regression with 2.6.12 kernel

I'm seeing 12 new failures with a 2.6.12 kernel:

/tset/LSB.os/procprim/exec_L/T.execl_L 7 FAIL
/tset/LSB.os/procprim/exec_L/T.execle_L 7 FAIL
/tset/LSB.os/procprim/exec_L/T.execlp_L 7 FAIL
/tset/LSB.os/procprim/exec_L/T.execv_L 7 FAIL
/tset/LSB.os/procprim/exec_L/T.execve_L 7 FAIL
/tset/LSB.os/procprim/exec_L/T.execvp_L 7 FAIL
/tset/POSIX.os/procprim/exec/T.execl 30 FAIL
/tset/POSIX.os/procprim/exec/T.execle 30 FAIL
/tset/POSIX.os/procprim/exec/T.execlp 30 FAIL
/tset/POSIX.os/procprim/exec/T.execv 30 FAIL
/tset/POSIX.os/procprim/exec/T.execve 30 FAIL
/tset/POSIX.os/procprim/exec/T.execvp 30 FAIL

The "7" tests all return:

For ITIMER_REAL: it_value.tv_sec reset incorrectly by exec:
expected value 0; observed value 0

The "30" tests all return:

Following execl(...), time to alarm was cleared
Following execl(...), alarm not set correctly
Expected value 250; Observed value 0

If I roll back to a 2.6.11 kernel without changing anything else, they all 
pass. (this is on ia32).

I'm digging through the kernel source now, but thought I'd toss this out 
if anyone else has already found this issue and might have a patch.

-- 
Stew Benedict
Mandriva
Mandrakesoft becomes Mandriva