ftrace: modulize the number of CPU buffers
Steven Rostedt [Mon, 12 May 2008 19:21:00 +0000 (21:21 +0200)]
Currently ftrace allocates a trace buffer for every possible CPU.
Work is being done to change it to only online CPUs and add hooks
to hotplug CPUS.

This patch lays out the infrastructure for such a change.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

kernel/trace/trace.c

index 290e9da..5da391c 100644 (file)
 unsigned long __read_mostly    tracing_max_latency = (cycle_t)ULONG_MAX;
 unsigned long __read_mostly    tracing_thresh;
 
+static unsigned long __read_mostly     tracing_nr_buffers;
+static cpumask_t __read_mostly         tracing_buffer_mask;
+
+#define for_each_tracing_cpu(cpu)      \
+       for_each_cpu_mask(cpu, tracing_buffer_mask)
+
 /* dummy trace to disable tracing */
 static struct tracer no_tracer __read_mostly = {
        .name           = "none",
@@ -328,7 +334,7 @@ update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
        WARN_ON_ONCE(!irqs_disabled());
        __raw_spin_lock(&ftrace_max_lock);
        /* clear out all the previous traces */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                data = tr->data[i];
                flip_trace(max_tr.data[i], data);
                tracing_reset(data);
@@ -352,7 +358,7 @@ update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 
        WARN_ON_ONCE(!irqs_disabled());
        __raw_spin_lock(&ftrace_max_lock);
-       for_each_possible_cpu(i)
+       for_each_tracing_cpu(i)
                tracing_reset(max_tr.data[i]);
 
        flip_trace(max_tr.data[cpu], data);
@@ -398,7 +404,7 @@ int register_tracer(struct tracer *type)
                 * internal tracing to verify that everything is in order.
                 * If we fail, we do not register this tracer.
                 */
-               for_each_possible_cpu(i) {
+               for_each_tracing_cpu(i) {
                        data = tr->data[i];
                        if (!head_page(data))
                                continue;
@@ -417,7 +423,7 @@ int register_tracer(struct tracer *type)
                        goto out;
                }
                /* Only reset on passing, to avoid touching corrupted buffers */
-               for_each_possible_cpu(i) {
+               for_each_tracing_cpu(i) {
                        data = tr->data[i];
                        if (!head_page(data))
                                continue;
@@ -847,7 +853,7 @@ find_next_entry(struct trace_iterator *iter, int *ent_cpu)
        int next_cpu = -1;
        int cpu;
 
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                if (!head_page(tr->data[cpu]))
                        continue;
                ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
@@ -972,7 +978,7 @@ static void *s_start(struct seq_file *m, loff_t *pos)
                iter->prev_ent = NULL;
                iter->prev_cpu = -1;
 
-               for_each_possible_cpu(i) {
+               for_each_tracing_cpu(i) {
                        iter->next_idx[i] = 0;
                        iter->next_page[i] = NULL;
                }
@@ -1089,7 +1095,7 @@ print_trace_header(struct seq_file *m, struct trace_iterator *iter)
        if (type)
                name = type->name;
 
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                if (head_page(tr->data[cpu])) {
                        total += tr->data[cpu]->trace_idx;
                        if (tr->data[cpu]->trace_idx > tr->entries)
@@ -1519,7 +1525,7 @@ static int trace_empty(struct trace_iterator *iter)
        struct trace_array_cpu *data;
        int cpu;
 
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                data = iter->tr->data[cpu];
 
                if (head_page(data) && data->trace_idx &&
@@ -1831,7 +1837,7 @@ tracing_cpumask_write(struct file *filp, const char __user *ubuf,
 
        raw_local_irq_disable();
        __raw_spin_lock(&ftrace_max_lock);
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                /*
                 * Increase/decrease the disabled counter if we are
                 * about to flip a bit in the cpumask:
@@ -2308,7 +2314,7 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
        ftrace_enabled = 0;
 #endif
        smp_wmb();
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                data = iter->tr->data[cpu];
 
                if (!head_page(data) || !data->trace_idx)
@@ -2605,7 +2611,7 @@ static int trace_alloc_page(void)
        int i;
 
        /* first allocate a page for each CPU */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                array = (void *)__get_free_page(GFP_KERNEL);
                if (array == NULL) {
                        printk(KERN_ERR "tracer: failed to allocate page"
@@ -2630,7 +2636,7 @@ static int trace_alloc_page(void)
        }
 
        /* Now that we successfully allocate a page per CPU, add them */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                data = global_trace.data[i];
                page = list_entry(pages.next, struct page, lru);
                list_del_init(&page->lru);
@@ -2666,7 +2672,7 @@ static int trace_free_page(void)
        int ret = 0;
 
        /* free one page from each buffer */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                data = global_trace.data[i];
                p = data->trace_pages.next;
                if (p == &data->trace_pages) {
@@ -2717,8 +2723,12 @@ __init static int tracer_alloc_buffers(void)
 
        global_trace.ctrl = tracer_enabled;
 
+       /* TODO: make the number of buffers hot pluggable with CPUS */
+       tracing_nr_buffers = num_possible_cpus();
+       tracing_buffer_mask = cpu_possible_map;
+
        /* Allocate the first page for all buffers */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
                max_tr.data[i] = &per_cpu(max_data, i);