Description of problem: The _exit() function in libc(IA32) calls syscall 252 (exit_group) which is unimplemented in the IA64 emulation. exit() is called anyway, so this may not have major adverse effects. However, the wrong syscall generates an ugly syslog message. Moreover, there are many other unimplemented ia32 syscalls, some of which are called on a regular basis by ia32 applications. Version-Release number of selected component (if applicable): 2.3.2-95.6 How reproducible: always Steps to Reproduce: 1. compile "Hello World" C Program on i386 (using shared libs) 2. run generated executable on ia64 Actual results: /var/log/messages entry: kernel: IA32 syscall #252 issued, maybe we should implement it Expected results: No error message Additional info: There are a number of other syscalls that cause similar error messages, I have seen #258 (set_tid_address) and #270 (tgkill).
I know you want to mark this as a duplicate of BUG 107116. Before you do so, hgave a look at this code: GLIBC_SOURCES/sysdeps/unix/sysv/linux/_exit.c void _exit (status) int status; { while (1) { #ifdef __NR_exit_group INLINE_SYSCALL (exit_group, 1, status); #endif INLINE_SYSCALL (exit, 1, status); #ifdef ABORT_INSTRUCTION ABORT_INSTRUCTION; #endif } } Looking at this code, it seems that this file includes wrong kernel headers suggesting that the exit_group() syscall exists. IMO there'd be an easy way to handle this problem: include correct kernel headers that do not define NR_exit_group and other NR_ macros for unimplemented syscalls on ia64/ia32.
exit_group is implemented on i386, therefore it is of course in i386 headers. Compatibility libraries certainly shouldn't differ from the real ones, it really either needs to be implemented in the kernel, or implemented in the emulator.
I fail to see why a compatibility library should invoke unimplemented system calls. Actually, I consider that pretty dangerous. To my understanding, what mainly matters is the ABI exposed by the library to applications, not the kernel interface.