xref: /linux/tools/sched_ext/include/scx/common.bpf.h (revision d86adb4fc0655a0867da811d000df75d2a325ef6)
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 #include "vmlinux.h"
11 #include <bpf/bpf_helpers.h>
12 #include <bpf/bpf_tracing.h>
13 #include <asm-generic/errno.h>
14 #include "user_exit_info.h"
15 
16 #define PF_WQ_WORKER			0x00000020	/* I'm a workqueue worker */
17 #define PF_KTHREAD			0x00200000	/* I am a kernel thread */
18 #define PF_EXITING			0x00000004
19 #define CLOCK_MONOTONIC			1
20 
21 /*
22  * Earlier versions of clang/pahole lost upper 32bits in 64bit enums which can
23  * lead to really confusing misbehaviors. Let's trigger a build failure.
24  */
25 static inline void ___vmlinux_h_sanity_check___(void)
26 {
27 	_Static_assert(SCX_DSQ_FLAG_BUILTIN,
28 		       "bpftool generated vmlinux.h is missing high bits for 64bit enums, upgrade clang and pahole");
29 }
30 
31 s32 scx_bpf_create_dsq(u64 dsq_id, s32 node) __ksym;
32 s32 scx_bpf_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags, bool *is_idle) __ksym;
33 void scx_bpf_dispatch(struct task_struct *p, u64 dsq_id, u64 slice, u64 enq_flags) __ksym;
34 void scx_bpf_dispatch_vtime(struct task_struct *p, u64 dsq_id, u64 slice, u64 vtime, u64 enq_flags) __ksym;
35 u32 scx_bpf_dispatch_nr_slots(void) __ksym;
36 void scx_bpf_dispatch_cancel(void) __ksym;
37 bool scx_bpf_consume(u64 dsq_id) __ksym;
38 u32 scx_bpf_reenqueue_local(void) __ksym;
39 void scx_bpf_kick_cpu(s32 cpu, u64 flags) __ksym;
40 s32 scx_bpf_dsq_nr_queued(u64 dsq_id) __ksym;
41 void scx_bpf_destroy_dsq(u64 dsq_id) __ksym;
42 void scx_bpf_exit_bstr(s64 exit_code, char *fmt, unsigned long long *data, u32 data__sz) __ksym __weak;
43 void scx_bpf_error_bstr(char *fmt, unsigned long long *data, u32 data_len) __ksym;
44 void scx_bpf_dump_bstr(char *fmt, unsigned long long *data, u32 data_len) __ksym __weak;
45 u32 scx_bpf_cpuperf_cap(s32 cpu) __ksym __weak;
46 u32 scx_bpf_cpuperf_cur(s32 cpu) __ksym __weak;
47 void scx_bpf_cpuperf_set(s32 cpu, u32 perf) __ksym __weak;
48 u32 scx_bpf_nr_cpu_ids(void) __ksym __weak;
49 const struct cpumask *scx_bpf_get_possible_cpumask(void) __ksym __weak;
50 const struct cpumask *scx_bpf_get_online_cpumask(void) __ksym __weak;
51 void scx_bpf_put_cpumask(const struct cpumask *cpumask) __ksym __weak;
52 const struct cpumask *scx_bpf_get_idle_cpumask(void) __ksym;
53 const struct cpumask *scx_bpf_get_idle_smtmask(void) __ksym;
54 void scx_bpf_put_idle_cpumask(const struct cpumask *cpumask) __ksym;
55 bool scx_bpf_test_and_clear_cpu_idle(s32 cpu) __ksym;
56 s32 scx_bpf_pick_idle_cpu(const cpumask_t *cpus_allowed, u64 flags) __ksym;
57 s32 scx_bpf_pick_any_cpu(const cpumask_t *cpus_allowed, u64 flags) __ksym;
58 bool scx_bpf_task_running(const struct task_struct *p) __ksym;
59 s32 scx_bpf_task_cpu(const struct task_struct *p) __ksym;
60 
61 static inline __attribute__((format(printf, 1, 2)))
62 void ___scx_bpf_bstr_format_checker(const char *fmt, ...) {}
63 
64 /*
65  * Helper macro for initializing the fmt and variadic argument inputs to both
66  * bstr exit kfuncs. Callers to this function should use ___fmt and ___param to
67  * refer to the initialized list of inputs to the bstr kfunc.
68  */
69 #define scx_bpf_bstr_preamble(fmt, args...)					\
70 	static char ___fmt[] = fmt;						\
71 	/*									\
72 	 * Note that __param[] must have at least one				\
73 	 * element to keep the verifier happy.					\
74 	 */									\
75 	unsigned long long ___param[___bpf_narg(args) ?: 1] = {};		\
76 										\
77 	_Pragma("GCC diagnostic push")						\
78 	_Pragma("GCC diagnostic ignored \"-Wint-conversion\"")			\
79 	___bpf_fill(___param, args);						\
80 	_Pragma("GCC diagnostic pop")						\
81 
82 /*
83  * scx_bpf_exit() wraps the scx_bpf_exit_bstr() kfunc with variadic arguments
84  * instead of an array of u64. Using this macro will cause the scheduler to
85  * exit cleanly with the specified exit code being passed to user space.
86  */
87 #define scx_bpf_exit(code, fmt, args...)					\
88 ({										\
89 	scx_bpf_bstr_preamble(fmt, args)					\
90 	scx_bpf_exit_bstr(code, ___fmt, ___param, sizeof(___param));		\
91 	___scx_bpf_bstr_format_checker(fmt, ##args);				\
92 })
93 
94 /*
95  * scx_bpf_error() wraps the scx_bpf_error_bstr() kfunc with variadic arguments
96  * instead of an array of u64. Invoking this macro will cause the scheduler to
97  * exit in an erroneous state, with diagnostic information being passed to the
98  * user.
99  */
100 #define scx_bpf_error(fmt, args...)						\
101 ({										\
102 	scx_bpf_bstr_preamble(fmt, args)					\
103 	scx_bpf_error_bstr(___fmt, ___param, sizeof(___param));			\
104 	___scx_bpf_bstr_format_checker(fmt, ##args);				\
105 })
106 
107 /*
108  * scx_bpf_dump() wraps the scx_bpf_dump_bstr() kfunc with variadic arguments
109  * instead of an array of u64. To be used from ops.dump() and friends.
110  */
111 #define scx_bpf_dump(fmt, args...)						\
112 ({										\
113 	scx_bpf_bstr_preamble(fmt, args)					\
114 	scx_bpf_dump_bstr(___fmt, ___param, sizeof(___param));			\
115 	___scx_bpf_bstr_format_checker(fmt, ##args);				\
116 })
117 
118 #define BPF_STRUCT_OPS(name, args...)						\
119 SEC("struct_ops/"#name)								\
120 BPF_PROG(name, ##args)
121 
122 #define BPF_STRUCT_OPS_SLEEPABLE(name, args...)					\
123 SEC("struct_ops.s/"#name)							\
124 BPF_PROG(name, ##args)
125 
126 /**
127  * RESIZABLE_ARRAY - Generates annotations for an array that may be resized
128  * @elfsec: the data section of the BPF program in which to place the array
129  * @arr: the name of the array
130  *
131  * libbpf has an API for setting map value sizes. Since data sections (i.e.
132  * bss, data, rodata) themselves are maps, a data section can be resized. If
133  * a data section has an array as its last element, the BTF info for that
134  * array will be adjusted so that length of the array is extended to meet the
135  * new length of the data section. This macro annotates an array to have an
136  * element count of one with the assumption that this array can be resized
137  * within the userspace program. It also annotates the section specifier so
138  * this array exists in a custom sub data section which can be resized
139  * independently.
140  *
141  * See RESIZE_ARRAY() for the userspace convenience macro for resizing an
142  * array declared with RESIZABLE_ARRAY().
143  */
144 #define RESIZABLE_ARRAY(elfsec, arr) arr[1] SEC("."#elfsec"."#arr)
145 
146 /**
147  * MEMBER_VPTR - Obtain the verified pointer to a struct or array member
148  * @base: struct or array to index
149  * @member: dereferenced member (e.g. .field, [idx0][idx1], .field[idx0] ...)
150  *
151  * The verifier often gets confused by the instruction sequence the compiler
152  * generates for indexing struct fields or arrays. This macro forces the
153  * compiler to generate a code sequence which first calculates the byte offset,
154  * checks it against the struct or array size and add that byte offset to
155  * generate the pointer to the member to help the verifier.
156  *
157  * Ideally, we want to abort if the calculated offset is out-of-bounds. However,
158  * BPF currently doesn't support abort, so evaluate to %NULL instead. The caller
159  * must check for %NULL and take appropriate action to appease the verifier. To
160  * avoid confusing the verifier, it's best to check for %NULL and dereference
161  * immediately.
162  *
163  *	vptr = MEMBER_VPTR(my_array, [i][j]);
164  *	if (!vptr)
165  *		return error;
166  *	*vptr = new_value;
167  *
168  * sizeof(@base) should encompass the memory area to be accessed and thus can't
169  * be a pointer to the area. Use `MEMBER_VPTR(*ptr, .member)` instead of
170  * `MEMBER_VPTR(ptr, ->member)`.
171  */
172 #define MEMBER_VPTR(base, member) (typeof((base) member) *)			\
173 ({										\
174 	u64 __base = (u64)&(base);						\
175 	u64 __addr = (u64)&((base) member) - __base;				\
176 	_Static_assert(sizeof(base) >= sizeof((base) member),			\
177 		       "@base is smaller than @member, is @base a pointer?");	\
178 	asm volatile (								\
179 		"if %0 <= %[max] goto +2\n"					\
180 		"%0 = 0\n"							\
181 		"goto +1\n"							\
182 		"%0 += %1\n"							\
183 		: "+r"(__addr)							\
184 		: "r"(__base),							\
185 		  [max]"i"(sizeof(base) - sizeof((base) member)));		\
186 	__addr;									\
187 })
188 
189 /**
190  * ARRAY_ELEM_PTR - Obtain the verified pointer to an array element
191  * @arr: array to index into
192  * @i: array index
193  * @n: number of elements in array
194  *
195  * Similar to MEMBER_VPTR() but is intended for use with arrays where the
196  * element count needs to be explicit.
197  * It can be used in cases where a global array is defined with an initial
198  * size but is intended to be be resized before loading the BPF program.
199  * Without this version of the macro, MEMBER_VPTR() will use the compile time
200  * size of the array to compute the max, which will result in rejection by
201  * the verifier.
202  */
203 #define ARRAY_ELEM_PTR(arr, i, n) (typeof(arr[i]) *)				\
204 ({										\
205 	u64 __base = (u64)arr;							\
206 	u64 __addr = (u64)&(arr[i]) - __base;					\
207 	asm volatile (								\
208 		"if %0 <= %[max] goto +2\n"					\
209 		"%0 = 0\n"							\
210 		"goto +1\n"							\
211 		"%0 += %1\n"							\
212 		: "+r"(__addr)							\
213 		: "r"(__base),							\
214 		  [max]"r"(sizeof(arr[0]) * ((n) - 1)));			\
215 	__addr;									\
216 })
217 
218 
219 /*
220  * BPF declarations and helpers
221  */
222 
223 /* list and rbtree */
224 #define __contains(name, node) __attribute__((btf_decl_tag("contains:" #name ":" #node)))
225 #define private(name) SEC(".data." #name) __hidden __attribute__((aligned(8)))
226 
227 void *bpf_obj_new_impl(__u64 local_type_id, void *meta) __ksym;
228 void bpf_obj_drop_impl(void *kptr, void *meta) __ksym;
229 
230 #define bpf_obj_new(type) ((type *)bpf_obj_new_impl(bpf_core_type_id_local(type), NULL))
231 #define bpf_obj_drop(kptr) bpf_obj_drop_impl(kptr, NULL)
232 
233 void bpf_list_push_front(struct bpf_list_head *head, struct bpf_list_node *node) __ksym;
234 void bpf_list_push_back(struct bpf_list_head *head, struct bpf_list_node *node) __ksym;
235 struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head) __ksym;
236 struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head) __ksym;
237 struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root,
238 				      struct bpf_rb_node *node) __ksym;
239 int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node,
240 			bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b),
241 			void *meta, __u64 off) __ksym;
242 #define bpf_rbtree_add(head, node, less) bpf_rbtree_add_impl(head, node, less, NULL, 0)
243 
244 struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root) __ksym;
245 
246 void *bpf_refcount_acquire_impl(void *kptr, void *meta) __ksym;
247 #define bpf_refcount_acquire(kptr) bpf_refcount_acquire_impl(kptr, NULL)
248 
249 /* task */
250 struct task_struct *bpf_task_from_pid(s32 pid) __ksym;
251 struct task_struct *bpf_task_acquire(struct task_struct *p) __ksym;
252 void bpf_task_release(struct task_struct *p) __ksym;
253 
254 /* cgroup */
255 struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level) __ksym;
256 void bpf_cgroup_release(struct cgroup *cgrp) __ksym;
257 struct cgroup *bpf_cgroup_from_id(u64 cgid) __ksym;
258 
259 /* css iteration */
260 struct bpf_iter_css;
261 struct cgroup_subsys_state;
262 extern int bpf_iter_css_new(struct bpf_iter_css *it,
263 			    struct cgroup_subsys_state *start,
264 			    unsigned int flags) __weak __ksym;
265 extern struct cgroup_subsys_state *
266 bpf_iter_css_next(struct bpf_iter_css *it) __weak __ksym;
267 extern void bpf_iter_css_destroy(struct bpf_iter_css *it) __weak __ksym;
268 
269 /* cpumask */
270 struct bpf_cpumask *bpf_cpumask_create(void) __ksym;
271 struct bpf_cpumask *bpf_cpumask_acquire(struct bpf_cpumask *cpumask) __ksym;
272 void bpf_cpumask_release(struct bpf_cpumask *cpumask) __ksym;
273 u32 bpf_cpumask_first(const struct cpumask *cpumask) __ksym;
274 u32 bpf_cpumask_first_zero(const struct cpumask *cpumask) __ksym;
275 void bpf_cpumask_set_cpu(u32 cpu, struct bpf_cpumask *cpumask) __ksym;
276 void bpf_cpumask_clear_cpu(u32 cpu, struct bpf_cpumask *cpumask) __ksym;
277 bool bpf_cpumask_test_cpu(u32 cpu, const struct cpumask *cpumask) __ksym;
278 bool bpf_cpumask_test_and_set_cpu(u32 cpu, struct bpf_cpumask *cpumask) __ksym;
279 bool bpf_cpumask_test_and_clear_cpu(u32 cpu, struct bpf_cpumask *cpumask) __ksym;
280 void bpf_cpumask_setall(struct bpf_cpumask *cpumask) __ksym;
281 void bpf_cpumask_clear(struct bpf_cpumask *cpumask) __ksym;
282 bool bpf_cpumask_and(struct bpf_cpumask *dst, const struct cpumask *src1,
283 		     const struct cpumask *src2) __ksym;
284 void bpf_cpumask_or(struct bpf_cpumask *dst, const struct cpumask *src1,
285 		    const struct cpumask *src2) __ksym;
286 void bpf_cpumask_xor(struct bpf_cpumask *dst, const struct cpumask *src1,
287 		     const struct cpumask *src2) __ksym;
288 bool bpf_cpumask_equal(const struct cpumask *src1, const struct cpumask *src2) __ksym;
289 bool bpf_cpumask_intersects(const struct cpumask *src1, const struct cpumask *src2) __ksym;
290 bool bpf_cpumask_subset(const struct cpumask *src1, const struct cpumask *src2) __ksym;
291 bool bpf_cpumask_empty(const struct cpumask *cpumask) __ksym;
292 bool bpf_cpumask_full(const struct cpumask *cpumask) __ksym;
293 void bpf_cpumask_copy(struct bpf_cpumask *dst, const struct cpumask *src) __ksym;
294 u32 bpf_cpumask_any_distribute(const struct cpumask *cpumask) __ksym;
295 u32 bpf_cpumask_any_and_distribute(const struct cpumask *src1,
296 				   const struct cpumask *src2) __ksym;
297 
298 /* rcu */
299 void bpf_rcu_read_lock(void) __ksym;
300 void bpf_rcu_read_unlock(void) __ksym;
301 
302 
303 /*
304  * Other helpers
305  */
306 
307 /* useful compiler attributes */
308 #define likely(x) __builtin_expect(!!(x), 1)
309 #define unlikely(x) __builtin_expect(!!(x), 0)
310 #define __maybe_unused __attribute__((__unused__))
311 
312 /*
313  * READ/WRITE_ONCE() are from kernel (include/asm-generic/rwonce.h). They
314  * prevent compiler from caching, redoing or reordering reads or writes.
315  */
316 typedef __u8  __attribute__((__may_alias__))  __u8_alias_t;
317 typedef __u16 __attribute__((__may_alias__)) __u16_alias_t;
318 typedef __u32 __attribute__((__may_alias__)) __u32_alias_t;
319 typedef __u64 __attribute__((__may_alias__)) __u64_alias_t;
320 
321 static __always_inline void __read_once_size(const volatile void *p, void *res, int size)
322 {
323 	switch (size) {
324 	case 1: *(__u8_alias_t  *) res = *(volatile __u8_alias_t  *) p; break;
325 	case 2: *(__u16_alias_t *) res = *(volatile __u16_alias_t *) p; break;
326 	case 4: *(__u32_alias_t *) res = *(volatile __u32_alias_t *) p; break;
327 	case 8: *(__u64_alias_t *) res = *(volatile __u64_alias_t *) p; break;
328 	default:
329 		barrier();
330 		__builtin_memcpy((void *)res, (const void *)p, size);
331 		barrier();
332 	}
333 }
334 
335 static __always_inline void __write_once_size(volatile void *p, void *res, int size)
336 {
337 	switch (size) {
338 	case 1: *(volatile  __u8_alias_t *) p = *(__u8_alias_t  *) res; break;
339 	case 2: *(volatile __u16_alias_t *) p = *(__u16_alias_t *) res; break;
340 	case 4: *(volatile __u32_alias_t *) p = *(__u32_alias_t *) res; break;
341 	case 8: *(volatile __u64_alias_t *) p = *(__u64_alias_t *) res; break;
342 	default:
343 		barrier();
344 		__builtin_memcpy((void *)p, (const void *)res, size);
345 		barrier();
346 	}
347 }
348 
349 #define READ_ONCE(x)					\
350 ({							\
351 	union { typeof(x) __val; char __c[1]; } __u =	\
352 		{ .__c = { 0 } };			\
353 	__read_once_size(&(x), __u.__c, sizeof(x));	\
354 	__u.__val;					\
355 })
356 
357 #define WRITE_ONCE(x, val)				\
358 ({							\
359 	union { typeof(x) __val; char __c[1]; } __u =	\
360 		{ .__val = (val) }; 			\
361 	__write_once_size(&(x), __u.__c, sizeof(x));	\
362 	__u.__val;					\
363 })
364 
365 /*
366  * log2_u32 - Compute the base 2 logarithm of a 32-bit exponential value.
367  * @v: The value for which we're computing the base 2 logarithm.
368  */
369 static inline u32 log2_u32(u32 v)
370 {
371         u32 r;
372         u32 shift;
373 
374         r = (v > 0xFFFF) << 4; v >>= r;
375         shift = (v > 0xFF) << 3; v >>= shift; r |= shift;
376         shift = (v > 0xF) << 2; v >>= shift; r |= shift;
377         shift = (v > 0x3) << 1; v >>= shift; r |= shift;
378         r |= (v >> 1);
379         return r;
380 }
381 
382 /*
383  * log2_u64 - Compute the base 2 logarithm of a 64-bit exponential value.
384  * @v: The value for which we're computing the base 2 logarithm.
385  */
386 static inline u32 log2_u64(u64 v)
387 {
388         u32 hi = v >> 32;
389         if (hi)
390                 return log2_u32(hi) + 32 + 1;
391         else
392                 return log2_u32(v) + 1;
393 }
394 
395 #include "compat.bpf.h"
396 
397 #endif	/* __SCX_COMMON_BPF_H */
398