linux_dsm_epyc7002/kernel/trace/ring_buffer_benchmark.c
Vaibhav Nagarnaik 7ea5906405 tracing: Use NUMA allocation for per-cpu ring buffer pages
The tracing ring buffer is a group of per-cpu ring buffers where
allocation and logging is done on a per-cpu basis. The events that are
generated on a particular CPU are logged in the corresponding buffer.
This is to provide wait-free writes between CPUs and good NUMA node
locality while accessing the ring buffer.

However, the allocation routines consider NUMA locality only for buffer
page metadata and not for the actual buffer page. This causes the pages
to be allocated on the NUMA node local to the CPU where the allocation
routine is running at the time.

This patch fixes the problem by using a NUMA node specific allocation
routine so that the pages are allocated from a NUMA node local to the
logging CPU.

I tested with the getuid_microbench from autotest. It is a simple binary
that calls getuid() in a loop and measures the average time for the
syscall to complete. The following command was used to test:
$ getuid_microbench 1000000

Compared the numbers found on kernel with and without this patch and
found that logging latency decreases by 30-50 ns/call.
tracing with non-NUMA allocation - 569 ns/call
tracing with NUMA allocation     - 512 ns/call

Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Michael Rubin <mrubin@google.com>
Cc: David Sharp <dhsharp@google.com>
Link: http://lkml.kernel.org/r/1304470602-20366-1-git-send-email-vnagarnaik@google.com
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2011-06-14 22:04:39 -04:00

489 lines
10 KiB
C

/*
* ring buffer tester and benchmark
*
* Copyright (C) 2009 Steven Rostedt <srostedt@redhat.com>
*/
#include <linux/ring_buffer.h>
#include <linux/completion.h>
#include <linux/kthread.h>
#include <linux/module.h>
#include <linux/time.h>
#include <asm/local.h>
struct rb_page {
u64 ts;
local_t commit;
char data[4080];
};
/* run time and sleep time in seconds */
#define RUN_TIME 10
#define SLEEP_TIME 10
/* number of events for writer to wake up the reader */
static int wakeup_interval = 100;
static int reader_finish;
static struct completion read_start;
static struct completion read_done;
static struct ring_buffer *buffer;
static struct task_struct *producer;
static struct task_struct *consumer;
static unsigned long read;
static int disable_reader;
module_param(disable_reader, uint, 0644);
MODULE_PARM_DESC(disable_reader, "only run producer");
static int write_iteration = 50;
module_param(write_iteration, uint, 0644);
MODULE_PARM_DESC(write_iteration, "# of writes between timestamp readings");
static int producer_nice = 19;
static int consumer_nice = 19;
static int producer_fifo = -1;
static int consumer_fifo = -1;
module_param(producer_nice, uint, 0644);
MODULE_PARM_DESC(producer_nice, "nice prio for producer");
module_param(consumer_nice, uint, 0644);
MODULE_PARM_DESC(consumer_nice, "nice prio for consumer");
module_param(producer_fifo, uint, 0644);
MODULE_PARM_DESC(producer_fifo, "fifo prio for producer");
module_param(consumer_fifo, uint, 0644);
MODULE_PARM_DESC(consumer_fifo, "fifo prio for consumer");
static int read_events;
static int kill_test;
#define KILL_TEST() \
do { \
if (!kill_test) { \
kill_test = 1; \
WARN_ON(1); \
} \
} while (0)
enum event_status {
EVENT_FOUND,
EVENT_DROPPED,
};
static enum event_status read_event(int cpu)
{
struct ring_buffer_event *event;
int *entry;
u64 ts;
event = ring_buffer_consume(buffer, cpu, &ts, NULL);
if (!event)
return EVENT_DROPPED;
entry = ring_buffer_event_data(event);
if (*entry != cpu) {
KILL_TEST();
return EVENT_DROPPED;
}
read++;
return EVENT_FOUND;
}
static enum event_status read_page(int cpu)
{
struct ring_buffer_event *event;
struct rb_page *rpage;
unsigned long commit;
void *bpage;
int *entry;
int ret;
int inc;
int i;
bpage = ring_buffer_alloc_read_page(buffer, cpu);
if (!bpage)
return EVENT_DROPPED;
ret = ring_buffer_read_page(buffer, &bpage, PAGE_SIZE, cpu, 1);
if (ret >= 0) {
rpage = bpage;
/* The commit may have missed event flags set, clear them */
commit = local_read(&rpage->commit) & 0xfffff;
for (i = 0; i < commit && !kill_test; i += inc) {
if (i >= (PAGE_SIZE - offsetof(struct rb_page, data))) {
KILL_TEST();
break;
}
inc = -1;
event = (void *)&rpage->data[i];
switch (event->type_len) {
case RINGBUF_TYPE_PADDING:
/* failed writes may be discarded events */
if (!event->time_delta)
KILL_TEST();
inc = event->array[0] + 4;
break;
case RINGBUF_TYPE_TIME_EXTEND:
inc = 8;
break;
case 0:
entry = ring_buffer_event_data(event);
if (*entry != cpu) {
KILL_TEST();
break;
}
read++;
if (!event->array[0]) {
KILL_TEST();
break;
}
inc = event->array[0] + 4;
break;
default:
entry = ring_buffer_event_data(event);
if (*entry != cpu) {
KILL_TEST();
break;
}
read++;
inc = ((event->type_len + 1) * 4);
}
if (kill_test)
break;
if (inc <= 0) {
KILL_TEST();
break;
}
}
}
ring_buffer_free_read_page(buffer, bpage);
if (ret < 0)
return EVENT_DROPPED;
return EVENT_FOUND;
}
static void ring_buffer_consumer(void)
{
/* toggle between reading pages and events */
read_events ^= 1;
read = 0;
while (!reader_finish && !kill_test) {
int found;
do {
int cpu;
found = 0;
for_each_online_cpu(cpu) {
enum event_status stat;
if (read_events)
stat = read_event(cpu);
else
stat = read_page(cpu);
if (kill_test)
break;
if (stat == EVENT_FOUND)
found = 1;
}
} while (found && !kill_test);
set_current_state(TASK_INTERRUPTIBLE);
if (reader_finish)
break;
schedule();
__set_current_state(TASK_RUNNING);
}
reader_finish = 0;
complete(&read_done);
}
static void ring_buffer_producer(void)
{
struct timeval start_tv;
struct timeval end_tv;
unsigned long long time;
unsigned long long entries;
unsigned long long overruns;
unsigned long missed = 0;
unsigned long hit = 0;
unsigned long avg;
int cnt = 0;
/*
* Hammer the buffer for 10 secs (this may
* make the system stall)
*/
trace_printk("Starting ring buffer hammer\n");
do_gettimeofday(&start_tv);
do {
struct ring_buffer_event *event;
int *entry;
int i;
for (i = 0; i < write_iteration; i++) {
event = ring_buffer_lock_reserve(buffer, 10);
if (!event) {
missed++;
} else {
hit++;
entry = ring_buffer_event_data(event);
*entry = smp_processor_id();
ring_buffer_unlock_commit(buffer, event);
}
}
do_gettimeofday(&end_tv);
cnt++;
if (consumer && !(cnt % wakeup_interval))
wake_up_process(consumer);
#ifndef CONFIG_PREEMPT
/*
* If we are a non preempt kernel, the 10 second run will
* stop everything while it runs. Instead, we will call
* cond_resched and also add any time that was lost by a
* rescedule.
*
* Do a cond resched at the same frequency we would wake up
* the reader.
*/
if (cnt % wakeup_interval)
cond_resched();
#endif
} while (end_tv.tv_sec < (start_tv.tv_sec + RUN_TIME) && !kill_test);
trace_printk("End ring buffer hammer\n");
if (consumer) {
/* Init both completions here to avoid races */
init_completion(&read_start);
init_completion(&read_done);
/* the completions must be visible before the finish var */
smp_wmb();
reader_finish = 1;
/* finish var visible before waking up the consumer */
smp_wmb();
wake_up_process(consumer);
wait_for_completion(&read_done);
}
time = end_tv.tv_sec - start_tv.tv_sec;
time *= USEC_PER_SEC;
time += (long long)((long)end_tv.tv_usec - (long)start_tv.tv_usec);
entries = ring_buffer_entries(buffer);
overruns = ring_buffer_overruns(buffer);
if (kill_test)
trace_printk("ERROR!\n");
if (!disable_reader) {
if (consumer_fifo < 0)
trace_printk("Running Consumer at nice: %d\n",
consumer_nice);
else
trace_printk("Running Consumer at SCHED_FIFO %d\n",
consumer_fifo);
}
if (producer_fifo < 0)
trace_printk("Running Producer at nice: %d\n",
producer_nice);
else
trace_printk("Running Producer at SCHED_FIFO %d\n",
producer_fifo);
/* Let the user know that the test is running at low priority */
if (producer_fifo < 0 && consumer_fifo < 0 &&
producer_nice == 19 && consumer_nice == 19)
trace_printk("WARNING!!! This test is running at lowest priority.\n");
trace_printk("Time: %lld (usecs)\n", time);
trace_printk("Overruns: %lld\n", overruns);
if (disable_reader)
trace_printk("Read: (reader disabled)\n");
else
trace_printk("Read: %ld (by %s)\n", read,
read_events ? "events" : "pages");
trace_printk("Entries: %lld\n", entries);
trace_printk("Total: %lld\n", entries + overruns + read);
trace_printk("Missed: %ld\n", missed);
trace_printk("Hit: %ld\n", hit);
/* Convert time from usecs to millisecs */
do_div(time, USEC_PER_MSEC);
if (time)
hit /= (long)time;
else
trace_printk("TIME IS ZERO??\n");
trace_printk("Entries per millisec: %ld\n", hit);
if (hit) {
/* Calculate the average time in nanosecs */
avg = NSEC_PER_MSEC / hit;
trace_printk("%ld ns per entry\n", avg);
}
if (missed) {
if (time)
missed /= (long)time;
trace_printk("Total iterations per millisec: %ld\n",
hit + missed);
/* it is possible that hit + missed will overflow and be zero */
if (!(hit + missed)) {
trace_printk("hit + missed overflowed and totalled zero!\n");
hit--; /* make it non zero */
}
/* Caculate the average time in nanosecs */
avg = NSEC_PER_MSEC / (hit + missed);
trace_printk("%ld ns per entry\n", avg);
}
}
static void wait_to_die(void)
{
set_current_state(TASK_INTERRUPTIBLE);
while (!kthread_should_stop()) {
schedule();
set_current_state(TASK_INTERRUPTIBLE);
}
__set_current_state(TASK_RUNNING);
}
static int ring_buffer_consumer_thread(void *arg)
{
while (!kthread_should_stop() && !kill_test) {
complete(&read_start);
ring_buffer_consumer();
set_current_state(TASK_INTERRUPTIBLE);
if (kthread_should_stop() || kill_test)
break;
schedule();
__set_current_state(TASK_RUNNING);
}
__set_current_state(TASK_RUNNING);
if (kill_test)
wait_to_die();
return 0;
}
static int ring_buffer_producer_thread(void *arg)
{
init_completion(&read_start);
while (!kthread_should_stop() && !kill_test) {
ring_buffer_reset(buffer);
if (consumer) {
smp_wmb();
wake_up_process(consumer);
wait_for_completion(&read_start);
}
ring_buffer_producer();
trace_printk("Sleeping for 10 secs\n");
set_current_state(TASK_INTERRUPTIBLE);
schedule_timeout(HZ * SLEEP_TIME);
__set_current_state(TASK_RUNNING);
}
if (kill_test)
wait_to_die();
return 0;
}
static int __init ring_buffer_benchmark_init(void)
{
int ret;
/* make a one meg buffer in overwite mode */
buffer = ring_buffer_alloc(1000000, RB_FL_OVERWRITE);
if (!buffer)
return -ENOMEM;
if (!disable_reader) {
consumer = kthread_create(ring_buffer_consumer_thread,
NULL, "rb_consumer");
ret = PTR_ERR(consumer);
if (IS_ERR(consumer))
goto out_fail;
}
producer = kthread_run(ring_buffer_producer_thread,
NULL, "rb_producer");
ret = PTR_ERR(producer);
if (IS_ERR(producer))
goto out_kill;
/*
* Run them as low-prio background tasks by default:
*/
if (!disable_reader) {
if (consumer_fifo >= 0) {
struct sched_param param = {
.sched_priority = consumer_fifo
};
sched_setscheduler(consumer, SCHED_FIFO, &param);
} else
set_user_nice(consumer, consumer_nice);
}
if (producer_fifo >= 0) {
struct sched_param param = {
.sched_priority = consumer_fifo
};
sched_setscheduler(producer, SCHED_FIFO, &param);
} else
set_user_nice(producer, producer_nice);
return 0;
out_kill:
if (consumer)
kthread_stop(consumer);
out_fail:
ring_buffer_free(buffer);
return ret;
}
static void __exit ring_buffer_benchmark_exit(void)
{
kthread_stop(producer);
if (consumer)
kthread_stop(consumer);
ring_buffer_free(buffer);
}
module_init(ring_buffer_benchmark_init);
module_exit(ring_buffer_benchmark_exit);
MODULE_AUTHOR("Steven Rostedt");
MODULE_DESCRIPTION("ring_buffer_benchmark");
MODULE_LICENSE("GPL");