Re: [PATCH 14/23] memory: ti-emif-pm: Fix cast to iomem pointer

[Krzysztof Kozlowski]

On Thu, Jul 23, 2020 at 11:14:02AM +0200, Arnd Bergmann wrote:
> On Thu, Jul 23, 2020 at 11:02 AM Krzysztof Kozlowski <krzk@xxxxxxxxxx> wrote:
> > On Thu, Jul 23, 2020 at 10:48:19AM +0200, Arnd Bergmann wrote:
> > > On Thu, Jul 23, 2020 at 9:39 AM Krzysztof Kozlowski <krzk@xxxxxxxxxx> wrote:
> > > >
> > > > Cast pointer to iomem memory properly to fix sparse warning:
> > > >
> > > >     drivers/memory/ti-emif-pm.c:251:38: warning: incorrect type in argument 1 (different address spaces)
> > > >     drivers/memory/ti-emif-pm.c:251:38:    expected void const volatile [noderef] __iomem *addr
> > > >     drivers/memory/ti-emif-pm.c:251:38:    got void *
> > > >
> > > > Signed-off-by: Krzysztof Kozlowski <krzk@xxxxxxxxxx>
> > > > ---
> > > >  drivers/memory/ti-emif-pm.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/memory/ti-emif-pm.c b/drivers/memory/ti-emif-pm.c
> > > > index 9c90f815ad3a..6c747c1e98cb 100644
> > > > --- a/drivers/memory/ti-emif-pm.c
> > > > +++ b/drivers/memory/ti-emif-pm.c
> > > > @@ -248,7 +248,7 @@ MODULE_DEVICE_TABLE(of, ti_emif_of_match);
> > > >  static int ti_emif_resume(struct device *dev)
> > > >  {
> > > >         unsigned long tmp =
> > > > -                       __raw_readl((void *)emif_instance->ti_emif_sram_virt);
> > > > +                       __raw_readl((void __iomem *)emif_instance->ti_emif_sram_virt);
> > > >
> > >
> > > Maybe this shouldn't even be __raw_readl(), but instead READ_ONCE()?
> >
> > Won't readl() be enough? Indeed it looks problematic.
> 
> readl() won't work on big-endian kernels, since this is a byte comparison.

Ah, right.

> 
> > > The other accesses in this file don't use MMIO wrappers either but just treat
> > > it as a pointer. The effect would be the same though.
> >
> > I think all the reads and writes are with readl() and writel().
> 
> I actually see only one other access:
> 
>         copy_addr = sram_exec_copy(emif_data->sram_pool_code,
>                                    (void *)emif_data->ti_emif_sram_virt,
>                                    &ti_emif_sram, ti_emif_sram_sz);
> 
> and this one ends up in a memcpy() that does not perform any byte
> swapping or barriers.

At least the barrier would come through mutex in sram_exec_copy() and
later spin locks for page table manipulation.

Anyway, I do not have the HW to test the changes or to confirm whether
this is real issue.  I guess the driver author/owner should follow up on
this report.

Best regards,
Krzysztof