Re: [Xen-devel] [PATCH] x86emul/test: disable pie for 64-bit builds

[Wei Liu <wei.liu2@xxxxxxxxxx>]

On Mon, Sep 25, 2017 at 05:54:41AM -0600, Jan Beulich wrote:
> >>> On 25.09.17 at 13:43, <wei.liu2@xxxxxxxxxx> wrote:
> > On Mon, Sep 25, 2017 at 05:35:05AM -0600, Jan Beulich wrote:
> >> >>> On 25.09.17 at 12:49, <wei.liu2@xxxxxxxxxx> wrote:
> >> > PIE may (and commonly will) result in the binary being loaded above
> >> > the 4Gb boundary, which can't work with at least the VZEROUPPER compat
> >> > mode test.
> >> > 
> >> > Reported-by: Wei Liu <wei.liu2@xxxxxxxxxx>
> >> > Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> >> > Signed-off-by: Wei Liu <wei.liu2@xxxxxxxxxx>
> >> > ---
> >> > Cc: Jan Beulich <jbeulich@xxxxxxxx>
> >> > Cc: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> >> > 
> >> > With this patch, vzeroupper passes, but one other test fails.
> >> > Testing SSE packed single 64-bit code sequence...[line 368] failed!
> >> 
> >> Feel free to mail me the binary again, albeit that one's going to
> >> be more difficult to debug without being able to see it myself.
> >> 
> >> > @@ -98,7 +98,9 @@ asm:
> >> >  
> >> >  asm/%: asm ;
> >> >  
> >> > -HOSTCFLAGS += $(CFLAGS_xeninclude) -I.
> >> > +HOSTCFLAGS-x86_64 :=
> >> > +$(call cc-option-add,HOSTCFLAGS,HOSTCC,-no-pie)
> >> > +HOSTCFLAGS += $(CFLAGS_xeninclude) -I. $(HOSTCFLAGS-$(XEN_COMPILE_ARCH))
> >> 
> >> I don't understand this change to my original patch: You now
> >> conditionally add -no-pie to HOSTCFLAGS (i.e. also for 32-bit builds),
> >> and HOSTCFLAGS-x86_64 remains empty. I also don't see why the
> > 
> > My bad. The flag should be conditionally added to HOSTCFLAGS-x86_64.
> > 
> >> addition needs to be conditional: In order to be able to build the
> >> entire test, a reasonably new tool chain is needed anyway (much
> >> newer than what we require for building everything else). And finally
> > 
> > It needs to be conditional because not all gcc versions support -no-pie.
> 
> You mean older one (which would be no problem, as said) or even
> up-to-date ones (due to the way they're being configured)?

Let me be precise because I don't know which version you count as old or
up-to-date.

Gcc <5.4 has -pie but no -no-pie. IIRC passing -no-pie will cause the
linker to return an error. I don't have a machine that old to verify it
though.

> 
> >> - is there a difference between -no-pie and -fno-PIE / -fno-pie?
> >> 
> > 
> > I can't tell the difference by reading the manpage TBH. I only know one
> > works while the other doesn't by trial and error.
> 
> Oh, that's certainly worthwhile adding to the description then.

After reading a bit more into the manual: the -no-pie option is for
the linker, while the -fno-pie option is for code generation.

The build rune is in fact using HOSTCC to link the executable, hence we
need -fno-pie.

I'm not sure why omitting -fno-PIE is not a problem, but it works.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel