Define __NR_futex to be __NR_futex_time64 on riscv32
Subsystems (2)
PATCH 1/1 Define __NR_futex to be __NR_futex_time64 on riscv32
RISCV glibc has decided to use 64bit time_t from get go unlike other 32bit architecture therefore aliasing __NR_futex to __NR_futex_time64 helps avoid the below errors on rv32
tst_checkpoint.c:99:17: error: use of undeclared identifier 'SYS_futex'
--- lib/tst_checkpoint.c | 4 ++++ testcases/kernel/syscalls/clone/clone08.c | 4 ++++ testcases/kernel/syscalls/futex/futextest.h | 4 ++++ 3 files changed, 12 insertions(+)
lib/tst_checkpoint.c +4 -0
@@ -21,6 +21,10 @@* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ +#if !defined(__NR_futex) && defined(__riscv) && __riscv_xlen ==32 +# define __NR_futex __NR_futex_time64 +#endif +#include <stdint.h> #include <limits.h> #include <errno.h>
testcases/kernel/syscalls/clone/clone08.c +4 -0
@@ -5,6 +5,10 @@* Author: Zeng Linggang <[email protected]> */ +#if !defined(__NR_futex) && defined(__riscv) && __riscv_xlen ==32 +# define __NR_futex __NR_futex_time64 +#endif +#define _GNU_SOURCE#include <stdlib.h> #include <stdio.h>
testcases/kernel/syscalls/futex/futextest.h +4 -0
@@ -34,6 +34,10 @@#ifndef _FUTEXTEST_H#define _FUTEXTEST_H +#if !defined(__NR_futex) && defined(__riscv) && __riscv_xlen ==32 +# define __NR_futex __NR_futex_time64 +#endif +#include <unistd.h> #include <sys/syscall.h> #include <sys/types.h>
Signature · 1 line
2.26.2
PATCH 1/1 Define __NR_futex to be __NR_futex_time64 on riscv32
Hi Khem,
Quoted 5 lines
RISCV glibc has decided to use 64bit
time_tfrom get go unlike other 32bit architecture therefore aliasing__NR_futexto__NR_futex_time64helps avoid the below errors on rv32tst_checkpoint.c:99:17: error: use of undeclared identifier '
SYS_futex'
Thanks for your fix.
BTW, out of curiosity, is Risc-v 32 bit merged into glibc master?
I found a patch from Alistair Francis from January, which implements what you claim:
https://sourceware.org/legacy-ml/libc-alpha/2020-01/msg00205.html
...
+/* RV32 and RV64 both use 64-bit time_t */
+#define __TIMESIZE 64
sysdeps/unix/sysv/linux/riscv/bits/typesizes.h +0 -0
...
(part of https://patches-gcc.linaro.org/project/glibc/list/?series=21554
patchset) but it hasn't been merged yet to master.
+ There is older patchset from Zong Li, not yet merged.
https://patches-gcc.linaro.org/cover/12952/
Quoted 12 lines
Signed-off-by Khem Raj <[email protected]>--- lib/tst_checkpoint.c | 4 ++++ testcases/kernel/syscalls/clone/clone08.c | 4 ++++ testcases/kernel/syscalls/futex/futextest.h | 4 ++++ 3 files changed, 12 insertions(+)lib/tst_checkpoint.c +0
-0
lib/tst_checkpoint.c line 26
Quoted 6 lines
lib/tst_checkpoint.c +3
-0@@ -21,6 +21,10 @@* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ +#if !defined(__NR_futex) && defined(__riscv) && __riscv_xlen ==32 +# define __NR_futex __NR_futex_time64 +#endif
I guess this should go to include/lapi/futex.h, so we don't repeat ourselves. (and clone08.c needs to include it, others already do).
...
Kind regards,
Petr
PATCH 1/1 Define __NR_futex to be __NR_futex_time64 on riscv32
Hi Khem,
Quoted 6 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+#if !defined(__NR_futex) && defined(__riscv) && __riscv_xlen == 32 +# define __NR_futex __NR_futex_time64 +#endifI guess this should go to
include/lapi/futex.h, so we don't repeat ourselves. (andclone08.cneeds to include it, others already do).
I've noticed that we still don't have risc-v support in include/lapi/syscalls/. IMHO adding it would be better, than fixing just one missing __NR numbers.
Kind regards,
Petr
PATCH 1/1 Define __NR_futex to be __NR_futex_time64 on riscv32
In reply to Petr VorelQuoted 7 lines · Petr Vorel
Hi Khem,
Quoted 3 lines
RISCV glibc has decided to use 64bit
time_tfrom get go unlike other 32bit architecture therefore aliasing__NR_futexto__NR_futex_time64helps avoid the below errors on rv32
Quoted 5 lines
Quoted 1 line
tst_checkpoint.c:99:17: error: use of undeclared identifier '
SYS_futex'Thanks for your fix.
BTW, out of curiosity, is Risc-v 32 bit merged into glibc master?
Not yet but its almost finalised and should be merged before 2.32 hopefully
sysdeps/unix/sysv/linux/riscv/bits/typesizes.h
Quoted 9 lines
I found a patch from Alistair Francis from January, which implements what you claim:
https://sourceware.org/legacy-ml/libc-alpha/2020-01/msg00205.html
...(quoted diff) +2
-0+/* RV32 and RV64 both use 64-bit time_t */ +#define __TIMESIZE 64sysdeps/unix/sysv/linux/riscv/bits/typesizes.h +0
-0
right
Quoted 2 lines
(part of https://patches-gcc.linaro.org/project/glibc/list/?series=21554
patchset) but it hasn't been merged yet to master.
Its being reviewed
Quoted 4 lines
+ There is older patchset from Zong Li, not yet merged.
https://patches-gcc.linaro.org/cover/12952/
above patches are newer versions of this one.
Quoted 15 lines
Quoted 14 lines
Signed-off-by Khem Raj <[email protected]>--- lib/tst_checkpoint.c | 4 ++++ testcases/kernel/syscalls/clone/clone08.c | 4 ++++ testcases/kernel/syscalls/futex/futextest.h | 4 ++++ 3 files changed, 12 insertions(+)lib/tst_checkpoint.c +0
-0@@ -21,6 +21,10 @@* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */
Quoted 7 lines
lib/tst_checkpoint.c line 25
Quoted 3 lines
lib/tst_checkpoint.c +3
-0+#if !defined(__NR_futex) && defined(__riscv) && __riscv_xlen ==32 +# define __NR_futex __NR_futex_time64 +#endif I guess this should go to
include/lapi/futex.h, so we don't repeat ourselves. (andclone08.cneeds to include it, others already do).
OK makes sense, will send v2 ?
Quoted 4 lines
...
Kind regards,
Petr
PATCH 1/1 Define __NR_futex to be __NR_futex_time64 on riscv32
Hi Khem,
sorry for a delay.
Quoted 11 lines
Quoted 9 lines
Quoted 5 lines
RISCV glibc has decided to use 64bit
time_tfrom get go unlike other 32bit architecture therefore aliasing__NR_futexto__NR_futex_time64helps avoid the below errors on rv32tst_checkpoint.c:99:17: error: use of undeclared identifier '
SYS_futex'Thanks for your fix.
BTW, out of curiosity, is Risc-v 32 bit merged into glibc master?
Not yet but its almost finalised and should be merged before 2.32 hopefully
Nice!
Quoted 5 lines
Quoted 2 lines
I guess this should go to
include/lapi/futex.h, so we don't repeat ourselves. (andclone08.cneeds to include it, others already do).OK makes sense, will send v2 ?
Yes please.
Kind regards,
Petr