[Bug testsuite/11514] nd_syscall.exp takes more than 10 times longer than syscall.exp testcase

["jistone at redhat dot com" <sourceware-bugzilla at sourceware dot org>]

------- Additional Comments From jistone at redhat dot com  2010-04-21 19:05 -------
(In reply to comment #10)
> commit bd6593518895fc613b973500c03a3b844bbe68c5
> Author: Mark Wielaard <mjw@redhat.com>
> Date:   Tue Apr 20 22:06:15 2010 +0200
> 
>     PR11514 Only expand kallsyms once for kprobe_derived_probe_group.
>     
>     kprobe_register is really expensive when used for multiple symbol
>     based probes since it will call kallsyms_expand_symbol over and over
>     to uncompress the kernel symbol table. So flip the comparison loop
>     inside out by looping once over the kernel symbol table.
>     
>     * tapsets.cxx (kprobe_derived_probe_group::emit_module_decls): Declare
>       kprobe_resolve helper function.
>       (kprobe_derived_probe_group::emit_module_init): Call kprobe_resolve
>       through kallsyms_on_each_symbol. Never register through symbol name,
>       just check that address has been filled in by kprobe_resolve.

There's a gotcha about this that I mentioned on IRC.  On most archs,
kprobe_lookup_name() is a simple wrapper around kallsyms_lookup_name(), so this
new optimization is perfectly valid.

But on PPC64, it's a little more complicated to deal with function descriptors.
 In some cases the returned address is actually a pointer to the real address,
and in other cases it tries to search again with a '.' added to the name.  We
need to deal with this too if we're bypassing kprobe_lookup_name().


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=11514

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