| Message ID | 20220512131002.26093-1-chrubis@suse.cz |
|---|---|
| State | Accepted |
| Headers | show |
| Series | [v2] waitid10: raise SIGFPE directly | expand |
On Thu, May 12, 2022 at 3:08 PM Cyril Hrubis <chrubis@suse.cz> wrote: > > The SIGFPE for division by zero is actually not send for quite a few > architectures (ARM for instance) and even on x86 and x86_64 we need to > work around compiler to make it generate code that actually triggers the > condition. > > So this patch fixes the test in the simplest way possible. the child > just directly raises SIGFPE instead. > > Signed-off-by: Cyril Hrubis <chrubis@suse.cz> Acked-by: Jan Stancek <jstancek@redhat.com>
On Thu, May 12, 2022 at 9:08 PM Cyril Hrubis <chrubis@suse.cz> wrote: > The SIGFPE for division by zero is actually not send for quite a few > architectures (ARM for instance) and even on x86 and x86_64 we need to > work around compiler to make it generate code that actually triggers the > condition. > > So this patch fixes the test in the simplest way possible. the child > just directly raises SIGFPE instead. > > Signed-off-by: Cyril Hrubis <chrubis@suse.cz> > --- > testcases/kernel/syscalls/waitid/waitid10.c | 8 ++------ > 1 file changed, 2 insertions(+), 6 deletions(-) > > diff --git a/testcases/kernel/syscalls/waitid/waitid10.c > b/testcases/kernel/syscalls/waitid/waitid10.c > index e75edd07e..388b31cc9 100644 > --- a/testcases/kernel/syscalls/waitid/waitid10.c > +++ b/testcases/kernel/syscalls/waitid/waitid10.c > @@ -24,12 +24,8 @@ static void run(void) > pid_t pidchild; > > pidchild = SAFE_FORK(); > - if (!pidchild) { > - volatile int a, zero = 0; > - > - a = 1 / zero; > - exit(a); > - } > + if (!pidchild) > + raise(SIGFPE); > Better to have code comments to explain the reason here, just for better readability. And vote for merging before the new release as well. Reviewed-by: Li Wang <liwang@redhat.com> > > TST_EXP_PASS(waitid(P_ALL, 0, infop, WEXITED)); > TST_EXP_EQ_LI(infop->si_pid, pidchild); > -- > 2.35.1 > > > -- > Mailing list info: https://lists.linux.it/listinfo/ltp > >
Hi! > Better to have code comments to explain the reason here, > just for better readability. What exactly should we comment there, something as: Triggering SIGFPE by invalid instruction is not always possible, some architectures does not trap division-by-zero at all and even when it's possible we would have to fight the compiler optimizations that have tendency to remove undefined operations. > And vote for merging before the new release as well. That's what I'm aiming for.
> Triggering SIGFPE by invalid instruction is not always possible, some > architectures does not trap division-by-zero at all and even when it's > possible we would have to fight the compiler optimizations that have > tendency to remove undefined operations. > Looks good.
Hi! Added comment and pushed, thanks for the reviews.
diff --git a/testcases/kernel/syscalls/waitid/waitid10.c b/testcases/kernel/syscalls/waitid/waitid10.c index e75edd07e..388b31cc9 100644 --- a/testcases/kernel/syscalls/waitid/waitid10.c +++ b/testcases/kernel/syscalls/waitid/waitid10.c @@ -24,12 +24,8 @@ static void run(void) pid_t pidchild; pidchild = SAFE_FORK(); - if (!pidchild) { - volatile int a, zero = 0; - - a = 1 / zero; - exit(a); - } + if (!pidchild) + raise(SIGFPE); TST_EXP_PASS(waitid(P_ALL, 0, infop, WEXITED)); TST_EXP_EQ_LI(infop->si_pid, pidchild);
The SIGFPE for division by zero is actually not send for quite a few architectures (ARM for instance) and even on x86 and x86_64 we need to work around compiler to make it generate code that actually triggers the condition. So this patch fixes the test in the simplest way possible. the child just directly raises SIGFPE instead. Signed-off-by: Cyril Hrubis <chrubis@suse.cz> --- testcases/kernel/syscalls/waitid/waitid10.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)