Skip to contentSkip to conversation

tracing: Updated changes for 3.12

PATCH New Steven Rostedt 1 participant 5 messages Sep 6, 2013

Subsystems (3)
lore
Share & tools
lore

Steven Rostedt

author maintainer
#0 lore raw

cover tracing: Updated changes for 3.12

I'm holding off on the rcu unsafe changes with perf and function tracing. We'll still get bug splats with unsafe rcu usage, but we need to work out a better solution than I was going to push for 3.12. It's too late to get things smooth, thus we need to wait till 3.13 to get something that is decent.

For now, root needs to be careful in how they trace functions with perf.

git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace.git
for-next

Head SHA1: a0a5a0561f63905fe94c49bc567615829f42ce1e

Alexander Z Lam (1):
tracing: Make tracing_cpumask available for all instances

H. Peter Anvin (1):
x86-32, ftrace: Fix static ftrace when early microcode is enabled

Steven Rostedt (Red Hat) (2):
ftrace: Fix a slight race in modifying what function callback gets traced ftrace/rcu: Do not trace debug_lockdep_rcu_enabled()

----

 arch/x86/kernel/entry_32.S |    3 +++
 kernel/rcupdate.c          |    2 +-
 kernel/trace/ftrace.c      |   17 ++++++++++++++++-
 kernel/trace/trace.c       |   37 ++++++++++++++++++++-----------------
 kernel/trace/trace.h       |    1 +
 5 files changed, 41 insertions(+), 19 deletions(-)
author maintainer
#1 lore raw

PATCH 1/4 tracing: Make tracing_cpumask available for all instances

From: Alexander Z Lam <[email protected]>

Allow tracer instances to disable tracing by cpu by moving the static global tracing_cpumask into trace_array.

Cc Vaibhav Nagarnaik <[email protected]>
Cc David Sharp <[email protected]>
Cc Alexander Z Lam <[email protected]>
---
 kernel/trace/trace.c |   37 ++++++++++++++++++++-----------------
 kernel/trace/trace.h |    1 +
 2 files changed, 21 insertions(+), 17 deletions(-)
kernel/trace/trace.c +20 -17
index 496f94d..7974ba2 100644
@@ -3166,11 +3166,6 @@ static const struct file_operations show_traces_fops = {
 };
 
 /*
- * Only trace on a CPU if the bitmask is set:
- */
-static cpumask_var_t tracing_cpumask;
-
-/*
  * The tracer itself will not take this lock, but still we want
  * to provide a consistent cpumask to user-space:
  */
@@ -3186,11 +3181,12 @@ static ssize_t
 tracing_cpumask_read(struct file *filp, char __user *ubuf,
 		     size_t count, loff_t *ppos)
 {
+	struct trace_array *tr = file_inode(filp)->i_private;
 	int len;
 
 	mutex_lock(&tracing_cpumask_update_lock);
 
-	len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
+	len = cpumask_scnprintf(mask_str, count, tr->tracing_cpumask);
 	if (count - len < 2) {
 		count = -EINVAL;
 		goto out_err;
@@ -3208,7 +3204,7 @@ static ssize_t
 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
 		      size_t count, loff_t *ppos)
 {
-	struct trace_array *tr = filp->private_data;
+	struct trace_array *tr = file_inode(filp)->i_private;
 	cpumask_var_t tracing_cpumask_new;
 	int err, cpu;
 
@@ -3228,12 +3224,12 @@ tracing_cpumask_write(struct file *filp, const char __user *ubuf,
 		 * Increase/decrease the disabled counter if we are
 		 * about to flip a bit in the cpumask:
 		 */
-		if (cpumask_test_cpu(cpu, tracing_cpumask) &&
+		if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
 				!cpumask_test_cpu(cpu, tracing_cpumask_new)) {
 			atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
 			ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
 		}
-		if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
+		if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
 				cpumask_test_cpu(cpu, tracing_cpumask_new)) {
 			atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
 			ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
@@ -3242,7 +3238,7 @@ tracing_cpumask_write(struct file *filp, const char __user *ubuf,
 	arch_spin_unlock(&ftrace_max_lock);
 	local_irq_enable();
 
-	cpumask_copy(tracing_cpumask, tracing_cpumask_new);
+	cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
 
 	mutex_unlock(&tracing_cpumask_update_lock);
 	free_cpumask_var(tracing_cpumask_new);
@@ -3256,9 +3252,10 @@ err_unlock:
 }
 
 static const struct file_operations tracing_cpumask_fops = {
-	.open		= tracing_open_generic,
+	.open		= tracing_open_generic_tr,
 	.read		= tracing_cpumask_read,
 	.write		= tracing_cpumask_write,
+	.release	= tracing_release_generic_tr,
 	.llseek		= generic_file_llseek,
 };
 
@@ -5938,6 +5935,11 @@ static int new_instance_create(const char *name)
 	if (!tr->name)
 		goto out_free_tr;
 
+	if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
+		goto out_free_tr;
+
+	cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
+
 	raw_spin_lock_init(&tr->start_lock);
 
 	tr->current_trace = &nop_trace;
@@ -5969,6 +5971,7 @@ static int new_instance_create(const char *name)
  out_free_tr:
 	if (tr->trace_buffer.buffer)
 		ring_buffer_free(tr->trace_buffer.buffer);
+	free_cpumask_var(tr->tracing_cpumask);
 	kfree(tr->name);
 	kfree(tr);
 
@@ -6098,6 +6101,9 @@ init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
 {
 	int cpu;
 
+	trace_create_file("tracing_cpumask", 0644, d_tracer,
+			  tr, &tracing_cpumask_fops);
+
 	trace_create_file("trace_options", 0644, d_tracer,
 			  tr, &tracing_iter_fops);
 
@@ -6147,9 +6153,6 @@ static __init int tracer_init_debugfs(void)
 
 	init_tracer_debugfs(&global_trace, d_tracer);
 
-	trace_create_file("tracing_cpumask", 0644, d_tracer,
-			&global_trace, &tracing_cpumask_fops);
-
 	trace_create_file("available_tracers", 0444, d_tracer,
 			&global_trace, &show_traces_fops);
 
@@ -6371,7 +6374,7 @@ __init static int tracer_alloc_buffers(void)
 	if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
 		goto out;
 
-	if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
+	if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
 		goto out_free_buffer_mask;
 
 	/* Only allocate trace_printk buffers if a trace_printk exists */
@@ -6386,7 +6389,7 @@ __init static int tracer_alloc_buffers(void)
 		ring_buf_size = 1;
 
 	cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
-	cpumask_copy(tracing_cpumask, cpu_all_mask);
+	cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
 
 	raw_spin_lock_init(&global_trace.start_lock);
 
@@ -6441,7 +6444,7 @@ out_free_cpumask:
 #ifdef CONFIG_TRACER_MAX_TRACE
 	free_percpu(global_trace.max_buffer.data);
 #endif
-	free_cpumask_var(tracing_cpumask);
+	free_cpumask_var(global_trace.tracing_cpumask);
 out_free_buffer_mask:
 	free_cpumask_var(tracing_buffer_mask);
 out:
kernel/trace/trace.h +1 -0
index afaae41..502fed7 100644
@@ -206,6 +206,7 @@ struct trace_array {
 	struct dentry		*event_dir;
 	struct list_head	systems;
 	struct list_head	events;
+	cpumask_var_t		tracing_cpumask; /* only trace on set CPUs */
 	int			ref;
 };
 
Signature · 1 line
1.7.10.4
author maintainer
#2 lore raw

PATCH 2/4 ftrace: Fix a slight race in modifying what function callback gets traced

In reply to Steven Rostedt

From: "Steven Rostedt (Red Hat)" <[email protected]>

There's a slight race when going from a list function to a non list function. That is, when only one callback is registered to the function tracer, it gets called directly by the mcount trampoline. But if this function has filters, it may be called by the wrong functions.

As the list ops callback that handles multiple callbacks that are registered to ftrace, it also handles what functions they call. While the transaction is taking place, use the list function always, and after all the updates are finished (only the functions that should be traced are being traced), then we can update the trampoline to call the function directly.

---
 kernel/trace/ftrace.c |   17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)
kernel/trace/ftrace.c +16 -1
index a6d098c..03cf44a 100644
@@ -1978,12 +1978,27 @@ int __weak ftrace_arch_code_modify_post_process(void)
 
 void ftrace_modify_all_code(int command)
 {
+	int update = command & FTRACE_UPDATE_TRACE_FUNC;
+
+	/*
+	 * If the ftrace_caller calls a ftrace_ops func directly,
+	 * we need to make sure that it only traces functions it
+	 * expects to trace. When doing the switch of functions,
+	 * we need to update to the ftrace_ops_list_func first
+	 * before the transition between old and new calls are set,
+	 * as the ftrace_ops_list_func will check the ops hashes
+	 * to make sure the ops are having the right functions
+	 * traced.
+	 */
+	if (update)
+		ftrace_update_ftrace_func(ftrace_ops_list_func);
+
 	if (command & FTRACE_UPDATE_CALLS)
 		ftrace_replace_code(1);
 	else if (command & FTRACE_DISABLE_CALLS)
 		ftrace_replace_code(0);
 
-	if (command & FTRACE_UPDATE_TRACE_FUNC)
+	if (update && ftrace_trace_function != ftrace_ops_list_func)
 		ftrace_update_ftrace_func(ftrace_trace_function);
 
 	if (command & FTRACE_START_FUNC_RET)
Signature · 1 line
1.7.10.4
author maintainer
#3 lore raw

PATCH 3/4 x86-32, ftrace: Fix static ftrace when early microcode is enabled

In reply to Steven Rostedt

From: "H. Peter Anvin" <[email protected]>

Early microcode loading runs C code before paging is enabled on 32 bits. Since ftrace puts a hook into every function, that hook needs to be safe to execute in the pre-paging environment. This is currently true for dynamic ftrace but not for static ftrace.

Static ftrace is obsolescent and assumed to not be performance-critical, so we can simply test that the stack pointer falls within the valid range of kernel addresses.

---
 arch/x86/kernel/entry_32.S |    3 +++
 1 file changed, 3 insertions(+)
arch/x86/kernel/entry_32.S +3 -0
index 2cfbc3a..f0dcb0c 100644
@@ -1176,6 +1176,9 @@ ftrace_restore_flags:
 #else /* ! CONFIG_DYNAMIC_FTRACE */
 
 ENTRY(mcount)
+	cmpl $__PAGE_OFFSET, %esp
+	jb ftrace_stub		/* Paging not enabled yet? */
+
 	cmpl $0, function_trace_stop
 	jne  ftrace_stub
 
Signature · 1 line
1.7.10.4
author maintainer
#4 lore raw

PATCH 4/4 ftrace/rcu: Do not trace debug_lockdep_rcu_enabled()

In reply to Steven Rostedt

From: "Steven Rostedt (Red Hat)" <[email protected]>

The function debug_lockdep_rcu_enabled() is part of the RCU lockdep debugging, and is called very frequently. I found that if I enable a lot of debugging and run the function graph tracer, this function can cause a live lock of the system.

We don't usually trace lockdep infrastructure, no need to trace this either.

---
 kernel/rcupdate.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/rcupdate.c +1 -1
index cce6ba8..4f20c6c 100644
@@ -122,7 +122,7 @@ struct lockdep_map rcu_sched_lock_map =
 	STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key);
 EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
 
-int debug_lockdep_rcu_enabled(void)
+int notrace debug_lockdep_rcu_enabled(void)
 {
 	return rcu_scheduler_active && debug_locks &&
 	       current->lockdep_recursion == 0;
Signature · 1 line
1.7.10.4