Re: [PATCH v2 2/4] mm: kmemleak: add rbtree for objects allocated with physical address

[Patrick Wang]

On 2022/6/6 22:38, Catalin Marinas wrote:
> On Fri, Jun 03, 2022 at 11:54:13AM +0800, Patrick Wang wrote:
> > @@ -536,27 +543,32 @@ static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
> >   }
> >   
> >   /*
> > - * Remove an object from the object_tree_root and object_list. Must be called
> > - * with the kmemleak_lock held _if_ kmemleak is still enabled.
> > + * Remove an object from the object_tree_root (or object_phys_tree_root)
> > + * and object_list. Must be called with the kmemleak_lock held _if_ kmemleak
> > + * is still enabled.
> >    */
> >   static void __remove_object(struct kmemleak_object *object)
> >   {
> > -	rb_erase(&object->rb_node, &object_tree_root);
> > +	rb_erase(&object->rb_node, object->flags & OBJECT_PHYS ?
> > +				   &object_phys_tree_root :
> > +				   &object_tree_root);
>
> This pattern appears in a few place, I guess it's better with a macro,
> say get_object_tree_root(object). But see how many are left, I have some
> comments below on reducing the diff.

Will do.

> > @@ -709,12 +724,12 @@ static void delete_object_full(unsigned long ptr)
> >    * delete it. If the memory block is partially freed, the function may create
> >    * additional metadata for the remaining parts of the block.
> >    */
> > -static void delete_object_part(unsigned long ptr, size_t size)
> > +static void delete_object_part(unsigned long ptr, size_t size, bool is_phys)
> >   {
> >   	struct kmemleak_object *object;
> >   	unsigned long start, end;
> >   
> > -	object = find_and_remove_object(ptr, 1);
> > +	object = find_and_remove_object(ptr, 1, is_phys);
> >   	if (!object) {
> >   #ifdef DEBUG
> >   		kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
>
> The previous patch introduced a check on object->flags for
> delete_object_part(). I think you can just use is_phys directly now when
> calling create_object().

Will do.

> > @@ -1275,7 +1290,7 @@ static void scan_block(void *_start, void *_end,
> >   		 * is still present in object_tree_root and object_list
> >   		 * (with updates protected by kmemleak_lock).
> >   		 */
> > -		object = lookup_object(pointer, 1);
> > +		object = lookup_object(pointer, 1, false);
> >   		if (!object)
> >   			continue;
> >   		if (object == scanned)
> > @@ -1299,7 +1314,7 @@ static void scan_block(void *_start, void *_end,
> >   		raw_spin_unlock(&object->lock);
> >   
> >   		if (excess_ref) {
> > -			object = lookup_object(excess_ref, 0);
> > +			object = lookup_object(excess_ref, 0, false);
> >   			if (!object)
> >   				continue;
> >   			if (object == scanned)
> > @@ -1728,7 +1743,7 @@ static int dump_str_object_info(const char *str)
> >   
> >   	if (kstrtoul(str, 0, &addr))
> >   		return -EINVAL;
> > -	object = find_and_get_object(addr, 0);
> > +	object = find_and_get_object(addr, 0, false);
> >   	if (!object) {
> >   		pr_info("Unknown object at 0x%08lx\n", addr);
> >   		return -EINVAL;
>
> I think find_and_get_object() is never called on a phys object, so you
> can probably simplify these a bit. Just add an is_phys argument where
> strictly necessary and maybe even add a separate function like
> lookup_object_phys() to reduce the other changes.

Will add lookup_object_phys() function and find_and_get_object_phys()
function. The find_and_get_object() function is also called in many
places.

Thanks,
Patrick