Eric Wong authored Jul 11, 2016 and Junio C Hamano committed Jul 12, 2016

Using the a Linux-kernel-derived doubly-linked list
implementation from the Userspace RCU library allows us to
enqueue and delete items from the object request queue in
constant time.

This change reduces enqueue times in the prefetch() function
where object request queue could grow to several thousand
objects.

I left out the list_for_each_entry* family macros from list.h
which relied on the __typeof__ operator as we support platforms
without it.  Thus, list_entry (aka "container_of") needs to be
called explicitly inside macro-wrapped for loops.

The downside is this costs us an additional pointer per object
request, but this is offset by reduced overhead on queue
operations leading to improved performance and shorter queue
depths.

Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>