event trace 跟踪 schedule switch的例子

1.背景

最近debug时，需要trace schedule switch的前后进程，所以写了一个event trace的程序。

2.代码

#include <linux/tracepoint.h>
#include <trace/events/sched.h>
#include <linux/memblock.h>
#include <linux/pfn.h>
#include <linux/err.h>
#include <linux/module.h>
#include <linux/io.h>
#include <asm/memory.h>
#include <linux/slab.h>
#include <linux/sched/task_stack.h>
#include <linux/stacktrace.h>

#define MAX_STACK_TRACE_DEPTH 64
#define SKIP_CHECKED_PFN

static unsigned long memory_monitor_addr = 0;
static unsigned long memory_monitor_size = 0;

static unsigned long checked_cnt = 0;
static unsigned long backup_stacktrace[16][MAX_STACK_TRACE_DEPTH];
#ifdef SKIP_CHECKED_PFN
static unsigned long skip_pfn[16];
#endif

static int _kprint_stack(struct task_struct *task, int saved)
{
    struct stack_trace trace;
    unsigned long *entries;
    int err;

    entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
    if (!entries)
        return –ENOMEM;

    trace.nr_entries = 0;
    trace.max_entries = MAX_STACK_TRACE_DEPTH;
    trace.entries = entries;
    trace.skip = 0;
    if (!err) {
        unsigned int i;
        save_stack_trace_tsk(task,&trace);
        for (i = 0; i < trace.nr_entries; i ++) {
            pr_cont(“[<%px>] %pS
”, (void*)entries[i], (void*)entries[i]);
            if (saved)
                backup_stacktrace[checked_cnt][i] = entries[i];
               }
}
kfree(entries);
return err;
}

static unsigned long end_pfn = 0;
static unsigned long start_pfn = 0;
static void sched_switch_probe(void *data, bool preempt, struct task_struct*prev, struct task_struct *next)
{
    void *va_pos = NULL;
    unsigned long cur_pfn;
    unsigned long mem_data;
#ifdef SKIP_CHECKED_PFN
    int i = 0;
    int skip = 0;
#endif

    if (start_pfn == 0 || end_pfn == 0)
        return;

    for (cur_pfn = start_pfn; cur_pfn < end_pfn; cur_pfn ++) {
#ifdef SKIP_CHECKED_PFN
        skip = 0;
        for (i = 0; i <= checked_cnt; i ++) {
            if (cur_pfn == skip_pfn[i]) {
                skip = 1;
                break;    
            }
               }
               if (skip)
            continue;
#endif
        va_pos = __va(PFN_PHYS(cur_pfn) + 0x520);
        mem_data = *(uint64_t*)va_pos >> 40;
        if (mem_data==0xb0ffff || mem_data==0xb0cccc || mem_data==0xb00000){
            pr_emerg(“address[%px]=%lx, CPU=%d
”, __pa(va_pos),*(uint64_t*)va_pos, smp_processor_id());
            pr_emerg(“----current process stack----
”);
            _kprint_stack(prev, 0);
            pr_emerg(“----next process stack----
”);
            _kprint_stack(next,0);
            dump_stack();    
        }
}
return;
}

static int __init get_memory_monitor_range(char *str)
{
    unsigned long addr, size;

    pr_info(“%s(%s)
”, __func__, str);
    if (!str)
        return 0;
    size = memparse(str, &str);
    if (*str != ‘@’)
        return 0;
    addr = memparse(str + 1, &str);

    if (addr == 0 || size == 0){
        pr_info (“memory monitor is disable, addr=%#x, size=%#x
”, addr, size);
        return 0;
}
pr_info (“memory monitor is enable, addr=%#x, size=%#x
”, addr, size);
memory_monitor_addr = addr;
memory_monitor_size = size;
return 0;
}
early_param(“sched_monitor_addr”, get_memory_monitor_range);

static int __init tracepoint_sched_switch_init(void)
{
    int ret = 0;

    if (memory_monitor_addr == 0 || memory_monitor_size == 0)
        return 0;
    start_pfn = PHYS_PFN(memory_monitor_addr);
    end_pfn = PHYS_PFN(memory_monitor_addr + memory_monitor_size);
    pr_info (“memory monitor: start_pfn=%#x, end_pfn=%#x
”, start_pfn, end_pfn);

    ret = register_trace_sched_switch(sched_switch_probe, NULL);
    if (ret)
        goto fail;
    pr_info (“register trace_sched_switch success.
”);
    return 0;
fail:
    pr_info (“register trace_sched_switch failed.
”);
}

static void __exit tracepoint_sched_switch_exit(void)
{
    pr_info(“%s:%d: exit.
”, __func__, __LINE__);
    unregister_trace_sched_switch(sched_switch_probe, NULL);
}

pure_initcall(tracepoint_sched_switch_init);
module_exit(tracepoint_sched_switch_exit);