"David S. Miller" wrote:
>
> ..
> kernel/timer.c's main data structures desperately want to be per-cpu
> or allocated at boot time also. It, as has been noted often on this
> list, is actually more bloat than the ipv4 statistics stuff. :-)
I have a bunch of patches under test which do that.
The tricky bit is that at present, the per-cpu data for all 32
CPUs is always allocated. We had to change that to only allocate
the secondary CPU's memory when the CPU is coming up.
This means that:
for (i = 0; i < NR_CPUS; i++)
play_with(per_cpu(something, i));
becomes a bug. Because the per-cpu memory for not-possible
CPUs is not allocated. I created a `for_each_possible_cpu(i)'
helper macro for that.
Another issue:
/* this is basically for_each_possible_cpu() */
for (i = 0; i < NR_CPUS; i++) {
if (cpu_possible(i)) {
play_with(per_cpu(something, i));
}
}
the above code will fail if it is run before smp_init(), because
smp_init() sets up cpu_callout_map() and the per-cpu memory.
We (Dipanker and I) fixed up all the callers.
The code works OK, on ia32. But we think ia64 needs more work.
That's 128k saved. There's another 128k-odd in a huge hashtable
in kernel/pid.c (even on uniprocessor) which needs a diet. Also
80k or so in runqueues. The runqueues are set up before smp_init(),
but I expect a cpu notifier and some reorg there will work. Haven't
done sched.c yet.
I suspect we waste more memory than this in dynamically allocated
hashtables and mempools though.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Oct 23 2002 - 22:00:49 EST