Commit 3e0856b63fe0e375a0951e05c2ef98bb2ebd9421 updated
__sg_alloc_table_from_pages to use the same API as linux, but modified
the loop condition when going over the pages in a sg list. Part of the
change included moving the sg_next call out of the for loop and into the
body, which causes an off by one error when traversing the list. Since
sg_next is called before the loop body it will skip the first element
and read one past the last element.
This caused panics when running PRIME with nvidia-drm as the off-by-one
issue causes a NULL dereference.