libbpf-tools: add CO-RE profile (#3782)

It is based on Brendan Gregg's BCC profile and much of the code for it is borrowed from offcputime.c and ruqlen.c.
This is an example of usage.

# ./profile -iK
Sampling at 49 Hertz of all threads by kernel stack... Hit Ctrl-C to end.
^C    
    _raw_spin_unlock_irq
    finish_task_switch
    __schedule
    schedule
    schedule_timeout
    msleep
    HDMI21_Tx_EARC_MainThread
    kthread
    ret_from_fork
    -                hdmi21_earc_eng (1427)
        1

    __pi___clean_dcache_area_poc
    dma_direct_map_page
    dma_direct_map_sg
    dma_buf_unified_map
    dma_buf_map_attachment
    kbase_mem_umm_map_attachment
    kbase_mem_import
    kbase_ioctl
    __arm64_compat_sys_ioctl
    el0_svc_common.constprop.0
    el0_svc_compat_handler
    el0_svc_compat
    -                surface-manager (1911)
        4

    __pi___clean_dcache_area_poc
    dma_direct_map_page
    dma_direct_map_sg
    dma_buf_unified_map
    dma_buf_map_attachment
    kbase_mem_umm_map_attachment
    kbase_mem_import
    kbase_ioctl
    __arm64_compat_sys_ioctl
    el0_svc_common.constprop.0
    el0_svc_compat_handler
    el0_svc_compat
    -                tAiToneService (2141)
        12

    [Missed Kernel Stack]
    -                WebAppMgr (3624)
        1

...

This is the result of using the -f (folded) option.

# ./profile -fdK
...
tCMState;low_mem_notify_threshold;__alloc_pages_nodemask;__get_free_pages;__pollwait;unix_poll;sock_poll;do_sys_poll;__arm64_sys_poll;el0_svc_common.constprop.0;el0_svc_compat_handler;el0_svc_compat 2
surface-manager;sock_poll;do_sys_poll;__arm64_sys_poll;el0_svc_common.constprop.0;el0_svc_compat_handler;el0_svc_compat 1
qml-runner;sock_poll;do_sys_poll;__arm64_sys_poll;el0_svc_common.constprop.0;el0_svc_compat_handler;el0_svc_compat 96
mali-cmar-backe;_raw_spin_unlock_irqrestore;kbase_pm_update_cores_state;kbase_pm_do_poweron;kbase_pm_update_active;kbase_hwaccess_pm_gpu_active;kbase_pm_context_active_handle_suspend;kbase_js_sched;kbase_jd_submit;kbase_ioctl;__arm64_compat_sys_ioctl;el0_svc_common.constprop.0;el0_svc_compat_handler;el0_svc_compat 2
surface-manager;__fget;__fget_light;__fdget;sockfd_lookup_light;__sys_sendmsg;__arm64_compat_sys_sendmsg;el0_svc_common.constprop.0;el0_svc_compat_handler;el0_svc_compat 1
...
WARNING: 4 stack traces could not be displayed. Consider increasing --stack-storage-size.

Signed-off-by: Eunseon Lee [email protected]

libbpf-tools/.gitignore

@@ -43,6 +43,7 @@
 /offcputime
 /oomkill
 /opensnoop
+/profile
 /readahead
 /runqlat
 /runqlen

libbpf-tools/Makefile

@@ -70,6 +70,7 @@ APPS = \
 	numamove \
 	offcputime \
 	oomkill \
+	profile \
 	readahead \
 	runqlat \
 	runqlen \

libbpf-tools/profile.bpf.c

@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
+/*
+ * Copyright (c) 2022 LG Electronics
+ *
+ * Based on profile from BCC by Brendan Gregg and others.
+ * 28-Dec-2021   Eunseon Lee   Created this.
+ */
+#include <vmlinux.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_core_read.h>
+#include <bpf/bpf_tracing.h>
+#include "profile.h"
+#include "maps.bpf.h"
+
+const volatile bool kernel_stacks_only = false;
+const volatile bool user_stacks_only = false;
+const volatile bool include_idle = false;
+const volatile pid_t targ_pid = -1;
+const volatile pid_t targ_tid = -1;
+
+struct {
+	__uint(type, BPF_MAP_TYPE_STACK_TRACE);
+	__type(key, u32);
+} stackmap SEC(".maps");
+
+struct {
+	__uint(type, BPF_MAP_TYPE_HASH);
+	__type(key, struct key_t);
+	__type(value, u64);
+	__uint(max_entries, MAX_ENTRIES);
+} counts SEC(".maps");
+
+SEC("perf_event")
+int do_perf_event(struct bpf_perf_event_data *ctx)
+{
+	u64 id = bpf_get_current_pid_tgid();
+	u32 pid = id >> 32;
+	u32 tid = id;
+	u64 *valp;
+	static const u64 zero;
+	struct key_t key = {};
+
+	if (!include_idle && tid == 0)
+		return 0;
+
+	if (targ_pid != -1 && targ_pid != pid)
+		return 0;
+
+	if (targ_tid != -1 && targ_tid != tid)
+		return 0;
+
+	key.pid = pid;
+	bpf_get_current_comm(&key.name, sizeof(key.name));
+
+	if (user_stacks_only)
+		key.kern_stack_id = -1;
+	else
+		key.kern_stack_id = bpf_get_stackid(&ctx->regs, &stackmap, 0);
+
+	if (kernel_stacks_only)
+		key.user_stack_id = -1;
+	else
+		key.user_stack_id = bpf_get_stackid(&ctx->regs, &stackmap,
+						    BPF_F_USER_STACK);
+
+	valp = bpf_map_lookup_or_try_init(&counts, &key, &zero);
+	if (valp)
+		__sync_fetch_and_add(valp, 1);
+
+	return 0;
+}
+
+char LICENSE[] SEC("license") = "GPL";