Re: [PATCH v3 1/2] drm: Fix fault format

[Christophe Leroy]

Le 04/10/2024 à 15:21, Arnd Bergmann a écrit :
> On Thu, Oct 3, 2024, at 15:29, Vincenzo Frascino wrote:
> > diff --git a/drivers/gpu/drm/i915/gt/intel_gt.c
> > b/drivers/gpu/drm/i915/gt/intel_gt.c
> > index a6c69a706fd7..352ef5e1c615 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_gt.c
> > +++ b/drivers/gpu/drm/i915/gt/intel_gt.c
> > @@ -308,7 +308,7 @@ static void gen6_check_faults(struct intel_gt *gt)
> >   		fault = GEN6_RING_FAULT_REG_READ(engine);
> >   		if (fault & RING_FAULT_VALID) {
> >   			gt_dbg(gt, "Unexpected fault\n"
> > -			       "\tAddr: 0x%08lx\n"
> > +			       "\tAddr: 0x%08x\n"
> >   			       "\tAddress space: %s\n"
> >   			       "\tSource ID: %d\n"
> >   			       "\tType: %d\n",
>
> Isn't the type of PAGE_MASK still architecture dependent?

Indeed when I commented that PAGE_MASK was type agnostic I was thinking 
about the powerpc PAGE_MASK:

#define PAGE_MASK      (~((1 << PAGE_SHIFT) - 1))

It should probably be possible to generalise it to all architectures.

But if you keep some PAGE_MASK with forced UL type just like:

#define PAGE_SIZE		(_AC(1, UL) << PAGE_SHIFT)
#define PAGE_MASK		(~(PAGE_SIZE-1))

Then that version of PAGE_MASK isn't agnostic and ANDing an int with 
that mask makes a long result.

> I think you need a cast to either 'int' or 'long' here to
> make the corresponding format string work across all
> architectures. With the current version of your patch 2/2,
> it looks like it has to be %x for architectures with
> 64-bit phys_addr_t, but %lx for the other ones.
>
> Changing the 'u32 fault' variable to 'unsigned long'
> would also work here.
>
>        Arnd