On Wed, Jul 02, 2025 at 10:15:06PM +0200, Thomas Gleixner wrote:

On Wed, Jul 02 2025 at 11:21, Terry Tritton wrote:

...

Futex_waitv can not accept old_timespec32 struct, so userspace should

sys_futex_wait()

...

convert it from 32bit to 64bit before syscall in 32bit compatible mode.

This fix is based off [1]

Link: https://lore.kernel.org/all/20231203235117.29677-1-wegao@suse.com/ [1]

Originally-by: Wei Gao wegao@suse.com Signed-off-by: Terry Tritton terry.tritton@linaro.org

---

Changes in v3:

• Fix signed-off-by chain but for real this time

Changes in v2:

• Fix signed-off-by chain

.../testing/selftests/futex/include/futex2test.h | 15 +++++++++++++++ 1 file changed, 15 insertions(+)

diff --git a/tools/testing/selftests/futex/include/futex2test.h b/tools/testing/selftests/futex/include/futex2test.h index ea79662405bc..6780e51eb2d6 100644 --- a/tools/testing/selftests/futex/include/futex2test.h +++ b/tools/testing/selftests/futex/include/futex2test.h @@ -55,6 +55,13 @@ struct futex32_numa { futex_t numa; }; +#if !defined(__LP64__) +struct timespec64 {

• int64_t tv_sec;
• int64_t tv_nsec;

+}; +#endif

Why not use __kernel_timespec which is the actual data type that syscall requests?

...

/**

• futex_waitv - Wait at multiple futexes, wake on any
• @waiters: Array of waiters

@@ -65,7 +72,15 @@ struct futex32_numa { static inline int futex_waitv(volatile struct futex_waitv *waiters, unsigned long nr_waiters, unsigned long flags, struct timespec *timo, clockid_t clockid) { +#if !defined(__LP64__)

• struct timespec64 timo64 = {0};
• timo64.tv_sec = timo->tv_sec;
• timo64.tv_nsec = timo->tv_nsec;
• return syscall(__NR_futex_waitv, waiters, nr_waiters, flags, &timo64, clockid);

+#else return syscall(__NR_futex_waitv, waiters, nr_waiters, flags, timo, clockid); +#endif

You simply can do

    struct __kernel_timespec ts = {
    	.tv_sec = timo->tv_sec,
            .tv_nsec = timo->tv_nsec,
    };

return syscall(__NR_futex_waitv, waiters, nr_waiters, flags, &ts, clockid); unconditionally. No?

Thanks for your valuable comments. Your change has more directly interacts with system calls and simplifies the code without conditional compilation. Only concern from my side is test on very old kernel(But i think this is not problem for kself test)

Thanks,

    tglx