>which is reserved by FS instances. Do you see? If "dummy" inodes are private,
>we can easily invent a way to allocate/release them from net and save
>lots of space and dereferences sock <-> socket.
Yes. I think that's exactly what Alexander was talking about.
>[..] The only things to do is close()/poll()/ioctl().
Of course the network can't drop the lock from there without breaking the
rest ;).
>(snmp counters etc.). I will convert them to cpu local vars in one hit,
Good idea, so you'll avoid the lock and you'll have to do the sum (without
locks too) only at /proc read time.
>BTW why not to create NR_CPUS bss sections?
So also NR_CPUS .data and "maybe" each one starting cacheline aligned,
right? ;). Yes looks like a good idea and would save lots of memory as
everybody is enlarging his structures to go cacheline aligned. Then we
could remove all the ugly stuff like the aligned_data union in sched.c...
Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/