This is the mail archive of the systemtap@sourceware.org mailing list for the systemtap project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug translator/10228] Add more vma-tracking for user space symbol/backtraces


------- Additional Comments From fche at redhat dot com  2009-06-04 11:18 -------
(In reply to comment #2)
> There is also -x <pid> which sets the target to the process with that id.
> Ideally both -c <exec> and -x <pid> imply -d for the target process, so that the
> symbol and unwind tables are read (and any depended shared libraries).

This one may be too hard to do, considering that -c <exec> would require
heuristics that analyze the command line, guess at <exec>'s shlib needs,
then perhaps guess what child processes it might exec/dlopen later.
Maybe. (See also bug #6880).

But -x is even harder since there is no info at all at translate time
proper.  One could assume <host>=<target> and <runtime>=<impending>
and poke at /proc/PID/maps but that seems wrong.

Or one could investigate staprun-time on-the-fly feeding of this data,
though this would open worm cans aplenty.


> A vma-tracker on all processes doesn't do much good if none of the symbol and
> unwind tables for those processes and shared libraries are loaded.

Well, at the very least the code can produce <module>+offset type symbol
addresses for such cases.  Anyway we're hoping to get better than this
extreme "all processes ... none of the ... tables" configuration typically.


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=10228

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]