Andi Kleen <[email protected]> wrote:
> additional_cpus is used to tell Linux that there are more (or less)
> CPUs hotpluggable than the BIOS declares at boot time. This is needed
> because the way the BIOS is declaring it is a Linux extension, not standard.
> But even if it was standard it would be good policy to be able
> to override it because as we all know BIOS can be wrong, as in
> declare far too many.
I don't have a machine with pluggable CPUs, but I'd imagine if you'd take out
some CPUs, the number of additional CPUs you can plug in will increase by the
same number (forcing me to change the kernel command lince if I do), while the
number of slots (and the number of possible CPUs) stays the same unless
somebody offers a new kind of CPU card. Therefore, I'd expect "maxcpus=" to be
the only interface I want for this purpose. Put in a value bigger than the
amount plugged in -> voila.
OTOH, looking at Thomas' patch, I'd guess it would not work as expected ...
and looking at the code seems to confirm this. Besides that, I'd possibly
want a way to limit the number of online CPUs at boot saying something like
"onlinecpus=", which would not limit the number of CPUs I can plug in.
On Mon, Oct 06, 2008 at 12:59:18PM +0200, Bodo Eggert wrote:
> I don't have a machine with pluggable CPUs, but I'd imagine if you'd take out
> some CPUs, the number of additional CPUs you can plug in will increase by the
> same number (forcing me to change the kernel command lince if I do), while the
It only costs you some memory to have more hotpluggable CPUs (on my
kernel here around 40k/possible CPU), so there's very little "force" in
practice unless you set excessive values.
Some other architectures invented a "possible_cpus=..." parameter to address
this. I presume this could be done on x86 too, but it doesn't seem
like a very pressing issue. Even if it was done it would be better
to keep additional_cpus for compatibility.
> number of slots (and the number of possible CPUs) stays the same unless
> somebody offers a new kind of CPU card. Therefore, I'd expect "maxcpus=" to be
> the only interface I want for this purpose. Put in a value bigger than the
> amount plugged in -> voila.
maxcpus=... doesn't affect the cpu_possible_mask. It really is only
good for limiting CPUs.
> OTOH, looking at Thomas' patch, I'd guess it would not work as expected ...
> and looking at the code seems to confirm this. Besides that, I'd possibly
> want a way to limit the number of online CPUs at boot saying something like
> "onlinecpus=", which would not limit the number of CPUs I can plug in.
That is exactly what maxcpus=... does.
Anyways I don't see the additional_cpus=... removal patch in tip
so hopefully things are fine now
-Andi
--
[email protected]