On 23/05/2019 19:54, Laurent Vivier wrote:
In current code, __NR_msgrcv and__NR_semtimedop are supposed to be defined if __NR_msgsnd is defined.But linux headers 5.2-rc1 for MIPS define __NR_msgsnd without defining __NR_semtimedop and it breaks the QEMU build. __NR_semtimedop is defined in asm-mips/unistd_n64.h and asm-mips/unistd_n32.h but not in asm-mips/unistd_o32.h. Commit d9cb4336159a ("linux headers: update against Linux 5.2-rc1") has updated asm-mips/unistd_o32.h and added __NR_msgsnd but not __NR_semtimedop. It introduces __NR_semtimedop_time64 instead. This patch fixes the problem by checking for each __NR_XXX symbol before defining the corresponding syscall. Fixes: d9cb4336159a ("linux headers: update against Linux 5.2-rc1") Reported-by: Philippe Mathieu-Daudé <phi...@redhat.com> Signed-off-by: Laurent Vivier <laur...@vivier.eu> --- linux-user/syscall.c | 24 ++++++++++++++++-------- 1 file changed, 16 insertions(+), 8 deletions(-)
This only fixes the problem at build time, but the changes in the kernel headers introduce also a regression at execution time:
if the host kernel doesn't implement the syscall, the syscall fails (ENOSYS) whereas it was working before because it was using ipc() instead. I have this problem with a Fedora 28 on ppc64 (5.0.16-100.fc28.ppc64) (LTP test msgctl07).
I'm preparing a fix. Thanks, Laurent
