| Message ID | 002601d055e3$89cc7740$9d6565c0$@rt-rk.com |
|---|---|
| State | New |
| Headers | show |
On 03 Mar 2015 19:54, Petar Jovanovic wrote: > For x86-32, it will break. So, as it appears, the test is not portable. > I can simplify it in the following way: > > --- /dev/null > +++ b/elf/tst-split-dynreloc.lds > @@ -0,0 +1,10 @@ > +SECTIONS > +{ > + .rela.dyn : { *(.rela.dyn) } > + .rel.dyn : { *(.rel.dyn) } > + . += 1000; > + .rela.plt : { *(.rela.plt) } > + .rel.plt : { *(.rel.plt) } > +} > +INSERT BEFORE .init; > > This would also remove the need for __((section(".bar"))) in the test. > What do you think? that seems kind of dicey. i think doing it explicitly like you were before is less risky ? i thought we used linker scripts in glibc tests more, but i'm not actually finding any. i'm not sure if we're trying to avoid them. there's no way to produce a test that doesn't use linker scripts but still triggers the problem you're fixing ? -mike
> that seems kind of dicey. i think doing it explicitly like you were before is less risky ? I would agree. > i thought we used linker scripts in glibc tests more, but i'm not actually finding any. i'm not sure if we're trying to avoid them. > there's no way to produce a test that doesn't use linker scripts but still triggers the problem you're fixing ? The problem itself will appear wihout linkers scripts too, but I do not know if we can have a small test with no use of linker scripts. We could also omit a test if none can fit the existing test suite? Regards, Petar -----Original Message----- From: Mike Frysinger [mailto:vapier@gentoo.org] Sent: Tuesday, March 3, 2015 10:48 PM To: Petar Jovanovic Cc: libc-alpha@sourceware.org Subject: Re: [PATCH v3] Fix dynamic linker issue with bind-now On 03 Mar 2015 19:54, Petar Jovanovic wrote: > For x86-32, it will break. So, as it appears, the test is not portable. > I can simplify it in the following way: > > --- /dev/null > +++ b/elf/tst-split-dynreloc.lds > @@ -0,0 +1,10 @@ > +SECTIONS > +{ > + .rela.dyn : { *(.rela.dyn) } > + .rel.dyn : { *(.rel.dyn) } > + . += 1000; > + .rela.plt : { *(.rela.plt) } > + .rel.plt : { *(.rel.plt) } > +} > +INSERT BEFORE .init; > > This would also remove the need for __((section(".bar"))) in the test. > What do you think? that seems kind of dicey. i think doing it explicitly like you were before is less risky ? i thought we used linker scripts in glibc tests more, but i'm not actually finding any. i'm not sure if we're trying to avoid them. there's no way to produce a test that doesn't use linker scripts but still triggers the problem you're fixing ? -mike
Mike, what do you suggest? Regards, Petar -----Original Message----- From: Petar Jovanovic [mailto:petar.jovanovic@rt-rk.com] Sent: Wednesday, March 4, 2015 2:11 AM To: 'Mike Frysinger' Cc: 'libc-alpha@sourceware.org' Subject: RE: [PATCH v3] Fix dynamic linker issue with bind-now > that seems kind of dicey. i think doing it explicitly like you were before is less risky ? I would agree. > i thought we used linker scripts in glibc tests more, but i'm not actually finding any. i'm not sure if we're trying to avoid them. > there's no way to produce a test that doesn't use linker scripts but still triggers the problem you're fixing ? The problem itself will appear wihout linkers scripts too, but I do not know if we can have a small test with no use of linker scripts. We could also omit a test if none can fit the existing test suite? Regards, Petar -----Original Message----- From: Mike Frysinger [mailto:vapier@gentoo.org] Sent: Tuesday, March 3, 2015 10:48 PM To: Petar Jovanovic Cc: libc-alpha@sourceware.org Subject: Re: [PATCH v3] Fix dynamic linker issue with bind-now On 03 Mar 2015 19:54, Petar Jovanovic wrote: > For x86-32, it will break. So, as it appears, the test is not portable. > I can simplify it in the following way: > > --- /dev/null > +++ b/elf/tst-split-dynreloc.lds > @@ -0,0 +1,10 @@ > +SECTIONS > +{ > + .rela.dyn : { *(.rela.dyn) } > + .rel.dyn : { *(.rel.dyn) } > + . += 1000; > + .rela.plt : { *(.rela.plt) } > + .rel.plt : { *(.rel.plt) } > +} > +INSERT BEFORE .init; > > This would also remove the need for __((section(".bar"))) in the test. > What do you think? that seems kind of dicey. i think doing it explicitly like you were before is less risky ? i thought we used linker scripts in glibc tests more, but i'm not actually finding any. i'm not sure if we're trying to avoid them. there's no way to produce a test that doesn't use linker scripts but still triggers the problem you're fixing ? -mike
Ping. Regards, Petar -----Original Message----- From: Petar Jovanovic [mailto:petar.jovanovic@rt-rk.com] Sent: Saturday, March 7, 2015 3:22 AM To: 'Mike Frysinger' Cc: 'libc-alpha@sourceware.org' Subject: RE: [PATCH v3] Fix dynamic linker issue with bind-now Mike, what do you suggest? Regards, Petar -----Original Message----- From: Petar Jovanovic [mailto:petar.jovanovic@rt-rk.com] Sent: Wednesday, March 4, 2015 2:11 AM To: 'Mike Frysinger' Cc: 'libc-alpha@sourceware.org' Subject: RE: [PATCH v3] Fix dynamic linker issue with bind-now > that seems kind of dicey. i think doing it explicitly like you were before is less risky ? I would agree. > i thought we used linker scripts in glibc tests more, but i'm not actually finding any. i'm not sure if we're trying to avoid them. > there's no way to produce a test that doesn't use linker scripts but still triggers the problem you're fixing ? The problem itself will appear wihout linkers scripts too, but I do not know if we can have a small test with no use of linker scripts. We could also omit a test if none can fit the existing test suite? Regards, Petar -----Original Message----- From: Mike Frysinger [mailto:vapier@gentoo.org] Sent: Tuesday, March 3, 2015 10:48 PM To: Petar Jovanovic Cc: libc-alpha@sourceware.org Subject: Re: [PATCH v3] Fix dynamic linker issue with bind-now On 03 Mar 2015 19:54, Petar Jovanovic wrote: > For x86-32, it will break. So, as it appears, the test is not portable. > I can simplify it in the following way: > > --- /dev/null > +++ b/elf/tst-split-dynreloc.lds > @@ -0,0 +1,10 @@ > +SECTIONS > +{ > + .rela.dyn : { *(.rela.dyn) } > + .rel.dyn : { *(.rel.dyn) } > + . += 1000; > + .rela.plt : { *(.rela.plt) } > + .rel.plt : { *(.rel.plt) } > +} > +INSERT BEFORE .init; > > This would also remove the need for __((section(".bar"))) in the test. > What do you think? that seems kind of dicey. i think doing it explicitly like you were before is less risky ? i thought we used linker scripts in glibc tests more, but i'm not actually finding any. i'm not sure if we're trying to avoid them. there's no way to produce a test that doesn't use linker scripts but still triggers the problem you're fixing ? -mike
On 03 Mar 2015 19:54, Petar Jovanovic wrote: > For x86-32, it will break. So, as it appears, the test is not portable. > I can simplify it in the following way: by break, do you mean it fails to link ? or it doesn't trigger the test ? -mike
-----Original Message----- From: Mike Frysinger [mailto:vapier@gentoo.org] Sent: Wednesday, March 11, 2015 5:09 AM To: Petar Jovanovic Cc: libc-alpha@sourceware.org Subject: Re: [PATCH v3] Fix dynamic linker issue with bind-now > by break, do you mean it fails to link ? or it doesn't trigger the test ? It fails to link with a linker error: "linker error: /usr/bin/ld: .rela.dyn not found for insert" What I can do is to move the Makefile part into sysdeps/x86_64/Makefile. This way, it will be triggered for x86-64 only. What do you say? Regards, Petar
Ping. -----Original Message----- From: Petar Jovanovic [mailto:petar.jovanovic@rt-rk.com] Sent: Thursday, March 12, 2015 5:37 PM To: 'Mike Frysinger' Cc: 'libc-alpha@sourceware.org' Subject: RE: [PATCH v3] Fix dynamic linker issue with bind-now -----Original Message----- From: Mike Frysinger [mailto:vapier@gentoo.org] Sent: Wednesday, March 11, 2015 5:09 AM To: Petar Jovanovic Cc: libc-alpha@sourceware.org Subject: Re: [PATCH v3] Fix dynamic linker issue with bind-now > by break, do you mean it fails to link ? or it doesn't trigger the test ? It fails to link with a linker error: "linker error: /usr/bin/ld: .rela.dyn not found for insert" What I can do is to move the Makefile part into sysdeps/x86_64/Makefile. This way, it will be triggered for x86-64 only. What do you say? Regards, Petar
Ping. -----Original Message----- From: Petar Jovanovic [mailto:petar.jovanovic@rt-rk.com] Sent: Wednesday, April 1, 2015 1:20 AM To: 'Mike Frysinger' Cc: 'libc-alpha@sourceware.org' Subject: RE: [PATCH v3] Fix dynamic linker issue with bind-now Ping. -----Original Message----- From: Petar Jovanovic [mailto:petar.jovanovic@rt-rk.com] Sent: Thursday, March 12, 2015 5:37 PM To: 'Mike Frysinger' Cc: 'libc-alpha@sourceware.org' Subject: RE: [PATCH v3] Fix dynamic linker issue with bind-now -----Original Message----- From: Mike Frysinger [mailto:vapier@gentoo.org] Sent: Wednesday, March 11, 2015 5:09 AM To: Petar Jovanovic Cc: libc-alpha@sourceware.org Subject: Re: [PATCH v3] Fix dynamic linker issue with bind-now > by break, do you mean it fails to link ? or it doesn't trigger the test ? It fails to link with a linker error: "linker error: /usr/bin/ld: .rela.dyn not found for insert" What I can do is to move the Makefile part into sysdeps/x86_64/Makefile. This way, it will be triggered for x86-64 only. What do you say? Regards, Petar
--- /dev/null +++ b/elf/tst-split-dynreloc.lds @@ -0,0 +1,10 @@ +SECTIONS +{ + .rela.dyn : { *(.rela.dyn) } + .rel.dyn : { *(.rel.dyn) } + . += 1000; + .rela.plt : { *(.rela.plt) } + .rel.plt : { *(.rel.plt) } +} +INSERT BEFORE .init; +
For x86-32, it will break. So, as it appears, the test is not portable. I can simplify it in the following way: This would also remove the need for __((section(".bar"))) in the test. What do you think? Regards, Petar -----Original Message----- From: Mike Frysinger [mailto:vapier@gentoo.org] Sent: Monday, March 2, 2015 6:39 PM To: Petar Jovanovic Cc: libc-alpha@sourceware.org Subject: Re: [PATCH v3] Fix dynamic linker issue with bind-now On 02 Mar 2015 17:13, Petar Jovanovic wrote: > --- /dev/null > +++ b/elf/tst-split-dynreloc.lds > @@ -0,0 +1,6 @@ > +SECTIONS > +{ > + .bar : { *(.bar) } > +} > +INSERT AFTER .rela.dyn; what about arches that don't use RELA relocations ? x86 (32-bit) should be such a target and easy for you to test. -mike