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 /* DL bandwidth that needs destination reservation for this attach. */ 171 u64 sum_migrate_dl_bw; 172 /* 173 * CPU used for temporary DL bandwidth allocation during attach; 174 * -1 if no DL bandwidth was allocated in the current attach. 175 */ 176 int dl_bw_cpu; 177 178 /* Invalid partition error code, not lock protected */ 179 enum prs_errcode prs_err; 180 181 /* Handle for cpuset.cpus.partition */ 182 struct cgroup_file partition_file; 183 184 #ifdef CONFIG_CPUSETS_V1 185 struct fmeter fmeter; /* memory_pressure filter */ 186 187 /* for custom sched domain */ 188 int relax_domain_level; 189 190 /* Used to merge intersecting subsets for generate_sched_domains */ 191 struct uf_node node; 192 #endif 193 }; 194 195 extern struct cpuset top_cpuset; 196 197 static inline struct cpuset *css_cs(struct cgroup_subsys_state *css) 198 { 199 return css ? container_of(css, struct cpuset, css) : NULL; 200 } 201 202 /* Retrieve the cpuset for a task */ 203 static inline struct cpuset *task_cs(struct task_struct *task) 204 { 205 return css_cs(task_css(task, cpuset_cgrp_id)); 206 } 207 208 static inline struct cpuset *parent_cs(struct cpuset *cs) 209 { 210 return css_cs(cs->css.parent); 211 } 212 213 /* convenient tests for these bits */ 214 static inline bool is_cpuset_online(struct cpuset *cs) 215 { 216 return css_is_online(&cs->css) && !css_is_dying(&cs->css); 217 } 218 219 static inline int is_cpu_exclusive(const struct cpuset *cs) 220 { 221 return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); 222 } 223 224 static inline int is_mem_exclusive(const struct cpuset *cs) 225 { 226 return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); 227 } 228 229 static inline int is_mem_hardwall(const struct cpuset *cs) 230 { 231 return test_bit(CS_MEM_HARDWALL, &cs->flags); 232 } 233 234 static inline int is_sched_load_balance(const struct cpuset *cs) 235 { 236 return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); 237 } 238 239 static inline int is_memory_migrate(const struct cpuset *cs) 240 { 241 return test_bit(CS_MEMORY_MIGRATE, &cs->flags); 242 } 243 244 static inline int is_spread_page(const struct cpuset *cs) 245 { 246 return test_bit(CS_SPREAD_PAGE, &cs->flags); 247 } 248 249 static inline int is_spread_slab(const struct cpuset *cs) 250 { 251 return test_bit(CS_SPREAD_SLAB, &cs->flags); 252 } 253 254 /* 255 * Helper routine for generate_sched_domains(). 256 * Do cpusets a, b have overlapping effective cpus_allowed masks? 257 */ 258 static inline int cpusets_overlap(struct cpuset *a, struct cpuset *b) 259 { 260 return cpumask_intersects(a->effective_cpus, b->effective_cpus); 261 } 262 263 static inline int nr_cpusets(void) 264 { 265 /* jump label reference count + the top-level cpuset */ 266 return static_key_count(&cpusets_enabled_key.key) + 1; 267 } 268 269 static inline bool cpuset_is_populated(struct cpuset *cs) 270 { 271 lockdep_assert_cpuset_lock_held(); 272 273 /* Cpusets in the process of attaching should be considered as populated */ 274 return cgroup_is_populated(cs->css.cgroup) || 275 cs->attach_in_progress; 276 } 277 278 /** 279 * cpuset_for_each_child - traverse online children of a cpuset 280 * @child_cs: loop cursor pointing to the current child 281 * @pos_css: used for iteration 282 * @parent_cs: target cpuset to walk children of 283 * 284 * Walk @child_cs through the online children of @parent_cs. Must be used 285 * with RCU read locked. 286 */ 287 #define cpuset_for_each_child(child_cs, pos_css, parent_cs) \ 288 css_for_each_child((pos_css), &(parent_cs)->css) \ 289 if (is_cpuset_online(((child_cs) = css_cs((pos_css))))) 290 291 /** 292 * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants 293 * @des_cs: loop cursor pointing to the current descendant 294 * @pos_css: used for iteration 295 * @root_cs: target cpuset to walk ancestor of 296 * 297 * Walk @des_cs through the online descendants of @root_cs. Must be used 298 * with RCU read locked. The caller may modify @pos_css by calling 299 * css_rightmost_descendant() to skip subtree. @root_cs is included in the 300 * iteration and the first node to be visited. 301 */ 302 #define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs) \ 303 css_for_each_descendant_pre((pos_css), &(root_cs)->css) \ 304 if (is_cpuset_online(((des_cs) = css_cs((pos_css))))) 305 306 void rebuild_sched_domains_locked(void); 307 void cpuset_callback_lock_irq(void); 308 void cpuset_callback_unlock_irq(void); 309 void cpuset_update_tasks_cpumask(struct cpuset *cs, struct cpumask *new_cpus); 310 void cpuset_update_tasks_nodemask(struct cpuset *cs); 311 int cpuset_update_flag(cpuset_flagbits_t bit, struct cpuset *cs, int turning_on); 312 ssize_t cpuset_write_resmask(struct kernfs_open_file *of, 313 char *buf, size_t nbytes, loff_t off); 314 int cpuset_common_seq_show(struct seq_file *sf, void *v); 315 void cpuset_full_lock(void); 316 void cpuset_full_unlock(void); 317 318 /* 319 * cpuset-v1.c 320 */ 321 #ifdef CONFIG_CPUSETS_V1 322 extern struct cftype cpuset1_files[]; 323 void cpuset1_update_task_spread_flags(struct cpuset *cs, 324 struct task_struct *tsk); 325 void cpuset1_update_tasks_flags(struct cpuset *cs); 326 void cpuset1_hotplug_update_tasks(struct cpuset *cs, 327 struct cpumask *new_cpus, nodemask_t *new_mems, 328 bool cpus_updated, bool mems_updated); 329 int cpuset1_validate_change(struct cpuset *cur, struct cpuset *trial); 330 bool cpuset1_cpus_excl_conflict(struct cpuset *cs1, struct cpuset *cs2); 331 void cpuset1_init(struct cpuset *cs); 332 void cpuset1_online_css(struct cgroup_subsys_state *css); 333 int cpuset1_generate_sched_domains(cpumask_var_t **domains, 334 struct sched_domain_attr **attributes); 335 336 #else 337 static inline void cpuset1_update_task_spread_flags(struct cpuset *cs, 338 struct task_struct *tsk) {} 339 static inline void cpuset1_update_tasks_flags(struct cpuset *cs) {} 340 static inline void cpuset1_hotplug_update_tasks(struct cpuset *cs, 341 struct cpumask *new_cpus, nodemask_t *new_mems, 342 bool cpus_updated, bool mems_updated) {} 343 static inline int cpuset1_validate_change(struct cpuset *cur, 344 struct cpuset *trial) { return 0; } 345 static inline bool cpuset1_cpus_excl_conflict(struct cpuset *cs1, 346 struct cpuset *cs2) { return false; } 347 static inline void cpuset1_init(struct cpuset *cs) {} 348 static inline void cpuset1_online_css(struct cgroup_subsys_state *css) {} 349 static inline int cpuset1_generate_sched_domains(cpumask_var_t **domains, 350 struct sched_domain_attr **attributes) { return 0; }; 351 352 #endif /* CONFIG_CPUSETS_V1 */ 353 354 #endif /* __CPUSET_INTERNAL_H */ 355