Re: Customization of LKET

[Li Guanglei <guanglei at cn dot ibm dot com>]

Li Guanglei ??:
> Frank Ch. Eigler ??:
>
> lket_trace_extra is actually a wrapper around printf 
> [...]
> I am thinking of revert my changes of lket_trace_extra() to
> translator.

I though of another way of allowing the user add arbitrary extra trace 
data using just printf. It has the same convenience as deprecated 
lket_trace_extra(), no flag/%2n is needed, no changes to translator is 
needed, and user can call printf() multiple times. I implemented it 
and tested it for a while, and it worked fine for me.

And I think we can also add a function register_user_event() whose 
signature is:
  function register_user_event(grpid:long, hookid:long, fmt:string,
     name:string)
Then event register provides a way to tell the post-processing app the 
data/name of the extra trace data

The following shows an example of the above:

probe addevent.regevt
{
    register_user_event(GROUP_NETDEV, HOOKID_NETDEV_RECEIVE,
	"STRING:INT8:INT16", "sample string:sample id1:sample id2")
}

probe addevent.netdev.receive
{
    printf("%0s%1b%2b", "sample str", 11, 112)
}

Basically, this new way of user customizable tracing is done by a 
function named update_record() which will is called in a epilogue mode 
alias so it will be the last one to be called and it will update the 
total length of the trace data according to _stp_pbuf_len[cpu], 
including those logged by _lket_trace() and printf(). Every trace 
record has the format of:

total_len, sys_len(without user trace data), groupid, hookid, ...

> it shows that the size of context will increase from 6608 to 51408  if 
> MAXSTRINGLEN is defined as 1024. Seems acceptable, right?
I want also delete backtrace() function now. If a user want to print 
backtrace, he can just use printf and backtrace() function, e.g:
  printf("%1b%0s", 11, backtrace())

Any comments?

- Guanglei