Hi, On Fri, 2015-10-16 at 11:36 -0400, Camm Maguire wrote: > James Cowgill <[email protected]> writes: > > On Thu, 2015-10-15 at 14:02 -0400, Camm Maguire wrote: > > > In any case, I have this issue isolated in a qemu emulator, and can > > > foresee a fix. I have a remaining question. On mips32, it seems one > > > can safely assume the caller restores its own gp value after a call. On > > > mips64 this seems to be left up to the callee. Is this correct? Is > > > this affected by any compiler switches? > > > > I can answer that :) > > > > In n64 gp is callee saved, whereas in o32 it is caller saved, so you > > are correct. Since this is an absolute requirement of the ABI, I doubt > > there are any compiler switches for it. > > > > http://techpubs.sgi.com/library/dynaweb_docs/0640/SGI_Developer/books/Mpro_n32_ABI/sgi_html/ch02.html > > > > And thanks again! With gcl 2.6.12-25, I ntend to enforce LD_BIND_NOW on > mips64 only, as the veneer would have to push and pop at least gp and ra > regsiters in all cases just to support the stub, and this is of course > slower. Is there any objection to your knowledge to binaries linked > with -Wl,-z -Wl,now?
Using LD_BIND_NOW should work fine with mips64el - I don't know of any issues with it. James
signature.asc
Description: This is a digitally signed message part
