Subject: Re: Musings on ld.elf_so and symbol lookup caching
From: David Laight <firstname.lastname@example.org>
Date: 10/04/2002 20:18:18
On Fri, Oct 04, 2002 at 06:45:45PM +0000, Charles M. Hannum wrote:
> > You can't - because there are no system calls involved in symbol
> > fixup, neither does ktrace timestamp its tracepoints.
> > The timestamp was VERY useful in sorting out where the
> > bottlenecks were.
> Actually, the write(2)s to output the message are system calls, and in
> fact ktrace(1) does timestamp its output. I use both of those
> properties frequently.
I have used ktrace (and truss on SVR4/Solaris) with debugging
message turned on as well. However the output isn't exactly
easy to read - especially when there are the massive reams of
it that the RTLD_DEBUG_RELOC trace generates for something
like mozilla starting up.
Also I (made a resonable attempt) at removing the cost of
the printf and timestamp (between 4 and 5 usecs) from the trace.
I don't see a problem with such isolated changes.
David Laight: email@example.com