Hi Hauke & Thomas,
On Tue, Jun 11, 2019 at 10:19:47AM +0200, Thomas Bogendoerfer wrote:
> On Tue, Jun 11, 2019 at 01:41:21AM +0200, Hauke Mehrtens wrote:
> > On 5/28/19 7:05 PM, Paul Burton wrote:
> > > diff --git a/arch/mips/mm/mmap.c b/arch/mips/mm/mmap.c
> > > index 2f616ebeb7e0..7755a1fad05a 100644
> > > --- a/arch/mips/mm/mmap.c
> > > +++ b/arch/mips/mm/mmap.c
> > > @@ -203,6 +203,11 @@ unsigned long arch_randomize_brk(struct mm_struct *mm)
> > >
> > > int __virt_addr_valid(const volatile void *kaddr)
> > > {
> > > + unsigned long vaddr = (unsigned long)vaddr;
>
> the second vaddr should be better kaddr
D'oh..! Right you are...
Returning false all the time is enough to silence the hardened usercopy
warnings but clearly not the right behaviour.
> > Someone complained that this compiled to a constant "return 0" for him:
> > https://bugs.openwrt.org/index.php?do=details&task_id=2305#comment6554
> >
> > I just checked this on a unmodified 5.2-rc4 with the xway_defconfig and
> > I get this:
> >
> > 0001915c <__virt_addr_valid>:
> > 1915c: 03e00008 jr ra
> > 19160: 00001025 move v0,zero
> >
> > Is this intended?
>
> I don't think so. Interesting what the compiler decides to do here.
Yes, this is equivalent to using uninitialized_var() but I'm surprised
the code got discarded entirely...
Thanks,
Paul
