(int cpu) |
| 176 | { |
| 177 | return cpu_to_node(cpu); |
| 178 | } |
| 179 | #endif |
| 180 | |
| 181 | #endif /* [!]CONFIG_HAVE_MEMORYLESS_NODES */ |
| 182 | |
| 183 | #ifndef topology_physical_package_id |
| 184 | #define topology_physical_package_id(cpu) ((void)(cpu), -1) |
| 185 | #endif |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 186 | #ifndef topology_die_id |
| 187 | #define topology_die_id(cpu) ((void)(cpu), -1) |
| 188 | #endif |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 189 | #ifndef topology_core_id |
| 190 | #define topology_core_id(cpu) ((void)(cpu), 0) |
| 191 | #endif |
| 192 | #ifndef topology_sibling_cpumask |
| 193 | #define topology_sibling_cpumask(cpu) cpumask_of(cpu) |
| 194 | #endif |
| 195 | #ifndef topology_core_cpumask |
| 196 | #define topology_core_cpumask(cpu) cpumask_of(cpu) |
| 197 | #endif |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 198 | #ifndef topology_die_cpumask |
| 199 | #define topology_die_cpumask(cpu) cpumask_of(cpu) |
| 200 | #endif |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 201 | |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame] | 202 | #if defined(CONFIG_SCHED_SMT) && !defined(cpu_smt_mask) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 203 | static inline const struct cpumask *cpu_smt_mask(int cpu) |
| 204 | { |
| 205 | return topology_sibling_cpumask(cpu); |
| 206 | } |
| 207 | #endif |
| 208 | |
| 209 | static inline const struct cpumask *cpu_cpu_mask(int cpu) |
| 210 | { |
| 211 | return cpumask_of_node(cpu_to_node(cpu)); |
| 212 | } |
| 213 | |
| 214 | |
| 215 | #endif /* _LINUX_TOPOLOGY_H */ |