[v9,1/2] Y2038: Add 64-bit time for all architectures
Commit Message
glibc support for 64-bit time_t on 32-bit architectures
will involve:
- Using 64-bit times inside glibc, with conversions
to and from 32-bit times taking place as necessary
for interfaces using such times.
- Adding 64-bit-time support in the glibc public API.
This support should be dynamic, i.e. glibc should
provide both 32-bit and 64-bit implementations and
let user code choose at compile time whether to use
the 32-bit or 64-bit interfaces.
This requires a glibc-internal name for a type for times
that are always 64-bit.
To determine whether the default time_t interfaces are 32-bit
and so need conversions, or are 64-bit and so are compatible
with the internal 64-bit type without conversions, a macro
giving the size of the default time_t is also required.
This macro is called __TIMESIZE.
Based on __TIMESIZE, a new macro is defined, __TIME64_T_TYPE,
which is always the right __*_T_TYPE to hold a 64-bit-time.
__TIME64_T_TYPE equals __TIME_T_TYPE if __TIMESIZE equals 64
and equals __SQUAD_T_TYPE otherwise.
__time64_t can then replace uses of internal_time_t.
* bit/time64.h: New file.
* bits/timesize: (__TIMESIZE): New macro.
* include/time.h: replace internal_time_t with __time64_t.
* posix/bits/types (__time64_t): Add.
* stdlib/Makefile: Add bits/time64.h to includes.
* stdlib/Makefile: Add bits/timesize.h to includes.
* sysdeps/unix/sysv/linux/x86/bits/time64.h: New file.
* sysdeps/unix/sysv/linux/x86/bits/timesize.h (__TIMESIZE): New macro.
---
bits/time64.h | 36 +++++++++++++++++++
bits/timesize.h | 22 ++++++++++++
include/time.h | 7 +---
posix/bits/types.h | 8 +++++
stdlib/Makefile | 2 +-
sysdeps/unix/sysv/linux/x86/bits/time64.h | 39 +++++++++++++++++++++
sysdeps/unix/sysv/linux/x86/bits/timesize.h | 25 +++++++++++++
time/tzfile.c | 18 +++++-----
8 files changed, 141 insertions(+), 16 deletions(-)
create mode 100644 bits/time64.h
create mode 100644 bits/timesize.h
create mode 100644 sysdeps/unix/sysv/linux/x86/bits/time64.h
create mode 100644 sysdeps/unix/sysv/linux/x86/bits/timesize.h
Comments
On Wed, 10 Oct 2018, Albert ARIBAUD (3ADEV) wrote:
> bits/time64.h | 36 +++++++++++++++++++
> bits/timesize.h | 22 ++++++++++++
> include/time.h | 7 +---
> posix/bits/types.h | 8 +++++
> stdlib/Makefile | 2 +-
> sysdeps/unix/sysv/linux/x86/bits/time64.h | 39 +++++++++++++++++++++
> sysdeps/unix/sysv/linux/x86/bits/timesize.h | 25 +++++++++++++
My understanding from review comments on previous versions of this patch
was that the generic bits/time64.h was in fact fully correct for x32, so
no x86-specific version of that header should be needed - only the
x86-specific version of bits/timesize.h.
Hi Joseph,
On Wed, 10 Oct 2018 16:22:49 +0000, Joseph Myers
<joseph@codesourcery.com> wrote :
> On Wed, 10 Oct 2018, Albert ARIBAUD (3ADEV) wrote:
>
> > bits/time64.h | 36 +++++++++++++++++++
> > bits/timesize.h | 22 ++++++++++++
> > include/time.h | 7 +---
> > posix/bits/types.h | 8 +++++
> > stdlib/Makefile | 2 +-
> > sysdeps/unix/sysv/linux/x86/bits/time64.h | 39 +++++++++++++++++++++
> > sysdeps/unix/sysv/linux/x86/bits/timesize.h | 25 +++++++++++++
>
> My understanding from review comments on previous versions of this patch
> was that the generic bits/time64.h was in fact fully correct for x32, so
> no x86-specific version of that header should be needed - only the
> x86-specific version of bits/timesize.h.
That was not my understanding, which was as follows:
- We need to determine the size of the target architecture time_t type
in order to decide whether we need to define a 64-bit time type.
- we do not have the time_t size readily available, so we must infer it
based on some other type size.
- The best candidate is the word size, which matches the time_t size for
all architectures ; that is the most general rule.
- However, for x32 the word size is 32 yet the time_t size is 64, so
x32 needs an exception.
Cordialement,
Albert ARIBAUD
3ADEV
On Thu, 11 Oct 2018, Albert ARIBAUD wrote:
> - We need to determine the size of the target architecture time_t type
> in order to decide whether we need to define a 64-bit time type.
Yes.
> - we do not have the time_t size readily available, so we must infer it
> based on some other type size.
That happens in bits/timesize.h.
Once bits/timesize.h exists, the value from there can be used in
bits/time64.h, so only one bits/time64.h should be needed.
This patch could be split in two. The first would add bits/timesize.h
(two variants), and maybe also use __TIMESIZE in bits/msq-pad.h and remove
the x86-specific version of that header. The second would add
bits/time64.h (one variant).
Hi Joseph,
On Thu, 11 Oct 2018 14:15:33 +0000, Joseph Myers
<joseph@codesourcery.com> wrote :
> On Thu, 11 Oct 2018, Albert ARIBAUD wrote:
>
> > - We need to determine the size of the target architecture time_t type
> > in order to decide whether we need to define a 64-bit time type.
>
> Yes.
>
> > - we do not have the time_t size readily available, so we must infer it
> > based on some other type size.
>
> That happens in bits/timesize.h.
>
> Once bits/timesize.h exists, the value from there can be used in
> bits/time64.h, so only one bits/time64.h should be needed.
Ok.
> This patch could be split in two. The first would add bits/timesize.h
> (two variants), and maybe also use __TIMESIZE in bits/msq-pad.h and remove
> the x86-specific version of that header. The second would add
> bits/time64.h (one variant).
I assume you mean bits/msq.h, not bits/msq-pad.h.
But shouldn't modifications to bits/msq.h be put in a separate,
dedicated patch?
Cordialement,
Albert ARIBAUD
3ADEV
On Thu, 11 Oct 2018, Albert ARIBAUD wrote:
> > This patch could be split in two. The first would add bits/timesize.h
> > (two variants), and maybe also use __TIMESIZE in bits/msq-pad.h and remove
> > the x86-specific version of that header. The second would add
> > bits/time64.h (one variant).
>
> I assume you mean bits/msq.h, not bits/msq-pad.h.
No, I mean bits/msq-pad.h. As of today there is only one bits/msq.h for
systems using the Linux kernel, and the only reason for the x86-specific
variant of bits/msq-pad.h to exist is the lack of __TIMESIZE to use in the
generic bits/msq-pad.h.
Hi Joseph,
Le Thu, 11 Oct 2018 15:22:57 +0000, Joseph Myers
<joseph@codesourcery.com> a écrit :
> On Thu, 11 Oct 2018, Albert ARIBAUD wrote:
>
> > > This patch could be split in two. The first would add bits/timesize.h
> > > (two variants), and maybe also use __TIMESIZE in bits/msq-pad.h and remove
> > > the x86-specific version of that header. The second would add
> > > bits/time64.h (one variant).
> >
> > I assume you mean bits/msq.h, not bits/msq-pad.h.
>
> No, I mean bits/msq-pad.h. As of today there is only one bits/msq.h for
> systems using the Linux kernel, and the only reason for the x86-specific
> variant of bits/msq-pad.h to exist is the lack of __TIMESIZE to use in the
> generic bits/msq-pad.h.
Sorry, my repo was lagging one commit behind origin/master.
Cordialement,
Albert ARIBAUD
3ADEV
Hi Joseph,
On Thu, 11 Oct 2018 15:22:57 +0000, Joseph Myers
<joseph@codesourcery.com> wrote :
> On Thu, 11 Oct 2018, Albert ARIBAUD wrote:
>
> > > This patch could be split in two. The first would add bits/timesize.h
> > > (two variants), and maybe also use __TIMESIZE in bits/msq-pad.h and remove
> > > the x86-specific version of that header. The second would add
> > > bits/time64.h (one variant).
> >
> > I assume you mean bits/msq.h, not bits/msq-pad.h.
>
> No, I mean bits/msq-pad.h. As of today there is only one bits/msq.h for
> systems using the Linux kernel, and the only reason for the x86-specific
> variant of bits/msq-pad.h to exist is the lack of __TIMESIZE to use in the
> generic bits/msq-pad.h.
For consistency, should we not replace __WORDSIZE with __TIMESIZE in all
msq-pad.h files rather than just sysdeps/unix/sysv/linux/bits/msq-pad.h?
Cordialement,
Albert ARIBAUD
3ADEV
On Thu, 11 Oct 2018, Albert ARIBAUD wrote:
> > No, I mean bits/msq-pad.h. As of today there is only one bits/msq.h for
> > systems using the Linux kernel, and the only reason for the x86-specific
> > variant of bits/msq-pad.h to exist is the lack of __TIMESIZE to use in the
> > generic bits/msq-pad.h.
>
> For consistency, should we not replace __WORDSIZE with __TIMESIZE in all
> msq-pad.h files rather than just sysdeps/unix/sysv/linux/bits/msq-pad.h?
Sure, that would be reasonable.
new file mode 100644
@@ -0,0 +1,36 @@
+/* bits/time64.h -- underlying types for __time64_t. Generic version.
+ 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/>. */
+
+#ifndef _BITS_TYPES_H
+# error "Never include <bits/time64.h> directly; use <sys/types.h> instead."
+#endif
+
+#ifndef _BITS_TIME64_H
+#define _BITS_TIME64_H 1
+
+/* Define __TIME64_T_TYPE so that it is always a 64-bit type. */
+
+#if __TIMESIZE == 64
+/* If we already have 64-bit time type then use it. */
+# define __TIME64_T_TYPE __TIME_T_TYPE
+#else
+/* Define a 64-bit time type alongsize the 32-bit one. */
+# define __TIME64_T_TYPE __SQUAD_TYPE
+#endif
+
+#endif /* bits/time64.h */
new file mode 100644
@@ -0,0 +1,22 @@
+/* Bit size of the time_t type at glibc build time, general case.
+ 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 <bits/wordsize.h>
+
+/* Size in bits of the 'time_t' type of the default ABI. */
+#define __TIMESIZE __WORDSIZE
@@ -26,10 +26,6 @@ extern __typeof (clock_getcpuclockid) __clock_getcpuclockid;
/* Now define the internal interfaces. */
struct tm;
-/* time_t variant for representing time zone data, independent of
- time_t. */
-typedef __int64_t internal_time_t;
-
/* Defined in mktime.c. */
extern const unsigned short int __mon_yday[2][13] attribute_hidden;
@@ -43,7 +39,7 @@ extern int __use_tzfile attribute_hidden;
extern void __tzfile_read (const char *file, size_t extra,
char **extrap) attribute_hidden;
-extern void __tzfile_compute (internal_time_t timer, int use_localtime,
+extern void __tzfile_compute (__time64_t timer, int use_localtime,
long int *leap_correct, int *leap_hit,
struct tm *tp) attribute_hidden;
extern void __tzfile_default (const char *std, const char *dst,
@@ -101,7 +97,6 @@ extern char * __strptime_internal (const char *rp, const char *fmt,
extern double __difftime (time_t time1, time_t time0);
-
/* Use in the clock_* functions. Size of the field representing the
actual clock ID. */
#define CLOCK_IDFIELD_SIZE 3
@@ -25,6 +25,7 @@
#include <features.h>
#include <bits/wordsize.h>
+#include <bits/timesize.h>
/* Convenience types. */
typedef unsigned char __u_char;
@@ -138,6 +139,7 @@ __extension__ typedef unsigned long long int __uintmax_t;
# error
#endif
#include <bits/typesizes.h> /* Defines __*_T_TYPE macros. */
+#include <bits/time64.h> /* Defines __TIME*_T_TYPE macros. */
__STD_TYPE __DEV_T_TYPE __dev_t; /* Type of device numbers. */
@@ -211,6 +213,12 @@ __STD_TYPE __U32_TYPE __socklen_t;
It is not currently necessary for this to be machine-specific. */
typedef int __sig_atomic_t;
+#if __TIMESIZE == 64
+# define __time64_t __time_t
+#else
+__STD_TYPE __TIME64_T_TYPE __time64_t; /* Seconds since the Epoch. */
+#endif
+
#undef __STD_TYPE
#endif /* bits/types.h */
@@ -29,7 +29,7 @@ headers := stdlib.h bits/stdlib.h bits/stdlib-ldbl.h bits/stdlib-float.h \
ucontext.h sys/ucontext.h bits/indirect-return.h \
alloca.h fmtmsg.h \
bits/stdlib-bsearch.h sys/random.h bits/stdint-intn.h \
- bits/stdint-uintn.h
+ bits/stdint-uintn.h bits/time64.h bits/timesize.h \
routines := \
atof atoi atol atoll \
new file mode 100644
@@ -0,0 +1,39 @@
+/* bits/time64.h -- underlying types for __time64_t. Linux/x86-64 version.
+ 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/>. */
+
+#ifndef _BITS_TYPES_H
+# error "Never include <bits/time64.h> directly; use <sys/types.h> instead."
+#endif
+
+#ifndef _BITS_TIME64_H
+#define _BITS_TIME64_H 1
+
+/* Define __TIME64_T_TYPE so that it is always a 64-bit type. */
+
+#if defined __x86_64__ && defined __ILP32__
+/* For x32, time is 64-bit even though word size is 32-bit. */
+# define __TIME64_T_TYPE __SQUAD_TYPE
+#elif __TIMESIZE == 64
+/* If we already have 64-bit time type then use it. */
+# define __TIME64_T_TYPE __TIME_T_TYPE
+#else
+/* Define a 64-bit time type alongsize the 32-bit one. */
+# define __TIME64_T_TYPE __SQUAD_TYPE
+#endif
+
+#endif /* bits/time64.h */
new file mode 100644
@@ -0,0 +1,25 @@
+/* Bit size of the time_t type at glibc build time, x86-64 and x32 case.
+ 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/>. */
+
+#if defined __x86_64__ && defined __ILP32__
+/* For x32, time is 64-bit even though word size is 32-bit. */
+# define __TIMESIZE 64
+#else
+/* For others, time size is word size. */
+# define __TIMESIZE __WORDSIZE
+#endif
@@ -44,12 +44,12 @@ struct ttinfo
struct leap
{
- internal_time_t transition; /* Time the transition takes effect. */
+ __time64_t transition; /* Time the transition takes effect. */
long int change; /* Seconds of correction to apply. */
};
static size_t num_transitions;
-libc_freeres_ptr (static internal_time_t *transitions);
+libc_freeres_ptr (static __time64_t *transitions);
static unsigned char *type_idxs;
static size_t num_types;
static struct ttinfo *types;
@@ -113,8 +113,8 @@ __tzfile_read (const char *file, size_t extra, char **extrap)
size_t tzspec_len;
char *new = NULL;
- _Static_assert (sizeof (internal_time_t) == 8,
- "internal_time_t must be eight bytes");
+ _Static_assert (sizeof (__time64_t) == 8,
+ "__time64_t must be eight bytes");
__use_tzfile = 0;
@@ -217,9 +217,9 @@ __tzfile_read (const char *file, size_t extra, char **extrap)
if (__builtin_expect (num_transitions
> ((SIZE_MAX - (__alignof__ (struct ttinfo) - 1))
- / (sizeof (internal_time_t) + 1)), 0))
+ / (sizeof (__time64_t) + 1)), 0))
goto lose;
- total_size = num_transitions * (sizeof (internal_time_t) + 1);
+ total_size = num_transitions * (sizeof (__time64_t) + 1);
total_size = ((total_size + __alignof__ (struct ttinfo) - 1)
& ~(__alignof__ (struct ttinfo) - 1));
types_idx = total_size;
@@ -276,7 +276,7 @@ __tzfile_read (const char *file, size_t extra, char **extrap)
goto lose;
type_idxs = (unsigned char *) transitions + (num_transitions
- * sizeof (internal_time_t));
+ * sizeof (__time64_t));
types = (struct ttinfo *) ((char *) transitions + types_idx);
zone_names = (char *) types + num_types * sizeof (struct ttinfo);
leaps = (struct leap *) ((char *) transitions + leaps_idx);
@@ -578,7 +578,7 @@ __tzfile_default (const char *std, const char *dst,
}
void
-__tzfile_compute (internal_time_t timer, int use_localtime,
+__tzfile_compute (__time64_t timer, int use_localtime,
long int *leap_correct, int *leap_hit,
struct tm *tp)
{
@@ -667,7 +667,7 @@ __tzfile_compute (internal_time_t timer, int use_localtime,
initial search spot from it. Half of a gregorian year
has on average 365.2425 * 86400 / 2 = 15778476 seconds.
The value i can be truncated if size_t is smaller than
- internal_time_t, but this is harmless because it is just
+ __time64_t, but this is harmless because it is just
a guess. */
i = (transitions[num_transitions - 1] - timer) / 15778476;
if (i < num_transitions)