and
: Personally, I think we should just forget about the abomination unless
: somebody comes up with something that makes it really simple and doesn't
: impact the rest of the memory management code negatively. Almost nobody
: really uses shared anonymous mappings - and I think the people who do are
: seriously misguided.
Linus, you try to cheat yourself. First of all, Linux already has
anonymous shared memory (sysv shm despite of broken interface)
and its implementation is so pathetic, that it is impossible
to make it uglier :-)
Linux need good shared memory in any case. Well, the fact that nobody
wants to work seriously on linux VM is not a good reason
to blame on the concept :-) :-)
Actually, the simplest solution is to attach all shared memory areas
to common fake inodes (keeping backpointers to these areas), and as the first
step to swap them separately by dumb sysv shm algorithm.
It will not break anything and, as by-product, will implement posix shm.
Generally, I believe, that vm areas should refer not to inodes,
but to "paging/swapping policy entries", so that f.e. I could allocate
for a numeric task 265Mb array shared by two different processes
(one is for calculations, another is for services sort of long disk/network
communications) and declare that it must be swapped only by contiguous
2Mb chunks etc. etc. Making this thing once, we will not have any
silly limitations imposed by current implemenation.
I see no obstacles for implementing it, and see no bad effects
on normal paging.
BTW (not directly related topic) year ago when I blamed on Linux
helplessness for scientific tasks someone promised to repair it.
Well, nothing changed. What did occur?
Alexey Kuznetsov