Re: [PATCH] Hurd: Enable ifunc by default
On Wed, 13 Jan 2021, Thomas Schwinge wrote: > Hi! > > Thanks (and sorry for the delay), pushed "Hurd: Enable ifunc by default" > to master branch in commit e9cb89b936f831a02318d45fc4ddb06f7be55ae4, and > cherry-picked into releases/gcc-10 branch in commit > 92b131491c22eb4e4b663d226e9d97f1fd693063, releases/gcc-9 branch in commit > 0313ce139f4ca3c96db9dc82125ec9e4a167a224, releases/gcc-8 branch in commit > 975b0fa0f43e84bed3cb1b2b593132bc219f962c, see attached. I'm not sure what toolchain component the underlying bug is in, but this GCC commit (verified in the releases/gcc-10 case) results in a glibc build failure for i686-gnu with build-many-glibcs.py. https://sourceware.org/pipermail/libc-testresults/2021q1/007378.html The error is: /scratch/jmyers/glibc-bot/install/compilers/i686-gnu/lib/gcc/i686-glibc-gnu/11.0.0/../../../../i686-glibc-gnu/bin/ld: /scratch/jmyers/glibc-bot/build/compilers/i686-gnu/glibc/i686-gnu/elf/librtld.os: in function `hurd_file_name_lookup_retry': (.text+0x1e08e): undefined reference to `strncpy' collect2: error: ld returned 1 exit status Makefile:584: recipe for target '/scratch/jmyers/glibc-bot/build/compilers/i686-gnu/glibc/i686-gnu/elf/ld.so' failed make[3]: *** [/scratch/jmyers/glibc-bot/build/compilers/i686-gnu/glibc/i686-gnu/elf/ld.so] Error 1 -- Joseph S. Myers jos...@codesourcery.com
Re: [PATCH] Hurd: Enable ifunc by default
Hello, Joseph Myers, le lun. 18 janv. 2021 20:05:44 +, a ecrit: > On Wed, 13 Jan 2021, Thomas Schwinge wrote: > > Thanks (and sorry for the delay), pushed "Hurd: Enable ifunc by default" > > to master branch in commit e9cb89b936f831a02318d45fc4ddb06f7be55ae4, and > > cherry-picked into releases/gcc-10 branch in commit > > 92b131491c22eb4e4b663d226e9d97f1fd693063, releases/gcc-9 branch in commit > > 0313ce139f4ca3c96db9dc82125ec9e4a167a224, releases/gcc-8 branch in commit > > 975b0fa0f43e84bed3cb1b2b593132bc219f962c, see attached. > > I'm not sure what toolchain component the underlying bug is in, but this > GCC commit (verified in the releases/gcc-10 case) results in a glibc build > failure for i686-gnu with build-many-glibcs.py. > > https://sourceware.org/pipermail/libc-testresults/2021q1/007378.html > > The error is: > > /scratch/jmyers/glibc-bot/install/compilers/i686-gnu/lib/gcc/i686-glibc-gnu/11.0.0/../../../../i686-glibc-gnu/bin/ld: > > /scratch/jmyers/glibc-bot/build/compilers/i686-gnu/glibc/i686-gnu/elf/librtld.os: > in function `hurd_file_name_lookup_retry': > (.text+0x1e08e): undefined reference to `strncpy' Ah, I believe I had tested that (which is precisely why I asked for that commit to be done in gcc), but I'll have a look, thanks. Samuel
Re: [PATCH] Hurd: Enable ifunc by default
(leaving gcc out, it's really out of the story, it just happens to
expose support for ifunc)
Joseph Myers, le lun. 18 janv. 2021 20:05:44 +, a ecrit:
> /scratch/jmyers/glibc-bot/install/compilers/i686-gnu/lib/gcc/i686-glibc-gnu/11.0.0/../../../../i686-glibc-gnu/bin/ld:
>
> /scratch/jmyers/glibc-bot/build/compilers/i686-gnu/glibc/i686-gnu/elf/librtld.os:
> in function `hurd_file_name_lookup_retry':
> (.text+0x1e08e): undefined reference to `strncpy'
The story seems complex and related to the glibc rtld build rules.
- We need hurd/lookup-retry.c in the rtld
- It happens to use strncpy since ee11682d4f5 ("hurd: Fix strcpy calls")
- strncpy happens to have a hidden def in libc
- hurd/lookup-retry.os thus refers to __GI_strncpy rather than strncpy
- dl-allobjs.os thus includes strncpy-c.os (that provides __GI_strncpy),
and not strncpy.os (that provides strncpy)
- librtld.map thus contains strncpy-c.os only, and not strncpy.os
- thus no rtld-strncpy.os is getting built
- thus strncpy is not available in rtld
- but strncpy doesn't have a hidden def in rtld
- and thus hurd/rtld-lookup-retry.os refers to strncpy, not __GI_strncpy
- thus the link failure.
The base issue I see here is that dl-allobjs is based on libc-built
objects that might be using hidden defs while rtld-built objects might
not.
But we could also say that it's strncpy that should also have a hidden
def in rtld. The attached patch does this, is this ok?
Samuel
diff --git a/include/libc-symbols.h b/include/libc-symbols.h
index ea126ae70c..051d6b9492 100644
--- a/include/libc-symbols.h
+++ b/include/libc-symbols.h
@@ -856,6 +856,8 @@ for linking")
# define HIDDEN_BUILTIN_JUMPTARGET(name) HIDDEN_JUMPTARGET(name)
#endif
+#define rtld_hidden_builtin_proto(name, attrs...) rtld_hidden_proto (name,
##attrs)
+
#if IS_IN (libutil)
# define libutil_hidden_proto(name, attrs...) hidden_proto (name, ##attrs)
# define libutil_hidden_tls_proto(name, attrs...) \
diff --git a/include/string.h b/include/string.h
index 81dab39891..27cbde47ef 100644
--- a/include/string.h
+++ b/include/string.h
@@ -144,6 +144,7 @@ libc_hidden_builtin_proto (strcspn)
libc_hidden_builtin_proto (strlen)
libc_hidden_builtin_proto (strncmp)
libc_hidden_builtin_proto (strncpy)
+rtld_hidden_builtin_proto (strncpy)
libc_hidden_builtin_proto (strpbrk)
libc_hidden_builtin_proto (stpcpy)
libc_hidden_builtin_proto (strrchr)
Re: [PATCH] Hurd: Enable ifunc by default
On Mon, Jan 18, 2021 at 4:04 PM Samuel Thibault via Libc-alpha
wrote:
>
> (leaving gcc out, it's really out of the story, it just happens to
> expose support for ifunc)
>
> Joseph Myers, le lun. 18 janv. 2021 20:05:44 +, a ecrit:
> > /scratch/jmyers/glibc-bot/install/compilers/i686-gnu/lib/gcc/i686-glibc-gnu/11.0.0/../../../../i686-glibc-gnu/bin/ld:
> >
> > /scratch/jmyers/glibc-bot/build/compilers/i686-gnu/glibc/i686-gnu/elf/librtld.os:
> > in function `hurd_file_name_lookup_retry':
> > (.text+0x1e08e): undefined reference to `strncpy'
>
> The story seems complex and related to the glibc rtld build rules.
>
> - We need hurd/lookup-retry.c in the rtld
> - It happens to use strncpy since ee11682d4f5 ("hurd: Fix strcpy calls")
> - strncpy happens to have a hidden def in libc
> - hurd/lookup-retry.os thus refers to __GI_strncpy rather than strncpy
> - dl-allobjs.os thus includes strncpy-c.os (that provides __GI_strncpy),
> and not strncpy.os (that provides strncpy)
> - librtld.map thus contains strncpy-c.os only, and not strncpy.os
> - thus no rtld-strncpy.os is getting built
> - thus strncpy is not available in rtld
> - but strncpy doesn't have a hidden def in rtld
> - and thus hurd/rtld-lookup-retry.os refers to strncpy, not __GI_strncpy
> - thus the link failure.
>
> The base issue I see here is that dl-allobjs is based on libc-built
> objects that might be using hidden defs while rtld-built objects might
> not.
>
> But we could also say that it's strncpy that should also have a hidden
> def in rtld. The attached patch does this, is this ok?
>
String function usage in ld.so must be carefully managed. Why doesn't
hurd ld.so build script mark strncpy as needed?
--
H.J.
Re: [PATCH] Hurd: Enable ifunc by default
H.J. Lu, le lun. 18 janv. 2021 16:17:16 -0800, a ecrit: > String function usage in ld.so must be carefully managed. Sure. > Why doesn't hurd ld.so build script mark strncpy as needed? I don't even know what "hurd ld.so build script" is. AFAIK there is not much hurd-specific there except including the RPC stubs ($(objpfx)librtld.map: $(rpcuserlibs:.so=_pic.a)). Samuel
Re: [PATCH] Hurd: Enable ifunc by default
Samuel Thibault, le mar. 19 janv. 2021 01:21:05 +0100, a ecrit: > H.J. Lu, le lun. 18 janv. 2021 16:17:16 -0800, a ecrit: > > Why doesn't hurd ld.so build script mark strncpy as needed? > > I don't even know what "hurd ld.so build script" is. More precisely, I don't see why hurd should be doing anything special while everything is already taken care of by elf/Makefile's librtld.map file scripts. Samuel
Re: [PATCH] Hurd: Enable ifunc by default
On Mon, Jan 18, 2021 at 4:24 PM Samuel Thibault wrote: > > Samuel Thibault, le mar. 19 janv. 2021 01:21:05 +0100, a ecrit: > > H.J. Lu, le lun. 18 janv. 2021 16:17:16 -0800, a ecrit: > > > Why doesn't hurd ld.so build script mark strncpy as needed? > > > > I don't even know what "hurd ld.so build script" is. > > More precisely, I don't see why hurd should be doing anything special > while everything is already taken care of by elf/Makefile's librtld.map > file scripts. > I am taking a look. -- H.J.
[PATCH] Hurd: Add rtld-strncpy-c.c
On Mon, Jan 18, 2021 at 4:40 PM H.J. Lu wrote: > > On Mon, Jan 18, 2021 at 4:24 PM Samuel Thibault > wrote: > > > > Samuel Thibault, le mar. 19 janv. 2021 01:21:05 +0100, a ecrit: > > > H.J. Lu, le lun. 18 janv. 2021 16:17:16 -0800, a ecrit: > > > > Why doesn't hurd ld.so build script mark strncpy as needed? > > > > > > I don't even know what "hurd ld.so build script" is. > > > > More precisely, I don't see why hurd should be doing anything special > > while everything is already taken care of by elf/Makefile's librtld.map > > file scripts. > > > > I am taking a look. > Try this. I have no idea if it really works on hurd. -- H.J. From a73010aaab13c5cd67ded77091601b8cd01e3fcf Mon Sep 17 00:00:00 2001 From: "H.J. Lu" Date: Mon, 18 Jan 2021 18:19:16 -0800 Subject: [PATCH] Hurd: Add rtld-strncpy-c.c All IFUNC functions which are used in ld.so must have a rtld version if the IFUNC version isn't safe to use in ld.so. --- sysdeps/mach/hurd/i386/i686/multiarch/rtld-strncpy-c.c | 1 + 1 file changed, 1 insertion(+) create mode 100644 sysdeps/mach/hurd/i386/i686/multiarch/rtld-strncpy-c.c diff --git a/sysdeps/mach/hurd/i386/i686/multiarch/rtld-strncpy-c.c b/sysdeps/mach/hurd/i386/i686/multiarch/rtld-strncpy-c.c new file mode 100644 index 00..b759c999eb --- /dev/null +++ b/sysdeps/mach/hurd/i386/i686/multiarch/rtld-strncpy-c.c @@ -0,0 +1 @@ +#include -- 2.29.2
