uml: stop gdb from deleting breakpoints when running UML

Sometimes when UML is debugged gdb miss breakpoints.

When process traced by gdb do fork, debugger remove breakpoints from
child address space. There is possibility to trace more than one fork,
but this not work with UML, I guess (only guess) there is a deadlock -
gdb waits for UML and UML waits for gdb.

When clone() is called with SIGCHLD and CLONE_VM flags, gdb see this
as PTRACE_EVENT_FORK not as PTRACE_EVENT_CLONE and remove breakpoints
from child and at the same time from traced process, because either
have the same address space.

Maybe it is possible to do fix in gdb, but I'm not sure if there is
easy way to find out if traced and child processes share memory. So I
do fix for UML, it simply do not call clone() with both SIGCHLD and
CLONE_VM flags together.  Additionally __WALL flag is used for
waitpid() to assure not miss clone and normal process events.

[ jdike - checkpatch fixes ]

Signed-off-by: Stanislaw Gruszka <stf_xl@wp.pl>
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 1 insertions, 1 deletions

diff --git a/arch/um/os-Linux/drivers/tuntap_user.c b/arch/um/os-Linux/drivers/tuntap_user.c
index 6c55d3c8ead8..1037a3b6386e 100644
--- a/arch/um/os-Linux/drivers/tuntap_user.c
+++ b/arch/um/os-Linux/drivers/tuntap_user.c
@@ -107,7 +107,7 @@ static int tuntap_open_tramp(char *gate, int *fd_out, int me, int remote,
 		       "errno = %d\n", errno);
 		return err;
 	}
-	CATCH_EINTR(waitpid(pid, NULL, 0));
+	helper_wait(pid, 0, "tuntap_open_tramp");
 
 	cmsg = CMSG_FIRSTHDR(&msg);
 	if (cmsg == NULL) {