Re: [PATCH drm-next v6 02/13] drm: manager to keep track of GPUs VA mappings

[Danilo Krummrich]

On 7/7/23 13:00, Boris Brezillon wrote:
> On Fri, 30 Jun 2023 00:25:18 +0200
> Danilo Krummrich <dakr@xxxxxxxxxx> wrote:
>
> > +/**
> > + * drm_gpuva_for_each_va_range - iternator to walk over a range of &drm_gpuvas
> > + * @va__: &drm_gpuva structure to assign to in each iteration step
> > + * @mgr__: &drm_gpuva_manager to walk over
> > + * @start__: starting offset, the first gpuva will overlap this
> > + * @end__: ending offset, the last gpuva will start before this (but may
> > + * overlap)
> > + *
> > + * This iterator walks over all &drm_gpuvas in the &drm_gpuva_manager that lie
> > + * between @start__ and @end__. It is implemented similarly to list_for_each(),
> > + * but is using the &drm_gpuva_manager's internal interval tree to accelerate
> > + * the search for the starting &drm_gpuva, and hence isn't safe against removal
> > + * of elements. It assumes that @end__ is within (or is the upper limit of) the
> > + * &drm_gpuva_manager. This iterator does not skip over the &drm_gpuva_manager's
> > + * @kernel_alloc_node.
> > + */
> > +#define drm_gpuva_for_each_va_range(va__, mgr__, start__, end__) \
> > +	for (va__ = drm_gpuva_find_first((mgr__), (start__), (end__)); \
>
> drm_gpuva_find_first() takes the range size as its last argument, not
> the range end:
>
> 	for (va__ = drm_gpuva_find_first((mgr__), (start__), (end__) - (start__)); \

Good catch! Originally this was

drm_gpuva_it_iter_first(&(mgr)->rb.tree, (start__), (end__) - 1)

but then I changed it since I did not want to expose the interval tree 
functions directly.

> > +	     va__ && (va__->va.addr < (end__)) && \
> > +	     !list_entry_is_head(va__, &(mgr__)->rb.list, rb.entry); \
> > +	     va__ = list_next_entry(va__, rb.entry))
>
> If you define:
>
> static inline struct drm_gpuva *
> drm_gpuva_next(struct drm_gpuva *va)
> {
> 	if (va && !list_is_last(&va->rb.entry, &va->mgr->rb.list))
> 		return list_next_entry(va, rb.entry);
>
> 	return NULL;
> } >
> the for loop becomes a bit more readable:

Yes, it would. However, I don't want it to be confused with 
drm_gpuva_find_next(). Maybe I should rename the latter to something 
like drm_gpuva_find_next_neighbor() then.

> 	for (va__ = drm_gpuva_find_first((mgr__), (start__), (end__) - (start__)); \
> 	     va__ && (va__->va.addr < (end__)); \
> 	     va__ = drm_gpuva_next(va__))
>
> > +
> > +/**
> > + * drm_gpuva_for_each_va_range_safe - iternator to safely walk over a range of
> > + * &drm_gpuvas
> > + * @va__: &drm_gpuva to assign to in each iteration step
> > + * @next__: another &drm_gpuva to use as temporary storage
> > + * @mgr__: &drm_gpuva_manager to walk over
> > + * @start__: starting offset, the first gpuva will overlap this
> > + * @end__: ending offset, the last gpuva will start before this (but may
> > + * overlap)
> > + *
> > + * This iterator walks over all &drm_gpuvas in the &drm_gpuva_manager that lie
> > + * between @start__ and @end__. It is implemented similarly to
> > + * list_for_each_safe(), but is using the &drm_gpuva_manager's internal interval
> > + * tree to accelerate the search for the starting &drm_gpuva, and hence is safe
> > + * against removal of elements. It assumes that @end__ is within (or is the
> > + * upper limit of) the &drm_gpuva_manager. This iterator does not skip over the
> > + * &drm_gpuva_manager's @kernel_alloc_node.
> > + */
> > +#define drm_gpuva_for_each_va_range_safe(va__, next__, mgr__, start__, end__) \
> > +	for (va__ = drm_gpuva_find_first((mgr__), (start__), (end__)), \
> > +	     next__ = va ? list_next_entry(va__, rb.entry) : NULL; \
> > +	     va__ && (va__->va.addr < (end__)) && \
> > +	     !list_entry_is_head(va__, &(mgr__)->rb.list, rb.entry); \
> > +	     va__ = next__, next__ = list_next_entry(va__, rb.entry))
>
> And this is the safe version using the drm_gpuva_next() helper:
>
> 	for (va__ = drm_gpuva_find_first((mgr__), (start__), (end__) - (start__)), \
> 	     next__ = drm_gpuva_next(va__); \
> 	     va__ && (va__->va.addr < (end__)); \
> 	     va__ = next__, next__ = drm_gpuva_next(va__))
>
> Those changes fixed an invalid pointer access I had in the sm_unmap()
> path.

Sorry you did run into this bug.

- Danilo