Messages in this thread | | | Date | Tue, 12 Jan 2010 12:57:15 -0800 (PST) | From | Linus Torvalds <> | Subject | Re: [RFC PATCH 1/5] use nr_cpus= to set nr_cpu_ids early |
| |
On Tue, 12 Jan 2010, Christoph Lameter wrote: > > Trouble is that NR_CPUS is used all over the place. If nr_cpu_ids < > NR_CPUS at boot then there is a danger of for loops to NR_CPUS going out > of bounds. > > CONFIG_NR_CPUS is used for various bitmaps so that seems to be okay. > > drivers/acpi/numa.c:acpi_numa_init probably should use nr_cpu_ids > instead now. > > There is an octeon driver in staging that has some issues with NR_CPUS > as well.
I suspect that this is the real reason for the current behavior of 'maxcpus=', and that if all of those issues get fixed we could probably make maxcpus do what Yinghai's new 'nr_cpus=' does.
So in a perfect world, CONFIG_NR_CPUS's would never really be used, except for some fundamental static allocations/limits that are too painful to try to make dynamic.
I doubt that anybody really _cares_ about the "you can add them later" behavior of the current maxcpus thing, and I suspect that the nr_cpus semantics is what people generally would have expected.
Linus
| |