Use a new worker, with help of the create_idle struct
to fork the idle thread. We now have two workers, the first
of them triggered by __smp_prepare_cpu. But the later is
going away soon.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

As we now boot cpus from cpu_up, we don't need it.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

After all the infrastructure work, we're now prepared
to boot the cpus from cpu_up, and not from prepare_cpus.
So the difference between cold boot and hotplug is effectively
over, and the functions are used to the purposes they're meant to.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

It was okay when cpus were cold booted before this point.
But with the new state machine, they will not have arrived to
the trampoline yet. zapping low mappings will have the bad effect
of breaking it completely after paging enablement
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Only call schedule_work if keventd is already running.
This is already the way x86_64 does
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

it is redundant, since it is already done by set_cpu_sibling_map()
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

this does not exist, so it will be an empty macro
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

With this, remove its late_initcall marker from nmi_32.c
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Two more files goes away. nmi_64.h and nmi_32.h gives birth
to nmi.h
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

use mach_traps when it is supposed to be used.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

By this time, they are already set in init routines
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

We move it to apic_32.c, since it's irq related anyway,
and only called from that file.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

We do it and also fix conflicts, which makes x86_64 automatically
closer to i386
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Do it and also fix conflicts, which automatically makes
x86_64 look closer to i386
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

this patch allows x86_64 to use subarch mach_ headers
in practice, since x86_64 does not have any subarch, it
will use mach_default. But it will allow for substantially
less code duplication
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

the cpu count is changed accordingly: now, what matters is
online cpus.
Also, we add those functions for x86_64
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

It will ease integration for x86_64
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impressing friends is a very important thing.
Do it in a separate function to make it even more
explicit, and ease integration.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

weighting a map will do.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Signed-off-by: NIngo Molnar <mingo@elte.hu>

This is the way x86_64 does, and complement the already
present patch that does the bios cpu to apicid mapping here
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

We fill the per-cpu (or array) that maps
bios cpu id to apicid in mpparse_32.c, the way x86_64 does
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

We use the same routing as x86_64, moved now to setup.c.
Just with a few ifdefs inside.
Note that this routing uses prefill_possible_map().
It has the very nice side effect of allowing hotplugging of
cpus that are marked as present but disabled by acpi bios.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

missing a semicolon
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Instead of declaring them inside of X86_64 ifdef, do it
unconditionally
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

this will serve as a reference as to whether or not to
use the per_cpu variables in mpparse. Done the same way
as x86_64
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

This mapping already exists in x86_64, just provide it for
i386
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

take them off smp_{32,64}.h and move to smp.h
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Done to match x86_64
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

We have already removed the only condition that could fail here.
so just don't test for any return value
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Do tests before do_boot_cpu in native_cpu_up for i386.
Tests are a little bit broader than originally, and are the
same as x86_64. Test for smp_callin is not applicable right now
and is deferred.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Put it in a disable_smp() function, as x86_64 does
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Isolate all sanity checking in a smp_sanity_check()
function as x86_64 does.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

goal is to have i386 and x86_64 closer, so we
add barriers to match
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

This patch does not change the behaviour of x86_64, since APIC_INTEGRATED
is always defined as (1). But the code now matches exactly i386 version
(well, this part of the code, at least)
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

it is always integrated, so define as 1.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

it does not need to be inside lock. Do the way i386 does.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

This is done so we call setup_secondary_clock() in the same place x86_64
does. A separate patch for this is appearantly not needed. But clock
initialization is such a delicate thing, that it's safer to do this way
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

the call to idle is guaranteed to do it.
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

This matches x86_64 behaviour, which is a superior one IMHO
Signed-off-by: NGlauber Costa <gcosta@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>