1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 3 #ifndef __CPUSET_INTERNAL_H 4 #define __CPUSET_INTERNAL_H 5 6 #include <linux/cgroup.h> 7 #include <linux/cpu.h> 8 #include <linux/cpumask.h> 9 #include <linux/cpuset.h> 10 #include <linux/spinlock.h> 11 #include <linux/union_find.h> 12 #include <linux/sched/isolation.h> 13 14 /* See "Frequency meter" comments, below. */ 15 16 struct fmeter { 17 int cnt; /* unprocessed events count */ 18 int val; /* most recent output value */ 19 time64_t time; /* clock (secs) when val computed */ 20 spinlock_t lock; /* guards read or write of above */ 21 }; 22 23 /* 24 * Invalid partition error code 25 */ 26 enum prs_errcode { 27 PERR_NONE = 0, 28 PERR_INVCPUS, 29 PERR_INVPARENT, 30 PERR_NOTPART, 31 PERR_NOTEXCL, 32 PERR_NOCPUS, 33 PERR_HOTPLUG, 34 PERR_CPUSEMPTY, 35 PERR_HKEEPING, 36 PERR_ACCESS, 37 PERR_REMOTE, 38 }; 39 40 /* bits in struct cpuset flags field */ 41 typedef enum { 42 CS_CPU_EXCLUSIVE, 43 CS_MEM_EXCLUSIVE, 44 CS_MEM_HARDWALL, 45 CS_MEMORY_MIGRATE, 46 CS_SCHED_LOAD_BALANCE, 47 CS_SPREAD_PAGE, 48 CS_SPREAD_SLAB, 49 } cpuset_flagbits_t; 50 51 /* The various types of files and directories in a cpuset file system */ 52 53 typedef enum { 54 FILE_MEMORY_MIGRATE, 55 FILE_CPULIST, 56 FILE_MEMLIST, 57 FILE_EFFECTIVE_CPULIST, 58 FILE_EFFECTIVE_MEMLIST, 59 FILE_SUBPARTS_CPULIST, 60 FILE_EXCLUSIVE_CPULIST, 61 FILE_EFFECTIVE_XCPULIST, 62 FILE_ISOLATED_CPULIST, 63 FILE_CPU_EXCLUSIVE, 64 FILE_MEM_EXCLUSIVE, 65 FILE_MEM_HARDWALL, 66 FILE_SCHED_LOAD_BALANCE, 67 FILE_PARTITION_ROOT, 68 FILE_SCHED_RELAX_DOMAIN_LEVEL, 69 FILE_MEMORY_PRESSURE_ENABLED, 70 FILE_MEMORY_PRESSURE, 71 FILE_SPREAD_PAGE, 72 FILE_SPREAD_SLAB, 73 } cpuset_filetype_t; 74 75 struct cpuset { 76 struct cgroup_subsys_state css; 77 78 unsigned long flags; /* "unsigned long" so bitops work */ 79 80 /* 81 * On default hierarchy: 82 * 83 * The user-configured masks can only be changed by writing to 84 * cpuset.cpus and cpuset.mems, and won't be limited by the 85 * parent masks. 86 * 87 * The effective masks is the real masks that apply to the tasks 88 * in the cpuset. They may be changed if the configured masks are 89 * changed or hotplug happens. 90 * 91 * effective_mask == configured_mask & parent's effective_mask, 92 * and if it ends up empty, it will inherit the parent's mask. 93 * 94 * 95 * On legacy hierarchy: 96 * 97 * The user-configured masks are always the same with effective masks. 98 */ 99 100 /* user-configured CPUs and Memory Nodes allow to tasks */ 101 cpumask_var_t cpus_allowed; 102 nodemask_t mems_allowed; 103 104 /* effective CPUs and Memory Nodes allow to tasks */ 105 cpumask_var_t effective_cpus; 106 nodemask_t effective_mems; 107 108 /* 109 * Exclusive CPUs dedicated to current cgroup (default hierarchy only) 110 * 111 * The effective_cpus of a valid partition root comes solely from its 112 * effective_xcpus and some of the effective_xcpus may be distributed 113 * to sub-partitions below & hence excluded from its effective_cpus. 114 * For a valid partition root, its effective_cpus have no relationship 115 * with cpus_allowed unless its exclusive_cpus isn't set. 116 * 117 * This value will only be set if either exclusive_cpus is set or 118 * when this cpuset becomes a local partition root. 119 */ 120 cpumask_var_t effective_xcpus; 121 122 /* 123 * Exclusive CPUs as requested by the user (default hierarchy only) 124 * 125 * Its value is independent of cpus_allowed and designates the set of 126 * CPUs that can be granted to the current cpuset or its children when 127 * it becomes a valid partition root. The effective set of exclusive 128 * CPUs granted (effective_xcpus) depends on whether those exclusive 129 * CPUs are passed down by its ancestors and not yet taken up by 130 * another sibling partition root along the way. 131 * 132 * If its value isn't set, it defaults to cpus_allowed. 133 */ 134 cpumask_var_t exclusive_cpus; 135 136 /* 137 * This is old Memory Nodes tasks took on. 138 * 139 * - top_cpuset.old_mems_allowed is initialized to mems_allowed. 140 * - A new cpuset's old_mems_allowed is initialized when some 141 * task is moved into it. 142 * - old_mems_allowed is used in cpuset_migrate_mm() when we change 143 * cpuset.mems_allowed and have tasks' nodemask updated, and 144 * then old_mems_allowed is updated to mems_allowed. 145 */ 146 nodemask_t old_mems_allowed; 147 148 /* 149 * Tasks are being attached to this cpuset. Used to prevent 150 * zeroing cpus/mems_allowed between ->can_attach() and ->attach(). 151 */ 152 int attach_in_progress; 153 154 /* partition root state */ 155 int partition_root_state; 156 157 /* 158 * Whether cpuset is a remote partition. 159 * It used to be a list anchoring all remote partitions — we can switch back 160 * to a list if we need to iterate over the remote partitions. 161 */ 162 bool remote_partition; 163 164 /* 165 * number of SCHED_DEADLINE tasks attached to this cpuset, so that we 166 * know when to rebuild associated root domain bandwidth information. 167 */ 168 int nr_deadline_tasks; 169 int nr_migrate_dl_tasks; 170 u64 sum_migrate_dl_bw; 171 172 /* Invalid partition error code, not lock protected */ 173 enum prs_errcode prs_err; 174 175 /* Handle for cpuset.cpus.partition */ 176 struct cgroup_file partition_file; 177 178 #ifdef CONFIG_CPUSETS_V1 179 struct fmeter fmeter; /* memory_pressure filter */ 180 181 /* for custom sched domain */ 182 int relax_domain_level; 183 184 /* Used to merge intersecting subsets for generate_sched_domains */ 185 struct uf_node node; 186 #endif 187 }; 188 189 extern struct cpuset top_cpuset; 190 191 static inline struct cpuset *css_cs(struct cgroup_subsys_state *css) 192 { 193 return css ? container_of(css, struct cpuset, css) : NULL; 194 } 195 196 /* Retrieve the cpuset for a task */ 197 static inline struct cpuset *task_cs(struct task_struct *task) 198 { 199 return css_cs(task_css(task, cpuset_cgrp_id)); 200 } 201 202 static inline struct cpuset *parent_cs(struct cpuset *cs) 203 { 204 return css_cs(cs->css.parent); 205 } 206 207 /* convenient tests for these bits */ 208 static inline bool is_cpuset_online(struct cpuset *cs) 209 { 210 return css_is_online(&cs->css) && !css_is_dying(&cs->css); 211 } 212 213 static inline int is_cpu_exclusive(const struct cpuset *cs) 214 { 215 return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); 216 } 217 218 static inline int is_mem_exclusive(const struct cpuset *cs) 219 { 220 return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); 221 } 222 223 static inline int is_mem_hardwall(const struct cpuset *cs) 224 { 225 return test_bit(CS_MEM_HARDWALL, &cs->flags); 226 } 227 228 static inline int is_sched_load_balance(const struct cpuset *cs) 229 { 230 return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); 231 } 232 233 static inline int is_memory_migrate(const struct cpuset *cs) 234 { 235 return test_bit(CS_MEMORY_MIGRATE, &cs->flags); 236 } 237 238 static inline int is_spread_page(const struct cpuset *cs) 239 { 240 return test_bit(CS_SPREAD_PAGE, &cs->flags); 241 } 242 243 static inline int is_spread_slab(const struct cpuset *cs) 244 { 245 return test_bit(CS_SPREAD_SLAB, &cs->flags); 246 } 247 248 /* 249 * Helper routine for generate_sched_domains(). 250 * Do cpusets a, b have overlapping effective cpus_allowed masks? 251 */ 252 static inline int cpusets_overlap(struct cpuset *a, struct cpuset *b) 253 { 254 return cpumask_intersects(a->effective_cpus, b->effective_cpus); 255 } 256 257 static inline int nr_cpusets(void) 258 { 259 /* jump label reference count + the top-level cpuset */ 260 return static_key_count(&cpusets_enabled_key.key) + 1; 261 } 262 263 static inline bool cpuset_is_populated(struct cpuset *cs) 264 { 265 lockdep_assert_cpuset_lock_held(); 266 267 /* Cpusets in the process of attaching should be considered as populated */ 268 return cgroup_is_populated(cs->css.cgroup) || 269 cs->attach_in_progress; 270 } 271 272 /** 273 * cpuset_for_each_child - traverse online children of a cpuset 274 * @child_cs: loop cursor pointing to the current child 275 * @pos_css: used for iteration 276 * @parent_cs: target cpuset to walk children of 277 * 278 * Walk @child_cs through the online children of @parent_cs. Must be used 279 * with RCU read locked. 280 */ 281 #define cpuset_for_each_child(child_cs, pos_css, parent_cs) \ 282 css_for_each_child((pos_css), &(parent_cs)->css) \ 283 if (is_cpuset_online(((child_cs) = css_cs((pos_css))))) 284 285 /** 286 * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants 287 * @des_cs: loop cursor pointing to the current descendant 288 * @pos_css: used for iteration 289 * @root_cs: target cpuset to walk ancestor of 290 * 291 * Walk @des_cs through the online descendants of @root_cs. Must be used 292 * with RCU read locked. The caller may modify @pos_css by calling 293 * css_rightmost_descendant() to skip subtree. @root_cs is included in the 294 * iteration and the first node to be visited. 295 */ 296 #define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs) \ 297 css_for_each_descendant_pre((pos_css), &(root_cs)->css) \ 298 if (is_cpuset_online(((des_cs) = css_cs((pos_css))))) 299 300 void rebuild_sched_domains_locked(void); 301 void cpuset_callback_lock_irq(void); 302 void cpuset_callback_unlock_irq(void); 303 void cpuset_update_tasks_cpumask(struct cpuset *cs, struct cpumask *new_cpus); 304 void cpuset_update_tasks_nodemask(struct cpuset *cs); 305 int cpuset_update_flag(cpuset_flagbits_t bit, struct cpuset *cs, int turning_on); 306 ssize_t cpuset_write_resmask(struct kernfs_open_file *of, 307 char *buf, size_t nbytes, loff_t off); 308 int cpuset_common_seq_show(struct seq_file *sf, void *v); 309 void cpuset_full_lock(void); 310 void cpuset_full_unlock(void); 311 312 /* 313 * cpuset-v1.c 314 */ 315 #ifdef CONFIG_CPUSETS_V1 316 extern struct cftype cpuset1_files[]; 317 void cpuset1_update_task_spread_flags(struct cpuset *cs, 318 struct task_struct *tsk); 319 void cpuset1_update_tasks_flags(struct cpuset *cs); 320 void cpuset1_hotplug_update_tasks(struct cpuset *cs, 321 struct cpumask *new_cpus, nodemask_t *new_mems, 322 bool cpus_updated, bool mems_updated); 323 int cpuset1_validate_change(struct cpuset *cur, struct cpuset *trial); 324 bool cpuset1_cpus_excl_conflict(struct cpuset *cs1, struct cpuset *cs2); 325 void cpuset1_init(struct cpuset *cs); 326 void cpuset1_online_css(struct cgroup_subsys_state *css); 327 int cpuset1_generate_sched_domains(cpumask_var_t **domains, 328 struct sched_domain_attr **attributes); 329 330 #else 331 static inline void cpuset1_update_task_spread_flags(struct cpuset *cs, 332 struct task_struct *tsk) {} 333 static inline void cpuset1_update_tasks_flags(struct cpuset *cs) {} 334 static inline void cpuset1_hotplug_update_tasks(struct cpuset *cs, 335 struct cpumask *new_cpus, nodemask_t *new_mems, 336 bool cpus_updated, bool mems_updated) {} 337 static inline int cpuset1_validate_change(struct cpuset *cur, 338 struct cpuset *trial) { return 0; } 339 static inline bool cpuset1_cpus_excl_conflict(struct cpuset *cs1, 340 struct cpuset *cs2) { return false; } 341 static inline void cpuset1_init(struct cpuset *cs) {} 342 static inline void cpuset1_online_css(struct cgroup_subsys_state *css) {} 343 static inline int cpuset1_generate_sched_domains(cpumask_var_t **domains, 344 struct sched_domain_attr **attributes) { return 0; }; 345 346 #endif /* CONFIG_CPUSETS_V1 */ 347 348 #endif /* __CPUSET_INTERNAL_H */ 349