Re: [tip:tracing/urgent] tracing: Allocate the ftrace event profile buffer dynamically

Re: [tip:tracing/urgent] tracing: Allocate the ftrace event profile buffer dynamically

[Steven Rostedt]

On Sat, 2009-09-19 at 07:58 +0000, tip-bot for Frederic Weisbecker
wrote:

> +/*
> + * We can't use a size but a type in alloc_percpu()
> + * So let's create a dummy type that matches the desired size
> + */
> +typedef struct {char buf[FTRACE_MAX_PROFILE_SIZE];} profile_buf_t;
> +
>  static int ftrace_profile_enable_event(struct ftrace_event_call *event)
>  {
> +	char *buf;
> +	int ret;
> +
>  	if (atomic_inc_return(&event->profile_count))
>  		return 0;
>  
> -	return event->profile_enable();
> +	buf = (char *)alloc_percpu(profile_buf_t);
> +	if (!buf)
> +		return -ENOMEM;

Ingo,

Did you pull in the version that allocates a buffer for each event? I
thought Frederic made just a global per cpu buffer that all events can
use. The buffer is just a temporary storage that will be too big to put
on the stack.

-- Steve

> +
> +	rcu_assign_pointer(event->profile_buf, buf);
> +
> +	buf = (char *)alloc_percpu(profile_buf_t);
> +	if (!buf) {
> +		ret = -ENOMEM;
> +		goto fail_nmi;
> +	}
> +
> +	rcu_assign_pointer(event->profile_buf_nmi, buf);
> +
> +	ret = event->profile_enable();
> +	if (!ret)
> +		return 0;
> +
> +	kfree(event->profile_buf_nmi);
> +fail_nmi:
> +	kfree(event->profile_buf);
> +
> +	return ret;
>  }
>  

Re: [tip:tracing/urgent] tracing: Allocate the ftrace event profile buffer dynamically

[Steven Rostedt]

On Sat, 2009-09-19 at 10:06 +0000, tip-bot for Frederic Weisbecker
wrote:
> Commit-ID:  20ab4425a77a1f34028cc6ce57053c22c184ba5f
> Gitweb:     http://git.kernel.org/tip/20ab4425a77a1f34028cc6ce57053c22c184ba5f
> Author:     Frederic Weisbecker <fweisbec@gmail.com>
> AuthorDate: Fri, 18 Sep 2009 06:10:28 +0200
> Committer:  Frederic Weisbecker <fweisbec@gmail.com>
> CommitDate: Fri, 18 Sep 2009 07:25:44 +0200
> 
> tracing: Allocate the ftrace event profile buffer dynamically
> 
> Currently the trace event profile buffer is allocated in the stack. But
> this may be too much for the stack, as the events can have large
> statically defined field size and can also grow with dynamic arrays.
> 
> Allocate two per cpu buffer for all profiled events. The first cpu
> buffer is used to host every non-nmi context traces. It is protected
> by disabling the interrupts while writing and committing the trace.
> 
> The second buffer is reserved for nmi. So that there is no race between
> them and the first buffer.
> 
> The whole write/commit section is rcu protected because we release
> these buffers while deactivating the last profiling trace event.
> 
> v2: Move the buffers from trace_event to be global, as pointed by
>     Steven Rostedt.

OK, now I'm confused ;-)

This version looks like the correct one.

-- Steve

> 
> v3: Fix the syscall events to handle the profiling buffer races
>     by disabling interrupts, now that the buffers are globals.
> 
> Suggested-by: Steven Rostedt <rostedt@goodmis.org>
> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Li Zefan <lizf@cn.fujitsu.com>
> Cc: Jason Baron <jbaron@redhat.com>
> Cc: Masami Hiramatsu <mhiramat@redhat.com>
> Cc: Ingo Molnar <mingo@elte.hu>
> 

Re: [tip:tracing/urgent] tracing: Allocate the ftrace event profile buffer dynamically

[Ingo Molnar]

* Steven Rostedt <rostedt@goodmis.org> wrote:

> On Sat, 2009-09-19 at 07:58 +0000, tip-bot for Frederic Weisbecker
> wrote:
> 
> > +/*
> > + * We can't use a size but a type in alloc_percpu()
> > + * So let's create a dummy type that matches the desired size
> > + */
> > +typedef struct {char buf[FTRACE_MAX_PROFILE_SIZE];} profile_buf_t;
> > +
> >  static int ftrace_profile_enable_event(struct ftrace_event_call *event)
> >  {
> > +	char *buf;
> > +	int ret;
> > +
> >  	if (atomic_inc_return(&event->profile_count))
> >  		return 0;
> >  
> > -	return event->profile_enable();
> > +	buf = (char *)alloc_percpu(profile_buf_t);
> > +	if (!buf)
> > +		return -ENOMEM;
> 
> Ingo,
> 
> Did you pull in the version that allocates a buffer for each event? I 
> thought Frederic made just a global per cpu buffer that all events can 
> use. The buffer is just a temporary storage that will be too big to 
> put on the stack.

Yeah - i had v1 briefly - then pulled in the later (v3) one.

	Ingo