Re: [PATCH 2/2] tracing: Allocate the ftrace event profile buffer dynamically

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, Sep 18, 2009 at 06:24:14AM +0200, Frederic Weisbecker wrote:
>  static void prof_syscall_enter(struct pt_regs *regs, long id)
>  {
> -	struct syscall_trace_enter *rec;
>  	struct syscall_metadata *sys_data;
> +	struct syscall_trace_enter *rec;
> +	char *raw_data;
>  	int syscall_nr;
>  	int size;
> +	int cpu;
>  
>  	syscall_nr = syscall_get_nr(current, regs);
>  	if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
> @@ -402,20 +404,39 @@ static void prof_syscall_enter(struct pt_regs *regs, long id)
>  	size = ALIGN(size + sizeof(u32), sizeof(u64));
>  	size -= sizeof(u32);
>  
> -	do {
> -		char raw_data[size];
> +	if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
> +		      "profile buffer not large enough"))
> +		return;
> +
> +	/*
> +	 * We are not in nmi. Also we can't be preempted by a sysenter event
> +	 * from interrupt, so we can safely take this buffer without
> +	 * masking interrupts. But we still need rcu_read_lock to protect
> +	 * against concurrent buffer release.
> +	 */
> +	rcu_read_lock();



Oh...now that we use a global buffer, the buffer is not safe anymore against
interrupts or NMIs.

Looks like I will need a third iteration.

Sorry for the noise...