Re: [PATCH v3] LoongArch: fix missing trap for enabled exceptions on narrowing operation

On Tue, 2026-07-07 at 15:06 +0100, Yury Khrustalev wrote:
> On Tue, Jul 07, 2026 at 01:46:09PM +0100, Yury Khrustalev wrote:
> > On Tue, Mar 24, 2026 at 03:22:59PM +0800, Xi Ruoyao wrote:
> > > The libc_feupdateenv_test macro is supposed to trap when the trap for a
> > > previously held exception is enabled.  But
> > > libc_feupdateenv_test_loongarch wasn't doing it properly: the comment
> > > claims "setting of the cause bits" would cause "the hardware to generate
> > > the exception" but that's simply not true for the LoongArch movgr2fcsr
> > > instruction.
> > > 
> > > To fix the issue, we need to call __feraiseexcept in case a held exception
> > > is enabled to trap.
> > > 
> > > Reviewed-by: caiyinyu <[email protected]>
> > > Signed-off-by: Xi Ruoyao <[email protected]>
> > > ---
> > > 
> > > ...
> > > 
> > > diff --git a/math/test-narrowing-trap.c b/math/test-narrowing-trap.c
> > > new file mode 100644
> > > index 0000000000..a483c4bf1a
> > > --- /dev/null
> > > +++ b/math/test-narrowing-trap.c
> > > 
> > > ...
> > > 
> > > +static int
> > > +do_test (void)
> > > +{
> > > +  pid_t pid;
> > > +
> > > +  if (!EXCEPTION_ENABLE_SUPPORTED (FE_INVALID))
> > > +    FAIL_UNSUPPORTED ("feenableexcept (FE_INVALID) not supported");
> > 
> > On AArch64 we have
> > 
> >   #define EXCEPTION_ENABLE_SUPPORTED(EXCEPT)	((EXCEPT) == 0)
> > 
> > Which means that this test will always be unsupported even if
> > 
> >   feenableexcept (FE_INVALID)
> > 
> > would return 0 (compare with math/test-fenv-preserve.c for example).
> > 
> > I think this check is not correct.
> > 
> > Thanks,
> > Yury
> 
> This simple fix should work I think.
> 
> diff --git a/math/test-narrowing-trap.c b/math/test-narrowing-trap.c
> index 6d21357913..5871cb33fc 100644
> --- a/math/test-narrowing-trap.c
> +++ b/math/test-narrowing-trap.c
> @@ -29,14 +29,14 @@ do_test (void)
>  #ifdef FE_INVALID
>    pid_t pid;
>  
> -  if (!EXCEPTION_ENABLE_SUPPORTED (FE_INVALID))
> -    FAIL_UNSUPPORTED ("feenableexcept (FE_INVALID) not supported");
> -
>    pid = xfork ();
>    if (pid == 0)
>      {
>        int r = feenableexcept (FE_INVALID);
>  
> +      if (!EXCEPTION_ENABLE_SUPPORTED (FE_INVALID) && (r == -1))

The parentheses around r == -1 is redundant.  Otherwise OK to me.

> +        FAIL_UNSUPPORTED ("feenableexcept (FE_INVALID) not supported");
> +
>        TEST_COMPARE (r, 0);
>        fdiv (0.0, 0.0);
>        _exit (0);

-- 
Xi Ruoyao <[email protected]>