The following define in libc/misc/sysvipc/ipc.h effectively makes uclibc wrappers OR user's *ctl commands with 0x100 before passing to kernel. However this flag is not expected and not decoded on kernel side. The latter causes all user's commands passed to *ctl fuctions fail with EINVAL result.

#if __WORDSIZE == 32 || defined __alpha__ || defined __mips__
# define __IPC_64	0x100
#else
# define __IPC_64	0x0
#endif

The problem affects the following SYSV fuctions: semctl(), msgctl(), shmctl().
ipc/sem.c:
SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
{
[...]
   switch(cmd) { <-- 0x100 is not removed, goes directly to default case and returns -EINVAL

ipc/shm.c:
SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
{
[...]
   version = ipc_parse_version(&cmd); <-- 0x100 is removed iff __ARCH_WANT_IPC_PARSE_VERSION is defined (which is not always the case)

ipc/msg.c:
SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
{
[...]
   version = ipc_parse_version(&cmd); <-- 0x100 is removed iff __ARCH_WANT_IPC_PARSE_VERSION is defined (which is not always the case)

A quick and dirty fix is to define __IPC_64 0x0 unconditionally as modern kernels do not seem to use old 32 bit structures internally any more (checked 2.6 and 3.x trunk). So does user space.

A better solution is to look at kernel's version to check if 0x100 is actually needed.