Re: [PATCH] drm/i915/gt: Avoid redundant pointer validity check

[Deepak R Varma <drv@mailo.com>]

On Mon, Feb 06, 2023 at 10:33:13AM +0000, Matthew Auld wrote:
> On 06/02/2023 09:45, Tvrtko Ursulin wrote:
> > 
> > Hi,
> > 
> > Adding Matt & Thomas as potential candidates to review.
> > 
> > Regards,
> > 
> > Tvrtko
> > 
> > On 03/02/2023 19:30, Deepak R Varma wrote:
> > > The macro definition of gen6_for_all_pdes() expands to a for loop such
> > > that it breaks when the page table is null. Hence there is no need to
> > > again test validity of the page table entry pointers in the pde list.
> > > This change is identified using itnull.cocci semantic patch.
> > > 
> > > Signed-off-by: Deepak R Varma <drv@mailo.com>
> > > ---
> > > Please note: Proposed change is compile tested only.
> > > 
> > >   drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 5 ++---
> > >   1 file changed, 2 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
> > > b/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
> > > index 5aaacc53fa4c..787b9e6d9f59 100644
> > > --- a/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
> > > +++ b/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
> > > @@ -258,8 +258,7 @@ static void gen6_ppgtt_free_pd(struct gen6_ppgtt
> > > *ppgtt)
> > >       u32 pde;
> > >       gen6_for_all_pdes(pt, pd, pde)
> > > -        if (pt)
> > > -            free_pt(&ppgtt->base.vm, pt);
> > > +        free_pt(&ppgtt->base.vm, pt);
> > >   }
> > >   static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
> > > @@ -304,7 +303,7 @@ static void pd_vma_unbind(struct
> > > i915_address_space *vm,
> > >       /* Free all no longer used page tables */
> > >       gen6_for_all_pdes(pt, ppgtt->base.pd, pde) {
> > > -        if (!pt || atomic_read(&pt->used))
> > > +        if (atomic_read(&pt->used))
> 
> Wow, I was really confused trying to remember how this all works.
> 
> The gen6_for_all_pdes() does:
> 
> (pt = i915_pt_entry(pd, iter), true)
> 
> So NULL pt is expected, and does not 'break' here, since 'true' is always
> the value that decides whether to terminate the loop. So this patch would
> lead to NULL ptr deref, AFAICT.

Hello Matt,
I understand it now. I was misreading the true as part of the function argument.
Could you please also comment if the implementation of gen6_ppgtt_free_pd() in
the same file is safe? It doesn't appear to have an check on pt validity here.

Thank you,
deepak.

> 
> 
> 
> > >               continue;
> > >           free_pt(&ppgtt->base.vm, pt);