Re: [PATCH 3/6] uprobes/tracing: Make uprobe_{trace,perf}_print() uretprobe-friendly

[Srikar Dronamraju <srikar@linux.vnet.ibm.com>]

* Oleg Nesterov <oleg@redhat.com> [2013-04-01 18:08:51]:

> Change uprobe_trace_print() and uprobe_perf_print() to check
> is_ret_probe() and fill ring_buffer_event accordingly.
> 
> Also change uprobe_trace_func() and uprobe_perf_func() to not
> _print() if is_ret_probe() is true. Note that we keep ->handler()
> nontrivial even for uretprobe, we need this for filtering and for
> other potential extensions.
> 
> Signed-off-by: Oleg Nesterov <oleg@redhat.com>
> ---
>  kernel/trace/trace_uprobe.c |   42 +++++++++++++++++++++++++++++++++---------
>  1 files changed, 33 insertions(+), 9 deletions(-)
> 
> diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
> index e91a354..db2718a 100644
> --- a/kernel/trace/trace_uprobe.c
> +++ b/kernel/trace/trace_uprobe.c
> @@ -515,15 +515,26 @@ static void uprobe_trace_print(struct trace_uprobe *tu,
>  	int size, i;
>  	struct ftrace_event_call *call = &tu->call;
> 
> -	size = SIZEOF_TRACE_ENTRY(1) + tu->size;
> +	if (is_ret_probe(tu))

One nit:
Here and couple of places below .. we could check for func instead of
is_ret_probe() right?
Or is there an advantage of checking is_ret_probe() over func?

> +		size = SIZEOF_TRACE_ENTRY(2);
> +	else
> +		size = SIZEOF_TRACE_ENTRY(1);
> +
>  	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
> -						  size, 0, 0);
> +						  size + tu->size, 0, 0);
>  	if (!event)
>  		return;
> 
>  	entry = ring_buffer_event_data(event);
> -	entry->vaddr[0] = instruction_pointer(regs);
> -	data = DATAOF_TRACE_ENTRY(entry, 1);
> +	if (is_ret_probe(tu)) {
> +		entry->vaddr[0] = func;
> +		entry->vaddr[1] = instruction_pointer(regs);
> +		data = DATAOF_TRACE_ENTRY(entry, 2);
> +	} else {
> +		entry->vaddr[0] = instruction_pointer(regs);
> +		data = DATAOF_TRACE_ENTRY(entry, 1);
> +	}
> +
>  	for (i = 0; i < tu->nr_args; i++)
>  		call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
> 
> @@ -534,7 +545,8 @@ static void uprobe_trace_print(struct trace_uprobe *tu,
>  /* uprobe handler */
>  static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs)
>  {
> -	uprobe_trace_print(tu, 0, regs);
> +	if (!is_ret_probe(tu))
> +		uprobe_trace_print(tu, 0, regs);

Should this hunk be in the previous patch?

Also something for the future:
Most times a user uses a return probe, the user probably wants to probe
the function entry too. So should we extend the abi from p+r to
p+r+..<something else> to mean it traces both function entry and return.
Esp given that uretprobe has been elegantly been designed to make this a
possibility.


>  	return 0;
>  }
> 
> @@ -783,7 +795,11 @@ static void uprobe_perf_print(struct trace_uprobe *tu,
>  	void *data;
>  	int size, rctx, i;
> 
> -	size = SIZEOF_TRACE_ENTRY(1);
> +	if (is_ret_probe(tu))
> +		size = SIZEOF_TRACE_ENTRY(2);
> +	else
> +		size = SIZEOF_TRACE_ENTRY(1);
> +
>  	size = ALIGN(size + tu->size + sizeof(u32), sizeof(u64)) - sizeof(u32);
>  	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
>  		return;
> @@ -794,8 +810,15 @@ static void uprobe_perf_print(struct trace_uprobe *tu,
>  		goto out;
> 
>  	ip = instruction_pointer(regs);
> -	entry->vaddr[0] = ip;
> -	data = DATAOF_TRACE_ENTRY(entry, 1);
> +	if (is_ret_probe(tu)) {
> +		entry->vaddr[0] = func;
> +		entry->vaddr[1] = ip;
> +		data = DATAOF_TRACE_ENTRY(entry, 2);
> +	} else {
> +		entry->vaddr[0] = ip;
> +		data = DATAOF_TRACE_ENTRY(entry, 1);
> +	}
> +
>  	for (i = 0; i < tu->nr_args; i++)
>  		call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
> 
> @@ -811,7 +834,8 @@ static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
>  	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
>  		return UPROBE_HANDLER_REMOVE;
> 
> -	uprobe_perf_print(tu, 0, regs);
> +	if (!is_ret_probe(tu))
> +		uprobe_perf_print(tu, 0, regs);
>  	return 0;
>  }
> 
> -- 
> 1.5.5.1
>