Hi, Hamza,
On Mon, Jul 12, 2021 at 10:33:38PM -0400, Hamza Mahfooz wrote:
> @@ -1467,16 +1467,20 @@ rmap_walk_init_level(struct slot_rmap_walk_iterator *iterator, int level)
>
> static void
> slot_rmap_walk_init(struct slot_rmap_walk_iterator *iterator,
> - struct kvm_memory_slot *slot, int start_level,
> + const struct kvm_memory_slot *slot, int start_level,
> int end_level, gfn_t start_gfn, gfn_t end_gfn)
> {
> - iterator->slot = slot;
> - iterator->start_level = start_level;
> - iterator->end_level = end_level;
> - iterator->start_gfn = start_gfn;
> - iterator->end_gfn = end_gfn;
> + struct slot_rmap_walk_iterator iter = {
> + .slot = slot,
> + .start_gfn = start_gfn,
> + .end_gfn = end_gfn,
> + .start_level = start_level,
> + .end_level = end_level,
> + };
> +
> + rmap_walk_init_level(&iter, iterator->start_level);
Here it should be s/iterator->//.
>
> - rmap_walk_init_level(iterator, iterator->start_level);
> + memcpy(iterator, &iter, sizeof(struct slot_rmap_walk_iterator));
> }
This patch breaks kvm/queue with above issue. Constify of kvm_memory_slot
pointer should have nothing to do with this so at least it should need a
separate patch. At the meantime I also don't understand why memcpy() here,
which seems to be even slower..
--
Peter Xu
