Re: [PATCH v3 09/16] elf: Move vDSO setup to rtld (BZ#24967) - regression found

[Adhemerval Zanella <adhemerval dot zanella at linaro dot org>]

On 07/01/2020 13:13, Lukasz Majewski wrote:
> 
> Unfortunately this patch caused regression on my ARMv7 32 bit setup.
> Please find filled in bugzilla entry:
> 
> https://sourceware.org/bugzilla/show_bug.cgi?id=25352

Are you sure you are testing it properly? As indicated by Andreas, it
seems a mismatch between ld.so and libc.so.  I just setup a debian 9 
VM with both a 4.9 and 5.4 kernel and a simple test:

--
#include <time.h>

int main ()
{
  clock_gettime (CLOCK_REALTIME, &(struct timespec) {});
  clock_gettime (CLOCK_MONOTONIC, &(struct timespec) {});
  clock_gettime (CLOCK_BOOTTIME, &(struct timespec) {});

  return 0;
}
--

does show the expected results.  On 4.9 kernel I am seeing:

15:36:06 syscall_0x193(0, 0xbed9a3d8, 0xc8, 0, 0xbed9a3d8, 0) = -1 ENOSYS (Function not implemented)
15:36:06 clock_gettime(CLOCK_BOOTTIME, {tv_sec=1303, tv_nsec=570795799}) = 0
15:36:06 exit_group(0)                  = ?
15:36:06 +++ exited with 0 +++

While on a 5.4 kernel:

syscall_0x193(0, 0xbedfb3e8, 0xc8, 0, 0xbedfb3e8, 0) = 0
syscall_0x193(0x1, 0xbedfb3e8, 0xc8, 0, 0xbedfb3e8, 0x1) = 0
syscall_0x193(0x7, 0xbedfb3e8, 0xc8, 0, 0xbedfb3e8, 0x7) = 0

(arm clock_gettime64 vDSO was added on 5.5).

Attachment: signature.asc
Description: OpenPGP digital signature