platform/kernel/linux-rpi.git
15 years agox86, voyager: move Voyager-specific defines to voyager.h
Ingo Molnar [Sat, 31 Jan 2009 01:16:04 +0000 (02:16 +0100)]
x86, voyager: move Voyager-specific defines to voyager.h

They dont belong into the generic headers.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up spurious vector sanity check
Ingo Molnar [Sat, 31 Jan 2009 01:06:50 +0000 (02:06 +0100)]
x86, apic: clean up spurious vector sanity check

Move the spurious vector sanity check to the place where it's
defined - out of a .c file.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: unify the APIC vector enumeration
Ingo Molnar [Sat, 31 Jan 2009 01:10:03 +0000 (02:10 +0100)]
x86, apic: unify the APIC vector enumeration

Most of the vector layout on 32-bit and 64-bit is identical now,
so eliminate the duplicated enumeration of the vectors.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: update copyrights
Ingo Molnar [Sat, 31 Jan 2009 01:03:42 +0000 (02:03 +0100)]
x86: update copyrights

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up header section
Ingo Molnar [Sat, 31 Jan 2009 00:59:14 +0000 (01:59 +0100)]
x86, apic: clean up header section

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, irq: add LOCAL_PERF_VECTOR
Ingo Molnar [Sat, 31 Jan 2009 01:23:27 +0000 (02:23 +0100)]
x86, irq: add LOCAL_PERF_VECTOR

Add a slot for the performance monitoring interrupt. Not yet used
by any subsystem - but the hardware has it. (This eases integration
with performance monitoring code.)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic, 32-bit: add self-IPI methods
Ingo Molnar [Fri, 30 Jan 2009 22:42:18 +0000 (23:42 +0100)]
x86, apic, 32-bit: add self-IPI methods

Impact: fix rare crash on 32-bit

The 32-bit APIC drivers had their send_IPI_self vectors set to NULL,
but ioapic_retrigger_irq() depends on it being always set. Fix it.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: don't treat bigsmp as non-standard
Yinghai Lu [Thu, 29 Jan 2009 22:19:22 +0000 (14:19 -0800)]
x86: don't treat bigsmp as non-standard

just like 64 bit switch from flat logical APIC messages to
flat physical mode automatically.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: seperate default_send_IPI_mask_sequence/allbutself from logical
Yinghai Lu [Fri, 30 Jan 2009 03:31:49 +0000 (19:31 -0800)]
x86: seperate default_send_IPI_mask_sequence/allbutself from logical

Impact: 32-bit should use logical version

there are two version: for default_send_IPI_mask_sequence/allbutself
one in ipi.h and one in ipi.c for 32bit

it seems .h version overwrote ipi.c for a while.

restore it so 32 bit could use its old logical version.
also remove dupicated functions in .c

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: fix compiling with 64bit with def_to_bigsmp
Yinghai Lu [Fri, 30 Jan 2009 03:30:04 +0000 (19:30 -0800)]
x86: fix compiling with 64bit with def_to_bigsmp

only need to do cut off with 32bit

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic unification: remove left over files
Yinghai Lu [Fri, 30 Jan 2009 03:29:24 +0000 (19:29 -0800)]
x86, apic unification: remove left over files

Impact: cleanup

remove unused files

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: allow more than 8 cpus to be used on 32-bit
Yinghai Lu [Thu, 29 Jan 2009 23:14:46 +0000 (15:14 -0800)]
x86: allow more than 8 cpus to be used on 32-bit

X86_PC is the only remaining 'sub' architecture, so we dont need
it anymore.

This also cleans up a few spurious references to X86_PC in the
driver space - those certainly should be X86.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: unify genapic code, unify subarchitectures, remove old subarchitecture code...
Suresh Siddha [Thu, 29 Jan 2009 20:11:08 +0000 (12:11 -0800)]
x86: unify genapic code, unify subarchitectures, remove old subarchitecture code, xapic fix

xapic fix for 32bit platform with less than 8 cpu's.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: move to the X86_32_NON_STANDARD code section
Ingo Molnar [Tue, 27 Jan 2009 17:46:23 +0000 (18:46 +0100)]
x86/Voyager: move to the X86_32_NON_STANDARD code section

Make Voyager depend on X86_32_NON_STANDARD - it is a non-standard 32-bit
SMP architecture.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: rename X86_GENERICARCH to X86_32_NON_STANDARD
Ingo Molnar [Tue, 27 Jan 2009 17:43:09 +0000 (18:43 +0100)]
x86: rename X86_GENERICARCH to X86_32_NON_STANDARD

X86_GENERICARCH is a misnomer - it contains non-PC 32-bit architectures
that are not included in the default build.

Rename it to X86_32_NON_STANDARD.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: remove the subarch menu
Ingo Molnar [Tue, 27 Jan 2009 17:31:41 +0000 (18:31 +0100)]
x86: remove the subarch menu

Remove the subarch menu and standardize on X86_PC.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: move X86_VSMP from subarch menu
Ingo Molnar [Tue, 27 Jan 2009 17:29:13 +0000 (18:29 +0100)]
x86: move X86_VSMP from subarch menu

Move X86_VSMP out of the subarch menu - this way it can be enabled
together with standard PC support as well, in the same kernel.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: move non-standard 32-bit platform Kconfig entries
Ingo Molnar [Tue, 27 Jan 2009 17:24:57 +0000 (18:24 +0100)]
x86: move non-standard 32-bit platform Kconfig entries

- make X86_GENERICARCH depend X86_NON_STANDARD

- move X86_SUMMIT, X86_ES7000 and X86_BIGSMP out of the subarchitecture
  menu and under this option

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: move VOYAGER to the NON_STANDARD_PLATFORM section
Ingo Molnar [Tue, 27 Jan 2009 17:20:09 +0000 (18:20 +0100)]
x86: move VOYAGER to the NON_STANDARD_PLATFORM section

Move X86_ELAN (old, NCR hw platform built on Intel CPUs) from the
subarchitecture menu to the non-standard-platform section.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: move ELAN to the NON_STANDARD_PLATFORM section
Ingo Molnar [Tue, 27 Jan 2009 17:18:25 +0000 (18:18 +0100)]
x86: move ELAN to the NON_STANDARD_PLATFORM section

Move X86_ELAN (old, AMD based web-boxes) from the subarchitecture
menu to the non-standard-platform section.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: cleanup, introduce CONFIG_NON_STANDARD_PLATFORMS
Ingo Molnar [Tue, 27 Jan 2009 17:11:43 +0000 (18:11 +0100)]
x86: cleanup, introduce CONFIG_NON_STANDARD_PLATFORMS

Introduce a Y/N Kconfig option for non-PC x86 platforms.

Make VisWS, RDC321 and SGI/UV depend on this.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove ISA quirk
Ingo Molnar [Tue, 27 Jan 2009 16:51:37 +0000 (17:51 +0100)]
x86/Voyager: remove ISA quirk

Voyager has this ISA quirk (because Voyager has no ISA support):

 config ISA
  bool "ISA support"
depends on !X86_VOYAGER

There's a ton of x86 hardware that does not support ISA, and because
most ISA drivers cannot auto-detect in a safe way, the convention in
the kernel has always been to not enable ISA drivers if they are not
needed.

Voyager users can do likewise - no need for a Kconfig quirk.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove power management Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:49:22 +0000 (17:49 +0100)]
x86/Voyager: remove power management Kconfig quirk

Voyager has this PM/ACPI Kconfig quirk:

 menu "Power management and ACPI options"
depends on !X86_VOYAGER

Most of the PM features are auto-detect so they should be safe to run
on just about any hardware. (If not, those instances need fixing.)

In any case, if a kernel is built for Voyager, the power management
options can be disabled.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove HOTPLUG_CPU Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:47:24 +0000 (17:47 +0100)]
x86/Voyager: remove HOTPLUG_CPU Kconfig quirk

Voyager has this Kconfig quirk:

 config HOTPLUG_CPU
  bool "Support for hot-pluggable CPUs"
depends on SMP && HOTPLUG && !X86_VOYAGER

But this exception will be moot once Voyager starts using the
generic x86 code.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove MCE quirk
Ingo Molnar [Tue, 27 Jan 2009 16:45:17 +0000 (17:45 +0100)]
x86/Voyager: remove MCE quirk

If no MCE code is desired on Voyager hw then the solution
is to turn them off in the .config - and to extend the MCE
code to not initialize on Voyager.

Remove the build-time quirk.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove APIC/IO-APIC Kbuild quirk
Ingo Molnar [Tue, 27 Jan 2009 16:40:48 +0000 (17:40 +0100)]
x86/Voyager: remove APIC/IO-APIC Kbuild quirk

The lapic/ioapic code properly auto-detects and is safe to run on CPUs that
have no local APIC. (or which have their lapic turned off in the hardware)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove PARAVIRT Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:38:46 +0000 (17:38 +0100)]
x86/Voyager: remove PARAVIRT Kconfig quirk

Remove this Kconfig quirk:

 config PARAVIRT
  bool "Enable paravirtualization code"
depends on !X86_VOYAGER
  help

Voyager support built into a kernel does not preclude paravirt support.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove KVM_GUEST quirk
Ingo Molnar [Tue, 27 Jan 2009 16:37:33 +0000 (17:37 +0100)]
x86/Voyager: remove KVM_GUEST quirk

Voyager has this quirk currently:

 config KVM_GUEST
  bool "KVM Guest support"
  select PARAVIRT
depends on !X86_VOYAGER

Voyager support built into a kernel image does not exclude
KVM paravirt guest support - so remove this quirk.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove KVM_CLOCK quirk
Ingo Molnar [Tue, 27 Jan 2009 16:36:31 +0000 (17:36 +0100)]
x86/Voyager: remove KVM_CLOCK quirk

Voyager has this build-time quirk to exclude KVM_CLOCK:

  bool "KVM paravirtualized clock"
  select PARAVIRT
  select PARAVIRT_CLOCK
depends on !X86_VOYAGER

Voyager support built into a kernel image does not exclude
KVM paravirt clock support - so remove this quirk.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove VMI Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:34:56 +0000 (17:34 +0100)]
x86/Voyager: remove VMI Kconfig quirk

x86/Voyager has this build-time quirk:

  bool "VMI Guest support"
  select PARAVIRT
  depends on X86_32
depends on !X86_VOYAGER

Since VMI is auto-detected (and Voyager will be auto-detected) there's no
reason for this quirk.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/VisWS: remove Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:33:34 +0000 (17:33 +0100)]
x86/VisWS: remove Kconfig quirk

VisWS has this quirk currently:

 config X86_VISWS
  bool "SGI 320/540 (Visual Workstation)"
depends on X86_32 && PCI && !X86_VOYAGER && X86_MPPARSE && PCI_GODIRECT

The !Voyager quirk is unnecessary.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove X86_FIND_SMP_CONFIG Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:28:08 +0000 (17:28 +0100)]
x86/Voyager: remove X86_FIND_SMP_CONFIG Kconfig quirk

x86/Voyager had this Kconfig quirk:

 config X86_FIND_SMP_CONFIG
def_bool y
depends on X86_MPPARSE || X86_VOYAGER

Which splits off the find_smp_config() callback into a build-time quirk.

Voyager should use the existing x86_quirks.mach_find_smp_config() callback
to introduce SMP-config quirks. NUMAQ-32 and VISWS already use this.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove X86_BIOS_REBOOT Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:17:55 +0000 (17:17 +0100)]
x86/Voyager: remove X86_BIOS_REBOOT Kconfig quirk

Voyager has this Kconfig quirk:

config X86_BIOS_REBOOT
bool
depends on !X86_VOYAGER
default y

Voyager should use the existing machine_ops.emergency_restart reboot
quirk mechanism instead of a build-time quirk.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove X86_HT Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:16:16 +0000 (17:16 +0100)]
x86/Voyager: remove X86_HT Kconfig quirk

Voyager has this Kconfig quirk:

depends on (X86_32 && !X86_VOYAGER) || X86_64

That is unnecessary as HT support is CPUID driven and explicitly
 enumerated.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: generalize boot_cpu_id
Ingo Molnar [Tue, 27 Jan 2009 16:13:05 +0000 (17:13 +0100)]
x86: generalize boot_cpu_id

x86/Voyager can boot on non-zero processors. While that can probably
be fixed by properly remapping the physical CPU IDs, keep boot_cpu_id
for now for easier transition - and expand it to all of x86.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: replace CONFIG_X86_SMP with CONFIG_SMP
Ingo Molnar [Tue, 27 Jan 2009 16:07:08 +0000 (17:07 +0100)]
x86: replace CONFIG_X86_SMP with CONFIG_SMP

The x86/Voyager subarch used to have this distinction between
 'x86 SMP support' and 'Voyager SMP support':

 config X86_SMP
bool
depends on SMP && ((X86_32 && !X86_VOYAGER) || X86_64)

This is a pointless distinction - Voyager can (and already does) use
smp_ops to implement various SMP quirks it has - and it can be extended
more to cover all the specialities of Voyager.

So remove this complication in the Kconfig space.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove ARCH_SUSPEND_POSSIBLE Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:04:39 +0000 (17:04 +0100)]
x86/Voyager: remove ARCH_SUSPEND_POSSIBLE Kconfig quirk

Voyager has this Kconfig quirk for suspend/resume:

 config ARCH_SUSPEND_POSSIBLE
  def_bool y
  depends on !X86_VOYAGER

The proper mechanism to not suspend on a piece of hardware to disable
CONFIG_SUSPEND. Remove the quirk.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove HIBERNATION Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:03:24 +0000 (17:03 +0100)]
x86/Voyager: remove HIBERNATION Kconfig quirk

Voyager has this hibernation quirk:

 config ARCH_HIBERNATION_POSSIBLE
  def_bool y
depends on !SMP || !X86_VOYAGER

Hibernation is a generic facility provided on all x86 platforms. If it
is buggy on Voyager then that bug should be fixed - not worked around.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove KGDB Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:02:29 +0000 (17:02 +0100)]
x86/Voyager: remove KGDB Kconfig quirk

x86/Voyager has this KGDB quirk:

select HAVE_ARCH_KGDB if !X86_VOYAGER

This is completely pointless - there's nothing in KGDB that cannot work
on Voyager. Remove it.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove KVM Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 16:01:14 +0000 (17:01 +0100)]
x86/Voyager: remove KVM Kconfig quirk

Voyager and other subarchitectures have this Kconfig quirk:

  select HAVE_KVM if ((X86_32 && !X86_VOYAGER && !X86_VISWS && !X86_NUMAQ) || X86_64)

This is unnecessary, as KVM cleanly detects based on CPUID capabilities.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove NATSEMI Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 15:59:11 +0000 (16:59 +0100)]
x86/Voyager: remove NATSEMI Kconfig quirk

x86/Voyager has this quirk for SCx200 support:

 config SCx200
  tristate "NatSemi SCx200 support"
depends on !X86_VOYAGER

Remove it - Voyager users can disable drivers they dont need.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: remove MCA Kconfig quirk
Ingo Molnar [Tue, 27 Jan 2009 15:57:49 +0000 (16:57 +0100)]
x86/Voyager: remove MCA Kconfig quirk

Remove Voyager Kconfig quirk: just like any other hardware platform
users of Voyager systems can configure in the hardware drivers they need.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: disable Voyager temporarily
Ingo Molnar [Wed, 28 Jan 2009 18:55:34 +0000 (19:55 +0100)]
x86: disable Voyager temporarily

x86/Voyager does not build right now and it's unclear whether it will
be cleaned up and ported to the subarch-less 32-bit x86 code - so disable
it for now.

If it's fixed we'll re-enable it - or remove it after some time. There's
a very low number of systems running development kernels on x86/Voyager
currently. (one or two on the whole planet)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86/Voyager: clean up BROKEN Kconfig reference
Ingo Molnar [Tue, 27 Jan 2009 16:54:17 +0000 (17:54 +0100)]
x86/Voyager: clean up BROKEN Kconfig reference

CONFIG_BROKEN has been removed from the upstream kernel years ago,
but X86_VOYAGER still had a stale reference to it - remove it.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: remove subarchitecture support
Ingo Molnar [Wed, 28 Jan 2009 18:32:55 +0000 (19:32 +0100)]
x86: remove subarchitecture support

Remove the 32-bit subarchitecture support code.

All subarchitectures but Voyager have been converted. Voyager will be
done later or will be removed.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: move mach-default/*.h files to asm/
Ingo Molnar [Wed, 28 Jan 2009 18:34:09 +0000 (19:34 +0100)]
x86: move mach-default/*.h files to asm/

We are getting rid of subarchitecture support - move the hook files
to asm/. (These are now stale and should be replaced with more explicit
runtime mechanisms - but the transition is simpler this way.)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: remove subarchitecture support code
Ingo Molnar [Wed, 28 Jan 2009 18:11:44 +0000 (19:11 +0100)]
x86: remove subarchitecture support code

Remove remaining bits of the subarchitecture code. Now that all the
special platforms are runtime probed and runtime handled, we can remove
these facilities.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: remove mach_ipi.h
Ingo Molnar [Wed, 28 Jan 2009 18:14:52 +0000 (19:14 +0100)]
x86, smp: remove mach_ipi.h

Move mach_ipi.h definitions into genapic.h.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, bigsmp: consolidate header code
Ingo Molnar [Wed, 28 Jan 2009 18:19:12 +0000 (19:19 +0100)]
x86, bigsmp: consolidate header code

Move all the asm/bigsmp/*.h definitions into bigsmp_32.c.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, bigsmp: consolidate code
Ingo Molnar [Wed, 28 Jan 2009 18:04:37 +0000 (19:04 +0100)]
x86, bigsmp: consolidate code

Move all code to arch/x86/kernel/bigsmp_32.c.

With this it ceases to rely on any build-time subarch features.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, NUMAQ: Consolidate code
Ingo Molnar [Wed, 28 Jan 2009 18:01:05 +0000 (19:01 +0100)]
x86, NUMAQ: Consolidate code

Move all NUMAQ code into arch/x86/kernel/numaq.c.

With this it ceases to rely on any build-time subarch features.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, ES7000: Consolidate code
Ingo Molnar [Wed, 28 Jan 2009 18:01:05 +0000 (19:01 +0100)]
x86, ES7000: Consolidate code

Move all ES7000 code into arch/x86/kernel/es7000_32.c.

With this it ceases to rely on any build-time subarch features.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: remove mach_apic.h, fix
Suresh Siddha [Thu, 29 Jan 2009 01:52:57 +0000 (17:52 -0800)]
x86: remove mach_apic.h, fix

Use apic_read() instead of open-coded mmio.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: remove mach_apic.h
Ingo Molnar [Wed, 28 Jan 2009 16:55:37 +0000 (17:55 +0100)]
x86: remove mach_apic.h

Spread mach_apic.h definitions into genapic.h. (with some knock-on effects
on smp.h and apic.h.)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, summit: consolidate code, fix
Ingo Molnar [Thu, 29 Jan 2009 10:29:22 +0000 (11:29 +0100)]
x86, summit: consolidate code, fix

Build fix for !NUMA Summit.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, summit: consolidate code
Ingo Molnar [Wed, 28 Jan 2009 17:49:31 +0000 (18:49 +0100)]
x86, summit: consolidate code

Consolidate all the Summit code into a single file:
arch/x86/kernel/summit_32.c.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, numaq: consolidate code
Ingo Molnar [Wed, 28 Jan 2009 17:47:24 +0000 (18:47 +0100)]
x86, numaq: consolidate code

Move all the NUMAQ subarch definitions into numaq.c. With this it
ceases to depend on build-time subarch features.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: refactor ->wake_cpu
Ingo Molnar [Wed, 28 Jan 2009 16:50:18 +0000 (17:50 +0100)]
x86, smp: refactor ->wake_cpu

- remove macro wrappers

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: remove mach_apicdef.h
Ingo Molnar [Wed, 28 Jan 2009 16:36:56 +0000 (17:36 +0100)]
x86: remove mach_apicdef.h

Move its definitions into apic.h.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: remove mach_mpspec.h
Ingo Molnar [Wed, 28 Jan 2009 16:36:56 +0000 (17:36 +0100)]
x86: remove mach_mpspec.h

Move its definitions into mpspec.h.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: eliminate asm/mach-*/mach_mpparse.h
Ingo Molnar [Wed, 28 Jan 2009 16:29:27 +0000 (17:29 +0100)]
x86: eliminate asm/mach-*/mach_mpparse.h

Move the definition to mpparse.h.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: eliminate asm/mach-default/mach_wakecpu.h
Ingo Molnar [Wed, 28 Jan 2009 16:16:25 +0000 (17:16 +0100)]
x86, smp: eliminate asm/mach-default/mach_wakecpu.h

Spread mach_wakecpu.h's definitions into apic.h and genapic.h
and remove mach_wakecpu.h.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, ES7000: consolidate the APIC code
Ingo Molnar [Wed, 28 Jan 2009 16:09:58 +0000 (17:09 +0100)]
x86, ES7000: consolidate the APIC code

Consolidate all the ES7000 APIC code into arch/x86/mach-generic/es7000.c.

With this ES7000 ceases to rely on any subarchitecture include files.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: refactor ->inquire_remote_apic() methods
Ingo Molnar [Wed, 28 Jan 2009 15:31:52 +0000 (16:31 +0100)]
x86, smp: refactor ->inquire_remote_apic() methods

Nothing exciting - a few subarches dont want APIC remote reads to
be performed - the others are content with the default method.

 - extend the generic code to handle NULL methods

 - clear out dummy methods and replace them with NULL

 - clean up: remove wrapper macros, etc.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: remove ->restore_NMI_vector()
Ingo Molnar [Wed, 28 Jan 2009 15:43:47 +0000 (16:43 +0100)]
x86, smp: remove ->restore_NMI_vector()

Nothing actually restores the NMI vector - so remove this
logic altogether.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: refactor ->store/restore_NMI_vector() methods
Ingo Molnar [Wed, 28 Jan 2009 15:31:52 +0000 (16:31 +0100)]
x86, smp: refactor ->store/restore_NMI_vector() methods

Only NUMAQ does something substantial here, because it initializes
via NMIs (not via INIT as standard SMP startup) - so it needs to
store and restore the NMI vector.

 - extend the generic code to handle NULL methods

 - clear out dummy methods and replace them with NULL

 - clean up: remove wrapper macros, etc.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: refactor ->smp_callin_clear_local_apic() methods
Ingo Molnar [Wed, 28 Jan 2009 15:31:52 +0000 (16:31 +0100)]
x86, smp: refactor ->smp_callin_clear_local_apic() methods

Only NUMAQ does something substantial here, because it initializes
via NMIs (not via INIT as standard SMP startup) - so it needs to
reset the APIC.

 - extend the generic code to handle NULL methods

 - clear out dummy methods and replace them with NULL

 - clean up: remove wrapper macros, etc.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: refactor ->wait_for_init_deassert()
Ingo Molnar [Wed, 28 Jan 2009 15:21:32 +0000 (16:21 +0100)]
x86, smp: refactor ->wait_for_init_deassert()

- spread out the namespace on a per APIC driver basis

 - handle a NULL ->wait_for_init_deassert() as a 'dont wait' default method

 - remove NUMAQ and Summit handlers

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: set ->trampoline_phys_low/high on 64-bit too
Ingo Molnar [Wed, 28 Jan 2009 15:15:16 +0000 (16:15 +0100)]
x86: set ->trampoline_phys_low/high on 64-bit too

64-bit x86 has zero for ->trampoline_phys_low/high, but the smpboot
code can use these values - so it's better to set them up to their
correct values.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: clean up ->trampoline_phys_low/high handling
Ingo Molnar [Wed, 28 Jan 2009 15:09:23 +0000 (16:09 +0100)]
x86, smp: clean up ->trampoline_phys_low/high handling

- spread out the namespace on a per apic driver basis

 - remove wrapper macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: untangle the send_IPI_*() jungle
Ingo Molnar [Wed, 28 Jan 2009 14:42:24 +0000 (15:42 +0100)]
x86, apic: untangle the send_IPI_*() jungle

Our send_IPI_*() methods and definitions are a twisted mess: the same
symbol is defined to different things depending on .config details,
in a non-transparent way.

 - spread out the quirks into separately named per apic driver methods

 - prefix the standard PC methods with default_

 - get rid of wrapper macro obfuscation

 - clean up various details

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: refactor ->cpu_mask_to_apicid*()
Ingo Molnar [Wed, 28 Jan 2009 14:20:18 +0000 (15:20 +0100)]
x86, apic: refactor ->cpu_mask_to_apicid*()

- spread out the namespace on a per driver basis

 - clean up the functions

 - get rid of macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: get rid of *_APIC_ID_MASK definitions
Ingo Molnar [Wed, 28 Jan 2009 14:08:53 +0000 (15:08 +0100)]
x86, apic: get rid of *_APIC_ID_MASK definitions

Impact: cleanup

Remove the *_APIC_ID_MASK subarch definitions and move them straight
to the genapic driver initialization code.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: refactor ->apic_id_mask & APIC_ID_MASK
Ingo Molnar [Wed, 28 Jan 2009 13:59:17 +0000 (14:59 +0100)]
x86, apic: refactor ->apic_id_mask & APIC_ID_MASK

- spread out the namespace on a per driver basis

 - get rid of wrapper macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: refactor ->get_apic_id() & GET_APIC_ID()
Ingo Molnar [Wed, 28 Jan 2009 13:08:38 +0000 (14:08 +0100)]
x86, apic: refactor ->get_apic_id() & GET_APIC_ID()

- spread out the namespace on a per driver basis

 - get rid of macro wrappers

 - small cleanups

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: consolidate the ->mps_oem_check() code
Ingo Molnar [Wed, 28 Jan 2009 12:44:32 +0000 (13:44 +0100)]
x86: consolidate the ->mps_oem_check() code

- spread out the mps_oem_check() namespace on a per APIC driver basis

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, mpparse: call the generic quirk handlers early
Ingo Molnar [Wed, 28 Jan 2009 12:54:56 +0000 (13:54 +0100)]
x86, mpparse: call the generic quirk handlers early

Call all the registered MPS quirk handlers early. These methods scan
low RAM typically for specific signatures so are safe to be called
early.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, smp: clean up mps_oem_check()
Ingo Molnar [Wed, 28 Jan 2009 12:47:42 +0000 (13:47 +0100)]
x86, smp: clean up mps_oem_check()

Impact: cleanup

 - allow NULL ->mps_oem_check() entries

 - clean up the code flow

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: refactor ->phys_pkg_id()
Ingo Molnar [Wed, 28 Jan 2009 12:24:54 +0000 (13:24 +0100)]
x86, apic: refactor ->phys_pkg_id()

Refactor the ->phys_pkg_id() methods:

 - namespace separation

 - macro wrapper removal

 - open-coded calls to the methods in the generic code

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: unify phys_pkg_id()
Ingo Molnar [Wed, 28 Jan 2009 12:31:22 +0000 (13:31 +0100)]
x86, apic: unify phys_pkg_id()

- unify the call signature of 64-bit to that of 32-bit

 - clean up the types all around

 - clean up namespace contamination

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, es7000: clean up es7000_enable_apic_mode()
Ingo Molnar [Wed, 28 Jan 2009 12:15:06 +0000 (13:15 +0100)]
x86, es7000: clean up es7000_enable_apic_mode()

- eliminate the needless es7000_enable_apic_mode() complication which
  was not apparent prior the namespace cleanups

- clean up the control flow in es7000_enable_apic_mode()

- other cleanups

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: refactor ->enable_apic_mode() subarch methods
Ingo Molnar [Wed, 28 Jan 2009 11:43:18 +0000 (12:43 +0100)]
x86: refactor ->enable_apic_mode() subarch methods

Only ES7000 has a real ->enable_apic_mode() method, the other
subarchitectures define it but keep it empty.

So mark the vector as NULL, extend the generic code to handle
NULL -setup_portio_remap() entries and remove all the empty
handlers.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: refactor ->check_phys_apicid_present() subarch methods
Ingo Molnar [Wed, 28 Jan 2009 11:43:18 +0000 (12:43 +0100)]
x86: refactor ->check_phys_apicid_present() subarch methods

- spread out the namespace to per driver methods

 - extend it to 64-bit as well so that we can use
   apic->check_phys_apicid_present() unconditionally

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: refactor ->setup_portio_remap() subarch methods
Ingo Molnar [Wed, 28 Jan 2009 11:43:18 +0000 (12:43 +0100)]
x86: refactor ->setup_portio_remap() subarch methods

Only NUMAQ has a real ->setup_portio_remap() method, the other
subarchitectures define it but keep it empty.

So mark the vector as NULL, extend the generic code to handle
NULL -setup_portio_remap() entries and remove all the empty
handlers.

Also move the NUMAQ method from the header file into the
 apic driver .c file.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up ->apicid_to_cpu_present()
Ingo Molnar [Wed, 28 Jan 2009 05:50:47 +0000 (06:50 +0100)]
x86, apic: clean up ->apicid_to_cpu_present()

- separate the namespace

 - remove macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up ->cpu_present_to_apicid()
Ingo Molnar [Wed, 28 Jan 2009 05:50:47 +0000 (06:50 +0100)]
x86, apic: clean up ->cpu_present_to_apicid()

- separate the namespace

 - remove macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up ->cpu_to_logical_apicid()
Ingo Molnar [Wed, 28 Jan 2009 05:50:47 +0000 (06:50 +0100)]
x86, apic: clean up ->cpu_to_logical_apicid()

- separate the namespace

 - remove macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up ->apicid_to_node()
Ingo Molnar [Wed, 28 Jan 2009 05:50:47 +0000 (06:50 +0100)]
x86, apic: clean up ->apicid_to_node()

- separate the namespace

 - remove macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: streamline the ->multi_timer_check() quirk
Ingo Molnar [Wed, 28 Jan 2009 06:17:26 +0000 (07:17 +0100)]
x86, apic: streamline the ->multi_timer_check() quirk

only NUMAQ uses this quirk: to prevent the timer IRQ from being added
on secondary nodes.

All other genapic templates can have a NULL ->multi_timer_check()
callback.

Also, extend the generic code to treat a NULL pointer accordingly.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up ->setup_apic_routing()
Ingo Molnar [Wed, 28 Jan 2009 05:50:47 +0000 (06:50 +0100)]
x86, apic: clean up ->setup_apic_routing()

- separate the namespace

 - remove macros

 - remove namespace clash on 64-bit

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up ->ioapic_phys_id_map()
Ingo Molnar [Wed, 28 Jan 2009 05:50:47 +0000 (06:50 +0100)]
x86, apic: clean up ->ioapic_phys_id_map()

- separate the namespace

 - remove macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up ->init_apic_ldr()
Ingo Molnar [Wed, 28 Jan 2009 05:50:47 +0000 (06:50 +0100)]
x86, apic: clean up ->init_apic_ldr()

- separate the namespace

 - remove macros

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up ->vector_allocation_domain()
Ingo Molnar [Wed, 28 Jan 2009 05:50:47 +0000 (06:50 +0100)]
x86, apic: clean up ->vector_allocation_domain()

- separate the namespace

 - remove macros

 - move the default vector-allocation-domain to mach-generic

 - fix whitespace damage

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: remove no_balance_irq and no_ioapic_check flags
Ingo Molnar [Wed, 28 Jan 2009 04:57:56 +0000 (05:57 +0100)]
x86, apic: remove no_balance_irq and no_ioapic_check flags

These flags are completely unused. (the in-kernel IRQ balancer has
been removed from the upstream kernel.)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up check_apicid*() callbacks
Ingo Molnar [Wed, 28 Jan 2009 04:41:42 +0000 (05:41 +0100)]
x86, apic: clean up check_apicid*() callbacks

Clean up these methods - to make it clearer which function is
used in which case.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: rename genapic::apic_destination_logical to genapic::dest_logical
Ingo Molnar [Wed, 28 Jan 2009 04:29:25 +0000 (05:29 +0100)]
x86, apic: rename genapic::apic_destination_logical to genapic::dest_logical

This field name was unreasonably long - shorten it.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: clean up the APIC_DEST_LOGICAL logic
Ingo Molnar [Wed, 28 Jan 2009 04:13:04 +0000 (05:13 +0100)]
x86: clean up the APIC_DEST_LOGICAL logic

Impact: cleanup

The bigsmp and es7000 subarchitectures un-defined APIC_DEST_LOGICAL in
a rather nasty way by re-defining it to zero. That is infinitely
fragile and makes it very hard to see what to code really does in
a given context. The very same constant has different meanings and
values - depending on which subarch is enabled.

Untangle this mess by never undefining the constant, but instead
propagating the right values into the genapic driver templates.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: rename ->ESR_DISABLE to ->disable_esr
Ingo Molnar [Wed, 28 Jan 2009 04:08:44 +0000 (05:08 +0100)]
x86: rename ->ESR_DISABLE to ->disable_esr

the ->ESR_DISABLE shouting variant was used to enable the esr_disable
macro wrappers. Those ugly macros are removed now so we can rename
->ESR_DISABLE to ->disable_esr

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86: clean up esr_disable() methods
Ingo Molnar [Wed, 28 Jan 2009 04:01:41 +0000 (05:01 +0100)]
x86: clean up esr_disable() methods

Impact: cleanup

Most subarchitectures want to disable the APIC ESR (Error Status Register),
because they generally have hardware hacks that wrap standard CPUs into
a bigger system and hence the APIC bus is quite non-standard and weirdnesses
(lockups) have been seen with ESR reporting.

Remove the esr_disable macros and put the desired flag into each
subarchitecture's genapic template directly.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agox86, apic: clean up / remove TARGET_CPUS
Ingo Molnar [Wed, 28 Jan 2009 03:32:51 +0000 (04:32 +0100)]
x86, apic: clean up / remove TARGET_CPUS

Impact: cleanup

use apic->target_cpus() directly instead of the TARGET_CPUS wrapper.

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