This seems to work:

use strict;
use warnings;

my $start = 'start:';
my $end = 'stop:';

$_ = join '', <DATA>;
@_ = m/\Q$start\E(.*?)\Q$end\E/gs;
print $_[-1];

__DATA__
something start: some text stop: other text here start: again somethin
+g stop: may be some text here start: and finally stop: and maybe some
+thing here
[download]

This might not be terribly efficient however, if there happen to be a lot of matches. It might be better to start with a rindex:

pos = rindex($_, $start);
m/\Q$start\E(.*?)\Q$end\E/gs;
print $1;
[download]