ftrace: user raw_spin_lock in tracing
authorSteven Rostedt <rostedt@goodmis.org>
Mon, 12 May 2008 19:20:55 +0000 (21:20 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Fri, 23 May 2008 19:14:11 +0000 (21:14 +0200)
Lock debugging enabled cause huge performance problems for tracing. Having
the lock verification happening for every function that is called
because mcount calls spin_lock can cripple the system.

This patch converts the spin_locks used by ftrace into raw_spin_locks.

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
kernel/trace/trace.h

index 736dcfb3ed01316b461820f02a0ec7ba06e1ee5e..3009aafa4dde74252f52ea6043d691b11693add4 100644 (file)
@@ -133,7 +133,8 @@ static const char *trace_options[] = {
        NULL
 };
 
-static DEFINE_SPINLOCK(ftrace_max_lock);
+static raw_spinlock_t ftrace_max_lock =
+       (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 
 /*
  * Copy the new maximum trace into the separate maximum-trace
@@ -335,7 +336,7 @@ update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
        int i;
 
        WARN_ON_ONCE(!irqs_disabled());
-       spin_lock(&ftrace_max_lock);
+       __raw_spin_lock(&ftrace_max_lock);
        /* clear out all the previous traces */
        for_each_possible_cpu(i) {
                data = tr->data[i];
@@ -344,7 +345,7 @@ update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
        }
 
        __update_max_tr(tr, tsk, cpu);
-       spin_unlock(&ftrace_max_lock);
+       __raw_spin_unlock(&ftrace_max_lock);
 }
 
 /**
@@ -360,7 +361,7 @@ update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
        int i;
 
        WARN_ON_ONCE(!irqs_disabled());
-       spin_lock(&ftrace_max_lock);
+       __raw_spin_lock(&ftrace_max_lock);
        for_each_possible_cpu(i)
                tracing_reset(max_tr.data[i]);
 
@@ -368,7 +369,7 @@ update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
        tracing_reset(data);
 
        __update_max_tr(tr, tsk, cpu);
-       spin_unlock(&ftrace_max_lock);
+       __raw_spin_unlock(&ftrace_max_lock);
 }
 
 int register_tracer(struct tracer *type)
@@ -652,13 +653,15 @@ trace_function(struct trace_array *tr, struct trace_array_cpu *data,
        struct trace_entry *entry;
        unsigned long irq_flags;
 
-       spin_lock_irqsave(&data->lock, irq_flags);
+       raw_local_irq_save(irq_flags);
+       __raw_spin_lock(&data->lock);
        entry                   = tracing_get_trace_entry(tr, data);
        tracing_generic_entry_update(entry, flags);
        entry->type             = TRACE_FN;
        entry->fn.ip            = ip;
        entry->fn.parent_ip     = parent_ip;
-       spin_unlock_irqrestore(&data->lock, irq_flags);
+       __raw_spin_unlock(&data->lock);
+       raw_local_irq_restore(irq_flags);
 }
 
 void
@@ -678,14 +681,16 @@ __trace_special(void *__tr, void *__data,
        struct trace_entry *entry;
        unsigned long irq_flags;
 
-       spin_lock_irqsave(&data->lock, irq_flags);
+       raw_local_irq_save(irq_flags);
+       __raw_spin_lock(&data->lock);
        entry                   = tracing_get_trace_entry(tr, data);
        tracing_generic_entry_update(entry, 0);
        entry->type             = TRACE_SPECIAL;
        entry->special.arg1     = arg1;
        entry->special.arg2     = arg2;
        entry->special.arg3     = arg3;
-       spin_unlock_irqrestore(&data->lock, irq_flags);
+       __raw_spin_unlock(&data->lock);
+       raw_local_irq_restore(irq_flags);
 
        trace_wake_up();
 }
@@ -725,7 +730,8 @@ tracing_sched_switch_trace(struct trace_array *tr,
        struct trace_entry *entry;
        unsigned long irq_flags;
 
-       spin_lock_irqsave(&data->lock, irq_flags);
+       raw_local_irq_save(irq_flags);
+       __raw_spin_lock(&data->lock);
        entry                   = tracing_get_trace_entry(tr, data);
        tracing_generic_entry_update(entry, flags);
        entry->type             = TRACE_CTX;
@@ -736,7 +742,8 @@ tracing_sched_switch_trace(struct trace_array *tr,
        entry->ctx.next_prio    = next->prio;
        entry->ctx.next_state   = next->state;
        __trace_stack(tr, data, flags, 4);
-       spin_unlock_irqrestore(&data->lock, irq_flags);
+       __raw_spin_unlock(&data->lock);
+       raw_local_irq_restore(irq_flags);
 }
 
 void
@@ -749,7 +756,8 @@ tracing_sched_wakeup_trace(struct trace_array *tr,
        struct trace_entry *entry;
        unsigned long irq_flags;
 
-       spin_lock_irqsave(&data->lock, irq_flags);
+       raw_local_irq_save(irq_flags);
+       __raw_spin_lock(&data->lock);
        entry                   = tracing_get_trace_entry(tr, data);
        tracing_generic_entry_update(entry, flags);
        entry->type             = TRACE_WAKE;
@@ -760,7 +768,8 @@ tracing_sched_wakeup_trace(struct trace_array *tr,
        entry->ctx.next_prio    = wakee->prio;
        entry->ctx.next_state   = wakee->state;
        __trace_stack(tr, data, flags, 5);
-       spin_unlock_irqrestore(&data->lock, irq_flags);
+       __raw_spin_unlock(&data->lock);
+       raw_local_irq_restore(irq_flags);
 
        trace_wake_up();
 }
@@ -1824,7 +1833,8 @@ tracing_cpumask_write(struct file *filp, const char __user *ubuf,
        if (err)
                goto err_unlock;
 
-       spin_lock_irq(&ftrace_max_lock);
+       raw_local_irq_disable();
+       __raw_spin_lock(&ftrace_max_lock);
        for_each_possible_cpu(cpu) {
                /*
                 * Increase/decrease the disabled counter if we are
@@ -1839,7 +1849,8 @@ tracing_cpumask_write(struct file *filp, const char __user *ubuf,
                        atomic_dec(&global_trace.data[cpu]->disabled);
                }
        }
-       spin_unlock_irq(&ftrace_max_lock);
+       __raw_spin_unlock(&ftrace_max_lock);
+       raw_local_irq_enable();
 
        tracing_cpumask = tracing_cpumask_new;
 
@@ -2299,7 +2310,7 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
 
        for_each_cpu_mask(cpu, mask) {
                data = iter->tr->data[cpu];
-               spin_lock(&data->lock);
+               __raw_spin_lock(&data->lock);
        }
 
        while (find_next_entry_inc(iter) != NULL) {
@@ -2320,7 +2331,7 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
 
        for_each_cpu_mask(cpu, mask) {
                data = iter->tr->data[cpu];
-               spin_unlock(&data->lock);
+               __raw_spin_unlock(&data->lock);
        }
 
        for_each_cpu_mask(cpu, mask) {
@@ -2538,8 +2549,7 @@ static int trace_alloc_page(void)
        /* Now that we successfully allocate a page per CPU, add them */
        for_each_possible_cpu(i) {
                data = global_trace.data[i];
-               spin_lock_init(&data->lock);
-               lockdep_set_class(&data->lock, &data->lock_key);
+               data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
                page = list_entry(pages.next, struct page, lru);
                list_del_init(&page->lru);
                list_add_tail(&page->lru, &data->trace_pages);
@@ -2547,8 +2557,7 @@ static int trace_alloc_page(void)
 
 #ifdef CONFIG_TRACER_MAX_TRACE
                data = max_tr.data[i];
-               spin_lock_init(&data->lock);
-               lockdep_set_class(&data->lock, &data->lock_key);
+               data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
                page = list_entry(pages.next, struct page, lru);
                list_del_init(&page->lru);
                list_add_tail(&page->lru, &data->trace_pages);
index 98cbfd05d75477b92ae58803fcdfa1e2c582f697..25cba28eb9baa104323088eae62bc5ca6c53ec3a 100644 (file)
@@ -76,7 +76,7 @@ struct trace_entry {
 struct trace_array_cpu {
        struct list_head        trace_pages;
        atomic_t                disabled;
-       spinlock_t              lock;
+       raw_spinlock_t          lock;
        struct lock_class_key   lock_key;
 
        /* these fields get copied into max-trace: */