Patchwork Use VDSO interface for gettimeofday on aarch64

login
register
mail settings
Submitter Steve Ellcey
Date May 15, 2018, 11:07 p.m.
Message ID <1526425669.8676.49.camel@cavium.com>
Download mbox | patch
Permalink /patch/27277/
State New
Headers show

Comments

Steve Ellcey - May 15, 2018, 11:07 p.m.
Here is an updated version of the gettimeofday vdso patch.  I tried to
address as many of the questions as I could.  I added two tests,
tst-gettimeofday and tst-gettimeofday2.  tst-gettimeofday2 just 
includes tst-gettimeofday but is tested with 'LD_BIND_NOW=1'.

I think defining __gettimeofday to be __redirect___gettimeofday and
including sys/time.h is the best way to handle things.  Even
if we did a explicit definition of __redirect__gettimeofday we
would need to include sys/time.h to get the timeval and timezone
structures and if we didn't redefine __gettimeofday when doing
that the definition of __gettimeofday in the header file would
conflict with the one defined by libc_ifunc_hidden.  This is how
x86 and powerpc handle it in their gettimeofday functions and also
how aarch64 handles the definition/redefinition of memcpy in
sysdeps/aarch64/multiarch/memcpy.c.

I am not sure how to handle the static linked binary issue you raised,
I have been copying what x86/powerpc did and if they or some other
platform has solved this then I would be interested in how to
do it.  Since PR 19767 is still open I assume it hasn't been fixed
anywhere yet and I would rather not try to deal with it in this
patch.

I changed __gettimeofday_syscall to __gettimeofday_vsyscall and
I got rid of the redefinition of libc_hidden_def by just calling
__hidden_ver1 directly (in the shared case).

There do seem to be places where libc calls gettimeofday (nis/nis_call.c,
login/logwtmp.c, resolv/gai_suspend.c, others).  Most of them call
__gettimeofday but some just call gettimeofday.  I am not sure what
if anything needs to be done with these calls, they don't seem to
have changed when x86 or powerpc made their gettimeofday/vdso changes..

Steve Ellcey
sellcey@cavium.com


This is what I would use for a commit message:


aarch64: Use an ifunc/VDSO to implement gettimeofday in shared glibc.

This patch uses an ifunc to implement gettimeofday in the shared libc.
If the kernel supports the VDSO interface we use it, otherwise we use
the old method of a vsyscall.  The static version of gettimeofday 
continues to use a syscall.


2018-05-15  Steve Ellcey  <sellcey@caviumnetworks.com>

	* sysdeps/unix/sysv/linux/aarch64/gettimeofday.c: New file.
	* posix/tst-gettimeofday.c: New file.
	* posix/tst-gettimeofday2.c: New file.
	* posix/Makefile (tests): Add new tests to list.
Szabolcs Nagy - May 16, 2018, 10:44 a.m.
On 16/05/18 00:07, Steve Ellcey wrote:
> Here is an updated version of the gettimeofday vdso patch.  I tried to
> address as many of the questions as I could.  I added two tests,
> tst-gettimeofday and tst-gettimeofday2.  tst-gettimeofday2 just
> includes tst-gettimeofday but is tested with 'LD_BIND_NOW=1'.
> 
> I think defining __gettimeofday to be __redirect___gettimeofday and
> including sys/time.h is the best way to handle things.  Even
> if we did a explicit definition of __redirect__gettimeofday we
> would need to include sys/time.h to get the timeval and timezone
> structures and if we didn't redefine __gettimeofday when doing
> that the definition of __gettimeofday in the header file would
> conflict with the one defined by libc_ifunc_hidden.  This is how
> x86 and powerpc handle it in their gettimeofday functions and also
> how aarch64 handles the definition/redefinition of memcpy in
> sysdeps/aarch64/multiarch/memcpy.c.
> 
> I am not sure how to handle the static linked binary issue you raised,
> I have been copying what x86/powerpc did and if they or some other
> platform has solved this then I would be interested in how to
> do it.  Since PR 19767 is still open I assume it hasn't been fixed
> anywhere yet and I would rather not try to deal with it in this
> patch.
> 
> I changed __gettimeofday_syscall to __gettimeofday_vsyscall and
> I got rid of the redefinition of libc_hidden_def by just calling
> __hidden_ver1 directly (in the shared case).
> 

This is OK.

> There do seem to be places where libc calls gettimeofday (nis/nis_call.c,
> login/logwtmp.c, resolv/gai_suspend.c, others).  Most of them call
> __gettimeofday but some just call gettimeofday.  I am not sure what
> if anything needs to be done with these calls, they don't seem to
> have changed when x86 or powerpc made their gettimeofday/vdso changes..
> 

What i wanted to know/document is that internal libc.so
calls don't go via the ifunc resolver, but call the
vsyscall and this is the only reason why it should remain
a vsyscall instead of a syscall as far as i can see
(otherwise if ifunc already checked the vdso then there
would be no point doing that in vsyscall too)

The other thing that would be nice to document is that
why this change is safe for gettimeofday but not clock_gettime.
(former does not have to set errno other than EFAULT but that
case never works with vdso anyway, so the gettimeofday vdso
function is a complete implementation, while clock_gettime
has to deal with errno after the vdso call)

> Steve Ellcey
> sellcey@cavium.com
> 
> 
> This is what I would use for a commit message:
> 
> 
> aarch64: Use an ifunc/VDSO to implement gettimeofday in shared glibc.
> 
> This patch uses an ifunc to implement gettimeofday in the shared libc.

please add here something like

"This is faster compared to the vsyscall mechanism that has to
check a global pointer, demangle it and call it indirectly when
the VDSO is present. Resolving the gettimeofday symbol directly
to the VDSO code is safe because there are no failures that the
libc has to handle by setting errno like in a generic vsyscall."

> If the kernel supports the VDSO interface we use it, otherwise we use
> the old method of a vsyscall.  The static version of gettimeofday
> continues to use a syscall.
> 
> 
> 2018-05-15  Steve Ellcey<sellcey@caviumnetworks.com>
> 
> 	* sysdeps/unix/sysv/linux/aarch64/gettimeofday.c: New file.
> 	* posix/tst-gettimeofday.c: New file.
> 	* posix/tst-gettimeofday2.c: New file.
> 	* posix/Makefile (tests): Add new tests to list.
> 

> --- a/posix/tst-gettimeofday.c
> +++ b/posix/tst-gettimeofday.c
...
> +
> +#include <errno.h>
> +#include <stdio.h>
> +#include <unistd.h>
> +#include <sys/time.h>
> +#include <time.h>
> +
> +
> +/* Test that nanosleep() does sleep.  */

this seems to be from the nanosleep test,
i think that can be reused instead of copied, just
add a comment that it is used for gettimeofday testing
too and in the testcase that is run with BIND_NOW
add a comment that gettimeofday may be ifunc resolved
on targets with VDSO.

(a better test is probably checking if &gettimeofday
is indeed in vdso and not in libc.so, although that is
tricky: it may point to the plt in the main executable,
so the test has to be a shared lib, then bindnow is not
needed, but the test would be target specific)

> +static int
> +do_test (void)
> +{
> +  /* Current time.  */
> +  struct timeval tv1;
> +  TEMP_FAILURE_RETRY (gettimeofday (&tv1, NULL));
> +
> +  /* Sleep for one second to make sure time changes.  */
> +  TEMP_FAILURE_RETRY (sleep (1));
> +
...

> +#ifdef SHARED
> +
> +# define __gettimeofday __redirect___gettimeofday
> +# include <sys/time.h>
> +# undef __gettimeofday
> +# define HAVE_VSYSCALL
> +# include <dl-vdso.h>
> +# include <sysdep-vdso.h>
> +

i'd add a comment here:

/* Used as a fallback in the ifunc resolver if VDSO is not available
    and for libc.so internal __gettimeofday calls.  */

> +static int
> +__gettimeofday_vsyscall (struct timeval *tv, struct timezone *tz)
> +{
> +  return INLINE_VSYSCALL (gettimeofday, 2, tv, tz);
> +}
> +
> +/* PREPARE_VERSION will need an __LP64__ ifdef when ILP32 support
> +   goes in.  See _libc_vdso_platform_setup in
> +   sysdeps/unix/sysv/linux/aarch64/init-first.c.  */
> +
> +# undef INIT_ARCH
> +# define INIT_ARCH() \
> +	   PREPARE_VERSION (linux_version, "LINUX_2.6.39", 123718537); \
> +	   void *vdso_gettimeofday = \
> +	     _dl_vdso_vsym ("__kernel_gettimeofday", &linux_version);
> +
> +libc_ifunc_hidden (__redirect___gettimeofday, __gettimeofday,
> +                    vdso_gettimeofday ?: (void *) __gettimeofday_vsyscall)
> +
> +__hidden_ver1 (__gettimeofday_vsyscall, __GI___gettimeofday,
> +	       __gettimeofday_vsyscall);
> +
> +#else
> +
> +# include <sys/time.h>
> +# include <sysdep.h>
> +int
> +__gettimeofday (struct timeval *tv, struct timezone *tz)
> +{
> +  return INLINE_SYSCALL (gettimeofday, 2, tv, tz);
> +}
> +libc_hidden_def (__gettimeofday)
> +
> +#endif
> +
> +weak_alias (__gettimeofday, gettimeofday)
> +libc_hidden_weak (gettimeofday)
>
Adhemerval Zanella Netto - May 16, 2018, 11:51 a.m.
On 16/05/2018 07:44, Szabolcs Nagy wrote:
> On 16/05/18 00:07, Steve Ellcey wrote:
>> Here is an updated version of the gettimeofday vdso patch.  I tried to
>> address as many of the questions as I could.  I added two tests,
>> tst-gettimeofday and tst-gettimeofday2.  tst-gettimeofday2 just
>> includes tst-gettimeofday but is tested with 'LD_BIND_NOW=1'.
>>
>> I think defining __gettimeofday to be __redirect___gettimeofday and
>> including sys/time.h is the best way to handle things.  Even
>> if we did a explicit definition of __redirect__gettimeofday we
>> would need to include sys/time.h to get the timeval and timezone
>> structures and if we didn't redefine __gettimeofday when doing
>> that the definition of __gettimeofday in the header file would
>> conflict with the one defined by libc_ifunc_hidden.  This is how
>> x86 and powerpc handle it in their gettimeofday functions and also
>> how aarch64 handles the definition/redefinition of memcpy in
>> sysdeps/aarch64/multiarch/memcpy.c.
>>
>> I am not sure how to handle the static linked binary issue you raised,
>> I have been copying what x86/powerpc did and if they or some other
>> platform has solved this then I would be interested in how to
>> do it.  Since PR 19767 is still open I assume it hasn't been fixed
>> anywhere yet and I would rather not try to deal with it in this
>> patch.
>>
>> I changed __gettimeofday_syscall to __gettimeofday_vsyscall and
>> I got rid of the redefinition of libc_hidden_def by just calling
>> __hidden_ver1 directly (in the shared case).
>>
> 
> This is OK.
> 
>> There do seem to be places where libc calls gettimeofday (nis/nis_call.c,
>> login/logwtmp.c, resolv/gai_suspend.c, others).  Most of them call
>> __gettimeofday but some just call gettimeofday.  I am not sure what
>> if anything needs to be done with these calls, they don't seem to
>> have changed when x86 or powerpc made their gettimeofday/vdso changes..
>>
> 
> What i wanted to know/document is that internal libc.so
> calls don't go via the ifunc resolver, but call the
> vsyscall and this is the only reason why it should remain
> a vsyscall instead of a syscall as far as i can see
> (otherwise if ifunc already checked the vdso then there
> would be no point doing that in vsyscall too)
> 
> The other thing that would be nice to document is that
> why this change is safe for gettimeofday but not clock_gettime.
> (former does not have to set errno other than EFAULT but that
> case never works with vdso anyway, so the gettimeofday vdso
> function is a complete implementation, while clock_gettime
> has to deal with errno after the vdso call)

As I put previously keep in mind that different that x86 and powerpc 
implementations, where the vDSO symbol does not fail; the arm64 vDSO 
implements a syscall fallback in case of underlying hardware requires 
an out-of-line counter access (arch_timer_enable_workaround).

Using a ifunc accessors to call vDSO directly will result in a slight
different semantic since generic implementation (kernel/time/time.c)
might return EFAULT in some cases (which won't be handled by ifunc
implementation).  This should not be an issue since POSIX [1] defines
no error code should reserved for the symbol, but it might trigger
some test in LTP.

> 
>> Steve Ellcey
>> sellcey@cavium.com
>>
>>
>> This is what I would use for a commit message:
>>
>>
>> aarch64: Use an ifunc/VDSO to implement gettimeofday in shared glibc.
>>
>> This patch uses an ifunc to implement gettimeofday in the shared libc.
> 
> please add here something like
> 
> "This is faster compared to the vsyscall mechanism that has to
> check a global pointer, demangle it and call it indirectly when
> the VDSO is present. Resolving the gettimeofday symbol directly
> to the VDSO code is safe because there are no failures that the
> libc has to handle by setting errno like in a generic vsyscall."
> 
>> If the kernel supports the VDSO interface we use it, otherwise we use
>> the old method of a vsyscall.  The static version of gettimeofday
>> continues to use a syscall.
>>
>>
>> 2018-05-15  Steve Ellcey<sellcey@caviumnetworks.com>
>>
>>     * sysdeps/unix/sysv/linux/aarch64/gettimeofday.c: New file.
>>     * posix/tst-gettimeofday.c: New file.
>>     * posix/tst-gettimeofday2.c: New file.
>>     * posix/Makefile (tests): Add new tests to list.
>>
> 
>> --- a/posix/tst-gettimeofday.c
>> +++ b/posix/tst-gettimeofday.c
> ...
>> +
>> +#include <errno.h>
>> +#include <stdio.h>
>> +#include <unistd.h>
>> +#include <sys/time.h>
>> +#include <time.h>
>> +
>> +
>> +/* Test that nanosleep() does sleep.  */
> 
> this seems to be from the nanosleep test,
> i think that can be reused instead of copied, just
> add a comment that it is used for gettimeofday testing
> too and in the testcase that is run with BIND_NOW
> add a comment that gettimeofday may be ifunc resolved
> on targets with VDSO.
> 
> (a better test is probably checking if &gettimeofday
> is indeed in vdso and not in libc.so, although that is
> tricky: it may point to the plt in the main executable,
> so the test has to be a shared lib, then bindnow is not
> needed, but the test would be target specific)
> 
>> +static int
>> +do_test (void)
>> +{
>> +  /* Current time.  */
>> +  struct timeval tv1;
>> +  TEMP_FAILURE_RETRY (gettimeofday (&tv1, NULL));
>> +
>> +  /* Sleep for one second to make sure time changes.  */
>> +  TEMP_FAILURE_RETRY (sleep (1));
>> +
> ...
> 
>> +#ifdef SHARED
>> +
>> +# define __gettimeofday __redirect___gettimeofday
>> +# include <sys/time.h>
>> +# undef __gettimeofday
>> +# define HAVE_VSYSCALL
>> +# include <dl-vdso.h>
>> +# include <sysdep-vdso.h>
>> +
> 
> i'd add a comment here:
> 
> /* Used as a fallback in the ifunc resolver if VDSO is not available
>    and for libc.so internal __gettimeofday calls.  */
> 
>> +static int
>> +__gettimeofday_vsyscall (struct timeval *tv, struct timezone *tz)
>> +{
>> +  return INLINE_VSYSCALL (gettimeofday, 2, tv, tz);
>> +}
>> +
>> +/* PREPARE_VERSION will need an __LP64__ ifdef when ILP32 support
>> +   goes in.  See _libc_vdso_platform_setup in
>> +   sysdeps/unix/sysv/linux/aarch64/init-first.c.  */
>> +
>> +# undef INIT_ARCH
>> +# define INIT_ARCH() \
>> +       PREPARE_VERSION (linux_version, "LINUX_2.6.39", 123718537); \
>> +       void *vdso_gettimeofday = \
>> +         _dl_vdso_vsym ("__kernel_gettimeofday", &linux_version);
>> +
>> +libc_ifunc_hidden (__redirect___gettimeofday, __gettimeofday,
>> +                    vdso_gettimeofday ?: (void *) __gettimeofday_vsyscall)
>> +
>> +__hidden_ver1 (__gettimeofday_vsyscall, __GI___gettimeofday,
>> +           __gettimeofday_vsyscall);
>> +
>> +#else
>> +
>> +# include <sys/time.h>
>> +# include <sysdep.h>
>> +int
>> +__gettimeofday (struct timeval *tv, struct timezone *tz)
>> +{
>> +  return INLINE_SYSCALL (gettimeofday, 2, tv, tz);
>> +}
>> +libc_hidden_def (__gettimeofday)
>> +
>> +#endif
>> +
>> +weak_alias (__gettimeofday, gettimeofday)
>> +libc_hidden_weak (gettimeofday)
>>
>
Szabolcs Nagy - May 16, 2018, 12:03 p.m.
On 16/05/18 12:51, Adhemerval Zanella wrote:
> On 16/05/2018 07:44, Szabolcs Nagy wrote:
>> The other thing that would be nice to document is that
>> why this change is safe for gettimeofday but not clock_gettime.
>> (former does not have to set errno other than EFAULT but that
>> case never works with vdso anyway, so the gettimeofday vdso
>> function is a complete implementation, while clock_gettime
>> has to deal with errno after the vdso call)
> 
> As I put previously keep in mind that different that x86 and powerpc
> implementations, where the vDSO symbol does not fail; the arm64 vDSO
> implements a syscall fallback in case of underlying hardware requires
> an out-of-line counter access (arch_timer_enable_workaround).
> 
> Using a ifunc accessors to call vDSO directly will result in a slight
> different semantic since generic implementation (kernel/time/time.c)
> might return EFAULT in some cases (which won't be handled by ifunc
> implementation).  This should not be an issue since POSIX [1] defines
> no error code should reserved for the symbol, but it might trigger
> some test in LTP.
> 

sorry i missed this comment,

if the vdso can fail with -EFAULT then this optimization is not valid
(gettimeofday must either return 0 or -1).

if userspace can test whatever the kernel is testing there, that
could be added to the ifunc resolver and only use vdso if we know
it does not fail (i don't know if this is possible).
Szabolcs Nagy - May 16, 2018, 12:12 p.m.
On 16/05/18 13:03, Szabolcs Nagy wrote:
> On 16/05/18 12:51, Adhemerval Zanella wrote:
>> On 16/05/2018 07:44, Szabolcs Nagy wrote:
>>> The other thing that would be nice to document is that
>>> why this change is safe for gettimeofday but not clock_gettime.
>>> (former does not have to set errno other than EFAULT but that
>>> case never works with vdso anyway, so the gettimeofday vdso
>>> function is a complete implementation, while clock_gettime
>>> has to deal with errno after the vdso call)
>>
>> As I put previously keep in mind that different that x86 and powerpc
>> implementations, where the vDSO symbol does not fail; the arm64 vDSO
>> implements a syscall fallback in case of underlying hardware requires
>> an out-of-line counter access (arch_timer_enable_workaround).
>>
>> Using a ifunc accessors to call vDSO directly will result in a slight
>> different semantic since generic implementation (kernel/time/time.c)
>> might return EFAULT in some cases (which won't be handled by ifunc
>> implementation).  This should not be an issue since POSIX [1] defines
>> no error code should reserved for the symbol, but it might trigger
>> some test in LTP.
>>
> 
> sorry i missed this comment,
> 
> if the vdso can fail with -EFAULT then this optimization is not valid
> (gettimeofday must either return 0 or -1).
> 
> if userspace can test whatever the kernel is testing there, that
> could be added to the ifunc resolver and only use vdso if we know
> it does not fail (i don't know if this is possible).
> 

hm in posix EFAULT is not a requirement, it's probably
undefined behaviour if gettimeofday is called with
wrong pointers. (this is borderline since linux man
pages document the EFAULT, but obviously that cannot
work for vdso anyway)

then the optimization is ok, but we definitely need
to document why we think directly going to the vdso
is safe.
Adhemerval Zanella Netto - May 16, 2018, 1:23 p.m.
On 16/05/2018 09:12, Szabolcs Nagy wrote:
> On 16/05/18 13:03, Szabolcs Nagy wrote:
>> On 16/05/18 12:51, Adhemerval Zanella wrote:
>>> On 16/05/2018 07:44, Szabolcs Nagy wrote:
>>>> The other thing that would be nice to document is that
>>>> why this change is safe for gettimeofday but not clock_gettime.
>>>> (former does not have to set errno other than EFAULT but that
>>>> case never works with vdso anyway, so the gettimeofday vdso
>>>> function is a complete implementation, while clock_gettime
>>>> has to deal with errno after the vdso call)
>>>
>>> As I put previously keep in mind that different that x86 and powerpc
>>> implementations, where the vDSO symbol does not fail; the arm64 vDSO
>>> implements a syscall fallback in case of underlying hardware requires
>>> an out-of-line counter access (arch_timer_enable_workaround).
>>>
>>> Using a ifunc accessors to call vDSO directly will result in a slight
>>> different semantic since generic implementation (kernel/time/time.c)
>>> might return EFAULT in some cases (which won't be handled by ifunc
>>> implementation).  This should not be an issue since POSIX [1] defines
>>> no error code should reserved for the symbol, but it might trigger
>>> some test in LTP.
>>>
>>
>> sorry i missed this comment,
>>
>> if the vdso can fail with -EFAULT then this optimization is not valid
>> (gettimeofday must either return 0 or -1).
>>
>> if userspace can test whatever the kernel is testing there, that
>> could be added to the ifunc resolver and only use vdso if we know
>> it does not fail (i don't know if this is possible).
>>
> 
> hm in posix EFAULT is not a requirement, it's probably
> undefined behaviour if gettimeofday is called with
> wrong pointers. (this is borderline since linux man
> pages document the EFAULT, but obviously that cannot
> work for vdso anyway)
> 
> then the optimization is ok, but we definitely need
> to document why we think directly going to the vdso
> is safe.

I think the optimization is fine as well, we had previous discussion about some
syscall semantics returning EFAULT [1] (which also lead to LTP test fixes [2]),
and the generic idea was since is UB to rely on EFAULT return code (for instance
a call to 'gettimeofday (0, 0)') it is fine to glibc also change semantic if it
finds suitable.

[1] https://sourceware.org/ml/libc-alpha/2017-10/msg00830.html
[2] https://github.com/linux-test-project/ltp/commit/259db6fed55f88ab32a0875e66803eee44d298be
Steve Ellcey - May 22, 2018, 8:25 p.m.
On Wed, 2018-05-16 at 11:44 +0100, Szabolcs Nagy wrote:
> On 16/05/18 00:07, Steve Ellcey wrote:
> > 
> > There do seem to be places where libc calls gettimeofday
> > (nis/nis_call.c,
> > login/logwtmp.c, resolv/gai_suspend.c, others).  Most of them call
> > __gettimeofday but some just call gettimeofday.  I am not sure what
> > if anything needs to be done with these calls, they don't seem to
> > have changed when x86 or powerpc made their gettimeofday/vdso
> > changes..
> > 
> What i wanted to know/document is that internal libc.so
> calls don't go via the ifunc resolver, but call the
> vsyscall and this is the only reason why it should remain
> a vsyscall instead of a syscall as far as i can see
> (otherwise if ifunc already checked the vdso then there
> would be no point doing that in vsyscall too)
> 
> The other thing that would be nice to document is that
> why this change is safe for gettimeofday but not clock_gettime.
> (former does not have to set errno other than EFAULT but that
> case never works with vdso anyway, so the gettimeofday vdso
> function is a complete implementation, while clock_gettime
> has to deal with errno after the vdso call)

Szabolcs,

I am having trouble figuring out how to document these issues.  The
problem is that I don't understand the details of the
syscall/vsyscall/vdso interfaces well enough to add intelligent
comments to the code that would address your comments.

Any chance you could add the comments that you think are needed and
check the patch in?  It seems like you and Adhemerval are both happy
with the code, it is just the comments (and the test if you want to use
the nanosleep test instead of adding the new ones) that need fixing.

Steve Ellcey
sellcey@cavium.com

Patch

diff --git a/posix/Makefile b/posix/Makefile
index e9730ee..2cd3a44 100644
--- a/posix/Makefile
+++ b/posix/Makefile
@@ -95,7 +95,8 @@  tests		:= test-errno tstgetopt testfnm runtests runptests \
 		   tst-posix_spawn-fd tst-posix_spawn-setsid \
 		   tst-posix_fadvise tst-posix_fadvise64 \
 		   tst-sysconf-empty-chroot tst-glob_symlinks tst-fexecve \
-		   tst-glob-tilde test-ssize-max
+		   tst-glob-tilde test-ssize-max \
+		   tst-gettimeofday tst-gettimeofday2
 tests-internal	:= bug-regex5 bug-regex20 bug-regex33 \
 		   tst-rfc3484 tst-rfc3484-2 tst-rfc3484-3 \
 		   tst-glob_lstat_compat
@@ -257,6 +258,8 @@  tst-boost-ARGS = BOOST.tests
 bug-glob1-ARGS = "$(objpfx)"
 tst-execvp3-ARGS = --test-dir=$(objpfx)
 
+tst-gettimeofday2-ENV = LD_BIND_NOW=1
+
 testcases.h: TESTS TESTS2C.sed
 	LC_ALL=C sed -f TESTS2C.sed < $< > $@T
 	mv -f $@T $@
diff --git a/posix/tst-gettimeofday.c b/posix/tst-gettimeofday.c
index e69de29..703c1ce 100644
--- a/posix/tst-gettimeofday.c
+++ b/posix/tst-gettimeofday.c
@@ -0,0 +1,55 @@ 
+/* Copyright (C) 2018 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include <errno.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/time.h>
+#include <time.h>
+
+
+/* Test that nanosleep() does sleep.  */
+static int
+do_test (void)
+{
+  /* Current time.  */
+  struct timeval tv1;
+  TEMP_FAILURE_RETRY (gettimeofday (&tv1, NULL));
+
+  /* Sleep for one second to make sure time changes.  */
+  TEMP_FAILURE_RETRY (sleep (1));
+
+  /* At least one second must have passed.  */
+  struct timeval tv2;
+  TEMP_FAILURE_RETRY (gettimeofday (&tv2, NULL));
+
+  tv2.tv_sec -= tv1.tv_sec;
+  tv2.tv_usec -= tv1.tv_usec;
+  if (tv2.tv_usec < 0)
+    --tv2.tv_sec;
+
+  if (tv2.tv_sec < 1)
+    {
+      puts ("sleep didn't sleep long enough or gettimeofday is broken");
+      return 1;
+    }
+
+  return 0;
+}
+
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"
diff --git a/posix/tst-gettimeofday2.c b/posix/tst-gettimeofday2.c
index e69de29..6c08761 100644
--- a/posix/tst-gettimeofday2.c
+++ b/posix/tst-gettimeofday2.c
@@ -0,0 +1 @@ 
+#include "tst-gettimeofday.c"
diff --git a/sysdeps/unix/sysv/linux/aarch64/gettimeofday.c b/sysdeps/unix/sysv/linux/aarch64/gettimeofday.c
index e69de29..5b6ba70 100644
--- a/sysdeps/unix/sysv/linux/aarch64/gettimeofday.c
+++ b/sysdeps/unix/sysv/linux/aarch64/gettimeofday.c
@@ -0,0 +1,68 @@ 
+/* Copyright (C) 2018 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public License as
+   published by the Free Software Foundation; either version 2.1 of the
+   License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+/* Get the current time of day and timezone information,
+   putting it into *tv and *tz.  If tz is null, *tz is not filled.
+   Returns 0 on success, -1 on errors.  */
+
+#ifdef SHARED
+
+# define __gettimeofday __redirect___gettimeofday
+# include <sys/time.h>
+# undef __gettimeofday
+# define HAVE_VSYSCALL
+# include <dl-vdso.h>
+# include <sysdep-vdso.h>
+
+static int
+__gettimeofday_vsyscall (struct timeval *tv, struct timezone *tz)
+{
+  return INLINE_VSYSCALL (gettimeofday, 2, tv, tz);
+}
+
+/* PREPARE_VERSION will need an __LP64__ ifdef when ILP32 support
+   goes in.  See _libc_vdso_platform_setup in
+   sysdeps/unix/sysv/linux/aarch64/init-first.c.  */
+
+# undef INIT_ARCH
+# define INIT_ARCH() \
+	   PREPARE_VERSION (linux_version, "LINUX_2.6.39", 123718537); \
+	   void *vdso_gettimeofday = \
+	     _dl_vdso_vsym ("__kernel_gettimeofday", &linux_version);
+
+libc_ifunc_hidden (__redirect___gettimeofday, __gettimeofday,
+                    vdso_gettimeofday ?: (void *) __gettimeofday_vsyscall)
+
+__hidden_ver1 (__gettimeofday_vsyscall, __GI___gettimeofday,
+	       __gettimeofday_vsyscall);
+
+#else
+
+# include <sys/time.h>
+# include <sysdep.h>
+int
+__gettimeofday (struct timeval *tv, struct timezone *tz)
+{
+  return INLINE_SYSCALL (gettimeofday, 2, tv, tz);
+}
+libc_hidden_def (__gettimeofday)
+
+#endif
+
+weak_alias (__gettimeofday, gettimeofday)
+libc_hidden_weak (gettimeofday)