xref: /linux/tools/sched_ext/include/scx/common.bpf.h (revision 02baaa67d9afc2e56c6e1ac6a1fb1f1dd2be366f)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Copyright (c) 2022 Meta Platforms, Inc. and affiliates.
4  * Copyright (c) 2022 Tejun Heo <tj@kernel.org>
5  * Copyright (c) 2022 David Vernet <dvernet@meta.com>
6  */
7 #ifndef __SCX_COMMON_BPF_H
8 #define __SCX_COMMON_BPF_H
9 
10 /*
11  * The generated kfunc prototypes in vmlinux.h are missing address space
12  * attributes which cause build failures. For now, suppress the generated
13  * prototypes. See https://github.com/sched-ext/scx/issues/1111.
14  */
15 #define BPF_NO_KFUNC_PROTOTYPES
16 
17 #ifdef LSP
18 #define __bpf__
19 #include "../vmlinux.h"
20 #else
21 #include "vmlinux.h"
22 #endif
23 
24 #include <bpf/bpf_helpers.h>
25 #include <bpf/bpf_tracing.h>
26 #include <asm-generic/errno.h>
27 #include "user_exit_info.bpf.h"
28 #include "enum_defs.autogen.h"
29 
30 #define PF_IDLE				0x00000002	/* I am an IDLE thread */
31 #define PF_IO_WORKER			0x00000010	/* Task is an IO worker */
32 #define PF_WQ_WORKER			0x00000020	/* I'm a workqueue worker */
33 #define PF_KCOMPACTD			0x00010000      /* I am kcompactd */
34 #define PF_KSWAPD			0x00020000      /* I am kswapd */
35 #define PF_KTHREAD			0x00200000	/* I am a kernel thread */
36 #define PF_EXITING			0x00000004
37 #define CLOCK_MONOTONIC			1
38 
39 #ifndef NR_CPUS
40 #define NR_CPUS 1024
41 #endif
42 
43 #ifndef NUMA_NO_NODE
44 #define	NUMA_NO_NODE	(-1)
45 #endif
46 
47 extern int LINUX_KERNEL_VERSION __kconfig;
48 extern const char CONFIG_CC_VERSION_TEXT[64] __kconfig __weak;
49 extern const char CONFIG_LOCALVERSION[64] __kconfig __weak;
50 
51 /*
52  * Earlier versions of clang/pahole lost upper 32bits in 64bit enums which can
53  * lead to really confusing misbehaviors. Let's trigger a build failure.
54  */
___vmlinux_h_sanity_check___(void)55 static inline void ___vmlinux_h_sanity_check___(void)
56 {
57 	_Static_assert(SCX_DSQ_FLAG_BUILTIN,
58 		       "bpftool generated vmlinux.h is missing high bits for 64bit enums, upgrade clang and pahole");
59 }
60 
61 s32 scx_bpf_create_dsq(u64 dsq_id, s32 node) __ksym;
62 s32 scx_bpf_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags, bool *is_idle) __ksym;
63 s32 __scx_bpf_select_cpu_and(struct task_struct *p, const struct cpumask *cpus_allowed,
64 			     struct scx_bpf_select_cpu_and_args *args) __ksym __weak;
65 bool __scx_bpf_dsq_insert_vtime(struct task_struct *p, struct scx_bpf_dsq_insert_vtime_args *args) __ksym __weak;
66 u32 scx_bpf_dispatch_nr_slots(void) __ksym;
67 void scx_bpf_dispatch_cancel(void) __ksym;
68 void scx_bpf_kick_cpu(s32 cpu, u64 flags) __ksym;
69 s32 scx_bpf_dsq_nr_queued(u64 dsq_id) __ksym;
70 void scx_bpf_destroy_dsq(u64 dsq_id) __ksym;
71 struct task_struct *scx_bpf_dsq_peek(u64 dsq_id) __ksym __weak;
72 int bpf_iter_scx_dsq_new(struct bpf_iter_scx_dsq *it, u64 dsq_id, u64 flags) __ksym __weak;
73 struct task_struct *bpf_iter_scx_dsq_next(struct bpf_iter_scx_dsq *it) __ksym __weak;
74 void bpf_iter_scx_dsq_destroy(struct bpf_iter_scx_dsq *it) __ksym __weak;
75 void scx_bpf_exit_bstr(s64 exit_code, char *fmt, unsigned long long *data, u32 data__sz) __ksym __weak;
76 void scx_bpf_error_bstr(char *fmt, unsigned long long *data, u32 data_len) __ksym;
77 void scx_bpf_dump_bstr(char *fmt, unsigned long long *data, u32 data_len) __ksym __weak;
78 u32 scx_bpf_cpuperf_cap(s32 cpu) __ksym __weak;
79 u32 scx_bpf_cpuperf_cur(s32 cpu) __ksym __weak;
80 void scx_bpf_cpuperf_set(s32 cpu, u32 perf) __ksym __weak;
81 u32 scx_bpf_nr_node_ids(void) __ksym __weak;
82 u32 scx_bpf_nr_cpu_ids(void) __ksym __weak;
83 int scx_bpf_cpu_node(s32 cpu) __ksym __weak;
84 const struct cpumask *scx_bpf_get_possible_cpumask(void) __ksym __weak;
85 const struct cpumask *scx_bpf_get_online_cpumask(void) __ksym __weak;
86 void scx_bpf_put_cpumask(const struct cpumask *cpumask) __ksym __weak;
87 const struct cpumask *scx_bpf_get_idle_cpumask_node(int node) __ksym __weak;
88 const struct cpumask *scx_bpf_get_idle_cpumask(void) __ksym;
89 const struct cpumask *scx_bpf_get_idle_smtmask_node(int node) __ksym __weak;
90 const struct cpumask *scx_bpf_get_idle_smtmask(void) __ksym;
91 void scx_bpf_put_idle_cpumask(const struct cpumask *cpumask) __ksym;
92 bool scx_bpf_test_and_clear_cpu_idle(s32 cpu) __ksym;
93 s32 scx_bpf_pick_idle_cpu_node(const cpumask_t *cpus_allowed, int node, u64 flags) __ksym __weak;
94 s32 scx_bpf_pick_idle_cpu(const cpumask_t *cpus_allowed, u64 flags) __ksym;
95 s32 scx_bpf_pick_any_cpu_node(const cpumask_t *cpus_allowed, int node, u64 flags) __ksym __weak;
96 s32 scx_bpf_pick_any_cpu(const cpumask_t *cpus_allowed, u64 flags) __ksym;
97 bool scx_bpf_task_running(const struct task_struct *p) __ksym;
98 s32 scx_bpf_task_cpu(const struct task_struct *p) __ksym;
99 struct rq *scx_bpf_cpu_rq(s32 cpu) __ksym;
100 struct rq *scx_bpf_locked_rq(void) __ksym;
101 struct task_struct *scx_bpf_cpu_curr(s32 cpu) __ksym __weak;
102 u64 scx_bpf_now(void) __ksym __weak;
103 void scx_bpf_events(struct scx_event_stats *events, size_t events__sz) __ksym __weak;
104 
105 /*
106  * Use the following as @it__iter when calling scx_bpf_dsq_move[_vtime]() from
107  * within bpf_for_each() loops.
108  */
109 #define BPF_FOR_EACH_ITER	(&___it)
110 
111 #define scx_read_event(e, name)							\
112 	(bpf_core_field_exists((e)->name) ? (e)->name : 0)
113 
114 static inline __attribute__((format(printf, 1, 2)))
___scx_bpf_bstr_format_checker(const char * fmt,...)115 void ___scx_bpf_bstr_format_checker(const char *fmt, ...) {}
116 
117 #define SCX_STRINGIFY(x) #x
118 #define SCX_TOSTRING(x) SCX_STRINGIFY(x)
119 
120 /*
121  * Helper macro for initializing the fmt and variadic argument inputs to both
122  * bstr exit kfuncs. Callers to this function should use ___fmt and ___param to
123  * refer to the initialized list of inputs to the bstr kfunc.
124  */
125 #define scx_bpf_bstr_preamble(fmt, args...)					\
126 	static char ___fmt[] = fmt;						\
127 	/*									\
128 	 * Note that __param[] must have at least one				\
129 	 * element to keep the verifier happy.					\
130 	 */									\
131 	unsigned long long ___param[___bpf_narg(args) ?: 1] = {};		\
132 										\
133 	_Pragma("GCC diagnostic push")						\
134 	_Pragma("GCC diagnostic ignored \"-Wint-conversion\"")			\
135 	___bpf_fill(___param, args);						\
136 	_Pragma("GCC diagnostic pop")
137 
138 /*
139  * scx_bpf_exit() wraps the scx_bpf_exit_bstr() kfunc with variadic arguments
140  * instead of an array of u64. Using this macro will cause the scheduler to
141  * exit cleanly with the specified exit code being passed to user space.
142  */
143 #define scx_bpf_exit(code, fmt, args...)					\
144 ({										\
145 	scx_bpf_bstr_preamble(fmt, args)					\
146 	scx_bpf_exit_bstr(code, ___fmt, ___param, sizeof(___param));		\
147 	___scx_bpf_bstr_format_checker(fmt, ##args);				\
148 })
149 
150 /*
151  * scx_bpf_error() wraps the scx_bpf_error_bstr() kfunc with variadic arguments
152  * instead of an array of u64. Invoking this macro will cause the scheduler to
153  * exit in an erroneous state, with diagnostic information being passed to the
154  * user. It appends the file and line number to aid debugging.
155  */
156 #define scx_bpf_error(fmt, args...)						\
157 ({										\
158 	scx_bpf_bstr_preamble(							\
159 		__FILE__ ":" SCX_TOSTRING(__LINE__) ": " fmt, ##args)		\
160 	scx_bpf_error_bstr(___fmt, ___param, sizeof(___param));			\
161 	___scx_bpf_bstr_format_checker(						\
162 		__FILE__ ":" SCX_TOSTRING(__LINE__) ": " fmt, ##args);		\
163 })
164 
165 /*
166  * scx_bpf_dump() wraps the scx_bpf_dump_bstr() kfunc with variadic arguments
167  * instead of an array of u64. To be used from ops.dump() and friends.
168  */
169 #define scx_bpf_dump(fmt, args...)						\
170 ({										\
171 	scx_bpf_bstr_preamble(fmt, args)					\
172 	scx_bpf_dump_bstr(___fmt, ___param, sizeof(___param));			\
173 	___scx_bpf_bstr_format_checker(fmt, ##args);				\
174 })
175 
176 /*
177  * scx_bpf_dump_header() is a wrapper around scx_bpf_dump that adds a header
178  * of system information for debugging.
179  */
180 #define scx_bpf_dump_header()							\
181 ({										\
182 	scx_bpf_dump("kernel: %d.%d.%d %s\ncc: %s\n",				\
183 		     LINUX_KERNEL_VERSION >> 16,				\
184 		     LINUX_KERNEL_VERSION >> 8 & 0xFF,				\
185 		     LINUX_KERNEL_VERSION & 0xFF,				\
186 		     CONFIG_LOCALVERSION,					\
187 		     CONFIG_CC_VERSION_TEXT);					\
188 })
189 
190 #define BPF_STRUCT_OPS(name, args...)						\
191 SEC("struct_ops/"#name)								\
192 BPF_PROG(name, ##args)
193 
194 #define BPF_STRUCT_OPS_SLEEPABLE(name, args...)					\
195 SEC("struct_ops.s/"#name)							\
196 BPF_PROG(name, ##args)
197 
198 /**
199  * RESIZABLE_ARRAY - Generates annotations for an array that may be resized
200  * @elfsec: the data section of the BPF program in which to place the array
201  * @arr: the name of the array
202  *
203  * libbpf has an API for setting map value sizes. Since data sections (i.e.
204  * bss, data, rodata) themselves are maps, a data section can be resized. If
205  * a data section has an array as its last element, the BTF info for that
206  * array will be adjusted so that length of the array is extended to meet the
207  * new length of the data section. This macro annotates an array to have an
208  * element count of one with the assumption that this array can be resized
209  * within the userspace program. It also annotates the section specifier so
210  * this array exists in a custom sub data section which can be resized
211  * independently.
212  *
213  * See RESIZE_ARRAY() for the userspace convenience macro for resizing an
214  * array declared with RESIZABLE_ARRAY().
215  */
216 #define RESIZABLE_ARRAY(elfsec, arr) arr[1] SEC("."#elfsec"."#arr)
217 
218 /**
219  * MEMBER_VPTR - Obtain the verified pointer to a struct or array member
220  * @base: struct or array to index
221  * @member: dereferenced member (e.g. .field, [idx0][idx1], .field[idx0] ...)
222  *
223  * The verifier often gets confused by the instruction sequence the compiler
224  * generates for indexing struct fields or arrays. This macro forces the
225  * compiler to generate a code sequence which first calculates the byte offset,
226  * checks it against the struct or array size and add that byte offset to
227  * generate the pointer to the member to help the verifier.
228  *
229  * Ideally, we want to abort if the calculated offset is out-of-bounds. However,
230  * BPF currently doesn't support abort, so evaluate to %NULL instead. The caller
231  * must check for %NULL and take appropriate action to appease the verifier. To
232  * avoid confusing the verifier, it's best to check for %NULL and dereference
233  * immediately.
234  *
235  *	vptr = MEMBER_VPTR(my_array, [i][j]);
236  *	if (!vptr)
237  *		return error;
238  *	*vptr = new_value;
239  *
240  * sizeof(@base) should encompass the memory area to be accessed and thus can't
241  * be a pointer to the area. Use `MEMBER_VPTR(*ptr, .member)` instead of
242  * `MEMBER_VPTR(ptr, ->member)`.
243  */
244 #ifndef MEMBER_VPTR
245 #define MEMBER_VPTR(base, member) (typeof((base) member) *)			\
246 ({										\
247 	u64 __base = (u64)&(base);						\
248 	u64 __addr = (u64)&((base) member) - __base;				\
249 	_Static_assert(sizeof(base) >= sizeof((base) member),			\
250 		       "@base is smaller than @member, is @base a pointer?");	\
251 	asm volatile (								\
252 		"if %0 <= %[max] goto +2\n"					\
253 		"%0 = 0\n"							\
254 		"goto +1\n"							\
255 		"%0 += %1\n"							\
256 		: "+r"(__addr)							\
257 		: "r"(__base),							\
258 		  [max]"i"(sizeof(base) - sizeof((base) member)));		\
259 	__addr;									\
260 })
261 #endif /* MEMBER_VPTR */
262 
263 /**
264  * ARRAY_ELEM_PTR - Obtain the verified pointer to an array element
265  * @arr: array to index into
266  * @i: array index
267  * @n: number of elements in array
268  *
269  * Similar to MEMBER_VPTR() but is intended for use with arrays where the
270  * element count needs to be explicit.
271  * It can be used in cases where a global array is defined with an initial
272  * size but is intended to be be resized before loading the BPF program.
273  * Without this version of the macro, MEMBER_VPTR() will use the compile time
274  * size of the array to compute the max, which will result in rejection by
275  * the verifier.
276  */
277 #ifndef ARRAY_ELEM_PTR
278 #define ARRAY_ELEM_PTR(arr, i, n) (typeof(arr[i]) *)				\
279 ({										\
280 	u64 __base = (u64)arr;							\
281 	u64 __addr = (u64)&(arr[i]) - __base;					\
282 	asm volatile (								\
283 		"if %0 <= %[max] goto +2\n"					\
284 		"%0 = 0\n"							\
285 		"goto +1\n"							\
286 		"%0 += %1\n"							\
287 		: "+r"(__addr)							\
288 		: "r"(__base),							\
289 		  [max]"r"(sizeof(arr[0]) * ((n) - 1)));			\
290 	__addr;									\
291 })
292 #endif /* ARRAY_ELEM_PTR */
293 
294 /*
295  * BPF declarations and helpers
296  */
297 
298 /* list and rbtree */
299 #define __contains(name, node) __attribute__((btf_decl_tag("contains:" #name ":" #node)))
300 #define private(name) SEC(".data." #name) __hidden __attribute__((aligned(8)))
301 
302 void *bpf_obj_new_impl(__u64 local_type_id, void *meta) __ksym;
303 void bpf_obj_drop_impl(void *kptr, void *meta) __ksym;
304 
305 #define bpf_obj_new(type) ((type *)bpf_obj_new_impl(bpf_core_type_id_local(type), NULL))
306 #define bpf_obj_drop(kptr) bpf_obj_drop_impl(kptr, NULL)
307 
308 int bpf_list_push_front_impl(struct bpf_list_head *head,
309 				    struct bpf_list_node *node,
310 				    void *meta, __u64 off) __ksym;
311 #define bpf_list_push_front(head, node) bpf_list_push_front_impl(head, node, NULL, 0)
312 
313 int bpf_list_push_back_impl(struct bpf_list_head *head,
314 				   struct bpf_list_node *node,
315 				   void *meta, __u64 off) __ksym;
316 #define bpf_list_push_back(head, node) bpf_list_push_back_impl(head, node, NULL, 0)
317 
318 struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head) __ksym;
319 struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head) __ksym;
320 struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root,
321 				      struct bpf_rb_node *node) __ksym;
322 int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node,
323 			bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b),
324 			void *meta, __u64 off) __ksym;
325 #define bpf_rbtree_add(head, node, less) bpf_rbtree_add_impl(head, node, less, NULL, 0)
326 
327 struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root) __ksym;
328 
329 void *bpf_refcount_acquire_impl(void *kptr, void *meta) __ksym;
330 #define bpf_refcount_acquire(kptr) bpf_refcount_acquire_impl(kptr, NULL)
331 
332 /* task */
333 struct task_struct *bpf_task_from_pid(s32 pid) __ksym;
334 struct task_struct *bpf_task_acquire(struct task_struct *p) __ksym;
335 void bpf_task_release(struct task_struct *p) __ksym;
336 
337 /* cgroup */
338 struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level) __ksym;
339 void bpf_cgroup_release(struct cgroup *cgrp) __ksym;
340 struct cgroup *bpf_cgroup_from_id(u64 cgid) __ksym;
341 
342 /* css iteration */
343 struct bpf_iter_css;
344 struct cgroup_subsys_state;
345 extern int bpf_iter_css_new(struct bpf_iter_css *it,
346 			    struct cgroup_subsys_state *start,
347 			    unsigned int flags) __weak __ksym;
348 extern struct cgroup_subsys_state *
349 bpf_iter_css_next(struct bpf_iter_css *it) __weak __ksym;
350 extern void bpf_iter_css_destroy(struct bpf_iter_css *it) __weak __ksym;
351 
352 /* cpumask */
353 struct bpf_cpumask *bpf_cpumask_create(void) __ksym;
354 struct bpf_cpumask *bpf_cpumask_acquire(struct bpf_cpumask *cpumask) __ksym;
355 void bpf_cpumask_release(struct bpf_cpumask *cpumask) __ksym;
356 u32 bpf_cpumask_first(const struct cpumask *cpumask) __ksym;
357 u32 bpf_cpumask_first_zero(const struct cpumask *cpumask) __ksym;
358 void bpf_cpumask_set_cpu(u32 cpu, struct bpf_cpumask *cpumask) __ksym;
359 void bpf_cpumask_clear_cpu(u32 cpu, struct bpf_cpumask *cpumask) __ksym;
360 bool bpf_cpumask_test_cpu(u32 cpu, const struct cpumask *cpumask) __ksym;
361 bool bpf_cpumask_test_and_set_cpu(u32 cpu, struct bpf_cpumask *cpumask) __ksym;
362 bool bpf_cpumask_test_and_clear_cpu(u32 cpu, struct bpf_cpumask *cpumask) __ksym;
363 void bpf_cpumask_setall(struct bpf_cpumask *cpumask) __ksym;
364 void bpf_cpumask_clear(struct bpf_cpumask *cpumask) __ksym;
365 bool bpf_cpumask_and(struct bpf_cpumask *dst, const struct cpumask *src1,
366 		     const struct cpumask *src2) __ksym;
367 void bpf_cpumask_or(struct bpf_cpumask *dst, const struct cpumask *src1,
368 		    const struct cpumask *src2) __ksym;
369 void bpf_cpumask_xor(struct bpf_cpumask *dst, const struct cpumask *src1,
370 		     const struct cpumask *src2) __ksym;
371 bool bpf_cpumask_equal(const struct cpumask *src1, const struct cpumask *src2) __ksym;
372 bool bpf_cpumask_intersects(const struct cpumask *src1, const struct cpumask *src2) __ksym;
373 bool bpf_cpumask_subset(const struct cpumask *src1, const struct cpumask *src2) __ksym;
374 bool bpf_cpumask_empty(const struct cpumask *cpumask) __ksym;
375 bool bpf_cpumask_full(const struct cpumask *cpumask) __ksym;
376 void bpf_cpumask_copy(struct bpf_cpumask *dst, const struct cpumask *src) __ksym;
377 u32 bpf_cpumask_any_distribute(const struct cpumask *cpumask) __ksym;
378 u32 bpf_cpumask_any_and_distribute(const struct cpumask *src1,
379 				   const struct cpumask *src2) __ksym;
380 u32 bpf_cpumask_weight(const struct cpumask *cpumask) __ksym;
381 
382 int bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_words) __ksym;
383 int *bpf_iter_bits_next(struct bpf_iter_bits *it) __ksym;
384 void bpf_iter_bits_destroy(struct bpf_iter_bits *it) __ksym;
385 
386 #define def_iter_struct(name)							\
387 struct bpf_iter_##name {							\
388     struct bpf_iter_bits it;							\
389     const struct cpumask *bitmap;						\
390 };
391 
392 #define def_iter_new(name)							\
393 static inline int bpf_iter_##name##_new(					\
394 	struct bpf_iter_##name *it, const u64 *unsafe_ptr__ign, u32 nr_words)	\
395 {										\
396 	it->bitmap = scx_bpf_get_##name##_cpumask();				\
397 	return bpf_iter_bits_new(&it->it, (const u64 *)it->bitmap,		\
398 				 sizeof(struct cpumask) / 8);			\
399 }
400 
401 #define def_iter_next(name)							\
402 static inline int *bpf_iter_##name##_next(struct bpf_iter_##name *it) {		\
403 	return bpf_iter_bits_next(&it->it);					\
404 }
405 
406 #define def_iter_destroy(name)							\
407 static inline void bpf_iter_##name##_destroy(struct bpf_iter_##name *it) {	\
408 	scx_bpf_put_cpumask(it->bitmap);					\
409 	bpf_iter_bits_destroy(&it->it);						\
410 }
411 #define def_for_each_cpu(cpu, name) for_each_##name##_cpu(cpu)
412 
413 /// Provides iterator for possible and online cpus.
414 ///
415 /// # Example
416 ///
417 /// ```
418 /// static inline void example_use() {
419 ///     int *cpu;
420 ///
421 ///     for_each_possible_cpu(cpu){
422 ///         bpf_printk("CPU %d is possible", *cpu);
423 ///     }
424 ///
425 ///     for_each_online_cpu(cpu){
426 ///         bpf_printk("CPU %d is online", *cpu);
427 ///     }
428 /// }
429 /// ```
430 def_iter_struct(possible);
431 def_iter_new(possible);
432 def_iter_next(possible);
433 def_iter_destroy(possible);
434 #define for_each_possible_cpu(cpu) bpf_for_each(possible, cpu, NULL, 0)
435 
436 def_iter_struct(online);
437 def_iter_new(online);
438 def_iter_next(online);
439 def_iter_destroy(online);
440 #define for_each_online_cpu(cpu) bpf_for_each(online, cpu, NULL, 0)
441 
442 /*
443  * Access a cpumask in read-only mode (typically to check bits).
444  */
cast_mask(struct bpf_cpumask * mask)445 static __always_inline const struct cpumask *cast_mask(struct bpf_cpumask *mask)
446 {
447 	return (const struct cpumask *)mask;
448 }
449 
450 /*
451  * Return true if task @p cannot migrate to a different CPU, false
452  * otherwise.
453  */
is_migration_disabled(const struct task_struct * p)454 static inline bool is_migration_disabled(const struct task_struct *p)
455 {
456 	/*
457 	 * Testing p->migration_disabled in a BPF code is tricky because the
458 	 * migration is _always_ disabled while running the BPF code.
459 	 * The prolog (__bpf_prog_enter) and epilog (__bpf_prog_exit) for BPF
460 	 * code execution disable and re-enable the migration of the current
461 	 * task, respectively. So, the _current_ task of the sched_ext ops is
462 	 * always migration-disabled. Moreover, p->migration_disabled could be
463 	 * two or greater when a sched_ext ops BPF code (e.g., ops.tick) is
464 	 * executed in the middle of the other BPF code execution.
465 	 *
466 	 * Therefore, we should decide that the _current_ task is
467 	 * migration-disabled only when its migration_disabled count is greater
468 	 * than one. In other words, when  p->migration_disabled == 1, there is
469 	 * an ambiguity, so we should check if @p is the current task or not.
470 	 */
471 	if (bpf_core_field_exists(p->migration_disabled)) {
472 		if (p->migration_disabled == 1)
473 			return bpf_get_current_task_btf() != p;
474 		else
475 			return p->migration_disabled;
476 	}
477 	return false;
478 }
479 
480 /* rcu */
481 void bpf_rcu_read_lock(void) __ksym;
482 void bpf_rcu_read_unlock(void) __ksym;
483 
484 /*
485  * Time helpers, most of which are from jiffies.h.
486  */
487 
488 /**
489  * time_delta - Calculate the delta between new and old time stamp
490  * @after: first comparable as u64
491  * @before: second comparable as u64
492  *
493  * Return: the time difference, which is >= 0
494  */
time_delta(u64 after,u64 before)495 static inline s64 time_delta(u64 after, u64 before)
496 {
497 	return (s64)(after - before) > 0 ? (s64)(after - before) : 0;
498 }
499 
500 /**
501  * time_after - returns true if the time a is after time b.
502  * @a: first comparable as u64
503  * @b: second comparable as u64
504  *
505  * Do this with "<0" and ">=0" to only test the sign of the result. A
506  * good compiler would generate better code (and a really good compiler
507  * wouldn't care). Gcc is currently neither.
508  *
509  * Return: %true is time a is after time b, otherwise %false.
510  */
time_after(u64 a,u64 b)511 static inline bool time_after(u64 a, u64 b)
512 {
513 	return (s64)(b - a) < 0;
514 }
515 
516 /**
517  * time_before - returns true if the time a is before time b.
518  * @a: first comparable as u64
519  * @b: second comparable as u64
520  *
521  * Return: %true is time a is before time b, otherwise %false.
522  */
time_before(u64 a,u64 b)523 static inline bool time_before(u64 a, u64 b)
524 {
525 	return time_after(b, a);
526 }
527 
528 /**
529  * time_after_eq - returns true if the time a is after or the same as time b.
530  * @a: first comparable as u64
531  * @b: second comparable as u64
532  *
533  * Return: %true is time a is after or the same as time b, otherwise %false.
534  */
time_after_eq(u64 a,u64 b)535 static inline bool time_after_eq(u64 a, u64 b)
536 {
537 	return (s64)(a - b) >= 0;
538 }
539 
540 /**
541  * time_before_eq - returns true if the time a is before or the same as time b.
542  * @a: first comparable as u64
543  * @b: second comparable as u64
544  *
545  * Return: %true is time a is before or the same as time b, otherwise %false.
546  */
time_before_eq(u64 a,u64 b)547 static inline bool time_before_eq(u64 a, u64 b)
548 {
549 	return time_after_eq(b, a);
550 }
551 
552 /**
553  * time_in_range - Calculate whether a is in the range of [b, c].
554  * @a: time to test
555  * @b: beginning of the range
556  * @c: end of the range
557  *
558  * Return: %true is time a is in the range [b, c], otherwise %false.
559  */
time_in_range(u64 a,u64 b,u64 c)560 static inline bool time_in_range(u64 a, u64 b, u64 c)
561 {
562 	return time_after_eq(a, b) && time_before_eq(a, c);
563 }
564 
565 /**
566  * time_in_range_open - Calculate whether a is in the range of [b, c).
567  * @a: time to test
568  * @b: beginning of the range
569  * @c: end of the range
570  *
571  * Return: %true is time a is in the range [b, c), otherwise %false.
572  */
time_in_range_open(u64 a,u64 b,u64 c)573 static inline bool time_in_range_open(u64 a, u64 b, u64 c)
574 {
575 	return time_after_eq(a, b) && time_before(a, c);
576 }
577 
578 
579 /*
580  * Other helpers
581  */
582 
583 /* useful compiler attributes */
584 #ifndef likely
585 #define likely(x) __builtin_expect(!!(x), 1)
586 #endif
587 #ifndef unlikely
588 #define unlikely(x) __builtin_expect(!!(x), 0)
589 #endif
590 #ifndef __maybe_unused
591 #define __maybe_unused __attribute__((__unused__))
592 #endif
593 
594 /*
595  * READ/WRITE_ONCE() are from kernel (include/asm-generic/rwonce.h). They
596  * prevent compiler from caching, redoing or reordering reads or writes.
597  */
598 typedef __u8  __attribute__((__may_alias__))  __u8_alias_t;
599 typedef __u16 __attribute__((__may_alias__)) __u16_alias_t;
600 typedef __u32 __attribute__((__may_alias__)) __u32_alias_t;
601 typedef __u64 __attribute__((__may_alias__)) __u64_alias_t;
602 
__read_once_size(const volatile void * p,void * res,int size)603 static __always_inline void __read_once_size(const volatile void *p, void *res, int size)
604 {
605 	switch (size) {
606 	case 1: *(__u8_alias_t  *) res = *(volatile __u8_alias_t  *) p; break;
607 	case 2: *(__u16_alias_t *) res = *(volatile __u16_alias_t *) p; break;
608 	case 4: *(__u32_alias_t *) res = *(volatile __u32_alias_t *) p; break;
609 	case 8: *(__u64_alias_t *) res = *(volatile __u64_alias_t *) p; break;
610 	default:
611 		barrier();
612 		__builtin_memcpy((void *)res, (const void *)p, size);
613 		barrier();
614 	}
615 }
616 
__write_once_size(volatile void * p,void * res,int size)617 static __always_inline void __write_once_size(volatile void *p, void *res, int size)
618 {
619 	switch (size) {
620 	case 1: *(volatile  __u8_alias_t *) p = *(__u8_alias_t  *) res; break;
621 	case 2: *(volatile __u16_alias_t *) p = *(__u16_alias_t *) res; break;
622 	case 4: *(volatile __u32_alias_t *) p = *(__u32_alias_t *) res; break;
623 	case 8: *(volatile __u64_alias_t *) p = *(__u64_alias_t *) res; break;
624 	default:
625 		barrier();
626 		__builtin_memcpy((void *)p, (const void *)res, size);
627 		barrier();
628 	}
629 }
630 
631 /*
632  * __unqual_typeof(x) - Declare an unqualified scalar type, leaving
633  *			non-scalar types unchanged,
634  *
635  * Prefer C11 _Generic for better compile-times and simpler code. Note: 'char'
636  * is not type-compatible with 'signed char', and we define a separate case.
637  *
638  * This is copied verbatim from kernel's include/linux/compiler_types.h, but
639  * with default expression (for pointers) changed from (x) to (typeof(x)0).
640  *
641  * This is because LLVM has a bug where for lvalue (x), it does not get rid of
642  * an extra address_space qualifier, but does in case of rvalue (typeof(x)0).
643  * Hence, for pointers, we need to create an rvalue expression to get the
644  * desired type. See https://github.com/llvm/llvm-project/issues/53400.
645  */
646 #define __scalar_type_to_expr_cases(type) \
647 	unsigned type : (unsigned type)0, signed type : (signed type)0
648 
649 #define __unqual_typeof(x)                              \
650 	typeof(_Generic((x),                            \
651 		char: (char)0,                          \
652 		__scalar_type_to_expr_cases(char),      \
653 		__scalar_type_to_expr_cases(short),     \
654 		__scalar_type_to_expr_cases(int),       \
655 		__scalar_type_to_expr_cases(long),      \
656 		__scalar_type_to_expr_cases(long long), \
657 		default: (typeof(x))0))
658 
659 #define READ_ONCE(x)								\
660 ({										\
661 	union { __unqual_typeof(x) __val; char __c[1]; } __u =			\
662 		{ .__c = { 0 } };						\
663 	__read_once_size((__unqual_typeof(x) *)&(x), __u.__c, sizeof(x));	\
664 	__u.__val;								\
665 })
666 
667 #define WRITE_ONCE(x, val)							\
668 ({										\
669 	union { __unqual_typeof(x) __val; char __c[1]; } __u =			\
670 		{ .__val = (val) }; 						\
671 	__write_once_size((__unqual_typeof(x) *)&(x), __u.__c, sizeof(x));	\
672 	__u.__val;								\
673 })
674 
675 /*
676  * __calc_avg - Calculate exponential weighted moving average (EWMA) with
677  * @old and @new values. @decay represents how large the @old value remains.
678  * With a larger @decay value, the moving average changes slowly, exhibiting
679  * fewer fluctuations.
680  */
681 #define __calc_avg(old, new, decay) ({						\
682 	typeof(decay) thr = 1 << (decay);					\
683 	typeof(old) ret;							\
684 	if (((old) < thr) || ((new) < thr)) {					\
685 		if (((old) == 1) && ((new) == 0))				\
686 			ret = 0;						\
687 		else								\
688 			ret = ((old) - ((old) >> 1)) + ((new) >> 1);		\
689 	} else {								\
690 		ret = ((old) - ((old) >> (decay))) + ((new) >> (decay));	\
691 	}									\
692 	ret;									\
693 })
694 
695 /*
696  * log2_u32 - Compute the base 2 logarithm of a 32-bit exponential value.
697  * @v: The value for which we're computing the base 2 logarithm.
698  */
log2_u32(u32 v)699 static inline u32 log2_u32(u32 v)
700 {
701         u32 r;
702         u32 shift;
703 
704         r = (v > 0xFFFF) << 4; v >>= r;
705         shift = (v > 0xFF) << 3; v >>= shift; r |= shift;
706         shift = (v > 0xF) << 2; v >>= shift; r |= shift;
707         shift = (v > 0x3) << 1; v >>= shift; r |= shift;
708         r |= (v >> 1);
709         return r;
710 }
711 
712 /*
713  * log2_u64 - Compute the base 2 logarithm of a 64-bit exponential value.
714  * @v: The value for which we're computing the base 2 logarithm.
715  */
log2_u64(u64 v)716 static inline u32 log2_u64(u64 v)
717 {
718         u32 hi = v >> 32;
719         if (hi)
720                 return log2_u32(hi) + 32 + 1;
721         else
722                 return log2_u32(v) + 1;
723 }
724 
725 /*
726  * sqrt_u64 - Calculate the square root of value @x using Newton's method.
727  */
__sqrt_u64(u64 x)728 static inline u64 __sqrt_u64(u64 x)
729 {
730 	if (x == 0 || x == 1)
731 		return x;
732 
733 	u64 r = ((1ULL << 32) > x) ? x : (1ULL << 32);
734 
735 	for (int i = 0; i < 8; ++i) {
736 		u64 q = x / r;
737 		if (r <= q)
738 			break;
739 		r = (r + q) >> 1;
740 	}
741 	return r;
742 }
743 
744 /*
745  * Return a value proportionally scaled to the task's weight.
746  */
scale_by_task_weight(const struct task_struct * p,u64 value)747 static inline u64 scale_by_task_weight(const struct task_struct *p, u64 value)
748 {
749 	return (value * p->scx.weight) / 100;
750 }
751 
752 /*
753  * Return a value inversely proportional to the task's weight.
754  */
scale_by_task_weight_inverse(const struct task_struct * p,u64 value)755 static inline u64 scale_by_task_weight_inverse(const struct task_struct *p, u64 value)
756 {
757 	return value * 100 / p->scx.weight;
758 }
759 
760 
761 #include "compat.bpf.h"
762 #include "enums.bpf.h"
763 
764 #endif	/* __SCX_COMMON_BPF_H */
765