On 29/11/2016 10:33, Stefan Hajnoczi wrote: > On Mon, Nov 28, 2016 at 04:42:54PM +0100, Paolo Bonzini wrote: >> >> >> On 28/11/2016 16:38, Paolo Bonzini wrote: >>> >>> >>> On 28/11/2016 16:24, Adrian Bunk wrote: >>>> On Mon, Nov 28, 2016 at 04:05:33PM +0100, Paolo Bonzini wrote: >>>>> >>>>> >>>>> On 27/11/2016 17:28, Adrian Bunk wrote: >>>>>> Building qemu fails in distributions where gcc enables PIE >>>>>> by default (e.g. Debian unstable) with: >>>>>> /usr/bin/ld: -r and -pie may not be used together >>>>>> >>>>>> -r and -pie cannot be used together in the linker, >>>>>> and position independent is already relocatable. >>>>>> >>>>>> Use -r instead of -Wl,-r to avoid gcc passing -r to the >>>>>> linker when PIE is enabled. >>>>>> >>>>>> Signed-off-by: Adrian Bunk <b...@stusta.de> >>>>> >>>>> I think this is a bug in the linker. If the linker is producing >>>>> relocatable objects by default, it has no reason to refuse -r. Have you >>>>> tried asking the binutils folks about it too? >>>> >>>> The linker knows nothing about this default, gcc is passing -pie >>>> to the linker. >>> >>> The linker is receiving "-r -pie". It can satisfy the requirement of >>> producing a relocatable object by discarding the "-r", but it doesn't. >>> That'd be a linker bug. >>> >>> But in fact ELF makes PIE ET_DYN and relocatable ET_REL. That would >>> make the linker error the right thing, but then I don't understand what >>> you mean by "position independent is already relocatable". >> >> Aha, I looked at GCC source code and this is incorrect: "Use -r instead >> of -Wl,-r to avoid gcc passing -r to the linker when PIE is enabled". >> When GCC sees -r (as opposed to -Wl,-r) it does not pass -pie to the linker. > > Should we merge this patch for 2.8 with the commit updated to reflect this?
Yes, I've already sent a pull request for it. Paolo
