/*
* sched_task_trace.c -- record scheduling events to a byte stream
*/
#define NO_TASK_TRACE_DECLS
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/percpu.h>
#include <linux/hardirq.h>
#include <litmus/ftdev.h>
#include <litmus/litmus.h>
#include <litmus/sched_trace.h>
#include <litmus/feather_trace.h>
#include <litmus/ftdev.h>
#define NUM_EVENTS (1 << (CONFIG_SCHED_TASK_TRACE_SHIFT+11))
#define now() litmus_clock()
struct local_buffer {
struct st_event_record record[NUM_EVENTS];
char flag[NUM_EVENTS];
struct ft_buffer ftbuf;
};
DEFINE_PER_CPU(struct local_buffer, st_event_buffer);
static struct ftdev st_dev;
static int st_dev_can_open(struct ftdev *dev, unsigned int cpu)
{
return cpu_online(cpu) ? 0 : -ENODEV;
}
static int __init init_sched_task_trace(void)
{
struct local_buffer* buf;
int i, ok = 0, err;
printk("Allocated %u sched_trace_xxx() events per CPU "
"(buffer size: %d bytes)\n",
NUM_EVENTS, (int) sizeof(struct local_buffer));
err = ftdev_init(&st_dev, THIS_MODULE,
num_online_cpus(), "sched_trace");
if (err)
goto err_out;
for (i = 0; i < st_dev.minor_cnt; i++) {
buf = &per_cpu(st_event_buffer, i);
ok += init_ft_buffer(&buf->ftbuf, NUM_EVENTS,
sizeof(struct st_event_record),
buf->flag,
buf->record);
st_dev.minor[i].buf = &buf->ftbuf;
}
if (ok == st_dev.minor_cnt) {
st_dev.can_open = st_dev_can_open;
err = register_ftdev(&st_dev);
if (err)
goto err_dealloc;
} else {
err = -EINVAL;
goto err_dealloc;
}
return 0;
err_dealloc:
ftdev_exit(&st_dev);
err_out:
printk(KERN_WARNING "Could not register sched_trace module\n");
return err;
}
static void __exit exit_sched_task_trace(void)
{
ftdev_exit(&st_dev);
}
module_init(init_sched_task_trace);
module_exit(exit_sched_task_trace);
static inline struct st_event_record* get_record(u8 type, struct task_struct* t)
{
struct st_event_record* rec = NULL;
struct local_buffer* buf;
buf = &get_cpu_var(st_event_buffer);
if (ft_buffer_start_write(&buf->ftbuf, (void**) &rec)) {
rec->hdr.type = type;
rec->hdr.cpu = smp_processor_id();
rec->hdr.pid = t ? t->pid : 0;
rec->hdr.job = t ? t->rt_param.job_params.job_no : 0;
} else {
put_cpu_var(st_event_buffer);
}
/* rec will be NULL if it failed */
return rec;
}
static inline void put_record(struct st_event_record* rec)
{
struct local_buffer* buf;
buf = &__get_cpu_var(st_event_buffer);
ft_buffer_finish_write(&buf->ftbuf, rec);
put_cpu_var(st_event_buffer);
}
feather_callback void do_sched_trace_task_name(unsigned long id, unsigned long _task)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec = get_record(ST_NAME, t);
int i;
if (rec) {
for (i = 0; i < min(TASK_COMM_LEN, ST_NAME_LEN); i++)
rec->data.name.cmd[i] = t->comm[i];
put_record(rec);
}
}
feather_callback void do_sched_trace_task_param(unsigned long id, unsigned long _task)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec = get_record(ST_PARAM, t);
if (rec) {
rec->data.param.wcet = get_exec_cost(t);
rec->data.param.period = get_rt_period(t);
rec->data.param.phase = get_rt_phase(t);
rec->data.param.partition = get_partition(t);
rec->data.param.class = get_class(t);
put_record(rec);
}
}
feather_callback void do_sched_trace_task_release(unsigned long id, unsigned long _task)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec = get_record(ST_RELEASE, t);
if (rec) {
rec->data.release.release = get_release(t);
rec->data.release.deadline = get_deadline(t);
put_record(rec);
}
}
/* skipped: st_assigned_data, we don't use it atm */
feather_callback void do_sched_trace_task_switch_to(unsigned long id,
unsigned long _task)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec;
//if (is_realtime(t)) /* comment out to trace EVERYTHING */
{
rec = get_record(ST_SWITCH_TO, t);
if (rec) {
rec->data.switch_to.when = now();
rec->data.switch_to.exec_time = get_exec_time(t);
put_record(rec);
}
}
}
feather_callback void do_sched_trace_task_switch_away(unsigned long id,
unsigned long _task)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec;
//if (is_realtime(t)) /* comment out to trace EVERYTHING */
{
rec = get_record(ST_SWITCH_AWAY, t);
if (rec) {
rec->data.switch_away.when = now();
rec->data.switch_away.exec_time = get_exec_time(t);
put_record(rec);
}
}
}
feather_callback void do_sched_trace_task_completion(unsigned long id,
unsigned long _task,
unsigned long forced)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec = get_record(ST_COMPLETION, t);
if (rec) {
rec->data.completion.when = now();
rec->data.completion.forced = forced;
#ifdef LITMUS_NVIDIA
rec->data.completion.nv_int_count = (u16)atomic_read(&tsk_rt(t)->nv_int_count);
#endif
put_record(rec);
}
}
feather_callback void do_sched_trace_task_block(unsigned long id,
unsigned long _task)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec = get_record(ST_BLOCK, t);
if (rec) {
rec->data.block.when = now();
put_record(rec);
}
}
feather_callback void do_sched_trace_task_resume(unsigned long id,
unsigned long _task)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec = get_record(ST_RESUME, t);
if (rec) {
rec->data.resume.when = now();
put_record(rec);
}
}
feather_callback void do_sched_trace_sys_release(unsigned long id,
unsigned long _start)
{
lt_t *start = (lt_t*) _start;
struct st_event_record* rec = get_record(ST_SYS_RELEASE, NULL);
if (rec) {
rec->data.sys_release.when = now();
rec->data.sys_release.release = *start;
put_record(rec);
}
}
feather_callback void do_sched_trace_action(unsigned long id,
unsigned long _task,
unsigned long action)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record* rec = get_record(ST_ACTION, t);
if (rec) {
rec->data.action.when = now();
rec->data.action.action = action;
put_record(rec);
}
}
feather_callback void do_sched_trace_prediction_err(unsigned long id,
unsigned long _task,
unsigned long _distance,
unsigned long _rel_err)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record *rec = get_record(ST_PREDICTION_ERR, t);
if (rec) {
gpu_migration_dist_t* distance = (gpu_migration_dist_t*) _distance;
fp_t* rel_err = (fp_t*) _rel_err;
rec->data.prediction_err.distance = *distance;
rec->data.prediction_err.rel_err = rel_err->val;
put_record(rec);
}
}
feather_callback void do_sched_trace_migration(unsigned long id,
unsigned long _task,
unsigned long _mig_info)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record *rec = get_record(ST_MIGRATION, t);
if (rec) {
struct migration_info* mig_info = (struct migration_info*) _mig_info;
rec->hdr.extra = mig_info->distance;
rec->data.migration.observed = mig_info->observed;
rec->data.migration.estimated = mig_info->estimated;
put_record(rec);
}
}
feather_callback void do_sched_trace_tasklet_release(unsigned long id,
unsigned long _owner)
{
struct task_struct *t = (struct task_struct*) _owner;
struct st_event_record *rec = get_record(ST_TASKLET_RELEASE, t);
if (rec) {
rec->data.tasklet_release.when = now();
put_record(rec);
}
}
feather_callback void do_sched_trace_tasklet_begin(unsigned long id,
unsigned long _owner)
{
struct task_struct *t = (struct task_struct*) _owner;
struct st_event_record *rec = get_record(ST_TASKLET_BEGIN, t);
if (rec) {
rec->data.tasklet_begin.when = now();
if(!in_interrupt())
rec->data.tasklet_begin.exe_pid = current->pid;
else
rec->data.tasklet_begin.exe_pid = 0;
put_record(rec);
}
}
EXPORT_SYMBOL(do_sched_trace_tasklet_begin);
feather_callback void do_sched_trace_tasklet_end(unsigned long id,
unsigned long _owner,
unsigned long _flushed)
{
struct task_struct *t = (struct task_struct*) _owner;
struct st_event_record *rec = get_record(ST_TASKLET_END, t);
if (rec) {
rec->data.tasklet_end.when = now();
rec->data.tasklet_end.flushed = _flushed;
if(!in_interrupt())
rec->data.tasklet_end.exe_pid = current->pid;
else
rec->data.tasklet_end.exe_pid = 0;
put_record(rec);
}
}
EXPORT_SYMBOL(do_sched_trace_tasklet_end);
feather_callback void do_sched_trace_work_release(unsigned long id,
unsigned long _owner)
{
struct task_struct *t = (struct task_struct*) _owner;
struct st_event_record *rec = get_record(ST_WORK_RELEASE, t);
if (rec) {
rec->data.work_release.when = now();
put_record(rec);
}
}
feather_callback void do_sched_trace_work_begin(unsigned long id,
unsigned long _owner,
unsigned long _exe)
{
struct task_struct *t = (struct task_struct*) _owner;
struct st_event_record *rec = get_record(ST_WORK_BEGIN, t);
if (rec) {
struct task_struct *exe = (struct task_struct*) _exe;
rec->data.work_begin.exe_pid = exe->pid;
rec->data.work_begin.when = now();
put_record(rec);
}
}
EXPORT_SYMBOL(do_sched_trace_work_begin);
feather_callback void do_sched_trace_work_end(unsigned long id,
unsigned long _owner,
unsigned long _exe,
unsigned long _flushed)
{
struct task_struct *t = (struct task_struct*) _owner;
struct st_event_record *rec = get_record(ST_WORK_END, t);
if (rec) {
struct task_struct *exe = (struct task_struct*) _exe;
rec->data.work_end.exe_pid = exe->pid;
rec->data.work_end.flushed = _flushed;
rec->data.work_end.when = now();
put_record(rec);
}
}
EXPORT_SYMBOL(do_sched_trace_work_end);
feather_callback void do_sched_trace_eff_prio_change(unsigned long id,
unsigned long _task,
unsigned long _inh)
{
struct task_struct *t = (struct task_struct*) _task;
struct st_event_record *rec = get_record(ST_EFF_PRIO_CHANGE, t);
if (rec) {
struct task_struct *inh = (struct task_struct*) _inh;
rec->data.effective_priority_change.when = now();
rec->data.effective_priority_change.inh_pid = (inh != NULL) ?
inh->pid :
0xffff;
put_record(rec);
}
}
/* pray for no nesting of nv interrupts on same CPU... */
struct tracing_interrupt_map
{
int active;
int count;
unsigned long data[128]; // assume nesting less than 128...
unsigned long serial[128];
};
DEFINE_PER_CPU(struct tracing_interrupt_map, active_interrupt_tracing);
DEFINE_PER_CPU(u32, intCounter);
feather_callback void do_sched_trace_nv_interrupt_begin(unsigned long id,
unsigned long _device)
{
struct st_event_record *rec;
u32 serialNum;
{
u32* serial;
struct tracing_interrupt_map* int_map = &per_cpu(active_interrupt_tracing, smp_processor_id());
if(!int_map->active == 0xcafebabe)
{
int_map->count++;
}
else
{
int_map->active = 0xcafebabe;
int_map->count = 1;
}
//int_map->data[int_map->count-1] = _device;
serial = &per_cpu(intCounter, smp_processor_id());
*serial += num_online_cpus();
serialNum = *serial;
int_map->serial[int_map->count-1] = serialNum;
}
rec = get_record(ST_NV_INTERRUPT_BEGIN, NULL);
if(rec) {
u32 device = _device;
rec->data.nv_interrupt_begin.when = now();
rec->data.nv_interrupt_begin.device = device;
rec->data.nv_interrupt_begin.serialNumber = serialNum;
put_record(rec);
}
}
EXPORT_SYMBOL(do_sched_trace_nv_interrupt_begin);
/*
int is_interrupt_tracing_active(void)
{
struct tracing_interrupt_map* int_map = &per_cpu(active_interrupt_tracing, smp_processor_id());
if(int_map->active == 0xcafebabe)
return 1;
return 0;
}
*/
feather_callback void do_sched_trace_nv_interrupt_end(unsigned long id, unsigned long _device)
{
struct tracing_interrupt_map* int_map = &per_cpu(active_interrupt_tracing, smp_processor_id());
if(int_map->active == 0xcafebabe)
{
struct st_event_record *rec = get_record(ST_NV_INTERRUPT_END, NULL);
int_map->count--;
if(int_map->count == 0)
int_map->active = 0;
if(rec) {
u32 device = _device;
rec->data.nv_interrupt_end.when = now();
//rec->data.nv_interrupt_end.device = int_map->data[int_map->count];
rec->data.nv_interrupt_end.device = device;
rec->data.nv_interrupt_end.serialNumber = int_map->serial[int_map->count];
put_record(rec);
}
}
}
EXPORT_SYMBOL(do_sched_trace_nv_interrupt_end);