Re: [PATCH 1/7] x86/vdso: Respect COMPAT_32BIT_TIME

["H. Peter Anvin" <hpa@zytor.com>]

On 2026-03-05 01:24, Thomas Weißschuh wrote:
> 
>> Weak references would be a way to work around the link failures. 
> 
> I am still not sure where "the link failures" should be coming from.
> The only sense I can make out of it, is if somebody manually and directly links
> to vdso.so. Like in the following example:
> 
> $ cat test.c 
> #include <stdio.h>
> 
> #include <linux/time.h>
> 
> int __vdso_clock_gettime(__kernel_clockid_t clock, struct __kernel_timespec *ts);
> 
> int main(void)
> {
> 	struct __kernel_timespec ts;
> 	int ret;
> 
> 	printf("__vdso_clock_gettime=%p\n", __vdso_clock_gettime);
> 
> 	ret = __vdso_clock_gettime(CLOCK_REALTIME, &ts);
> 	printf("ret=%d\n", ret);
> }
> $ gcc test.c /lib/modules/$(uname -r)/vdso/vdso64.so
> $ ./a.out 
> __vdso_clock_gettime=0x7ff6ba2eeb80
> ret=0
> 
> This actually works on glibc (not on musl). But it is highly non-standard and
> relies on multiple implementation details. Furthermore it can fail to run on
> systems without a vDSO, as mentioned before.
> 
> Is this the usage pattern you have in mind?
> Do you know of anybody doing things this way?
> 

Yes, and yes, I do.

	-hpa