openwrt/staging/blogic.git
16 years agox86: fix section mismatch warning in mcheck/mce_amd_64.c
Sam Ravnborg [Wed, 30 Jan 2008 12:33:36 +0000 (13:33 +0100)]
x86: fix section mismatch warning in mcheck/mce_amd_64.c

Fix following warning:
WARNING: arch/x86/kernel/cpu/mcheck/built-in.o(.text+0x1584): Section mismatch: reference to .cpuinit.text:threshold_create_device in 'threshold_cpu_callback'

threshold_cpu_callback() is only used by threshold_cpu_notifier.
threshold_cpu_notifier is only used for cpu hot plug as it is registered
using register_hotcpu_notifier().

Mark them both __cpuinit to fix the warning.

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix section mismatch warning in mcheck/mce_64.c
Sam Ravnborg [Wed, 30 Jan 2008 12:33:36 +0000 (13:33 +0100)]
x86: fix section mismatch warning in mcheck/mce_64.c

Fix following warning:
WARNING: arch/x86/kernel/cpu/mcheck/built-in.o(.text+0x752): Section mismatch: reference to .cpuinit.text:mce_create_device in 'mce_cpu_callback'

mce_cpu_callback() is only used by mce_cpu_notofier.
The notifier is only used for hotplugable cpu's as it is
registered using register_hotcpu_notifier(),

Annotate them both __cpuinit to fix the warning.

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add the RDC machine specific reboot fixup
Florian Fainelli [Wed, 30 Jan 2008 12:33:36 +0000 (13:33 +0100)]
x86: add the RDC machine specific reboot fixup

The RDC R-321x SoC needs a reboot fixup which
uses its internal hardware watchdog set to
reset the CPU on next tick.

Signed-off-by: Florian Fainelli <florian.fainelli@telecomint.eu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add support for the RDC R-321x SoC
Florian Fainelli [Wed, 30 Jan 2008 12:33:36 +0000 (13:33 +0100)]
x86: add support for the RDC R-321x SoC

This patch adds support for the RDC R-321x system-on-chip,
also known as R-861x-(G). It uses the generic GPIO API and
has support for the on-chip hardware watchdog.

Build-fix from: Randy Dunlap <randy.dunlap@oracle.com>

Signed-off-by: Florian Fainelli <florian.fainelli@telecomint.eu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agopci: add PCI identifiers for the RDC devices
Florian Fainelli [Wed, 30 Jan 2008 12:33:36 +0000 (13:33 +0100)]
pci: add PCI identifiers for the RDC devices

This patch defines the PCI identifiers found in
the RDC R-321x System-on-Chip.

Signed-off-by: Florian Fainelli <florian.fainelli@telecomint.eu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add generic GPIO support to x86
Florian Fainelli [Wed, 30 Jan 2008 12:33:35 +0000 (13:33 +0100)]
x86: add generic GPIO support to x86

This patch adds the generic GPIO support to the x86
architecture. We do the same as for MIPS, we let
the machine override the gpio callbacks and provide
defaults one in mach-generic.

Signed-off-by: Florian Fainelli <florian.fainelli@telecomint.eu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: GEODE: update GPIO API to support setting multiple GPIOs at once
Andres Salomon [Wed, 30 Jan 2008 12:33:35 +0000 (13:33 +0100)]
x86: GEODE: update GPIO API to support setting multiple GPIOs at once

The existing Geode GPIO API only allows for updating one GPIO at once.  There
are instances where users want to update multiple GPIOs at once.  With the
current API, they are given two choices; either ignore the GPIO API:

      outl(0xc000, gpio_base + GPIO_OUTPUT_VAL);
      outl(0xc000, gpio_base + GPIO_OUTPUT_ENABLE);

Alternatively, call each GPIO update separately:

      geode_gpio_set(14, GPIO_OUTPUT_VAL);
      geode_gpio_set(15, GPIO_OUTPUT_VAL);
      geode_gpio_set(14, GPIO_OUTPUT_ENABLE);
      geode_gpio_set(15, GPIO_OUTPUT_ENABLE);

Neither are desirable.  This patch changes the GPIO API to allow for setting
of multiple GPIOs at once; rather than being passed an integer, we pass
a bitmask and provide a translation function.  The above code would now
look like this:

      geode_gpio_set(geode_gpio(14)|geode_gpio(15), GPIO_OUTPUT_VAL);
      geode_gpio_set(geode_gpio(14)|geode_gpio(15), GPIO_OUTPUT_ENABLE);

Since there are no upstream users of the GPIO API yet (afaik), best to
change this now.  This also adds a bit of sanity checking; it is no
longer possible to use a GPIO above 28.

Note the semantics of geode_gpio_isset() have changed:
geode_gpio_isset(geode_gpio(3)|geode_gpio(4), ...)
will only return true iff both GPIOs are set.

Signed-off-by: Andres Salomon <dilinger@debian.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agox86: fix cpu MHz reporting for AMD family 0x11
Andreas Herrmann3 [Wed, 30 Jan 2008 12:33:35 +0000 (13:33 +0100)]
x86: fix cpu MHz reporting for AMD family 0x11

Fix cpu MHz reporting for AMD family 0x11 when powernow-k8 is
disabled.

Just adhere to the CONSTANT_TSC feature bit for AMD CPUs when deciding
whether cpu_khz needs calibration. The additional check for CPU family
is not needed and prevents calibration for future CPUs.

Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: improve MTRR trimming messages
Ingo Molnar [Wed, 30 Jan 2008 12:33:35 +0000 (13:33 +0100)]
x86: improve MTRR trimming messages

improve the MTTR trimming messages and also trigger a WARN_ON()
so that kerneloops.org can pick it up and categorize it.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix NUMA emulation on 64-bit
Minoru Usui [Wed, 30 Jan 2008 12:33:35 +0000 (13:33 +0100)]
x86: fix NUMA emulation on 64-bit

I found a small bug of NUMA emulation code for x86_64. (CONFIG_NUMA_EMU)
If machine is non-NUMA, find_node_by_addr() should return
NUMA_NO_NODE, but current implementation code returns existent maximum
NUMA node number + 1.
This is not existent NUMA node number.

However, this behaviour does not affect NUMA emulation fortunately, because
acpi_fake_nodes() that is caller of find_node_by_addr() gets pxm
(proximity domain) by node_to_pxm() from non-existent NUMA node number
that was returned by find_node_by_addr().
node_to_pxm() returns PXM_INVAL that means illegal or non-existent
NUMA node number.

Signed-off-by: Minoru Usui <usui@mxm.nes.nec.co.jp>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agocpufreq: fix obvious condition statement error
Yi Yang [Wed, 30 Jan 2008 12:33:34 +0000 (13:33 +0100)]
cpufreq: fix obvious condition statement error

The function __cpufreq_set_policy in file drivers/cpufreq/cpufreq.c
has a very obvious error:

        if (policy->min > data->min && policy->min > policy->max) {
                ret = -EINVAL;
                goto error_out;
        }

This condtion statement is wrong because it returns -EINVAL only if
policy->min is greater than policy->max (in this case,
"policy->min > data->min" is true for ever.). In fact, it should
return -EINVAL as well if policy->max is less than data->min.

The correct condition should be:

if (policy->min > data->max || policy->max < data->min) {

The following test result testifies the above conclusion:

Before applying this patch:

[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2394000 1596000
[root@yangyi-dev /]# echo 1596000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
1596000
[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
1596000
[root@yangyi-dev /]# echo "2000000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
-bash: echo: write error: Invalid argument
[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
1596000
[root@yangyi-dev /]# echo "0" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
1596000
[root@yangyi-dev /]# echo "1595000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
1596000
[root@yangyi-dev /]#

After applying this patch:

[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2394000 1596000
[root@yangyi-dev /]# echo 1596000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
1596000
[root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
1596000
[root@localhost /]# echo "2000000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
-bash: echo: write error: Invalid argument
[root@localhost /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
1596000
[root@localhost /]# echo "0" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
-bash: echo: write error: Invalid argument
[root@localhost /]# echo "1595000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
-bash: echo: write error: Invalid argument
[root@localhost /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
1596000
[root@localhost /]# echo "1596000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
[root@localhost /]# echo "2394000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
[root@localhost /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
2394000
[root@localhost /]

Signed-off-by: Yi Yang <yi.y.yang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: early cpu_to_node fix in numa_64.c
travis@sgi.com [Wed, 30 Jan 2008 12:33:33 +0000 (13:33 +0100)]
x86: early cpu_to_node fix in numa_64.c

Both of these references to cpu_to_node() can potentially occur
before the "late" cpu_to_node map is setup.  Therefore, they
should be changed to use early_cpu_to_node().

Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: GEODE add the "mfgptfix" boot time option to fix MFGPT timers
Willy Tarreau [Wed, 30 Jan 2008 12:33:33 +0000 (13:33 +0100)]
x86: GEODE add the "mfgptfix" boot time option to fix MFGPT timers

The new "mfgptfix" boot command line option may be usd to fix MFGPT
timers on AMD Geode platforms when the BIOS has incorrectly applied
a workaround. TinyBIOS version 0.98 is known to be affected, 0.99
fixes the problem by letting the user disable the workaround.

Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove struct cpu_model_info
Hiroshi Shimamoto [Wed, 30 Jan 2008 12:33:33 +0000 (13:33 +0100)]
x86: remove struct cpu_model_info

No one uses struct cpu_model_info on x86_64 now.

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use KSYM_NAME_LEN
Cyrill Gorcunov [Wed, 30 Jan 2008 12:33:33 +0000 (13:33 +0100)]
x86: use KSYM_NAME_LEN

Use KSYM_NAME_LEN instead of numeric value

Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: should use array directly for early_ptr
Yinghai Lu [Wed, 30 Jan 2008 12:33:33 +0000 (13:33 +0100)]
x86: should use array directly for early_ptr

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove duplicated line about
Yinghai Lu [Wed, 30 Jan 2008 12:33:33 +0000 (13:33 +0100)]
x86: remove duplicated line about

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86/non-x86: percpu, node ids, apic ids x86.git fixup
Mike Travis [Wed, 30 Jan 2008 12:33:32 +0000 (13:33 +0100)]
x86/non-x86: percpu, node ids, apic ids x86.git fixup

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: left over fix for leak of early_ioremp in dmi_scan
Yinghai Lu [Wed, 30 Jan 2008 12:33:32 +0000 (13:33 +0100)]
x86: left over fix for leak of early_ioremp in dmi_scan

Signed-off-by: Yinghai Lu <yinghai@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86_32: trim memory by updating e820
Yinghai Lu [Wed, 30 Jan 2008 12:33:32 +0000 (13:33 +0100)]
x86_32: trim memory by updating e820

when MTRRs are not covering the whole e820 table, we need to trim the
RAM and need to update e820.

reuse some code on 64-bit as well.

here need to add early_get_cap and use it in early_cpu_detect, and move
mtrr_bp_init early.

The code successfully trimmed the memory map on Justin's system:

from:

 [    0.000000]  BIOS-e820: 0000000100000000 - 000000022c000000 (usable)

to:

 [    0.000000]   modified: 0000000100000000 - 0000000228000000 (usable)
 [    0.000000]   modified: 0000000228000000 - 000000022c000000 (reserved)

According to Justin it makes quite a difference:

|  When I boot the box without any trimming it acts like a 286 or 386,
|  takes about 10 minutes to boot (using raptor disks).

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Tested-by: Justin Piszcz <jpiszcz@lucidpixels.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix DEBUG_RODATA kconfig text
Ingo Molnar [Wed, 30 Jan 2008 12:33:32 +0000 (13:33 +0100)]
x86: fix DEBUG_RODATA kconfig text

fix kconfig text and make DEBUG_RODATA default.

this helps debugging quite a bit.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: constify function pointer tables
Jan Engelhardt [Wed, 30 Jan 2008 12:33:32 +0000 (13:33 +0100)]
x86: constify function pointer tables

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove depends on X86_32 from PARAVIRT & PARAVIRT_GUEST
Eduardo Pereira Habkost [Wed, 30 Jan 2008 12:33:32 +0000 (13:33 +0100)]
x86: remove depends on X86_32 from PARAVIRT & PARAVIRT_GUEST

With this, the paravirt_ops code can be enabled on x86_64 also.

Each guest implementation (Xen, VMI, lguest) now depends on X86_32. The
dependencies can be dropped for each one when they start to support
x86_64.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: allow enabling PARAVIRT without any guest implementation
Eduardo Pereira Habkost [Wed, 30 Jan 2008 12:33:32 +0000 (13:33 +0100)]
x86: allow enabling PARAVIRT without any guest implementation

This will allow people to enable the paravirt_ops code even when no
guest support is enabled, for broader testing of the paravirt_ops code.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove iBCS support
Andi Kleen [Wed, 30 Jan 2008 12:33:32 +0000 (13:33 +0100)]
x86: remove iBCS support

ibcs2 support has never been supported on 2.6 kernels as far as I know,
and if it has it must have been an external patch.  Anyways, if anybody
applies an external patch they could as well readd the ibcs checking
code to the ELF loader in the same patch.  But there is no reason to
keep this code running in all Linux kernels.  This will save at least
two strcmps each ELF execution.

No deprecation period because it could not have been used anyway.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agortc: use the IRQ callback interface in (old) RTC driver
Bernhard Walle [Wed, 30 Jan 2008 12:33:31 +0000 (13:33 +0100)]
rtc: use the IRQ callback interface in (old) RTC driver

the previous patch in the old RTC driver.  It also removes the direct
rtc_interrupt() call from arch/x86/kernel/hpetc.c so that there's finally no
(code) dependency to CONFIG_RTC in arch/x86/kernel/hpet.c.

Because of this, it's possible to compile the drivers/char/rtc.ko driver as
module and still use the HPET emulation functionality.  This is also expressed
in Kconfig.

Signed-off-by: Bernhard Walle <bwalle@suse.de>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Cc: Andi Kleen <ak@suse.de>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Robert Picco <Robert.Picco@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules
Bernhard Walle [Wed, 30 Jan 2008 12:33:28 +0000 (13:33 +0100)]
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules

enabled, then interrupts don't work for the rtc-cmos driver which results in
RTC_AIE*, RTC_PIE* and RTC_ALM being unusable.  This affects hwclock from
util-linux-ng at least on i386 since that uses RTC_PIE_ON.  (For x86-64, a
polling method is used for unknown reasons.)

This patch series now

  1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc
     driver uses to work around that problem,

  2. makes it possible to compile the old rtc driver as module, while still
     having CONFIG_HPET_EMULATE_RTC enabled and

  3. makes use of the exported functions in (1) in the new rtc-cmos driver.

This patch:

This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable
for kernel modules. It

  - exports the functions (EXPORT_SYMBOL_GPL()),
  - adds an interface to register the interrupt callback function
    instead of using only a fixed callback function and
  - replaces the rtc_get_rtc_time() function which depends on
    CONFIG_RTC with a call to get_rtc_time() which is defined in
    include/asm-generic/rtc.h.

The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is
removed by the next patch. After this, there's no (code) dependency of
this functions to CONFIG_RTC=y any more.

Signed-off-by: Bernhard Walle <bwalle@suse.de>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Cc: Andi Kleen <ak@suse.de>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Robert Picco <Robert.Picco@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix Section mismatch: reference to .init.text:lguest_entry
Sam Ravnborg [Wed, 30 Jan 2008 12:33:28 +0000 (13:33 +0100)]
x86: fix Section mismatch: reference to .init.text:lguest_entry

fix:

> WARNING: vmlinux.o(.data+0x4): Section mismatch: reference to .init.text:lguest_entry (between 'subarch_entries' and 'stack_start')

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86_32: always run the full set of paging state.
Ian Campbell [Wed, 30 Jan 2008 12:33:27 +0000 (13:33 +0100)]
x86_32: always run the full set of paging state.

I am preparing to convert the boot time page table to the kernels
native format.  To achieve that I need to enable PAE. Enabling PSE
and the no execute bit would not hurt.  So this patch modifies
the boot cpu path to execute all of the kernels enable code
if and only if we have the proper bits set in mmu_cr4_features.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Mika Penttilä <mika.penttila@kolumbus.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86_32: remove unnecessary use of %ebx as the boot cpu flag
Ian Campbell [Wed, 30 Jan 2008 12:33:27 +0000 (13:33 +0100)]
x86_32: remove unnecessary use of %ebx as the boot cpu flag

Currently in head_32.S there are two ways we test to see if we
are the boot cpu.  By looking at %ebx and by looking at the
static variable ready.  When changing things around I have
found that it gets tricky to preserve %ebx.  So this
patch just switches head.S over to the more reliable
test of always using ready.

Hopefully later we can kill these tests entirely.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Mika Penttilä <mika.penttila@kolumbus.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: change size of node ids from u8 to s16
travis@sgi.com [Wed, 30 Jan 2008 12:33:25 +0000 (13:33 +0100)]
x86: change size of node ids from u8 to s16

Change the size of node ids for X86_64 from u8 to s16 to
accomodate more than 32k nodes and allow for NUMA_NO_NODE
(-1) to be sign extended to int.

Cc: David Rientjes <rientjes@google.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Mike Travis <travis@sgi.com>
Reviewed-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: relax restrictions on setting CONFIG_NUMA on x86, #2
Mel Gorman [Wed, 30 Jan 2008 12:33:25 +0000 (13:33 +0100)]
x86: relax restrictions on setting CONFIG_NUMA on x86, #2

The FLATMEM memory model references a global mem_map and max_mapnr. This
is incompatible with how memory models used for NUMA view the world.
Builds fail if FLATMEM && NUMA are set on x86. This patch forbids that
combination of config items. This is consistent with x86_64
enforcements.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: make NUMA work on 32-bit
Mel Gorman [Wed, 30 Jan 2008 12:33:25 +0000 (13:33 +0100)]
x86: make NUMA work on 32-bit

The DISCONTIG memory model on x86 32 bit uses a remap allocator early
in boot. The objective is that portions of every node are mapped in to
the kernel virtual area (KVA) in place of ZONE_NORMAL so that node-local
allocations can be made for pgdat and mem_map structures.

With SPARSEMEM, the amount that is set aside is insufficient for all the
mem_maps to be allocated. During the boot process, it falls back to using
the bootmem allocator. This breaks assumptions that SPARSEMEM makes about
the layout of the mem_map in memory and results in a VM_BUG_ON triggering
due to pfn_to_page() returning garbage values.

This patch only enables the remap allocator for use with DISCONTIG.

Without SRAT support, a compile-error occurs because ACPI table parsing
functions are only available in x86-64. This patch also adds no-op stubs
and prints a warning message. What likely needs to be done is sharing
the table parsing functions between 32 and 64 bit if they are
compatible.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: make printk_address available on X86_32
Harvey Harrison [Wed, 30 Jan 2008 12:33:25 +0000 (13:33 +0100)]
x86: make printk_address available on X86_32

Small fomatting fixes to 64-bit as well, trailing whitespace
and extra semicolon, also move the ifdefs for CONFIG_KALLSYMS
into the function itself.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix early cpu_to_node panic from nr_free_zone_pages
Mike Travis [Wed, 30 Jan 2008 12:33:25 +0000 (13:33 +0100)]
x86: fix early cpu_to_node panic from nr_free_zone_pages

call early_cpu_to_node() since per_cpu(cpu_to_node_map) might not be setup
yet.

I also had to export x86_cpu_to_node_map_early_ptr because of some calls
from the network code to numa_node_id():

net/ipv4/netfilter/arp_tables.c:
net/ipv4/netfilter/ip_tables.c:
net/ipv4/netfilter/ip_tables.c:

Signed-off-by: Mike Travis <travis@sgi.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: reboot: remove inb_p usage
Alan Cox [Wed, 30 Jan 2008 12:33:25 +0000 (13:33 +0100)]
x86: reboot: remove inb_p usage

We are driving a motherboard port so use a 2uS explicit delay at this
point.

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoxen: fix section usage in xen-head.S and setup.c
Sam Ravnborg [Wed, 30 Jan 2008 12:33:25 +0000 (13:33 +0100)]
xen: fix section usage in xen-head.S and setup.c

additional section for .init.text appending a number.

A side effect of this was a section mismatch warning because modpost did
not recognize a .init.text section named .init.text.1: WARNING:
vmlinux.o(.text.head+0x247): Section mismatch: reference to
.init.text.1:start_kernel (between 'is386' and 'check_x87')

Fix this by hardcoding the "ax" in the pushsection.  Thanks to Torlaf for
reporting this.

Alan Modra provided the hint that made me able to locate the root cause of
this warning.  And Mike Frysinger told me how to properly fix it using
__INIT/__FINIT.

Fix following Section mismatch warning in addition:
WARNING: vmlinux.o(.text+0x14c8): Section mismatch: reference to .init.data:vsyscall_int80_start (between 'fiddle_vdso' and 'xen_setup_features')

fiddle_vdso was only used from a __init function - so declare it __init.

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Toralf Förster <toralf.foerster@gmx.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: export __supported_pte_mask
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: export __supported_pte_mask

export __supported_pte_mask variable as GPL symbol.
lguest is a user of it.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: export check_tsc_unstable
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: export check_tsc_unstable

Exporrt check_tsc_unstable function as GPL symbol. lguest is
a user of it.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: update reliability argument to printk_address
Harvey Harrison [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: update reliability argument to printk_address

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add vdso32-int80-syms.lds to .gitignore
Harvey Harrison [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: add vdso32-int80-syms.lds to .gitignore

One of the generated files was missed in gitignore.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: clean up paging_init()
Ingo Molnar [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: clean up paging_init()

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: get_cycles() fix
Ingo Molnar [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: get_cycles() fix

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: only support sparsemem
Yinghai Lu [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: only support sparsemem

sparsemem is only one supported, so could remove FLAT_NODE_MEM related,
that is only needed !SPARSEMEM

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Reviewed-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agogenirq: stackdump after the "Trying to free already-free IRQ" message
Ingo Molnar [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
genirq: stackdump after the "Trying to free already-free IRQ" message

these bugs are harder to find than they seem, a stackdump helps.

make it dependent on CONFIG_DEBUG_SHIRQ so that people can turn it off
if it annoys them.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix up alternatives with lockdep enabled
Ingo Molnar [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: fix up alternatives with lockdep enabled

An older binutils bug caused us to not fix up alternatives.
This problem involved mutex.c but we dont do lockdep section tricks
there anymore, so this workaround is moot. Keep the printk nevertheless,
just in case ... We can remove that later on.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add warning to check_tsc_warp()
Ingo Molnar [Wed, 30 Jan 2008 12:33:24 +0000 (13:33 +0100)]
x86: add warning to check_tsc_warp()

add warning to check_tsc_warp() - if get_cycles() does not progress.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: check_tsc_warp() slowness fix
Ingo Molnar [Wed, 30 Jan 2008 12:33:23 +0000 (13:33 +0100)]
x86: check_tsc_warp() slowness fix

100 million max # of loops is a bit too much - reduce it to 10 million.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove unneded casts
Jan Engelhardt [Wed, 30 Jan 2008 12:33:23 +0000 (13:33 +0100)]
x86: remove unneded casts

x86: remove unneeded casts

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use v8086_mode helper, trivial unification
Harvey Harrison [Wed, 30 Jan 2008 12:33:23 +0000 (13:33 +0100)]
x86: use v8086_mode helper, trivial unification

Use v8086_mode inline in fault_32.c, no functional change
also ifdef the section for 32-bit only and add to fault_64.c

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add debug of invalid per_cpu map accesses
Ingo Molnar [Wed, 30 Jan 2008 12:33:23 +0000 (13:33 +0100)]
x86: add debug of invalid per_cpu map accesses

dont crash survivable situations.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add debug of invalid per_cpu map accesses
travis@sgi.com [Wed, 30 Jan 2008 12:33:22 +0000 (13:33 +0100)]
x86: add debug of invalid per_cpu map accesses

Provide a means to trap usages of per_cpu map variables before
they are setup.  Define CONFIG_DEBUG_PER_CPU_MAPS to activate.

Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: change bios_cpu_apicid to percpu data variable fixup
travis@sgi.com [Wed, 30 Jan 2008 12:33:21 +0000 (13:33 +0100)]
x86: change bios_cpu_apicid to percpu data variable fixup

Change static bios_cpu_apicid array to a per_cpu data variable.
This includes using a static array used during initialization
similar to the way x86_cpu_to_apicid[] is handled.

There is one early use of bios_cpu_apicid in apic_is_clustered_box().
The other reference in cpu_present_to_apicid() is called after
smp_set_apicids() has setup the percpu version of bios_cpu_apicid.

Signed-off-by: Mike Travis <travis@sgi.com>
Reviewed-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix early NUMA-id access
Mike Travis [Wed, 30 Jan 2008 12:33:21 +0000 (13:33 +0100)]
x86: fix early NUMA-id access

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: change NR_CPUS arrays in numa_64 fixup
travis@sgi.com [Wed, 30 Jan 2008 12:33:21 +0000 (13:33 +0100)]
x86: change NR_CPUS arrays in numa_64 fixup

Change the following static arrays sized by NR_CPUS to
per_cpu data variables:

char cpu_to_node_map[NR_CPUS];

fixup:

  - Split cpu_to_node function into "early" and "late" versions
    so that x86_cpu_to_node_map_early_ptr is not EXPORT'ed and
    the cpu_to_node inline function is more streamlined.

  - This also involves setting up the percpu maps as early as possible.

  - Fix X86_32 NUMA build errors that previous version of this
    patch caused.

V2->V3:
    - add early_cpu_to_node function to keep cpu_to_node efficient
    - move and rename smp_set_apicids() to setup_percpu_maps()
    - call setup_percpu_maps() as early as possible

V1->V2:
    - Removed extraneous casts
    - Fix !NUMA builds with '#ifdef CONFIG_NUMA"

Signed-off-by: Mike Travis <travis@sgi.com>
Reviewed-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add generic clearcpuid=... option
Andi Kleen [Wed, 30 Jan 2008 12:33:21 +0000 (13:33 +0100)]
x86: add generic clearcpuid=... option

Add a generic option to clear any cpuid bit. I added it because it was
very easy to add with the new generic cpuid disable bitmap and perhaps
it will be useful in the future.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add noclflush option
Andi Kleen [Wed, 30 Jan 2008 12:33:21 +0000 (13:33 +0100)]
x86: add noclflush option

To disable CLFLUSH usage, especially in change_page_attr().

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: convert TSC disabling to generic cpuid disable bitmap
Andi Kleen [Wed, 30 Jan 2008 12:33:20 +0000 (13:33 +0100)]
x86: convert TSC disabling to generic cpuid disable bitmap

Fix from: Ian Campbell <ijc@hellion.org.uk>

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: don't disable RDTSC in userland for 32bit notsc
Andi Kleen [Wed, 30 Jan 2008 12:33:20 +0000 (13:33 +0100)]
x86: don't disable RDTSC in userland for 32bit notsc

Modern 32bit userland doesn't even boot when the TSC is disabled
because ld.so tends to contain RDTSCs.  So make notsc only effective for the
kernel, similar to 64bit.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: convert some existing cpuid disable options to new generic bitmap
Andi Kleen [Wed, 30 Jan 2008 12:33:20 +0000 (13:33 +0100)]
x86: convert some existing cpuid disable options to new generic bitmap

This convers nofxsr, mem=nopentium and nosep to use the new
generic cpuid disable bitmap instead of using own variables.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add framework to disable CPUID bits on the command line
Andi Kleen [Wed, 30 Jan 2008 12:33:20 +0000 (13:33 +0100)]
x86: add framework to disable CPUID bits on the command line

There are already various options to disable specific cpuid bits
on the command line. They all use their own variable. Add a generic
mask to make this easier in the future.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fill in missing pv_mmu_ops entries for PAGETABLE_LEVELS >= 3
Eduardo Habkost [Wed, 30 Jan 2008 12:33:20 +0000 (13:33 +0100)]
x86: fill in missing pv_mmu_ops entries for PAGETABLE_LEVELS >= 3

This finally makes paravirt-ops able to compile and boot under x86_64.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: don't set pagetable_setup_{start,done} hooks on 64-bit
Eduardo Habkost [Wed, 30 Jan 2008 12:33:20 +0000 (13:33 +0100)]
x86: don't set pagetable_setup_{start,done} hooks on 64-bit

paravirt_pagetable_setup_{start,done}() are not used (yet) under x86_64,
and native_pagetable_setup_{start,done}() don't exist on x86_64. So they
don't need to be set.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: include/asm-x86/paravirt.h: x86_64 mmu operations
Eduardo Habkost [Wed, 30 Jan 2008 12:33:20 +0000 (13:33 +0100)]
x86: include/asm-x86/paravirt.h: x86_64 mmu operations

Add .set_pgd field to pv_mmu_ops.

Implement pud_val(), __pud(), set_pgd(), pud_clear(), pgd_clear().

pud_clear() and pgd_clear() are implemented simply using set_pud()
and set_pmd(). They don't have a field at pv_mmu_ops.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: change function orders in paravirt.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: change function orders in paravirt.h

__pmd, pmd_val and set_pud are used before they are defined (as static)
We move them a little up in the file, so it doesn't happen.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: provide __parainstructions section
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: provide __parainstructions section

This patch adds the __parainstructions section to vmlinux.lds.S.
It's needed for the patching system.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add asm_offset PARAVIRT constants
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: add asm_offset PARAVIRT constants

This patch adds the constant PARAVIRT needs in asm_offsets_64.c

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fill pv_cpu_ops structure with cr8 fields
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: fill pv_cpu_ops structure with cr8 fields

This patch fills in the read and write cr8 fields with their
native version.

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: provide read and write cr8 paravirt hooks
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: provide read and write cr8 paravirt hooks

Since the cr8 manipulation functions ended up staying in the tree,
they can't be defined just when PARAVIRT is off: In this patch,
those functions are defined for the PARAVIRT case too.

[ mingo@elte.hu: fixes ]

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: puts read and write cr8 into pv_cpu_ops
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: puts read and write cr8 into pv_cpu_ops

This patch adds room for read and write_cr8 functions back in
pv_cpu_ops struct

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: put generic mm_hooks include into PARAVIRT
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: put generic mm_hooks include into PARAVIRT

With PARAVIRT, we actually have arch_{dup,exit}_mmap functions,
so we can't include the generic header

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: provide a native_init_IRQ function on 64-bit
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: provide a native_init_IRQ function on 64-bit

x86_64 lacks a native_init_IRQ() function, so we turn the arch's
init_IRQ() function into a native construct

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add stringify header
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:33:19 +0000 (13:33 +0100)]
x86: add stringify header

We use a __stringify construction at paravirt_patch_64.c.
It's better practice to include the stringify header directly

Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: checking aperture report for node instead
Yinghai Lu [Wed, 30 Jan 2008 12:33:18 +0000 (13:33 +0100)]
x86: checking aperture report for node instead

currently when gart iommu is enabled by BIOS or previous we got

"
Checking aperture...
CPU 0: aperture @4000000 size 64MB
CPU 1: aperture @4000000 size 64MB
"
we should use use Node instead.

we will get
"
Checking aperture...
Node 0: aperture @4000000 size 64MB
Node 1: aperture @4000000 size 64MB
"

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: move select_idle_routine() call after detect_ht()
Hiroshi Shimamoto [Wed, 30 Jan 2008 12:33:18 +0000 (13:33 +0100)]
x86: move select_idle_routine() call after detect_ht()

Move the select_idle_routine() call to after the detect_ht() call at
identify_cpu() on 64-bit.

This change is for printing the polling idle and HT enabled warning
message properly.

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: move warning message of polling idle and HT enabled
Hiroshi Shimamoto [Wed, 30 Jan 2008 12:33:18 +0000 (13:33 +0100)]
x86: move warning message of polling idle and HT enabled

The warning message at idle_setup() is never shown because
smp_num_sibling hasn't been updated at this point yet.

Move this polling idle and HT enabled warning to select_idle_routine().
I also implement this warning on 64-bit kernel.

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: msr for AMD Fam 10h mmio
Yinghai Lu [Wed, 30 Jan 2008 12:33:18 +0000 (13:33 +0100)]
x86: msr for AMD Fam 10h mmio

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix unconditional arch/x86/kernel/pcspeaker.o compiling
Michael Opdenacker [Wed, 30 Jan 2008 12:33:18 +0000 (13:33 +0100)]
x86: fix unconditional arch/x86/kernel/pcspeaker.o compiling

do not add the pcspkr platform device if pcspkr support is disabled.

Signed-off-by: Michael Opdenacker <michael@free-electrons.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: only call early_init_amd one time
Yinghai Lu [Wed, 30 Jan 2008 12:33:18 +0000 (13:33 +0100)]
x86: only call early_init_amd one time

Andi's patch
"
    x86: move X86_FEATURE_CONSTANT_TSC into early cpu feature detection

    Need this in the next patch in time_init and that happens early.

    This includes a minor fix on i386 where early_intel_workarounds()
    [which is now called early_init_intel] really executes early as
    the comments say.
"
calling early_init_amd in early_identify_cpu and identify_cpu two times.

this patch remove the one in identify_cpu

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, 32-bit: trim memory not covered by wb mtrrs
Jesse Barnes [Wed, 30 Jan 2008 12:33:18 +0000 (13:33 +0100)]
x86, 32-bit: trim memory not covered by wb mtrrs

On some machines, buggy BIOSes don't properly setup WB MTRRs to cover all
available RAM, meaning the last few megs (or even gigs) of memory will be
marked uncached.  Since Linux tends to allocate from high memory addresses
first, this causes the machine to be unusably slow as soon as the kernel
starts really using memory (i.e.  right around init time).

This patch works around the problem by scanning the MTRRs at boot and
figuring out whether the current end_pfn value (setup by early e820 code)
goes beyond the highest WB MTRR range, and if so, trimming it to match.  A
fairly obnoxious KERN_WARNING is printed too, letting the user know that
not all of their memory is available due to a likely BIOS bug.

Something similar could be done on i386 if needed, but the boot ordering
would be slightly different, since the MTRR code on i386 depends on the
boot_cpu_data structure being setup.

This patch fixes a bug in the last patch that caused the code to run on
non-Intel machines (AMD machines apparently don't need it and it's untested
on other non-Intel machines, so best keep it off).

Further enhancements and fixes from:

  Yinghai Lu <Yinghai.Lu@Sun.COM>
  Andi Kleen <ak@suse.de>

Signed-off-by: Jesse Barnes <jesse.barnes@intel.com>
Tested-by: Justin Piszcz <jpiszcz@lucidpixels.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: print which shared library/executable faulted in segfault etc. messages v3
Andi Kleen [Wed, 30 Jan 2008 12:33:18 +0000 (13:33 +0100)]
x86: print which shared library/executable faulted in segfault etc. messages v3

They now look like:

hal-resmgr[13791]: segfault at 3c rip 2b9c8caec182 rsp 7fff1e825d30 error 4 in libacl.so.1.1.0[2b9c8caea000+6000]

This makes it easier to pinpoint bugs to specific libraries.

And printing the offset into a mapping also always allows to find the
correct fault point in a library even with randomized mappings. Previously
there was no way to actually find the correct code address inside
the randomized mapping.

Relies on earlier patch to shorten the printk formats.

They are often now longer than 80 characters, but I think that's worth it.

[includes fix from Eric Dumazet to check d_path error value]

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: don't disable the APIC if it hasn't been mapped yet
Andi Kleen [Wed, 30 Jan 2008 12:33:17 +0000 (13:33 +0100)]
x86: don't disable the APIC if it hasn't been mapped yet

When the kernel panics early for some unrelated reason
there would be eventually an early exception inside panic because
clear_local_APIC tried to disable the not yet mapped APIC.
Check for that explicitely.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: optimize lock prefix switching to run less frequently
Andi Kleen [Wed, 30 Jan 2008 12:33:17 +0000 (13:33 +0100)]
x86: optimize lock prefix switching to run less frequently

On VMs implemented using JITs that cache translated code changing the lock
prefixes is a quite costly operation that forces the JIT to throw away and
retranslate a lot of code.

Previously a SMP kernel would rewrite the locks once for each CPU which
is quite unnecessary. This patch changes the code to never switch at boot in
 the normal case (SMP kernel booting with >1 CPU) or only once for SMP kernel
on UP.

This makes a significant difference in boot up performance on AMD SimNow!
Also I expect it to be a little faster on native systems too because a smp
switch does a lot of text_poke()s which each synchronize the pipeline.

v1->v2: Rename max_cpus
v1->v2: Fix off by one in UP check (Thomas Gleixner)

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: replace hard coded reservations in 64-bit early boot code with dynamic table
Andi Kleen [Wed, 30 Jan 2008 12:33:17 +0000 (13:33 +0100)]
x86: replace hard coded reservations in 64-bit early boot code with dynamic table

On x86-64 there are several memory allocations before bootmem. To avoid
them stomping on each other they used to be all hard coded in bad_area().
Replace this with an array that is filled as needed.

This cleans up the code considerably and allows to expand its use.

Cc: peterz@infradead.org
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify printk strings in fault_32|64.c
Harvey Harrison [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86: unify printk strings in fault_32|64.c

Adding the address of the faulting library missed removing a
line ending from X86_32.

Also update the shorter printk format for X86_32 in fault_64.c
to make it easier to se the remaining differences.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use shorter addresses in i386 segfault printks
Andi Kleen [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86: use shorter addresses in i386 segfault printks

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use the correct cpuid method to detect MWAIT support for C states
Andi Kleen [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86: use the correct cpuid method to detect MWAIT support for C states

Previously there was a AMD specific quirk to handle the case of
AMD Fam10h MWAIT not supporting any C states. But it turns out
that CPUID already has ways to detectly detect that without
using special quirks.

The new code simply checks if MWAIT supports at least C1 and doesn't
use it if it doesn't. No more vendor specific code.

Note this is does not simply clear MWAIT because MWAIT can be still
useful even without C states.

Credit goes to Ben Serebrin for pointing out the (nearly) obvious.

Cc: "Andreas Herrmann" <andreas.herrmann3@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: move MWAIT idle check to generic CPU initialization on 32-bit
Andi Kleen [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86: move MWAIT idle check to generic CPU initialization on 32-bit

Previously it was only run for Intel CPUs, but AMD Fam10h implements MWAIT too.

This matches 64bit behaviour.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: rename stack_pointer to kernel_trap_sp
Harvey Harrison [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86: rename stack_pointer to kernel_trap_sp

Choose a less generic name for such a special case.  Add
a comment explaining the odd use in X86_32.

Change the one user of stack_pointer.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: clean up ptrace.h
Harvey Harrison [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86: clean up ptrace.h

Leave definition of pt_regs in its own section, move all kernel
code to section afterwards, unify prototype definitions, has some
conditional prototypes to make it clear what was only defined in
32 and 64 bit.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify pt_regs accessors ptrace.h
Harvey Harrison [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86: unify pt_regs accessors ptrace.h

Unify the definiton of:
v8086_mode
user_mode
user_mode_vm
stack_pointer
instruction_pointer
frame_pointer

in ptrace.h to make it clear where the differences are between
32 and 64 bit.  Changes macros to static inlines as well.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: kdump failure
Hiroshi Shimamoto [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86: kdump failure

kdump needs ELF_CORE_COPY_REGS in crash_save_cpu().
This lack of the macro causes the following BUG.

 SysRq : Trigger a crashdump
 ------------[ cut here ]------------
 kernel BUG at include/linux/elfcore.h:105!
 invalid opcode: 0000 [1] PREEMPT SMP

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86_32: remove the useless NR_syscalls macro
Dmitri Vorobiev [Wed, 30 Jan 2008 12:33:16 +0000 (13:33 +0100)]
x86_32: remove the useless NR_syscalls macro

This is against current x86.git.

The size of the system call table for 32-bit x86 kernels is obtained by
compile-time calculation of the sys_call_table array, not from the value,
which the NR_syscalls macro expands to. This trivial patch removes the
fossil macro.

Manually tested by grepping the x86 files for the "NR_syscalls" string.
No relevant use cases found.

Build-tested using allyesconfig, allnoconfig and a couple of randconfig
instances. All builds successfully finished.

Runtime test performed using a stripped-down Debian-ish config. The system
booted successfully.

Signed-off-by: Dmitri Vorobiev <dmitri.vorobiev@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: 64-bit, remove redundant cpu_has_ definitions
Kyle McMartin [Wed, 30 Jan 2008 12:33:15 +0000 (13:33 +0100)]
x86: 64-bit, remove redundant cpu_has_ definitions

PSE, PGE, XMM, XMM2, and FXSR are defined as required features, and
will be optimized to a constant at compile time. Remove their redundant
definitions.

Signed-off-by: Kyle McMartin <kyle@mcmartin.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fixup NR-CPUS patch for numa
travis@sgi.com [Wed, 30 Jan 2008 12:33:15 +0000 (13:33 +0100)]
x86: fixup NR-CPUS patch for numa

This patch removes the EXPORT_SYMBOL for:

x86_cpu_to_node_map_init
x86_cpu_to_node_map_early_ptr

... thus fixing the section mismatch problem.

Also, the mem -> node hash lookup is fixed.

Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86/paravirt: make set_pud operation common
Jeremy Fitzhardinge [Wed, 30 Jan 2008 12:33:15 +0000 (13:33 +0100)]
x86/paravirt: make set_pud operation common

Remove duplicate set_pud()s.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86/paravirt: make set_pmd operation common
Jeremy Fitzhardinge [Wed, 30 Jan 2008 12:33:15 +0000 (13:33 +0100)]
x86/paravirt: make set_pmd operation common

Remove duplicate set_pmd()s.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86/paravirt: make set_pte operations common
Jeremy Fitzhardinge [Wed, 30 Jan 2008 12:33:15 +0000 (13:33 +0100)]
x86/paravirt: make set_pte operations common

Remove duplicate set_pte* operations.  PAE still needs to have special
variants of some of these because it can't atomically update a 64-bit
pte, so there's still some duplication.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86/paravirt: common implementation for pmd value ops
Jeremy Fitzhardinge [Wed, 30 Jan 2008 12:33:15 +0000 (13:33 +0100)]
x86/paravirt: common implementation for pmd value ops

Remove duplicate __pmd/pmd_val functions.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86/paravirt: common implementation for pgd value ops
Jeremy Fitzhardinge [Wed, 30 Jan 2008 12:33:15 +0000 (13:33 +0100)]
x86/paravirt: common implementation for pgd value ops

Remove duplicate __pgd/pgd_val functions.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>