Export __vdso_clock_gettime as GLIBC_PRIVATE
Commit Message
Hi,
We added vDSO support for time functions on i386 recently, but the
__vdso_clock_gettime symbol was kept as an internal symbol. This is
fine for glibc as is, but it breaks Fedora builds since rtkaio now
needs access to __vdso_clock_gettime. I could maintain a patch in
Fedora that exports this symbol as GLIBC_PRIVATE, but I reckon that it
might be more consistent to put it upstream since all of the
architectures that have vDSO time support export the symbol in
GLIBC_PRIVATE already.
Tested on i386 to verify that the build completes successfully. OK to
commit?
Siddhesh
* sysdeps/unix/sysv/linux/i386/Versions (GLIBC_PRIVATE): Add
__vdso_clock_gettime.
Comments
On Thu, Jan 22, 2015 at 12:05 AM, Siddhesh Poyarekar
<siddhesh@redhat.com> wrote:
> Hi,
>
> We added vDSO support for time functions on i386 recently, but the
> __vdso_clock_gettime symbol was kept as an internal symbol. This is
> fine for glibc as is, but it breaks Fedora builds since rtkaio now
> needs access to __vdso_clock_gettime. I could maintain a patch in
> Fedora that exports this symbol as GLIBC_PRIVATE, but I reckon that it
> might be more consistent to put it upstream since all of the
> architectures that have vDSO time support export the symbol in
> GLIBC_PRIVATE already.
>
> Tested on i386 to verify that the build completes successfully. OK to
> commit?
>
Can you explain why __clock_gettime isn't sufficient?
On Thu, 22 Jan 2015 04:44:14 -0800
"H.J. Lu" <hjl.tools@gmail.com> wrote:
> Can you explain why __clock_gettime isn't sufficient?
rtkaio includes a separate definition for clock_gettime. I guess I
should just patch through a redirect like clock_gettime in librt
instead of adding __vdso_clock_gettime. I withdraw this patch.
Thanks,
Siddhesh
@@ -47,5 +47,6 @@ libc {
}
GLIBC_PRIVATE {
__modify_ldt;
+ __vdso_clock_gettime;
}
}