Queries on SystemTap - with focus on User Space debugging

[Josh Stone]

On 05/26/2011 03:23 AM, Frank Ch. Eigler wrote:
> Hi -
> 
> On Thu, May 26, 2011 at 03:24:54PM +0530, Manjusha Ajay wrote:
>> [...]
>> probe kernel.function("send_sig") {
>> if ($sig == 11 && p && task_pid(p) == target()) {
>>      print_ubacktrace()
>>   }
>> }
>>
>> Also got a warning when I ran the above script
>>
>>> stap -c "./test_segfault' test.stp
>> WARNING:  read-only local variable 'p'  : identifier 'p' at test.stp:19:20
> 
> Ah yes, an appropriate warning: I meant $p instead of p above,
> since we're trying to extract the p parameter from the context
> of the send_sig function call.
> 
> It appears difficult to place an architecture-neutral kernel probe at
> the point where a page fault is about to be declared a SIGSEGV type
> error.  Might try kernel.function("force_sig_info") instead,
> filtering on $sig==11.

We have a predefined tapset that is supposed to cover all the different
ways signals can be sent, documented thusly:

/**
 * probe signal.send - Signal being sent to a process
 * Arguments:
 * @sig: The number of the signal
 * @sig_name: A string representation of the signal
 * @sig_pid: The PID of the process receiving the signal
 * @pid_name: The name of the signal recipient
 * @si_code: Indicates the signal type
 * @task: A task handle to the signal recipient
 * @sinfo: The address of siginfo struct
 * @shared: Indicates whether the signal is shared by the thread group
 * @send2queue: Indicates whether the signal is sent to an existing
 * sigqueue
 * @name: The name of the function used to send out the signal
 *
 * Context:
 *  The signal's sender.
 *
 */

So try something like:

  probe signal.send {
    if (sig == 11 && sig_pid == target()) {
      print_ubacktrace()
    }
  }