linux/tools/testing/selftests/bpf/progs/pyperf.h

359 lines
8.5 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
// Copyright (c) 2019 Facebook
#include <linux/sched.h>
#include <linux/ptrace.h>
#include <stdint.h>
#include <stddef.h>
#include <stdbool.h>
#include <linux/bpf.h>
#include <bpf/bpf_helpers.h>
selftests/bpf: add bpf_for_each(), bpf_for(), and bpf_repeat() macros Add bpf_for_each(), bpf_for(), and bpf_repeat() macros that make writing open-coded iterator-based loops much more convenient and natural. These macros utilize cleanup attribute to ensure proper destruction of the iterator and thanks to that manage to provide the ergonomics that is very close to C language's for() construct. Typical loop would look like: int i; int arr[N]; bpf_for(i, 0, N) { /* verifier will know that i >= 0 && i < N, so could be used to * directly access array elements with no extra checks */ arr[i] = i; } bpf_repeat() is very similar, but it doesn't expose iteration number and is meant as a simple "repeat action N times" loop: bpf_repeat(N) { /* whatever, N times */ } Note that `break` and `continue` statements inside the {} block work as expected. bpf_for_each() is a generalization over any kind of BPF open-coded iterator allowing to use for-each-like approach instead of calling low-level bpf_iter_<type>_{new,next,destroy}() APIs explicitly. E.g.: struct cgroup *cg; bpf_for_each(cgroup, cg, some, input, args) { /* do something with each cg */ } would call (not-yet-implemented) bpf_iter_cgroup_{new,next,destroy}() functions to form a loop over cgroups, where `some, input, args` are passed verbatim into constructor as bpf_iter_cgroup_new(&it, some, input, args). As a first demonstration, add pyperf variant based on the bpf_for() loop. Also clean up a few tests that either included bpf_misc.h header unnecessarily from the user-space, which is unsupported, or included it before any common types are defined (and thus leading to unnecessary compilation warnings, potentially). Signed-off-by: Andrii Nakryiko <andrii@kernel.org> Link: https://lore.kernel.org/r/20230308184121.1165081-6-andrii@kernel.org Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 10:41:18 -08:00
#include "bpf_misc.h"
#include "bpf_compiler.h"
#define FUNCTION_NAME_LEN 64
#define FILE_NAME_LEN 128
#define TASK_COMM_LEN 16
typedef struct {
int PyThreadState_frame;
int PyThreadState_thread;
int PyFrameObject_back;
int PyFrameObject_code;
int PyFrameObject_lineno;
int PyCodeObject_filename;
int PyCodeObject_name;
int String_data;
int String_size;
} OffsetConfig;
typedef struct {
uintptr_t current_state_addr;
uintptr_t tls_key_addr;
OffsetConfig offsets;
bool use_tls;
} PidData;
typedef struct {
uint32_t success;
} Stats;
typedef struct {
char name[FUNCTION_NAME_LEN];
char file[FILE_NAME_LEN];
} Symbol;
typedef struct {
uint32_t pid;
uint32_t tid;
char comm[TASK_COMM_LEN];
int32_t kernel_stack_id;
int32_t user_stack_id;
bool thread_current;
bool pthread_match;
bool stack_complete;
int16_t stack_len;
int32_t stack[STACK_MAX_LEN];
int has_meta;
int metadata;
char dummy_safeguard;
} Event;
typedef int pid_t;
typedef struct {
void* f_back; // PyFrameObject.f_back, previous frame
void* f_code; // PyFrameObject.f_code, pointer to PyCodeObject
void* co_filename; // PyCodeObject.co_filename
void* co_name; // PyCodeObject.co_name
} FrameData;
#ifdef SUBPROGS
__noinline
#else
__always_inline
#endif
static void *get_thread_state(void *tls_base, PidData *pidData)
{
void* thread_state;
int key;
bpf_probe_read_user(&key, sizeof(key), (void*)(long)pidData->tls_key_addr);
bpf_probe_read_user(&thread_state, sizeof(thread_state),
tls_base + 0x310 + key * 0x10 + 0x08);
return thread_state;
}
static __always_inline bool get_frame_data(void *frame_ptr, PidData *pidData,
FrameData *frame, Symbol *symbol)
{
// read data from PyFrameObject
bpf_probe_read_user(&frame->f_back,
sizeof(frame->f_back),
frame_ptr + pidData->offsets.PyFrameObject_back);
bpf_probe_read_user(&frame->f_code,
sizeof(frame->f_code),
frame_ptr + pidData->offsets.PyFrameObject_code);
// read data from PyCodeObject
if (!frame->f_code)
return false;
bpf_probe_read_user(&frame->co_filename,
sizeof(frame->co_filename),
frame->f_code + pidData->offsets.PyCodeObject_filename);
bpf_probe_read_user(&frame->co_name,
sizeof(frame->co_name),
frame->f_code + pidData->offsets.PyCodeObject_name);
// read actual names into symbol
if (frame->co_filename)
bpf_probe_read_user_str(&symbol->file,
sizeof(symbol->file),
frame->co_filename +
pidData->offsets.String_data);
if (frame->co_name)
bpf_probe_read_user_str(&symbol->name,
sizeof(symbol->name),
frame->co_name +
pidData->offsets.String_data);
return true;
}
struct {
__uint(type, BPF_MAP_TYPE_HASH);
__uint(max_entries, 1);
__type(key, int);
__type(value, PidData);
} pidmap SEC(".maps");
struct {
__uint(type, BPF_MAP_TYPE_HASH);
__uint(max_entries, 1);
__type(key, int);
__type(value, Event);
} eventmap SEC(".maps");
struct {
__uint(type, BPF_MAP_TYPE_HASH);
__uint(max_entries, 1);
__type(key, Symbol);
__type(value, int);
} symbolmap SEC(".maps");
struct {
__uint(type, BPF_MAP_TYPE_ARRAY);
__uint(max_entries, 1);
__type(key, int);
__type(value, Stats);
} statsmap SEC(".maps");
struct {
__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
__uint(max_entries, 32);
__uint(key_size, sizeof(int));
__uint(value_size, sizeof(int));
} perfmap SEC(".maps");
struct {
__uint(type, BPF_MAP_TYPE_STACK_TRACE);
__uint(max_entries, 1000);
__uint(key_size, sizeof(int));
__uint(value_size, sizeof(long long) * 127);
} stackmap SEC(".maps");
#ifdef USE_BPF_LOOP
struct process_frame_ctx {
int cur_cpu;
int32_t *symbol_counter;
void *frame_ptr;
FrameData *frame;
PidData *pidData;
Symbol *sym;
Event *event;
bool done;
};
static int process_frame_callback(__u32 i, struct process_frame_ctx *ctx)
{
int zero = 0;
void *frame_ptr = ctx->frame_ptr;
PidData *pidData = ctx->pidData;
FrameData *frame = ctx->frame;
int32_t *symbol_counter = ctx->symbol_counter;
int cur_cpu = ctx->cur_cpu;
Event *event = ctx->event;
Symbol *sym = ctx->sym;
if (frame_ptr && get_frame_data(frame_ptr, pidData, frame, sym)) {
int32_t new_symbol_id = *symbol_counter * 64 + cur_cpu;
int32_t *symbol_id = bpf_map_lookup_elem(&symbolmap, sym);
if (!symbol_id) {
bpf_map_update_elem(&symbolmap, sym, &zero, 0);
symbol_id = bpf_map_lookup_elem(&symbolmap, sym);
if (!symbol_id) {
ctx->done = true;
return 1;
}
}
if (*symbol_id == new_symbol_id)
(*symbol_counter)++;
barrier_var(i);
if (i >= STACK_MAX_LEN)
return 1;
event->stack[i] = *symbol_id;
event->stack_len = i + 1;
frame_ptr = frame->f_back;
}
return 0;
}
#endif /* USE_BPF_LOOP */
#ifdef GLOBAL_FUNC
__noinline
#elif defined(SUBPROGS)
static __noinline
#else
static __always_inline
#endif
int __on_event(struct bpf_raw_tracepoint_args *ctx)
{
uint64_t pid_tgid = bpf_get_current_pid_tgid();
pid_t pid = (pid_t)(pid_tgid >> 32);
PidData* pidData = bpf_map_lookup_elem(&pidmap, &pid);
if (!pidData)
return 0;
int zero = 0;
Event* event = bpf_map_lookup_elem(&eventmap, &zero);
if (!event)
return 0;
event->pid = pid;
event->tid = (pid_t)pid_tgid;
bpf_get_current_comm(&event->comm, sizeof(event->comm));
event->user_stack_id = bpf_get_stackid(ctx, &stackmap, BPF_F_USER_STACK);
event->kernel_stack_id = bpf_get_stackid(ctx, &stackmap, 0);
void* thread_state_current = (void*)0;
bpf_probe_read_user(&thread_state_current,
sizeof(thread_state_current),
(void*)(long)pidData->current_state_addr);
struct task_struct* task = (struct task_struct*)bpf_get_current_task();
void* tls_base = (void*)task;
void* thread_state = pidData->use_tls ? get_thread_state(tls_base, pidData)
: thread_state_current;
event->thread_current = thread_state == thread_state_current;
if (pidData->use_tls) {
uint64_t pthread_created;
uint64_t pthread_self;
bpf_probe_read_user(&pthread_self, sizeof(pthread_self),
tls_base + 0x10);
bpf_probe_read_user(&pthread_created,
sizeof(pthread_created),
thread_state +
pidData->offsets.PyThreadState_thread);
event->pthread_match = pthread_created == pthread_self;
} else {
event->pthread_match = 1;
}
if (event->pthread_match || !pidData->use_tls) {
void* frame_ptr;
FrameData frame;
Symbol sym = {};
int cur_cpu = bpf_get_smp_processor_id();
bpf_probe_read_user(&frame_ptr,
sizeof(frame_ptr),
thread_state +
pidData->offsets.PyThreadState_frame);
int32_t* symbol_counter = bpf_map_lookup_elem(&symbolmap, &sym);
if (symbol_counter == NULL)
return 0;
#ifdef USE_BPF_LOOP
struct process_frame_ctx ctx = {
.cur_cpu = cur_cpu,
.symbol_counter = symbol_counter,
.frame_ptr = frame_ptr,
.frame = &frame,
.pidData = pidData,
.sym = &sym,
.event = event,
};
bpf_loop(STACK_MAX_LEN, process_frame_callback, &ctx, 0);
if (ctx.done)
return 0;
#else
selftests/bpf: add bpf_for_each(), bpf_for(), and bpf_repeat() macros Add bpf_for_each(), bpf_for(), and bpf_repeat() macros that make writing open-coded iterator-based loops much more convenient and natural. These macros utilize cleanup attribute to ensure proper destruction of the iterator and thanks to that manage to provide the ergonomics that is very close to C language's for() construct. Typical loop would look like: int i; int arr[N]; bpf_for(i, 0, N) { /* verifier will know that i >= 0 && i < N, so could be used to * directly access array elements with no extra checks */ arr[i] = i; } bpf_repeat() is very similar, but it doesn't expose iteration number and is meant as a simple "repeat action N times" loop: bpf_repeat(N) { /* whatever, N times */ } Note that `break` and `continue` statements inside the {} block work as expected. bpf_for_each() is a generalization over any kind of BPF open-coded iterator allowing to use for-each-like approach instead of calling low-level bpf_iter_<type>_{new,next,destroy}() APIs explicitly. E.g.: struct cgroup *cg; bpf_for_each(cgroup, cg, some, input, args) { /* do something with each cg */ } would call (not-yet-implemented) bpf_iter_cgroup_{new,next,destroy}() functions to form a loop over cgroups, where `some, input, args` are passed verbatim into constructor as bpf_iter_cgroup_new(&it, some, input, args). As a first demonstration, add pyperf variant based on the bpf_for() loop. Also clean up a few tests that either included bpf_misc.h header unnecessarily from the user-space, which is unsupported, or included it before any common types are defined (and thus leading to unnecessary compilation warnings, potentially). Signed-off-by: Andrii Nakryiko <andrii@kernel.org> Link: https://lore.kernel.org/r/20230308184121.1165081-6-andrii@kernel.org Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 10:41:18 -08:00
#if defined(USE_ITER)
/* no for loop, no unrolling */
#elif defined(NO_UNROLL)
__pragma_loop_no_unroll
selftests/bpf: add bpf_for_each(), bpf_for(), and bpf_repeat() macros Add bpf_for_each(), bpf_for(), and bpf_repeat() macros that make writing open-coded iterator-based loops much more convenient and natural. These macros utilize cleanup attribute to ensure proper destruction of the iterator and thanks to that manage to provide the ergonomics that is very close to C language's for() construct. Typical loop would look like: int i; int arr[N]; bpf_for(i, 0, N) { /* verifier will know that i >= 0 && i < N, so could be used to * directly access array elements with no extra checks */ arr[i] = i; } bpf_repeat() is very similar, but it doesn't expose iteration number and is meant as a simple "repeat action N times" loop: bpf_repeat(N) { /* whatever, N times */ } Note that `break` and `continue` statements inside the {} block work as expected. bpf_for_each() is a generalization over any kind of BPF open-coded iterator allowing to use for-each-like approach instead of calling low-level bpf_iter_<type>_{new,next,destroy}() APIs explicitly. E.g.: struct cgroup *cg; bpf_for_each(cgroup, cg, some, input, args) { /* do something with each cg */ } would call (not-yet-implemented) bpf_iter_cgroup_{new,next,destroy}() functions to form a loop over cgroups, where `some, input, args` are passed verbatim into constructor as bpf_iter_cgroup_new(&it, some, input, args). As a first demonstration, add pyperf variant based on the bpf_for() loop. Also clean up a few tests that either included bpf_misc.h header unnecessarily from the user-space, which is unsupported, or included it before any common types are defined (and thus leading to unnecessary compilation warnings, potentially). Signed-off-by: Andrii Nakryiko <andrii@kernel.org> Link: https://lore.kernel.org/r/20230308184121.1165081-6-andrii@kernel.org Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 10:41:18 -08:00
#elif defined(UNROLL_COUNT)
__pragma_loop_unroll_count(UNROLL_COUNT)
selftests/bpf: Limit unroll_count for pyperf600 test LLVM commit [1] changed loop pragma behavior such that full loop unroll is always honored with user pragma. Previously, unroll count also depends on the unrolled code size. For pyperf600, without [1], the loop unroll count is 150. With [1], the loop unroll count is 600. The unroll count of 600 caused the program size close to 298k and this caused the following code is generated: 0: 7b 1a 00 ff 00 00 00 00 *(u64 *)(r10 - 256) = r1 ; uint64_t pid_tgid = bpf_get_current_pid_tgid(); 1: 85 00 00 00 0e 00 00 00 call 14 2: bf 06 00 00 00 00 00 00 r6 = r0 ; pid_t pid = (pid_t)(pid_tgid >> 32); 3: bf 61 00 00 00 00 00 00 r1 = r6 4: 77 01 00 00 20 00 00 00 r1 >>= 32 5: 63 1a fc ff 00 00 00 00 *(u32 *)(r10 - 4) = r1 6: bf a2 00 00 00 00 00 00 r2 = r10 7: 07 02 00 00 fc ff ff ff r2 += -4 ; PidData* pidData = bpf_map_lookup_elem(&pidmap, &pid); 8: 18 01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 r1 = 0 ll 10: 85 00 00 00 01 00 00 00 call 1 11: bf 08 00 00 00 00 00 00 r8 = r0 ; if (!pidData) 12: 15 08 15 e8 00 00 00 00 if r8 == 0 goto -6123 <LBB0_27588+0xffffffffffdae100> Note that insn 12 has a branch offset -6123 which is clearly illegal and will be rejected by the verifier. The negative offset is due to the branch range is greater than INT16_MAX. This patch changed the unroll count to be 150 to avoid above branch target insn out-of-range issue. Also the llvm is enhanced ([2]) to assert if the branch target insn is out of INT16 range. [1] https://reviews.llvm.org/D119148 [2] https://reviews.llvm.org/D123877 Signed-off-by: Yonghong Song <yhs@fb.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Link: https://lore.kernel.org/bpf/20220419043230.2928530-1-yhs@fb.com
2022-04-18 21:32:30 -07:00
#else
__pragma_loop_unroll_full
#endif /* NO_UNROLL */
/* Unwind python stack */
selftests/bpf: add bpf_for_each(), bpf_for(), and bpf_repeat() macros Add bpf_for_each(), bpf_for(), and bpf_repeat() macros that make writing open-coded iterator-based loops much more convenient and natural. These macros utilize cleanup attribute to ensure proper destruction of the iterator and thanks to that manage to provide the ergonomics that is very close to C language's for() construct. Typical loop would look like: int i; int arr[N]; bpf_for(i, 0, N) { /* verifier will know that i >= 0 && i < N, so could be used to * directly access array elements with no extra checks */ arr[i] = i; } bpf_repeat() is very similar, but it doesn't expose iteration number and is meant as a simple "repeat action N times" loop: bpf_repeat(N) { /* whatever, N times */ } Note that `break` and `continue` statements inside the {} block work as expected. bpf_for_each() is a generalization over any kind of BPF open-coded iterator allowing to use for-each-like approach instead of calling low-level bpf_iter_<type>_{new,next,destroy}() APIs explicitly. E.g.: struct cgroup *cg; bpf_for_each(cgroup, cg, some, input, args) { /* do something with each cg */ } would call (not-yet-implemented) bpf_iter_cgroup_{new,next,destroy}() functions to form a loop over cgroups, where `some, input, args` are passed verbatim into constructor as bpf_iter_cgroup_new(&it, some, input, args). As a first demonstration, add pyperf variant based on the bpf_for() loop. Also clean up a few tests that either included bpf_misc.h header unnecessarily from the user-space, which is unsupported, or included it before any common types are defined (and thus leading to unnecessary compilation warnings, potentially). Signed-off-by: Andrii Nakryiko <andrii@kernel.org> Link: https://lore.kernel.org/r/20230308184121.1165081-6-andrii@kernel.org Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 10:41:18 -08:00
#ifdef USE_ITER
int i;
bpf_for(i, 0, STACK_MAX_LEN) {
#else /* !USE_ITER */
for (int i = 0; i < STACK_MAX_LEN; ++i) {
selftests/bpf: add bpf_for_each(), bpf_for(), and bpf_repeat() macros Add bpf_for_each(), bpf_for(), and bpf_repeat() macros that make writing open-coded iterator-based loops much more convenient and natural. These macros utilize cleanup attribute to ensure proper destruction of the iterator and thanks to that manage to provide the ergonomics that is very close to C language's for() construct. Typical loop would look like: int i; int arr[N]; bpf_for(i, 0, N) { /* verifier will know that i >= 0 && i < N, so could be used to * directly access array elements with no extra checks */ arr[i] = i; } bpf_repeat() is very similar, but it doesn't expose iteration number and is meant as a simple "repeat action N times" loop: bpf_repeat(N) { /* whatever, N times */ } Note that `break` and `continue` statements inside the {} block work as expected. bpf_for_each() is a generalization over any kind of BPF open-coded iterator allowing to use for-each-like approach instead of calling low-level bpf_iter_<type>_{new,next,destroy}() APIs explicitly. E.g.: struct cgroup *cg; bpf_for_each(cgroup, cg, some, input, args) { /* do something with each cg */ } would call (not-yet-implemented) bpf_iter_cgroup_{new,next,destroy}() functions to form a loop over cgroups, where `some, input, args` are passed verbatim into constructor as bpf_iter_cgroup_new(&it, some, input, args). As a first demonstration, add pyperf variant based on the bpf_for() loop. Also clean up a few tests that either included bpf_misc.h header unnecessarily from the user-space, which is unsupported, or included it before any common types are defined (and thus leading to unnecessary compilation warnings, potentially). Signed-off-by: Andrii Nakryiko <andrii@kernel.org> Link: https://lore.kernel.org/r/20230308184121.1165081-6-andrii@kernel.org Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-03-08 10:41:18 -08:00
#endif
if (frame_ptr && get_frame_data(frame_ptr, pidData, &frame, &sym)) {
int32_t new_symbol_id = *symbol_counter * 64 + cur_cpu;
int32_t *symbol_id = bpf_map_lookup_elem(&symbolmap, &sym);
if (!symbol_id) {
bpf_map_update_elem(&symbolmap, &sym, &zero, 0);
symbol_id = bpf_map_lookup_elem(&symbolmap, &sym);
if (!symbol_id)
return 0;
}
if (*symbol_id == new_symbol_id)
(*symbol_counter)++;
event->stack[i] = *symbol_id;
event->stack_len = i + 1;
frame_ptr = frame.f_back;
}
}
#endif /* USE_BPF_LOOP */
event->stack_complete = frame_ptr == NULL;
} else {
event->stack_complete = 1;
}
Stats* stats = bpf_map_lookup_elem(&statsmap, &zero);
if (stats)
stats->success++;
event->has_meta = 0;
bpf_perf_event_output(ctx, &perfmap, 0, event, offsetof(Event, metadata));
return 0;
}
SEC("raw_tracepoint/kfree_skb")
int on_event(struct bpf_raw_tracepoint_args* ctx)
{
int ret = 0;
ret |= __on_event(ctx);
ret |= __on_event(ctx);
ret |= __on_event(ctx);
ret |= __on_event(ctx);
ret |= __on_event(ctx);
return ret;
}
char _license[] SEC("license") = "GPL";