well, one day, yes...
it doesn't take much, really
as a solution to the how-can-i-use-discontiguous-memory thread, too,
one day we should have in-memory images implicitly resident on the 'swapfile'
/dev/mem (with a swap_priority of 0, as distinct from the current -1, -2, ...)
then we could implement the above proposal with swapoff-like semantics,
removing an address range.
what happens to the procs/kernel_structures/etc living there?
same as what happens when a swap-space develops errors!
if you'rre paranoid enuff, you use mem as a RAID device.
not easy, to implement, but that's the generalisation you base it on.
but the easy part which comes with this generalisation is that one can
*add* memory --- the gap-at-16Mb and mem-bigger-than-64Mb problems
go away when you can just boot a vanilla kernel using at most 15Mb,
and then do a "swapon -priority=0 -lo=15Mb -hi=192Mb /dev/mem"
from the usual /etc/rc*.
anything with swap-prio >= 0 is real mem!
i can imagine handling that scarce resource of L2 cache as prio==1
and L1 cache as prio==2, and the goodness() function deriving its
current-proc-is-favoured policy from posession of a prio==3 resource
representing running-on-this-processor.
the using-fpu property is similar.
play with it.
i suspect linus et al already have this in mind,
that's what the -ve swap priorities immediately suggested to me....
Peter Swain ^..^ +61 2 9698 2322 (office)
swine@softway.com.au (00) +61 419 431 088 (mobile)
Evolutionary software, revolutionary results +61 2 9519 0171 (home)
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu