>>>>> On Mon, 16 Dec 2002 19:42:51 -0800, Richard Henderson <rth@twiddle.net> said: Richard> On Mon, Dec 16, 2002 at 06:15:28PM -0800, David Mosberger Richard> wrote: >> For someone who cares about modules and ia64 performance: you >> might want to consider to use "brl" in the PLT stub instead of >> the standardized code sequence. Richard> You'll have to figure out how to find the plt stub though. Richard> It's not marked in any way. The only thing you get a Richard> pointer to is the .IA_64.pltoff entry. Richard> For userland, the .plt stub has to use an indirect branch Richard> because the first invocation goes through the dynamic Richard> linker. Then we update the .pltoff entry and so subsequent Richard> calls go direct to the target. I assume you do early binding of the PLTs for kernel modules, right? If so, patching the PLT code stub is trivial (even in user-land you could do code patching, but that's obviously (a) not so trivial for multi-threaded programs and (b) hurts sharing). For user-level land, we should offer PLT-stub-inlining, but we discussed this in private already... --davidReceived on Mon Dec 16 19:52:29 2002
This archive was generated by hypermail 2.1.8 : 2005-08-02 09:20:11 EST