tree 3787e4408a969105a570e6044b9c45d268441665
parent b1c32770858844766af8fee1d32a6c92f57a2041
author Barret Rhoden <brho@cs.berkeley.edu> 1568648311 -0400
committer Barret Rhoden <brho@cs.berkeley.edu> 1570569071 -0400

arena: fix __arena_add() quantum alignment issue

__arena_add() was asserting that the span we were adding was quantum
'aligned.'  However, our source gives us allocations according to *its*
quantum, not our quantum.  If our source was e.g. bytes and we give out
pages, then our source could give us a span that does not match our
quantum.

This is fine, albeit a source of fragmentation.  We just make sure our
arena's bt tracks the object we want (on a quantum boundary and a
multiple of quantum).  span_bt will track whatever we actually got from
our source.

Another note: ALIGNED() checks on quantum were currently wrong - there's
nothing in the arena code that forced quantum to be a power of two.  At
least, I didn't see it, and don't want to restrict us to that yet.
Though we'll likely have issues with align, non-aligned quantums, and
qcaches.  For instance, if you ask for a quantum of 3, with an align of
64, the qcaches were told to do an align of 3 (quantum).

Signed-off-by: Barret Rhoden <brho@cs.berkeley.edu>