[RFA] Improve performance with lots of shared libraries

[Gary Benson <gbenson at redhat dot com>]

Hi all,

When inferiors load a lot of shared libraries gdb spends a lot of
time processing shared library breakpoint events.  These have a lot
of associated bookkeeping that is not easily avoided.

This patch modifies gdb to disable the shared library breakpoint
when it is not required, specifically:

 * when stop-on-solib-events is off,
 * when there are no pending breakpoints,
 * and when libthread_db is loaded

I have a simple test which dlopens 1000 shared libraries.  On my
machine it runs instantly outside of gdb, but takes roughly a minute
when run within gdb.  With this patch, it runs instantly in gdb too.

The idea for this was not mine, it was Jan Kratochvil's, and he
very definitely deserves the credit for it.  I wrote the code though,
so the bugs are all my fault :)

There are two things I'm not sure about with it as it stands.  One is
to do with program spaces: I noticed that breakpoints have a program
space, but breakpoint locations also have a program space.  Is the way
I have used these correct?

The other issue I have is the way it detects whether libthread_db is
loaded.  This should work fine on platforms that use a libthread_db,
but some platforms will have this optimization disabled.  Nothing will
get worse, but some platforms will not get better when they could.
Have I gone about this the wrong way?

I would definitely appreciate feedback from those of you who are using
gdb on applications with many shared libraries, to know if you are
getting any improvement.

Cheers,
Gary

-- 
http://gbenson.net/

Attachment: solib-break-disabler-1.patch
Description: Text document