Re: [RFC] Design + prototype: Multiple kprobes at an address - take 2]

[Maneesh Soni <maneesh at in dot ibm dot com>]

On Thu, Apr 07, 2005 at 09:01:46AM -0400, Ananth N Mavinakayanahalli wrote:
> 2.3.1	Registering a kprobe
> 
> 1. Kprobes are registered with a call to register_kprobe().
> 2. The kprobe infrastructure determines if a probe already exists at
>    the requested location. 
> 3. If this is a new kprobe, registration proceeds as it done
>    currently.
> 4. If a kprobe already exists at the requested address:
>    a. If the call is to register a jprobe, the new handler 
>       registration fails since we can't have a jprobe and a kprobe 
>       at the same address. Whether a registration is for a jprobe
>       or not is determined by the presence of a break_handler.
>    b. If the call is to register a kprobe and if the pre_handler
>       for the existing kprobe is not aggr_pre_handler, we then:
>       	1. Allocate a aggr_probe structure
> 	2. Populate the aggr_probe's kprobe structure with 
> 	   appropriate values for the addr, opcode, ainsn fields
> 	3. Set the handlers for the aggr_probe's kprobe to custom
> 	   handlers with list walk logic
> 	4. Add the existing kprobe to the aggr_probe's kprobes list
> 	5. Initialize the new kprobe
> 	6. Add the new kprobe to the aggr_probe's kprobes list
>    c. If the call is to register a kprobe and if the pre_handler
>       for the existing kprobe is aggr_pre_handler, it means that
>       we already have an aggr_probe defined at this address.
>       Unless the new probe to register is not a jprobe, it is 
>       initialized and added to the aggr_probe's "kprobes" list.
>       The registration fails if the new probe is a jprobe.
>       
> 2.3.2	Unregistering a kprobe
> 
> 1. Kprobes are unregistered by making a call to unregister_kprobe()
> 2. If a kprobe exists and its pre_handler is aggr_pre_handler(),
>    then, this is one of the possibly many kprobes at the address:
>    a. Delete the kprobe from the aggr_probe's kprobes list.
>    b. If the aggr_probe's kprobe list is empty, then we free the
>       aggr_probe structure.
> 3. If the kprobe to be unregistered is the only one at the address,
>    remove it as is done currently.
>  

How about keeping the kprobe core as it is by moving the existing register
and unregistering functionality to (un)register_kprobe_single(struct kprobe *)
and the multiple handler probes are handled in (un)register_kprobe(), somewhat 
like this

int register_kprobe(struct kprobe *kp)
{
	err = register_kprobe(kp);
	if (err == -EEXIST) {
		err = add_aggr_kprobe(kp);
	}

}

Thanks
Maneesh

-- 
Maneesh Soni
Linux Technology Center, 
IBM India Software Labs,
Bangalore, India
email: maneesh@in.ibm.com
Phone: 91-80-25044990