include/asm-x86/alternative.h: checkpatch cleanups - formatting only

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

include/asm-x86/acpi.h: checkpatch cleanups - formatting only

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

undo "x86: fix breakage of vSMP irq operations"

revert:

  "x86: fix breakage of vSMP irq operations"

the irqflags.h unification will solve this in a cleaner way.

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

x86: fix memtest print out

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: memtest bootparam

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: fix smpboot integration

>  yhlu@mpk:~/xx/xx/kernel/x86/linux-2.6> git-bisect bad
>  d1c707188ad646c8094cac9afb1738e7d0196ff2 is first bad commit
>  commit d1c707188ad646c8094cac9afb1738e7d0196ff2
>  Author: Glauber de Oliveira Costa <gcosta@redhat.com>
>  Date:   Wed Mar 19 14:25:53 2008 -0300
>
>     x86: include mach_apic.h in smpboot_64.c and smpboot.c
>
>     After the inclusion, a lot of files needs fixing for conflicts,
>     some of them in the headers themselves, to accomodate for both
>     i386 and x86_64 versions.
>
>     [ mingo@elte.hu: build fix ]
>
>     Signed-off-by: Glauber Costa <gcosta@redhat.com>
>     Signed-off-by: Ingo Molnar <mingo@elte.hu>
>
>  :040000 040000 19f574e64bb8003bbe984f3a8c1315db969dfdcd
>  6ffe96588c77bc936705599fa110107856201115 M      arch
>  :040000 040000 61269347ad4f384ed85cc87c4f2d004ed94492ac
>  8f5c713da25579a3cdf63db3d4c2f795261d0521 M      include
>  yhlu@mpk:~/xx/xx/kernel/x86/linux-2.6>
>

attached patch fixes that.

x86: relocate_kernel - use predefined PAGE_SIZE instead of own alias

This patch does clean up relocate_kernel_(32|64).S a bit by getting rid
of local PAGE_ALIGNED macro. We should use well-known PAGE_SIZE instead

Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: use same index for processor maps

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: allow NODES_SHIFT to be a config option on x86_64

Allow the maximum number of nodes in an x86_64 system to
be configurable. This patch does NOT change the default value
but allows the value to be a config option.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: increase max physical memory size of 64-bit

Increase the maximum physical address size of x86_64 system
to 44-bits. This is in preparation for future chips that
support larger physical memory sizes.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: bitops asm constraint fixes

This (simplified) piece of code didn't behave as expected due to
incorrect constraints in some of the bitops functions, when
X86_FEATURE_xxx is referring to other than the first long:

int test(struct cpuinfo_x86 *c) {
if (cpu_has(c, X86_FEATURE_xxx))
clear_cpu_cap(c, X86_FEATURE_xxx);
return cpu_has(c, X86_FEATURE_xxx);
}

I'd really like understand, though, what the policy of (not) having a
"memory" clobber in these operations is - currently, this appears to
be totally inconsistent. Also, many comments of the non-atomic
functions say those may also be re-ordered - this contradicts the use
of "asm volatile" in there, which again I'd like to understand.

As much as all of these, using 'int' for the 'nr' parameter and
'void *' for the 'addr' one is in conflict with
Documentation/atomic_ops.txt, especially because bt{,c,r,s} indeed
take the bit index as signed (which hence would really need special
precaution) and access the full 32 bits (if 'unsigned long' was used
properly here, 64 bits for x86-64) pointed at, so invalid uses like
referencing a 'char' array cannot currently be caught.

Finally, the code with and without this patch relies heavily on the
-fno-strict-aliasing compiler switch and I'm not certain this really
is a good idea.

In the light of all of this I'm sending this as RFC, as fixing the
above might warrant a much bigger patch...

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: fix ioapic bug again

un-revert:

 commit 4960c9df1407c2723459c60ff13e6efe0c209c62
 Author: Thomas Gleixner <tglx@linutronix.de>
 Date:   Tue Jan 22 10:23:01 2008 +0100

     Revert "x86: fix NMI watchdog & 'stopped time' problem"

         This reverts commit d4d25deca49ec2527a634557bf5a6cf449f85deb.

needs a proper fix though ...

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

i386: arch/x86/math-emu/reg_ld_str.c: fix warning

arch/x86/math-emu/reg_ld_str.c:380: warning: 'l[0]' may be used uninitialized in this function
arch/x86/math-emu/reg_ld_str.c:380: warning: 'l[1]' may be used uninitialized in this function

I can't actually spot the bug here.  There's one obvious place, but fixing
that didn't shut the warning up.

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

i386: arch/x86/math-emu/fpu_entry.c warning fix

arch/x86/math-emu/fpu_entry.c:555: warning: 'entry_sel_off.empty' is used uninitialized in this function

Presumably it's harmless, but I'll sleep better at night knowing that we
initialised it.

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT bug fix for attribute type check after reserve_memtype, debug

Make the PAT related printks in ioremap pr_debug.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT bug fix for attribute type check after reserve_memtype

Bug fixes for reserve_memtype() call in __ioremap and pci_mmap_page_range().
If reserve_memtype returns non-zero, then it is an error and subsequent free is
not required. Requested and returned prot value check should be done when
reserve_memtype returns success.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: pat cpu feature bit setting for known cpus

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: tom2 warning fix

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

x86: enable PAT for amd k8 and fam10h

make known_pat_cpu to think amd k8 and fam10h is ok too.

also make tom2 below to be WRBACK

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT fix

build fix for !CONFIG_MTRR.

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

x86: PAT infrastructure patch, documentation updates

Fix double help section in PAT Kconfig. Thanks to Randy Dunlap for catching
this bug.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: add PAT related debug prints

Adds debug prints at critical code. Adds enough info in dmesg to allow us to
do effective first round of analysis of any issues that may result due to PAT
patch series.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT add ioremap_wc() interface

Introduce ioremap_wc for wc remap.

(generic wrapper is in a later patch)

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT add set_memory_wc() interface

Add a set_memory_wc interface(), similar to set_memory_uc interface.
Callers has to call set_memory_uc, set_memory_wb and
set_memory_wc, set_memory_wb as pairs.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT use reserve free memtype in pci_mmap_page_range

Add reserve_memtype and free_memtype wrapper for pci_mmap_page_range. Free
is called on unmap, but identity map continues to be mapped as per
pci_mmap_page_range request, until next request for the same region calls
ioremap_change_attr(), which will go through without conflict. This way of
mapping is identical to one used in ioremap/iounmap.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT use reserve free memtype in set_memory_uc

Use reserve_memtype and free_memtype interfaces in set_memory_uc/set_memory_wb
interfaces to avoid aliasing.
Usage model of set_memory_uc and set_memory_wb is for RAM memory and users
will first call set_memory_uc and call set_memory_wb after use to reset the
attribute.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT use reserve free memtype in ioremap and iounmap

Use reserve_memtype and free_memtype interfaces in ioremap/iounmap to avoid
aliasing.

If there is an existing alias for the region, inherit the memory type from
the alias. If there are conflicting aliases for the entire region, then fail
ioremap.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT make ioremap_change_attr non-static

Make ioremap_change_attr() non-static and use prot_val in place of ioremap_mode.
This interface is used in subsequent PAT patches.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: revert ucminus change

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

x86: PAT infrastructure patch

Sets up pat_init() infrastructure.

PAT MSR has following setting.
PAT
|PCD
||PWT
|||
000 WB _PAGE_CACHE_WB
001 WC _PAGE_CACHE_WC
010 UC- _PAGE_CACHE_UC_MINUS
011 UC _PAGE_CACHE_UC

We are effectively changing WT from boot time setting to WC.
UC_MINUS is used to provide backward compatibility to existing /dev/mem
users(X).

reserve_memtype and free_memtype are new interfaces for maintaining alias-free
mapping. It is currently implemented in a simple way with a linked list and
not optimized. reserve and free tracks the effective memory type, as a result
of PAT and MTRR setting rather than what is actually requested in PAT.

pat_init piggy backs on mtrr_init as the rules for setting both pat and mtrr
are same.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: PAT documentation

Documentation about PAT related interfaces, intended usage and memory attribute
relationship.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: clean up aperture_64.c

Initializing to zero is generally bad idea, I hope it is right for
__init data, too.

Signed-off-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: allocate e820 resource struct all together

don't need to allocate that one by one

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: early memtest to find bad ram

do simple memtest after init_memory_mapping

use find_e820_area_size to find all ram range that is not reserved.

and do some simple bits test to find some bad ram.

if find some bad ram, use reserve_early to exclude that range.

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: processor.h - use PAGE_SIZE instead of numeric value

This patch replaces numeric constant with an appropriate macro

Also 0x800000000000UL is changed to bit shifting which is complement
to the code comment (thanks hpa for notice)

Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: Explicitly include required header files.

After an experimental cleanup of <linux/percpu.h>, these files were
exposed as invoking kmalloc() without including <linux/slab.h>.

Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: simplify sync_test_bit()

There really is no need for a redundant implementation here, just keep
the alternative name for allowing consumers to use consistent naming.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86, kprobes: correct post-eip value in post_hander()

I was trying to get the address of instruction to be executed
next after the kprobed instruction.  But regs->eip in post_handler()
contains value which is useless to the user. It's pre-corrected value.
This value is difficult to use without access to resume_execution(), which
is not exported anyway.
I moved the invocation of post_handler() to *after* resume_execution().
Now regs->eip contains meaningful value in post_handler().

I do not think this change breaks any backward-compatibility.
To make meaning of the old value, post_handler() would need access to
resume_execution() which is not exported.  I have difficulty to believe
that previous, uncorrected, regs->eip can be meaningfully used in
post_handler().

Signed-off-by: Yakov Lerner <iler.ml@gmail.com>
Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Acked-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: handle_vm86_trap cleanup

Use force_sig in handle_vm86_trap like other machine traps do.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: sys32_execve PT_DTRACE

The PT_DTRACE flag is meaningless and obsolete.
Don't touch it.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: ia32 ptrace vs -ENOSYS sysenter/syscall

The previous "x86_64 ia32 ptrace vs -ENOSYS" fix only covered
the int $0x80 system call entries.  This does the same fix
for the sysenter and syscall instruction paths.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: ptrace vs -ENOSYS

When we're stopped at syscall entry tracing, ptrace can change the %rax
value from -ENOSYS to something else.  If no system call is actually made
because the syscall number (now in orig_rax) is bad, then we now always
reset %rax to -ENOSYS again.

This changes it to leave the return value alone after entry tracing.
That way, the %rax value set by ptrace is there to be seen in user mode
(or in syscall exit tracing).  This is consistent with what the 32-bit
kernel does.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: ia32 ptrace vs -ENOSYS

When we're stopped at syscall entry tracing, ptrace can change the %eax
value from -ENOSYS to something else.  If no system call is actually made
because the syscall number (now in orig_eax) is bad, then the %eax value
set by ptrace should be returned to the user.  But, instead it gets reset
to -ENOSYS again.  This is a regression from the native 32-bit kernel.

This change fixes it by leaving the return value alone after entry tracing.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: remove the write-only timer_uses_ioapic_pin_0

This patch removes the write-only timer_uses_ioapic_pin_0
(gsi can't be <= 15 in the line of it's fake usage in mpparse_32.c).

Spotted by the GNU C compiler.

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: vsmp fix x86 vsmp fix is vsmp box cleanup

code got a bit smaller:

arch/x86/kernel/vsmp_64.o:

   text    data     bss     dec     hex filename
    205       4       0     209      d1 vsmp_64.o.before
    181       4       0     185      b9 vsmp_64.o.after

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

x86: apic_is_clustered_box to indicate unsynched TSC's on multiboard vSMP systems

Indicate TSCs are unreliable as time sources if the platform is
a multi chassi ScaleMP vSMPowered machine.

Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: vSMP: use pvops only if platform has the capability to support it

Re-arrange set_vsmp_pv_ops so that pv_ops are set only if
the platform has capability to support paravirtualized irq ops

Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: fix build breakage when PCI is define and PARAVIRT is not

- Fix the the build breakage when PARAVIRT is defined
  but PCI is not
  This fixes problem reported at:
http://marc.info/?l=linux-kernel&m=120525966600698&w=2
- Make is_vsmp_box() available even when PARAVIRT is not defined.
  This is needed to determine if tsc's are reliable as a time source
  even when PARAVIRT is not defined.
- split vsmp_init to use is_vsmp_box() and set_vsmp_pv_ops()
  set_vsmp_pv_ops will do nothing if PCI is not enabled in the config.

Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: vSMP: Fix is_vsmp_box()

is_vsmp_box() currently does not work on vSMPowered systems,  as pci cfg
space is not read correctly -- This patch fixes it.

Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: make struct mpc_config_translation NUMAQ-only

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: remove mpc_oem_bus_info()

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: remove mpc_oem_pci_bus()

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: remove mpc_apic_id()

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: rename gsi_start to gsi_base to match mpparse_32.c

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: limit scan to 1k of EBDA.

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: use get_bios_ebda in mpparse_64.c

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: add uniq_ioapic_id to mpparse_32.c

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: add bad_ioapic to mpparse_32.c

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: lindent mpparse_64.c

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: remove smpboot_32.c and smpboot_64.c

Remove the last leftovers from the files. Move the ones
that are still used to the files they belong, the others
that grep can't reach, simply throw away.

Merge comments ontop of file and that's it: smpboot integrated

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: move apicid mappings to smpboot.c

They are i386 specific (the x86_64 definitions live
elsewhere, and should remain there), so are enclosed around
an ifdef

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: merge cpu_exit_clear

this is the last remaining function in smpboot_32.c
Since it is i386 specific, move it around an ifdef to
smpboot.c

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: merge native_smp_prepare_cpus

With the previous changes, code for native_smp_prepare_cpus()
in i386 and x86_64 now look very similar. merge them into
smpboot.c. Minor differences are inside ifdef

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: introduce smpboot_clear_io_apic

x86_64 has two nr_ioapics = 0 statements. In 32-bit, it can be done
too. We do it through the smpboot_clear_io_apic() inline function,
to cope with subarchitectures (visws) that does not compile mpparse in

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: change x86_64 sanity checks to match i386.

They are mostly inocuous. APIC_INTEGRATED will expand to 1,
check_phys_apicid_present is checking for the same thing it was before,
etc. But the code is identical to i386 now, and will allow us to
integrate it.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: add extra sanity check

This test exists in x86_64 and also applies to i386. So we add it

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: change x86_64 native_smp_prepare_cpus to match i386

An APIC test is moved, and code is replaced by the mach-default
already defined function (smpboot_setup_io_apic).
setup_portio_remap() is added, but it is a nop in mach-default.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: assign nr_ioapics = 0 in smpboot_hooks.h

change smpboot_setup_io_apic() by to match x86_64 behaviour

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: additions to i386 native_smp_prepare_cpus.

Add function calls to native_smp_prepare_cpus in i386
to match x86_64

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: get rid of smp_boot_cpus

This patch get rid of smp_boot_cpus(), since it does not
boot any cpu anymore. Its code is split in a way to make
it closer to x86_64

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: use physical id when disabling smp

if smp configuration is not found at all, hook into 0.
This is done to match x86_64

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: merge native_smp_cpus_done

They look similar enough, and are merged. Only difference
(zap_low_mapping for i386) is inside ifdef

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: merge smp_prepare_boot_cpu

it is practically the same between arches now, so it is
moved to smpboot.c. Minor differences (gdt initialization)
live inside an ifdef

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: integrate start_secondary

It now looks the same between architectures, so we
merge it in smpboot.c. Minor differences goes inside
an ifdef

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: integrate do_boot_cpu

This is a very large patch, because it depends on a lot
of auxiliary static functions. But they all have been modified
to the point that they're sufficiently close now. So they're just
merged in smpboot.c

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: change boot_cpu_id to boot_cpu_physical_apicid

This is to match i386. The former name was cuter,
but the current is more meaningful and more general,
since cpu_id can be a logical id.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: move stack_start to smp.h

voyager would conflict with it, but the types are ultimately
compatible. So remove the extern definition from voyager_smp.c
in favour of the common one

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: move {un}map_cpu_to_logical_apicid to smpboot.c

Move map_cpu_to_logical_apicid() and unmap_cpu_to_logical_apicid()
to smpboot.c. They take together all the bunch of static functions
they rely upon

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: add callin tests to cpu_up

Now that we boot cpus here, callin_map has this meaning (same
as x86_64)

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: change wakeup_secondary name

wakeup_secondary_via_INIT => wakeup_secondary_cpu.

This is to match i386, where init is not always used.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: include mach_apic.h in smpboot_64.c and smpboot.c

After the inclusion, a lot of files needs fixing for conflicts,
some of them in the headers themselves, to accomodate for both
i386 and x86_64 versions.

[ mingo@elte.hu: build fix ]

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: call do_boot_cpu directly from native_cpu_up

We don't need __smp_prepare_cpu anymore.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: minor adjustments for do_boot_cpu

This patch provides minor adjustments for do_boot_cpus
in both architectures to allow for integration

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: calibrate delay with irqs enabled

We do it to make it close to x86_64. The later needs it,
otherwise the nmi watchdog can get into the scene and kill us
with a hammer.

Enabling irqs here used to trigger a bug in i386. This is because
time irq handling relies upon structures that are only initialized
after smp initcalls (More precisely, it will find
per_cpu(hrtimer_bases, cpu)->cb_pending list not initialized and crash)

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: provide an end_local_APIC_setup function

It splits setup_local_APIC in two, providing a function corresponding
to the ending part of it. As a side effect, smp_callin looks the same
between i386 and x86_64.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: wrap esr setting up in i386 in lapic_setup_esr

it is a little bit more complicated than x86_64 due to erratas and
other stuff, but its existance will ease integration

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: modify smp_callin in x86_64 to look like i386

We introduce empty macros just to make them look like the same

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: don't span a new worker in __smp_prepare_cpu

We can do it now that do_boot_cpu has its own worker.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: use create_idle struct in do_boot_cpu

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: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: get rid of commenced mask.

As we now boot cpus from cpu_up, we don't need it.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: boot cpus from cpu_up, instead of prepare_cpus

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: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: do not zap_low_mappings in __smp_prepare_cpus

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: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: schedule work only if keventd is already running

Only call schedule_work if keventd is already running.
This is already the way x86_64 does

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: don't initialize sibling and core maps during preparation

it is redundant, since it is already done by set_cpu_sibling_map()

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: call nmi_watchdog_default in i386

this does not exist, so it will be an empty macro

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: call check_nmi_watchdog explicitly in native_smp_cpus_done

With this, remove its late_initcall marker from nmi_32.c

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: unify nmi_32.h and nmi_64.h

Two more files goes away. nmi_64.h and nmi_32.h gives birth
to nmi.h

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: wipe get_nmi_reason out of nmi_64.h

use mach_traps when it is supposed to be used.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: don't set maps in native_smp_prepare_boot_cpu()

By this time, they are already set in init routines

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: move smp_intr_init away from smpboot_32.c

We move it to apic_32.c, since it's irq related anyway,
and only called from that file.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>