1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Copyright (c) 2024 Meta Platforms, Inc. and affiliates. 4 * Copyright (c) 2024 Tejun Heo <tj@kernel.org> 5 * Copyright (c) 2024 David Vernet <dvernet@meta.com> 6 */ 7 #ifndef __SCX_COMPAT_BPF_H 8 #define __SCX_COMPAT_BPF_H 9 10 #define __COMPAT_ENUM_OR_ZERO(__type, __ent) \ 11 ({ \ 12 __type __ret = 0; \ 13 if (bpf_core_enum_value_exists(__type, __ent)) \ 14 __ret = __ent; \ 15 __ret; \ 16 }) 17 18 /* v6.12: 819513666966 ("sched_ext: Add cgroup support") */ 19 struct cgroup *scx_bpf_task_cgroup___new(struct task_struct *p) __ksym __weak; 20 21 #define scx_bpf_task_cgroup(p) \ 22 (bpf_ksym_exists(scx_bpf_task_cgroup___new) ? \ 23 scx_bpf_task_cgroup___new((p)) : NULL) 24 25 /* 26 * v6.13: The verb `dispatch` was too overloaded and confusing. kfuncs are 27 * renamed to unload the verb. 28 * 29 * scx_bpf_dispatch_from_dsq() and friends were added during v6.12 by 30 * 4c30f5ce4f7a ("sched_ext: Implement scx_bpf_dispatch[_vtime]_from_dsq()"). 31 * 32 * v7.1: scx_bpf_dsq_move_to_local___v2() to add @enq_flags. 33 */ 34 bool scx_bpf_dsq_move_to_local___v2(u64 dsq_id, u64 enq_flags) __ksym __weak; 35 bool scx_bpf_dsq_move_to_local___v1(u64 dsq_id) __ksym __weak; 36 void scx_bpf_dsq_move_set_slice___new(struct bpf_iter_scx_dsq *it__iter, u64 slice) __ksym __weak; 37 void scx_bpf_dsq_move_set_vtime___new(struct bpf_iter_scx_dsq *it__iter, u64 vtime) __ksym __weak; 38 bool scx_bpf_dsq_move___new(struct bpf_iter_scx_dsq *it__iter, struct task_struct *p, u64 dsq_id, u64 enq_flags) __ksym __weak; 39 bool scx_bpf_dsq_move_vtime___new(struct bpf_iter_scx_dsq *it__iter, struct task_struct *p, u64 dsq_id, u64 enq_flags) __ksym __weak; 40 41 bool scx_bpf_consume___old(u64 dsq_id) __ksym __weak; 42 void scx_bpf_dispatch_from_dsq_set_slice___old(struct bpf_iter_scx_dsq *it__iter, u64 slice) __ksym __weak; 43 void scx_bpf_dispatch_from_dsq_set_vtime___old(struct bpf_iter_scx_dsq *it__iter, u64 vtime) __ksym __weak; 44 bool scx_bpf_dispatch_from_dsq___old(struct bpf_iter_scx_dsq *it__iter, struct task_struct *p, u64 dsq_id, u64 enq_flags) __ksym __weak; 45 bool scx_bpf_dispatch_vtime_from_dsq___old(struct bpf_iter_scx_dsq *it__iter, struct task_struct *p, u64 dsq_id, u64 enq_flags) __ksym __weak; 46 47 #define scx_bpf_dsq_move_to_local(dsq_id, enq_flags) \ 48 (bpf_ksym_exists(scx_bpf_dsq_move_to_local___v2) ? \ 49 scx_bpf_dsq_move_to_local___v2((dsq_id), (enq_flags)) : \ 50 (bpf_ksym_exists(scx_bpf_dsq_move_to_local___v1) ? \ 51 scx_bpf_dsq_move_to_local___v1((dsq_id)) : \ 52 scx_bpf_consume___old((dsq_id)))) 53 54 #define scx_bpf_dsq_move_set_slice(it__iter, slice) \ 55 (bpf_ksym_exists(scx_bpf_dsq_move_set_slice___new) ? \ 56 scx_bpf_dsq_move_set_slice___new((it__iter), (slice)) : \ 57 (bpf_ksym_exists(scx_bpf_dispatch_from_dsq_set_slice___old) ? \ 58 scx_bpf_dispatch_from_dsq_set_slice___old((it__iter), (slice)) : \ 59 (void)0)) 60 61 #define scx_bpf_dsq_move_set_vtime(it__iter, vtime) \ 62 (bpf_ksym_exists(scx_bpf_dsq_move_set_vtime___new) ? \ 63 scx_bpf_dsq_move_set_vtime___new((it__iter), (vtime)) : \ 64 (bpf_ksym_exists(scx_bpf_dispatch_from_dsq_set_vtime___old) ? \ 65 scx_bpf_dispatch_from_dsq_set_vtime___old((it__iter), (vtime)) : \ 66 (void)0)) 67 68 #define scx_bpf_dsq_move(it__iter, p, dsq_id, enq_flags) \ 69 (bpf_ksym_exists(scx_bpf_dsq_move___new) ? \ 70 scx_bpf_dsq_move___new((it__iter), (p), (dsq_id), (enq_flags)) : \ 71 (bpf_ksym_exists(scx_bpf_dispatch_from_dsq___old) ? \ 72 scx_bpf_dispatch_from_dsq___old((it__iter), (p), (dsq_id), (enq_flags)) : \ 73 false)) 74 75 #define scx_bpf_dsq_move_vtime(it__iter, p, dsq_id, enq_flags) \ 76 (bpf_ksym_exists(scx_bpf_dsq_move_vtime___new) ? \ 77 scx_bpf_dsq_move_vtime___new((it__iter), (p), (dsq_id), (enq_flags)) : \ 78 (bpf_ksym_exists(scx_bpf_dispatch_vtime_from_dsq___old) ? \ 79 scx_bpf_dispatch_vtime_from_dsq___old((it__iter), (p), (dsq_id), (enq_flags)) : \ 80 false)) 81 82 /* 83 * v6.15: 950ad93df2fc ("bpf: add kfunc for populating cpumask bits") 84 * 85 * Compat macro will be dropped on v6.19 release. 86 */ 87 int bpf_cpumask_populate(struct cpumask *dst, void *src, size_t src__sz) __ksym __weak; 88 89 #define __COMPAT_bpf_cpumask_populate(cpumask, src, size__sz) \ 90 (bpf_ksym_exists(bpf_cpumask_populate) ? \ 91 (bpf_cpumask_populate(cpumask, src, size__sz)) : -EOPNOTSUPP) 92 93 /* 94 * v6.19: Introduce lockless peek API for user DSQs. 95 * 96 * Preserve the following macro until v6.21. 97 */ 98 static inline struct task_struct *__COMPAT_scx_bpf_dsq_peek(u64 dsq_id) 99 { 100 struct task_struct *p = NULL; 101 struct bpf_iter_scx_dsq it; 102 103 if (bpf_ksym_exists(scx_bpf_dsq_peek)) 104 return scx_bpf_dsq_peek(dsq_id); 105 if (!bpf_iter_scx_dsq_new(&it, dsq_id, 0)) 106 p = bpf_iter_scx_dsq_next(&it); 107 bpf_iter_scx_dsq_destroy(&it); 108 return p; 109 } 110 111 /* 112 * v7.1: scx_bpf_sub_dispatch() for sub-sched dispatch. Preserve until 113 * we drop the compat layer for older kernels that lack the kfunc. 114 */ 115 bool scx_bpf_sub_dispatch___compat(u64 cgroup_id) __ksym __weak; 116 117 static inline bool scx_bpf_sub_dispatch(u64 cgroup_id) 118 { 119 if (bpf_ksym_exists(scx_bpf_sub_dispatch___compat)) 120 return scx_bpf_sub_dispatch___compat(cgroup_id); 121 return false; 122 } 123 124 /* 125 * v7.2: scx_bpf_cid_override() for explicit cpu->cid mapping. Ignore if 126 * missing. 127 */ 128 void scx_bpf_cid_override___compat(const s32 *cpu_to_cid, u32 cpu_to_cid__sz) __ksym __weak; 129 130 static inline void scx_bpf_cid_override(const s32 *cpu_to_cid, u32 cpu_to_cid__sz) 131 { 132 if (bpf_ksym_exists(scx_bpf_cid_override___compat)) 133 return scx_bpf_cid_override___compat(cpu_to_cid, cpu_to_cid__sz); 134 } 135 136 /** 137 * __COMPAT_is_enq_cpu_selected - Test if SCX_ENQ_CPU_SELECTED is on 138 * in a compatible way. We will preserve this __COMPAT helper until v6.16. 139 * 140 * @enq_flags: enqueue flags from ops.enqueue() 141 * 142 * Return: True if SCX_ENQ_CPU_SELECTED is turned on in @enq_flags 143 */ 144 static inline bool __COMPAT_is_enq_cpu_selected(u64 enq_flags) 145 { 146 #ifdef HAVE_SCX_ENQ_CPU_SELECTED 147 /* 148 * This is the case that a BPF code compiled against vmlinux.h 149 * where the enum SCX_ENQ_CPU_SELECTED exists. 150 */ 151 152 /* 153 * We should temporarily suspend the macro expansion of 154 * 'SCX_ENQ_CPU_SELECTED'. This avoids 'SCX_ENQ_CPU_SELECTED' being 155 * rewritten to '__SCX_ENQ_CPU_SELECTED' when 'SCX_ENQ_CPU_SELECTED' 156 * is defined in 'scripts/gen_enums.py'. 157 */ 158 #pragma push_macro("SCX_ENQ_CPU_SELECTED") 159 #undef SCX_ENQ_CPU_SELECTED 160 u64 flag; 161 162 /* 163 * When the kernel did not have SCX_ENQ_CPU_SELECTED, 164 * select_task_rq_scx() has never been skipped. Thus, this case 165 * should be considered that the CPU has already been selected. 166 */ 167 if (!bpf_core_enum_value_exists(enum scx_enq_flags, 168 SCX_ENQ_CPU_SELECTED)) 169 return true; 170 171 flag = bpf_core_enum_value(enum scx_enq_flags, SCX_ENQ_CPU_SELECTED); 172 return enq_flags & flag; 173 174 /* 175 * Once done, resume the macro expansion of 'SCX_ENQ_CPU_SELECTED'. 176 */ 177 #pragma pop_macro("SCX_ENQ_CPU_SELECTED") 178 #else 179 /* 180 * This is the case that a BPF code compiled against vmlinux.h 181 * where the enum SCX_ENQ_CPU_SELECTED does NOT exist. 182 */ 183 return true; 184 #endif /* HAVE_SCX_ENQ_CPU_SELECTED */ 185 } 186 187 188 #define scx_bpf_now() \ 189 (bpf_ksym_exists(scx_bpf_now) ? \ 190 scx_bpf_now() : \ 191 bpf_ktime_get_ns()) 192 193 /* 194 * v6.15: Introduce event counters. 195 * 196 * Preserve the following macro until v6.17. 197 */ 198 #define __COMPAT_scx_bpf_events(events, size) \ 199 (bpf_ksym_exists(scx_bpf_events) ? \ 200 scx_bpf_events(events, size) : ({})) 201 202 /* 203 * v6.15: Introduce NUMA-aware kfuncs to operate with per-node idle 204 * cpumasks. 205 * 206 * Preserve the following __COMPAT_scx_*_node macros until v6.17. 207 */ 208 #define __COMPAT_scx_bpf_nr_node_ids() \ 209 (bpf_ksym_exists(scx_bpf_nr_node_ids) ? \ 210 scx_bpf_nr_node_ids() : 1U) 211 212 #define __COMPAT_scx_bpf_cpu_node(cpu) \ 213 (bpf_ksym_exists(scx_bpf_cpu_node) ? \ 214 scx_bpf_cpu_node(cpu) : 0) 215 216 #define __COMPAT_scx_bpf_get_idle_cpumask_node(node) \ 217 (bpf_ksym_exists(scx_bpf_get_idle_cpumask_node) ? \ 218 scx_bpf_get_idle_cpumask_node(node) : \ 219 scx_bpf_get_idle_cpumask()) \ 220 221 #define __COMPAT_scx_bpf_get_idle_smtmask_node(node) \ 222 (bpf_ksym_exists(scx_bpf_get_idle_smtmask_node) ? \ 223 scx_bpf_get_idle_smtmask_node(node) : \ 224 scx_bpf_get_idle_smtmask()) 225 226 #define __COMPAT_scx_bpf_pick_idle_cpu_node(cpus_allowed, node, flags) \ 227 (bpf_ksym_exists(scx_bpf_pick_idle_cpu_node) ? \ 228 scx_bpf_pick_idle_cpu_node(cpus_allowed, node, flags) : \ 229 scx_bpf_pick_idle_cpu(cpus_allowed, flags)) 230 231 #define __COMPAT_scx_bpf_pick_any_cpu_node(cpus_allowed, node, flags) \ 232 (bpf_ksym_exists(scx_bpf_pick_any_cpu_node) ? \ 233 scx_bpf_pick_any_cpu_node(cpus_allowed, node, flags) : \ 234 scx_bpf_pick_any_cpu(cpus_allowed, flags)) 235 236 /* 237 * v6.18: Add a helper to retrieve the current task running on a CPU. 238 * 239 * Keep this helper available until v6.20 for compatibility. 240 */ 241 static inline struct task_struct *__COMPAT_scx_bpf_cpu_curr(int cpu) 242 { 243 struct rq *rq; 244 245 if (bpf_ksym_exists(scx_bpf_cpu_curr)) 246 return scx_bpf_cpu_curr(cpu); 247 248 rq = scx_bpf_cpu_rq(cpu); 249 250 return rq ? rq->curr : NULL; 251 } 252 253 /* 254 * v6.19: To work around BPF maximum parameter limit, the following kfuncs are 255 * replaced with variants that pack scalar arguments in a struct. Wrappers are 256 * provided to maintain source compatibility. 257 * 258 * v6.13: scx_bpf_dsq_insert_vtime() renaming is also handled here. See the 259 * block on dispatch renaming above for more details. 260 * 261 * The kernel will carry the compat variants until v6.23 to maintain binary 262 * compatibility. After v6.23 release, remove the compat handling and move the 263 * wrappers to common.bpf.h. 264 */ 265 s32 scx_bpf_select_cpu_and___compat(struct task_struct *p, s32 prev_cpu, u64 wake_flags, 266 const struct cpumask *cpus_allowed, u64 flags) __ksym __weak; 267 void scx_bpf_dispatch_vtime___compat(struct task_struct *p, u64 dsq_id, u64 slice, u64 vtime, u64 enq_flags) __ksym __weak; 268 void scx_bpf_dsq_insert_vtime___compat(struct task_struct *p, u64 dsq_id, u64 slice, u64 vtime, u64 enq_flags) __ksym __weak; 269 270 /** 271 * scx_bpf_select_cpu_and - Pick an idle CPU usable by task @p 272 * @p: task_struct to select a CPU for 273 * @prev_cpu: CPU @p was on previously 274 * @wake_flags: %SCX_WAKE_* flags 275 * @cpus_allowed: cpumask of allowed CPUs 276 * @flags: %SCX_PICK_IDLE* flags 277 * 278 * Inline wrapper that packs scalar arguments into a struct and calls 279 * __scx_bpf_select_cpu_and(). See __scx_bpf_select_cpu_and() for details. 280 */ 281 static inline s32 282 scx_bpf_select_cpu_and(struct task_struct *p, s32 prev_cpu, u64 wake_flags, 283 const struct cpumask *cpus_allowed, u64 flags) 284 { 285 if (bpf_core_type_exists(struct scx_bpf_select_cpu_and_args)) { 286 struct scx_bpf_select_cpu_and_args args = { 287 .prev_cpu = prev_cpu, 288 .wake_flags = wake_flags, 289 .flags = flags, 290 }; 291 292 return __scx_bpf_select_cpu_and(p, cpus_allowed, &args); 293 } else { 294 return scx_bpf_select_cpu_and___compat(p, prev_cpu, wake_flags, 295 cpus_allowed, flags); 296 } 297 } 298 299 /* 300 * scx_bpf_select_cpu_and() is now an inline wrapper. Use this instead of 301 * bpf_ksym_exists(scx_bpf_select_cpu_and) to test availability. 302 */ 303 #define __COMPAT_HAS_scx_bpf_select_cpu_and \ 304 (bpf_core_type_exists(struct scx_bpf_select_cpu_and_args) || \ 305 bpf_ksym_exists(scx_bpf_select_cpu_and___compat)) 306 307 /** 308 * scx_bpf_dsq_insert_vtime - Insert a task into the vtime priority queue of a DSQ 309 * @p: task_struct to insert 310 * @dsq_id: DSQ to insert into 311 * @slice: duration @p can run for in nsecs, 0 to keep the current value 312 * @vtime: @p's ordering inside the vtime-sorted queue of the target DSQ 313 * @enq_flags: SCX_ENQ_* 314 * 315 * Inline wrapper that packs scalar arguments into a struct and calls 316 * __scx_bpf_dsq_insert_vtime(). See __scx_bpf_dsq_insert_vtime() for details. 317 */ 318 static inline bool 319 scx_bpf_dsq_insert_vtime(struct task_struct *p, u64 dsq_id, u64 slice, u64 vtime, 320 u64 enq_flags) 321 { 322 if (bpf_core_type_exists(struct scx_bpf_dsq_insert_vtime_args)) { 323 struct scx_bpf_dsq_insert_vtime_args args = { 324 .dsq_id = dsq_id, 325 .slice = slice, 326 .vtime = vtime, 327 .enq_flags = enq_flags, 328 }; 329 330 return __scx_bpf_dsq_insert_vtime(p, &args); 331 } else if (bpf_ksym_exists(scx_bpf_dsq_insert_vtime___compat)) { 332 scx_bpf_dsq_insert_vtime___compat(p, dsq_id, slice, vtime, 333 enq_flags); 334 return true; 335 } else { 336 scx_bpf_dispatch_vtime___compat(p, dsq_id, slice, vtime, 337 enq_flags); 338 return true; 339 } 340 } 341 342 /* 343 * v6.19: scx_bpf_dsq_insert() now returns bool instead of void. Move 344 * scx_bpf_dsq_insert() decl to common.bpf.h and drop compat helper after v6.22. 345 * The extra ___compat suffix is to work around libbpf not ignoring __SUFFIX on 346 * kernel side. The entire suffix can be dropped later. 347 * 348 * v6.13: scx_bpf_dsq_insert() renaming is also handled here. See the block on 349 * dispatch renaming above for more details. 350 */ 351 bool scx_bpf_dsq_insert___v2___compat(struct task_struct *p, u64 dsq_id, u64 slice, u64 enq_flags) __ksym __weak; 352 void scx_bpf_dsq_insert___v1(struct task_struct *p, u64 dsq_id, u64 slice, u64 enq_flags) __ksym __weak; 353 void scx_bpf_dispatch___compat(struct task_struct *p, u64 dsq_id, u64 slice, u64 enq_flags) __ksym __weak; 354 355 static inline bool 356 scx_bpf_dsq_insert(struct task_struct *p, u64 dsq_id, u64 slice, u64 enq_flags) 357 { 358 if (bpf_ksym_exists(scx_bpf_dsq_insert___v2___compat)) { 359 return scx_bpf_dsq_insert___v2___compat(p, dsq_id, slice, enq_flags); 360 } else if (bpf_ksym_exists(scx_bpf_dsq_insert___v1)) { 361 scx_bpf_dsq_insert___v1(p, dsq_id, slice, enq_flags); 362 return true; 363 } else { 364 scx_bpf_dispatch___compat(p, dsq_id, slice, enq_flags); 365 return true; 366 } 367 } 368 369 /* 370 * v6.19: scx_bpf_task_set_slice() and scx_bpf_task_set_dsq_vtime() added to for 371 * sub-sched authority checks. Drop the wrappers and move the decls to 372 * common.bpf.h after v6.22. 373 */ 374 bool scx_bpf_task_set_slice___new(struct task_struct *p, u64 slice) __ksym __weak; 375 bool scx_bpf_task_set_dsq_vtime___new(struct task_struct *p, u64 vtime) __ksym __weak; 376 377 static inline void scx_bpf_task_set_slice(struct task_struct *p, u64 slice) 378 { 379 if (bpf_ksym_exists(scx_bpf_task_set_slice___new)) 380 scx_bpf_task_set_slice___new(p, slice); 381 else 382 p->scx.slice = slice; 383 } 384 385 static inline void scx_bpf_task_set_dsq_vtime(struct task_struct *p, u64 vtime) 386 { 387 if (bpf_ksym_exists(scx_bpf_task_set_dsq_vtime___new)) 388 scx_bpf_task_set_dsq_vtime___new(p, vtime); 389 else 390 p->scx.dsq_vtime = vtime; 391 } 392 393 /* 394 * v6.19: The new void variant can be called from anywhere while the older v1 395 * variant can only be called from ops.cpu_release(). The double ___ prefixes on 396 * the v2 variant need to be removed once libbpf is updated to ignore ___ prefix 397 * on kernel side. Drop the wrapper and move the decl to common.bpf.h after 398 * v6.22. 399 */ 400 u32 scx_bpf_reenqueue_local___v1(void) __ksym __weak; 401 void scx_bpf_reenqueue_local___v2___compat(void) __ksym __weak; 402 403 static inline bool __COMPAT_scx_bpf_reenqueue_local_from_anywhere(void) 404 { 405 return bpf_ksym_exists(scx_bpf_reenqueue_local___v2___compat); 406 } 407 408 static inline void scx_bpf_reenqueue_local(void) 409 { 410 if (__COMPAT_scx_bpf_reenqueue_local_from_anywhere()) 411 scx_bpf_reenqueue_local___v2___compat(); 412 else 413 scx_bpf_reenqueue_local___v1(); 414 } 415 416 /* 417 * v6.20: New scx_bpf_dsq_reenq() that allows re-enqueues on more DSQs. This 418 * will eventually deprecate scx_bpf_reenqueue_local(). 419 */ 420 void scx_bpf_dsq_reenq___compat(u64 dsq_id, u64 reenq_flags, const struct bpf_prog_aux *aux__prog) __ksym __weak; 421 422 static inline bool __COMPAT_has_generic_reenq(void) 423 { 424 return bpf_ksym_exists(scx_bpf_dsq_reenq___compat); 425 } 426 427 static inline void scx_bpf_dsq_reenq(u64 dsq_id, u64 reenq_flags) 428 { 429 if (bpf_ksym_exists(scx_bpf_dsq_reenq___compat)) 430 scx_bpf_dsq_reenq___compat(dsq_id, reenq_flags, NULL); 431 else if (dsq_id == SCX_DSQ_LOCAL && reenq_flags == 0) 432 scx_bpf_reenqueue_local(); 433 else 434 scx_bpf_error("kernel too old to reenqueue foreign local or user DSQs"); 435 } 436 437 /* 438 * Define sched_ext_ops. See compat.h::SCX_OPS_OPEN() for how backward 439 * compatibility is handled (this macro can be expanded to emit multiple 440 * variants for incompatible op changes; SCX_OPS_OPEN() handles purely 441 * additive changes at load time). 442 */ 443 #define SCX_OPS_DEFINE(__name, ...) \ 444 SEC(".struct_ops.link") \ 445 struct sched_ext_ops __name = { \ 446 __VA_ARGS__, \ 447 }; 448 449 /* 450 * Define a cid-form sched_ext_ops. Programs targeting this struct_ops type 451 * use cid-form callback signatures (select_cid, set_cmask, cid_online/offline, 452 * dispatch with cid arg, etc.) and may only call the cid-form scx_bpf_* 453 * kfuncs (kick_cid, task_cid, this_cid, ...). 454 */ 455 #define SCX_OPS_CID_DEFINE(__name, ...) \ 456 SEC(".struct_ops.link") \ 457 struct sched_ext_ops_cid __name = { \ 458 __VA_ARGS__, \ 459 }; 460 461 #endif /* __SCX_COMPAT_BPF_H */ 462