[PATCH] time.h: Change type of `tv_sec` in `timespec` to `int` from `_time_t`.
Without this change `timespec` is too large for the `clock_nanosleep` syscall, which translates to sleep(0) on 32bit systems when TIME64 is enabled (the default). Tested on MIPS. Signed-off-by: Nadav Tasher <tashernadav@gmail.com> --- include/time.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/include/time.h b/include/time.h index 1a1408990..25dfb9e2e 100644 --- a/include/time.h +++ b/include/time.h @@ -118,7 +118,7 @@ typedef __timer_t timer_t; has nanoseconds instead of microseconds. */ struct timespec { - __time_t tv_sec; /* Seconds. */ + int tv_sec; /* Seconds. */ long int tv_nsec; /* Nanoseconds. */ }; -- 2.34.1
Hi Nadav, do you have some test code showing the issue? best regards Waldemar Nadav Tasher wrote,
Without this change `timespec` is too large for the `clock_nanosleep` syscall, which translates to sleep(0) on 32bit systems when TIME64 is enabled (the default).
Tested on MIPS.
Signed-off-by: Nadav Tasher <tashernadav@gmail.com> --- include/time.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/time.h b/include/time.h index 1a1408990..25dfb9e2e 100644 --- a/include/time.h +++ b/include/time.h @@ -118,7 +118,7 @@ typedef __timer_t timer_t; has nanoseconds instead of microseconds. */ struct timespec { - __time_t tv_sec; /* Seconds. */ + int tv_sec; /* Seconds. */ long int tv_nsec; /* Nanoseconds. */ };
-- 2.34.1
_______________________________________________ devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
Hi Waldemar, The following snippet should do it when compiled statically for MIPS32 and emulated using qemu-user-static: #include <time.h> #include <stddef.h> #include <unistd.h> int main(int argc, char **argv) { time_t start = time(NULL); sleep(2); time_t end = time(NULL); return end != start; } When I strace the resulting binary, I saw: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=0, tv_nsec=2}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ? But I expected to see: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=2, tv_nsec=0}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ? kind regards Nadav Waldemar Brodkorb wrote,
Hi Nadav,
do you have some test code showing the issue?
best regards Waldemar
Nadav Tasher wrote,
Without this change `timespec` is too large for the `clock_nanosleep` syscall, which translates to sleep(0) on 32bit systems when TIME64 is enabled (the default).
Tested on MIPS.
Signed-off-by: Nadav Tasher <tashernadav@gmail.com> --- include/time.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/time.h b/include/time.h index 1a1408990..25dfb9e2e 100644 --- a/include/time.h +++ b/include/time.h @@ -118,7 +118,7 @@ typedef __timer_t timer_t; has nanoseconds instead of microseconds. */ struct timespec { - __time_t tv_sec; /* Seconds. */ + int tv_sec; /* Seconds. */ long int tv_nsec; /* Nanoseconds. */ };
-- 2.34.1
_______________________________________________ devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
I'll take a look сб, 21 дек. 2024 г. в 14:20, Nadav Tasher <tashernadav@gmail.com>:
Hi Waldemar,
The following snippet should do it when compiled statically for MIPS32 and emulated using qemu-user-static:
#include <time.h> #include <stddef.h> #include <unistd.h>
int main(int argc, char **argv) {
time_t start = time(NULL); sleep(2); time_t end = time(NULL);
return end != start; }
When I strace the resulting binary, I saw: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=0, tv_nsec=2}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
But I expected to see: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=2, tv_nsec=0}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
kind regards Nadav
Waldemar Brodkorb wrote,
Hi Nadav,
do you have some test code showing the issue?
best regards Waldemar
Nadav Tasher wrote,
Without this change `timespec` is too large for the `clock_nanosleep` syscall, which translates to sleep(0) on 32bit systems when TIME64 is enabled (the default).
Tested on MIPS.
Signed-off-by: Nadav Tasher <tashernadav@gmail.com> --- include/time.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/time.h b/include/time.h index 1a1408990..25dfb9e2e 100644 --- a/include/time.h +++ b/include/time.h @@ -118,7 +118,7 @@ typedef __timer_t timer_t; has nanoseconds instead of microseconds. */ struct timespec { - __time_t tv_sec; /* Seconds. */ + int tv_sec; /* Seconds. */ long int tv_nsec; /* Nanoseconds. */ };
-- 2.34.1
_______________________________________________ devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
Cannot reproduce it with the latest uclibc. Config in attachment сб, 21 дек. 2024 г. в 14:20, Nadav Tasher <tashernadav@gmail.com>:
Hi Waldemar,
The following snippet should do it when compiled statically for MIPS32 and emulated using qemu-user-static:
#include <time.h> #include <stddef.h> #include <unistd.h>
int main(int argc, char **argv) {
time_t start = time(NULL); sleep(2); time_t end = time(NULL);
return end != start; }
When I strace the resulting binary, I saw: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=0, tv_nsec=2}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
But I expected to see: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=2, tv_nsec=0}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
kind regards Nadav
Waldemar Brodkorb wrote,
Hi Nadav,
do you have some test code showing the issue?
best regards Waldemar
Nadav Tasher wrote,
Without this change `timespec` is too large for the `clock_nanosleep` syscall, which translates to sleep(0) on 32bit systems when TIME64 is enabled (the default).
Tested on MIPS.
Signed-off-by: Nadav Tasher <tashernadav@gmail.com> --- include/time.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/time.h b/include/time.h index 1a1408990..25dfb9e2e 100644 --- a/include/time.h +++ b/include/time.h @@ -118,7 +118,7 @@ typedef __timer_t timer_t; has nanoseconds instead of microseconds. */ struct timespec { - __time_t tv_sec; /* Seconds. */ + int tv_sec; /* Seconds. */ long int tv_nsec; /* Nanoseconds. */ };
-- 2.34.1
_______________________________________________ devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
Interesting - I have been able to reproduce the bug when compiling like so: make CROSS_COMPILE=mips-linux-gnu- TARGET_ARCH=mips KERNEL_HEADERS=~/linux-headers-4.19.325/include/ -j4 clean make CROSS_COMPILE=mips-linux-gnu- TARGET_ARCH=mips KERNEL_HEADERS=~/linux-headers-4.19.325/include/ -j4 Note that this also reproduces in the existing toolchain at https://toolchains.bootlin.com/downloads/releases/toolchains/mips32/tarballs... (this is actually how I encountered the bug). Dmitriy Chestnykh wrote:
Cannot reproduce it with the latest uclibc. Config in attachment
сб, 21 дек. 2024 г. в 14:20, Nadav Tasher <tashernadav@gmail.com>:
Hi Waldemar,
The following snippet should do it when compiled statically for MIPS32 and emulated using qemu-user-static:
#include <time.h> #include <stddef.h> #include <unistd.h>
int main(int argc, char **argv) {
time_t start = time(NULL); sleep(2); time_t end = time(NULL);
return end != start; }
When I strace the resulting binary, I saw: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=0, tv_nsec=2}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
But I expected to see: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=2, tv_nsec=0}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
kind regards Nadav
Waldemar Brodkorb wrote,
Hi Nadav,
do you have some test code showing the issue?
best regards Waldemar
Nadav Tasher wrote,
Without this change `timespec` is too large for the `clock_nanosleep` syscall, which translates to sleep(0) on 32bit systems when TIME64 is enabled (the default).
Tested on MIPS.
Signed-off-by: Nadav Tasher <tashernadav@gmail.com> --- include/time.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/time.h b/include/time.h index 1a1408990..25dfb9e2e 100644 --- a/include/time.h +++ b/include/time.h @@ -118,7 +118,7 @@ typedef __timer_t timer_t; has nanoseconds instead of microseconds. */ struct timespec { - __time_t tv_sec; /* Seconds. */ + int tv_sec; /* Seconds. */ long int tv_nsec; /* Nanoseconds. */ };
-- 2.34.1
_______________________________________________ devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
linux 4.x doesn't support time64 syscalls вс, 22 дек. 2024 г., 13:00 Nadav Tasher <tashernadav@gmail.com>:
Interesting - I have been able to reproduce the bug when compiling like so: make CROSS_COMPILE=mips-linux-gnu- TARGET_ARCH=mips KERNEL_HEADERS=~/linux-headers-4.19.325/include/ -j4 clean make CROSS_COMPILE=mips-linux-gnu- TARGET_ARCH=mips KERNEL_HEADERS=~/linux-headers-4.19.325/include/ -j4
Note that this also reproduces in the existing toolchain at https://toolchains.bootlin.com/downloads/releases/toolchains/mips32/tarballs... (this is actually how I encountered the bug).
Dmitriy Chestnykh wrote:
Cannot reproduce it with the latest uclibc. Config in attachment
сб, 21 дек. 2024 г. в 14:20, Nadav Tasher <tashernadav@gmail.com>:
Hi Waldemar,
The following snippet should do it when compiled statically for MIPS32 and emulated using qemu-user-static:
#include <time.h> #include <stddef.h> #include <unistd.h>
int main(int argc, char **argv) {
time_t start = time(NULL); sleep(2); time_t end = time(NULL);
return end != start; }
When I strace the resulting binary, I saw: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=0, tv_nsec=2}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
But I expected to see: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=2, tv_nsec=0}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
kind regards Nadav
Waldemar Brodkorb wrote,
Hi Nadav,
do you have some test code showing the issue?
best regards Waldemar
Nadav Tasher wrote,
Without this change `timespec` is too large for the `clock_nanosleep` syscall, which translates to sleep(0) on 32bit systems when TIME64 is enabled (the default).
Tested on MIPS.
Signed-off-by: Nadav Tasher <tashernadav@gmail.com> --- include/time.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/time.h b/include/time.h index 1a1408990..25dfb9e2e 100644 --- a/include/time.h +++ b/include/time.h @@ -118,7 +118,7 @@ typedef __timer_t timer_t; has nanoseconds instead of microseconds. */ struct timespec { - __time_t tv_sec; /* Seconds. */ + int tv_sec; /* Seconds. */ long int tv_nsec; /* Nanoseconds. */ };
-- 2.34.1
_______________________________________________ devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
The documentation of uClibc-ng does not mention linux support beyond 4.x. Maybe a check should be added to the header file enabling time64 support? Dmitriy Chestnykh wrote:
linux 4.x doesn't support time64 syscalls
вс, 22 дек. 2024 г., 13:00 Nadav Tasher <tashernadav@gmail.com>:
Interesting - I have been able to reproduce the bug when compiling like so: make CROSS_COMPILE=mips-linux-gnu- TARGET_ARCH=mips KERNEL_HEADERS=~/linux-headers-4.19.325/include/ -j4 clean make CROSS_COMPILE=mips-linux-gnu- TARGET_ARCH=mips KERNEL_HEADERS=~/linux-headers-4.19.325/include/ -j4
Note that this also reproduces in the existing toolchain at https://toolchains.bootlin.com/downloads/releases/toolchains/mips32/tarballs... (this is actually how I encountered the bug).
Dmitriy Chestnykh wrote:
Cannot reproduce it with the latest uclibc. Config in attachment
сб, 21 дек. 2024 г. в 14:20, Nadav Tasher <tashernadav@gmail.com>:
Hi Waldemar,
The following snippet should do it when compiled statically for MIPS32 and emulated using qemu-user-static:
#include <time.h> #include <stddef.h> #include <unistd.h>
int main(int argc, char **argv) {
time_t start = time(NULL); sleep(2); time_t end = time(NULL);
return end != start; }
When I strace the resulting binary, I saw: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=0, tv_nsec=2}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
But I expected to see: [pid 2271212] rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 [pid 2271212] clock_nanosleep(CLOCK_REALTIME, 0, {tv_sec=2, tv_nsec=0}, 0x7ffca592a250) = 0 [pid 2271212] brk(0x611d0d49b000) = 0x611d0d49b000 [pid 2271212] exit_group(1) = ?
kind regards Nadav
Waldemar Brodkorb wrote,
Hi Nadav,
do you have some test code showing the issue?
best regards Waldemar
Nadav Tasher wrote,
Without this change `timespec` is too large for the `clock_nanosleep` syscall, which translates to sleep(0) on 32bit systems when TIME64 is enabled (the default).
Tested on MIPS.
Signed-off-by: Nadav Tasher <tashernadav@gmail.com> --- include/time.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/time.h b/include/time.h index 1a1408990..25dfb9e2e 100644 --- a/include/time.h +++ b/include/time.h @@ -118,7 +118,7 @@ typedef __timer_t timer_t; has nanoseconds instead of microseconds. */ struct timespec { - __time_t tv_sec; /* Seconds. */ + int tv_sec; /* Seconds. */ long int tv_nsec; /* Nanoseconds. */ };
-- 2.34.1
_______________________________________________ devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
devel mailing list -- devel@uclibc-ng.org To unsubscribe send an email to devel-leave@uclibc-ng.org
participants (3)
-
Dmitriy Chestnykh -
Nadav Tasher -
Waldemar Brodkorb