[PATCH] i386/x86-64: adjust /proc/interrupts column headings

With (significantly) more than 10 CPUs online, the column headings
drifted off the positions of the column contents with growing CPU
numbers.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Fix race in cpu_local_* on preemptible kernels

When a process changes CPUs while doing the non atomic cpu_local_*
operations it might operate on the local_t of a different CPUs.

Fix that by disabling preemption.

Pointed out by Christopher Lameter

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Fix fast check in safe_smp_processor_id

The APIC ID returned by hard_smp_processor_id can be beyond
NR_CPUS and then overflow the x86_cpu_to_apic[] array.

Add a check for overflow. If it happens then the slow loop below
will catch.

Bug pointed out by Doug Thompson

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: x86_64 setup.c - printing cmp related boottime information

Getting phys_proc_id and cpu_core_id information to be printed at boot
time for AMD processors.  Also matching the Node related boot time
information that gets printed for Intel and AMD processors for NUMA
configurations.

Signed-off-by: Rohit Seth <rohitseth@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386/x86-64/ia64: Move polling flag into thread_info_status

During some profiling I noticed that default_idle causes a lot of
memory traffic. I think that is caused by the atomic operations
to clear/set the polling flag in thread_info. There is actually
no reason to make this atomic - only the idle thread does it
to itself, other CPUs only read it. So I moved it into ti->status.

Converted i386/x86-64/ia64 for now because that was the easiest
way to fix ACPI which also manipulates these flags in its idle
function.

Cc: Nick Piggin <npiggin@novell.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Len Brown <len.brown@intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Remove bogus RED-PEN comment in signal.c

No red zone possible/needed on the alternative stack.

It caused confusion.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: check_addr() cleanups

 - Use DMA_32BIT_MASK

 - Use %z for size_t

 - 80-cols

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: miscellaneous mm/init.c fixes

 - fix an off-by-one error in phys_pmd_init()
 - prevent phys_pmd_init() from removing mappings established earlier
 - fix the direct mapping early printk to in fact show the end of the range
 - remove an apparently orphan comment

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Mark mce_amd cpu notifier __cpuinit/__cpuinitdata

Cc: Jacob Shin <jacob.shin@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: mce_amd cleanup

Clean up mce_amd.c for readability and remove code no
longer needed.

Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: mce_amd support for family 0x10 processors

Add support for mce threshold registers found in future
AMD family 0x10 processors.  Backwards compatible with
family 0xF hardware.

AK: fixed build on !SMP

Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: mce_amd relocate sysfs files

Get rid of /sys/devices/system/threshold directory and move
mce_amd thresholding files into the machine sysfs directory --
/sys/devices/system/machinecheck.

AK: Fixed warning

Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: apic support for extended apic interrupt

Add support for extended APIC LVT found in future AMD processors.

Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Move VM86 config into arch/i386/Kconfig

Architecture specific configs like this have no business at all
in init/Kconfig. This prevents it from being set on x86-64

Pointed out by H.Peter Anvin

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386/x86-64: Fix isoimage when syslinux is in /usr/share

It's like this on SUSE systems.

Cc: hpa@zytor.com
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Update copyright in time.c

Update my copyright dates in arch/x86-64/kernel/time.c

Signed-off-by: Vojtech Pavlik <vojtech@suse.cz>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Explain why HPET T0_CMP register is written twice

After writing the CFG register, the first value written to the T0_CMP
register is the value at which next interrupt should be triggered, every
value after that sets the period of the interrupt. For that reason, the code
needs to write the value twice - to set both the phase and period.

[AK: I had already figured it out by myself, but it's still useful
to have a comment for this.]

Signed-off-by: Vojtech Pavlik <vojtech@suse.cz>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Make use of the *PER* constants in time.c

This patch makes use of the newly added conversion constants
in time.h to x86-64 time.c. The code gets significantly easier
to understand.

Signed-off-by: Vojtech Pavlik <vojtech@suse.cz>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Remove hack to manually enable HPET on AMD8111 southbridges

Remove #ifdefed code to manually enable HPET on AMD8111, where the
BIOS doesn't have ACPI HPET tables and doesn't enable it for us.

Signed-off-by: Vojtech Pavlik <vojtech@suse.cz>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Add X86_FEATURE_RDTSCP, fix rdtscp in /proc/cpuinfo

This patch adds the X86_FEATURE_RDTSCP #define, so that kernel code can
check for the feature easily and also fixes the location of the "rdtscp"
string in the cpuinfo tables.

Signed-off-by: Vojtech Pavlik <vojtech@suse.cz>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Rename oem_force_hpet_timer to apic_is_clustered_box

Rename oem_force_hpet_timer to apic_is_clustered_box, to give the
function a better fitting name - it really isn't at all about HPET.

Signed-off-by: Vojtech Pavlik <vojtech@suse.cz>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Add useful constants to time.h

In timekeeping code, one often does need to use conversion constants. Naming
these leads to code that's easier to understand, showing the reader between
which units the conversion is made.

Signed-off-by: Vojtech Pavlik <vojtech@suse.cz>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: moving phys_proc_id and cpu_core_id to cpuinfo_x86

Most of the fields of cpuinfo are defined in cpuinfo_x86 structure.
This patch moves the phys_proc_id and cpu_core_id for each processor to
cpuinfo_x86 structure as well.

Signed-off-by: Rohit Seth <rohitseth@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Calgary IOMMU - Calgary specific bits

This patch hooks Calgary into the build, the x86-64 IOMMU
initialization paths, and introduces the Calgary specific bits.  The
implementation draws inspiration from both PPC (which has support for
the same chip but requires firmware support which we don't have on
x86-64) and gart. Calgary is different from gart in that it support a
translation table per PHB, as opposed to the single gart aperture.

Changes from previous version:
 * Addition of boot-time disablement for bus-level translation/isolation
   (e.g, enable userspace DMA for things like X)
 * Usage of newer IOMMU abstraction functions

Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Jon Mason <jdmason@us.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Calgary IOMMU - IOMMU abstractions

This patch creates a new interface for IOMMUs by adding a centralized
location for IOMMU allocation (for translation tables/apertures) and
IOMMU initialization.  In creating these, code was moved around for
abstraction, uniformity, and consiceness.

Take note of the move of the iommu_setup bootarg parsing code to
__setup.  This is enabled by moving back the location of the aperture
allocation/detection to mem init (which while ugly, was already the
location of the swiotlb_init).

While a slight departure from the previous patch, I belive this provides
the true intention of the previous versions of the patch which changed
this code.  It also makes the addition of the upcoming calgary code much
cleaner than previous patches.

[AK: Removed one broken change. iommu_setup still has to be called
early]

Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Jon Mason <jdmason@us.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Calgary IOMMU - move valid_dma_direction into the callers

Based on Andi Kleen's comments on the original Calgary patch, move
valid_dma_direction into the calling functions.

Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Jon Mason <jdmason@us.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Calgary IOMMU - introduce iommu_detected

swiotlb relies on the gart specific iommu_aperture variable to know if
we discovered a hardware IOMMU before swiotlb initialization.  Introduce
iommu_detected to do the same thing, but in a HW IOMMU neutral manner,
in preparation for adding the Calgary HW IOMMU.

Signed-Off-By: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-Off-By: Jon Mason <jdmason@us.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Change assembly to use regular cpuid_count macro

Minor cleanup patch:

Replacing the asm statement with cpuid_count macro(which already
provides the same functionality).

Signed-off-by: Rohit Seth <rohitseth@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: use halt() instead of raw inline assembly

Use abstractions whenever possible.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Get rid of pud_offset_k / __pud_offset_k

pud_offset_k() equivalent to pud_offset() now.  Pointed out by Jan Beulich
Similar for __pud_offset_ok, which needs a small change in the callers.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386/x86-64: Consolidate arch/{i386,x86_64}/boot/compressed/misc.c

Clean up arch/{i386,x86_64}/boot/compressed/misc.c a bit to reduce their
differences. Should have zero effect on code generation.

Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: allow unwinder to build without module support

Add proper conditionals to be able to build with CONFIG_MODULES=n.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386/x86-64: fall back to old-style call trace if no unwinding

If no unwinding is possible at all for a certain exception instance,
fall back to the old style call trace instead of not showing any trace
at all.

Also, allow setting the stack trace mode at the command line.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386: reliable stack trace support i386 entry.S

To increase the usefulness of reliable stack unwinding, this adds CFI
unwind annotations to many low-level i386 routines.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386: reliable stack trace support (i386)

These are the i386-specific pieces to enable reliable stack traces. This is
going to be even more useful once CFI annotations get added to he assembly
code, namely to entry.S.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: reliable stack trace support (x86-64 syscall

Adjust the CFA offset for 64- and 32-bit syscall entries so that the five
slots pre-subtracted from the stack pointer do not appear to reside outside
of the current frame.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: reliable stack trace support (x86-64 IRQ stack

Change the switching to/from the IRQ stack so that unwind annotations can
be added for it without requiring CFA expressions.

AK: I cleaned it up a bit, making it unconditional and removing the
obsolete DEBUG_INFO full frame code.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: reliable stack trace support (x86-64)

These are the x86_64-specific pieces to enable reliable stack traces. The
only restriction with this is that it currently cannot unwind across the
interrupt->normal stack boundary, as that transition is lacking proper
annotation.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: reliable stack trace support

These are the generic bits needed to enable reliable stack traces based
on Dwarf2-like (.eh_frame) unwind information. Subsequent patches will
enable x86-64 and i386 to make use of this.

Thanks to Andi Kleen and Ingo Molnar, who pointed out several possibilities
for improvement.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: x86_86 msi miss one entry handler

  In x86_64 architecture, if device driver with msi function
gets 0xee vector by assign_irq_vector() function, system will
crash if this interrupt happens. It is because 0xee interrupt
entry is empty. This patch modifies this. This patch is based
on 2.6.17-rc6.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Rename IOMMU option, fix help and mark option embedded.

 - Rename the GART_IOMMU option to IOMMU to make clear it's not
   just for AMD
 - Rewrite the help text to better emphatise this fact
 - Make it an embedded option because too many people get it wrong.

To my astonishment I discovered the aacraid driver tests this
symbol directly. This looks quite broken to me - it's an internal
implementation detail of the PCI DMA API. Can the maintainer
please clarify what this test was intended to do?

Cc: linux-scsi@vger.kernel.org
Cc: alan@redhat.com
Cc: markh@osdl.org
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Make sure is_compat_task works early

Previously it would only work in the first 32bit system call, not during
early process setup.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: fix vector_lock deadlock in io_apic.c

Fix a potential deadlock scenario introduced by io_apic.c's new vector_lock
on i386 and x86_64.

Found by the locking correctness validator. The patch was boot-tested on
x86. For details of the deadlock scenario, see the validator output:

  ======================================================
  [ BUG: hard-safe -> hard-unsafe lock order detected! ]
  ------------------------------------------------------
  idle/1 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
   (msi_lock){....}, at: [<c04ff8d2>] startup_msi_irq_wo_maskbit+0x10/0x35

  and this task is already holding:
   (&irq_desc[i].lock){++..}, at: [<c015b924>] probe_irq_on+0x36/0x107
  which would create a new lock dependency:
   (&irq_desc[i].lock){++..} -> (msi_lock){....}

  but this new dependency connects a hard-irq-safe lock:
   (&irq_desc[i].lock){++..}
  ... which became hard-irq-safe at:
    [<c01468c4>] lockdep_acquire+0x68/0x84
    [<c10485e9>] _spin_lock+0x21/0x2f
    [<c015aff5>] __do_IRQ+0x3d/0x113
    [<c01062d3>] do_IRQ+0x8c/0xad

  to a hard-irq-unsafe lock:
   (vector_lock){--..}
  ... which became hard-irq-unsafe at:
  ...  [<c01468c4>] lockdep_acquire+0x68/0x84
    [<c10485e9>] _spin_lock+0x21/0x2f
    [<c011b5e8>] assign_irq_vector+0x34/0xc8
    [<c1aa82fa>] setup_IO_APIC+0x45a/0xcff
    [<c1aa56e3>] smp_prepare_cpus+0x5ea/0x8aa
    [<c010033f>] init+0x32/0x2cb
    [<c0102005>] kernel_thread_helper+0x5/0xb

  which could potentially lead to deadlocks!

  other info that might help us debug this:

  3 locks held by idle/1:
   #0:  (port_mutex){--..}, at: [<c067070d>] uart_add_one_port+0x61/0x289
   #1:  (&state->mutex){--..}, at: [<c067071f>] uart_add_one_port+0x73/0x289
   #2:  (&irq_desc[i].lock){++..}, at: [<c015b924>] probe_irq_on+0x36/0x107

  the hard-irq-safe lock's dependencies:
  -> (&irq_desc[i].lock){++..} ops: 9861 {
     initial-use  at:
                          [<c01468c4>] lockdep_acquire+0x68/0x84
                          [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                          [<c015b415>] setup_irq+0x9b/0x14d
                          [<c1aaa4c4>] time_init_hook+0xf/0x11
                          [<c1a9f320>] time_init+0x44/0x46
                          [<c1a9955f>] start_kernel+0x191/0x38f
                          [<c0100210>] 0xc0100210
     in-hardirq-W at:
                          [<c01468c4>] lockdep_acquire+0x68/0x84
                          [<c10485e9>] _spin_lock+0x21/0x2f
                          [<c015aff5>] __do_IRQ+0x3d/0x113
                          [<c01062d3>] do_IRQ+0x8c/0xad
     in-softirq-W at:
                          [<c01468c4>] lockdep_acquire+0x68/0x84
                          [<c10485e9>] _spin_lock+0x21/0x2f
                          [<c015aff5>] __do_IRQ+0x3d/0x113
                          [<c01062d3>] do_IRQ+0x8c/0xad
   }
   ... key      at: [<c1ea31e0>] irq_desc_lock_type+0x0/0x20
    -> (i8259A_lock){++..} ops: 5149 {
       initial-use  at:
                        [<c01468c4>] lockdep_acquire+0x68/0x84
                        [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                        [<c0108090>] init_8259A+0x11/0x8f
                        [<c1aa0d22>] init_ISA_irqs+0x12/0x4d
                        [<c1aaa4f0>] pre_intr_init_hook+0x8/0xa
                        [<c1aa0cb9>] init_IRQ+0xe/0x65
                        [<c1a99546>] start_kernel+0x178/0x38f
                        [<c0100210>] 0xc0100210
       in-hardirq-W at:
                        [<c01468c4>] lockdep_acquire+0x68/0x84
                        [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                        [<c0107fb0>] mask_and_ack_8259A+0x1b/0xcc
                        [<c015b007>] __do_IRQ+0x4f/0x113
                        [<c01062d3>] do_IRQ+0x8c/0xad
       in-softirq-W at:
                        [<c01468c4>] lockdep_acquire+0x68/0x84
                        [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                        [<c0107fb0>] mask_and_ack_8259A+0x1b/0xcc
                        [<c015b007>] __do_IRQ+0x4f/0x113
                        [<c01062d3>] do_IRQ+0x8c/0xad
     }
     ... key      at: [<c142f174>] i8259A_lock+0x14/0x40
   ... acquired at:
     [<c01468c4>] lockdep_acquire+0x68/0x84
     [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
     [<c0107eb2>] enable_8259A_irq+0x10/0x47
     [<c0107f12>] startup_8259A_irq+0x8/0xc
     [<c015b45e>] setup_irq+0xe4/0x14d
     [<c1aaa4c4>] time_init_hook+0xf/0x11
     [<c1a9f320>] time_init+0x44/0x46
     [<c1a9955f>] start_kernel+0x191/0x38f
     [<c0100210>] 0xc0100210

    -> (ioapic_lock){+...} ops: 122 {
       initial-use  at:
                        [<c01468c4>] lockdep_acquire+0x68/0x84
                        [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                        [<c1aa71db>] io_apic_get_version+0x16/0x55
                        [<c1aa5c73>] mp_register_ioapic+0xc6/0x127
                        [<c1aa382e>] acpi_parse_ioapic+0x2d/0x39
                        [<c1abe031>] acpi_table_parse_madt_family+0xb4/0x100
                        [<c1abe093>] acpi_table_parse_madt+0x16/0x18
                        [<c1aa3c8a>] acpi_boot_init+0x132/0x251
                        [<c1aa08ea>] setup_arch+0xd36/0xe37
                        [<c1a99434>] start_kernel+0x66/0x38f
                        [<c0100210>] 0xc0100210
       in-hardirq-W at:
                        [<c01468c4>] lockdep_acquire+0x68/0x84
                        [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                        [<c011bce1>] mask_IO_APIC_irq+0x11/0x31
                        [<c011c5cc>] ack_edge_ioapic_vector+0x31/0x41
                        [<c015b007>] __do_IRQ+0x4f/0x113
                        [<c01062d3>] do_IRQ+0x8c/0xad
     }
     ... key      at: [<c1432514>] ioapic_lock+0x14/0x3c
      -> (i8259A_lock){++..} ops: 5149 {
         initial-use  at:
                         [<c01468c4>] lockdep_acquire+0x68/0x84
                         [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                         [<c0108090>] init_8259A+0x11/0x8f
                         [<c1aa0d22>] init_ISA_irqs+0x12/0x4d
                         [<c1aaa4f0>] pre_intr_init_hook+0x8/0xa
                         [<c1aa0cb9>] init_IRQ+0xe/0x65
                         [<c1a99546>] start_kernel+0x178/0x38f
                         [<c0100210>] 0xc0100210
         in-hardirq-W at:
                         [<c01468c4>] lockdep_acquire+0x68/0x84
                         [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                         [<c0107fb0>] mask_and_ack_8259A+0x1b/0xcc
                         [<c015b007>] __do_IRQ+0x4f/0x113
                         [<c01062d3>] do_IRQ+0x8c/0xad
         in-softirq-W at:
                         [<c01468c4>] lockdep_acquire+0x68/0x84
                         [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
                         [<c0107fb0>] mask_and_ack_8259A+0x1b/0xcc
                         [<c015b007>] __do_IRQ+0x4f/0x113
                         [<c01062d3>] do_IRQ+0x8c/0xad
       }
       ... key      at: [<c142f174>] i8259A_lock+0x14/0x40
     ... acquired at:
     [<c01468c4>] lockdep_acquire+0x68/0x84
     [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
     [<c0107e6b>] disable_8259A_irq+0x10/0x47
     [<c011bdbd>] startup_edge_ioapic_vector+0x31/0x58
     [<c015b45e>] setup_irq+0xe4/0x14d
     [<c015b5a1>] request_irq+0xda/0xf9
     [<c1ac983a>] rtc_init+0x6a/0x1a7
     [<c0100457>] init+0x14a/0x2cb
     [<c0102005>] kernel_thread_helper+0x5/0xb

   ... acquired at:
     [<c01468c4>] lockdep_acquire+0x68/0x84
     [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
     [<c011bce1>] mask_IO_APIC_irq+0x11/0x31
     [<c011c5cc>] ack_edge_ioapic_vector+0x31/0x41
     [<c015b007>] __do_IRQ+0x4f/0x113
     [<c01062d3>] do_IRQ+0x8c/0xad

  the hard-irq-unsafe lock's dependencies:
  -> (vector_lock){--..} ops: 31 {
     initial-use  at:
                          [<c01468c4>] lockdep_acquire+0x68/0x84
                          [<c10485e9>] _spin_lock+0x21/0x2f
                          [<c011b5e8>] assign_irq_vector+0x34/0xc8
                          [<c1aa82fa>] setup_IO_APIC+0x45a/0xcff
                          [<c1aa56e3>] smp_prepare_cpus+0x5ea/0x8aa
                          [<c010033f>] init+0x32/0x2cb
                          [<c0102005>] kernel_thread_helper+0x5/0xb
     softirq-on-W at:
                          [<c01468c4>] lockdep_acquire+0x68/0x84
                          [<c10485e9>] _spin_lock+0x21/0x2f
                          [<c011b5e8>] assign_irq_vector+0x34/0xc8
                          [<c1aa82fa>] setup_IO_APIC+0x45a/0xcff
                          [<c1aa56e3>] smp_prepare_cpus+0x5ea/0x8aa
                          [<c010033f>] init+0x32/0x2cb
                          [<c0102005>] kernel_thread_helper+0x5/0xb
     hardirq-on-W at:
                          [<c01468c4>] lockdep_acquire+0x68/0x84
                          [<c10485e9>] _spin_lock+0x21/0x2f
                          [<c011b5e8>] assign_irq_vector+0x34/0xc8
                          [<c1aa82fa>] setup_IO_APIC+0x45a/0xcff
                          [<c1aa56e3>] smp_prepare_cpus+0x5ea/0x8aa
                          [<c010033f>] init+0x32/0x2cb
                          [<c0102005>] kernel_thread_helper+0x5/0xb
   }
   ... key      at: [<c1432574>] vector_lock+0x14/0x3c

  stack backtrace:
   [<c0104f36>] show_trace+0xd/0xf
   [<c010543e>] dump_stack+0x17/0x19
   [<c0144e34>] check_usage+0x1f6/0x203
   [<c0146395>] __lockdep_acquire+0x8c2/0xaa5
   [<c01468c4>] lockdep_acquire+0x68/0x84
   [<c10487f4>] _spin_lock_irqsave+0x2a/0x3a
   [<c04ff8d2>] startup_msi_irq_wo_maskbit+0x10/0x35
   [<c015b932>] probe_irq_on+0x44/0x107
   [<c0673d58>] serial8250_config_port+0x84b/0x986
   [<c06707b1>] uart_add_one_port+0x105/0x289
   [<c1ace54b>] serial8250_init+0xc3/0x10a
   [<c0100457>] init+0x14a/0x2cb
   [<c0102005>] kernel_thread_helper+0x5/0xb

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Jan Beulich <jbeulich@novell.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: remove unused gart header file

include/asm-x86_64/gart-mapping.h is only ever used in
arch/x86_64/kernel/setup.c and none of its contents are referenced.
Looks to be leftover cruft not removed in the dma_ops patch.

Signed-off-by: Jon Mason <jdmason@us.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Optimize bitmap_weight for small bitmaps

Use inline code bitmaps <= BITS_PER_LONG in bitmap_weight. This
gives _much_ better code.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Remove ia32_sys_call_table export

It was originally added for 2.4 oprofile, but 2.6 oprofile doesn't
need that anymore. Shouldn't be any use in tree anymore and it doesn't
make much sense to export the ia32 syscalls when the main syscalls
are not exported.

I think Adrian Bunk asked for removing it several times.

Also included hunk from Adrian to remove the .globl ia32_sys_call_table

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Remove long obsolete CVS

Early development of x86-64 Linux was in CVS, but that hasn't been
the case for a long time now. Remove the obsolete $Id$s.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: nmi watchdog header cleanup

Misc header cleanup for nmi watchdog.

Signed-off-by: Don Zickus <dzickus@redhat.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: fix unlikely profiling & vsyscalls on x86_64

fix unlikely profiling in vsyscalls ...

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: add END()/ENDPROC() annotations to entry.S

Since END()/ENDPROC() are now available, add respective annotations to
x86_64's entry.S. This should help debugging activities.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Add compat_printk and sysctl to turn off compat layer warnings

Sometimes e.g. with crashme the compat layer warnings can be noisy.
Add a way to turn them off by gating all output through compat_printk
that checks a global sysctl. The default is not changed.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Use -ENODEV in IOMMU initialization

Fix

initcall at 0xffffffff806c5b89: pci_iommu_init+0x0/0x53c(): returned with error code -1

Return -ENODEV instead when the IOMMU is not used.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386/x86-64: simplify ioapic_register_intr()

Simplify (remove duplication of) code in ioapic_register_intr().

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: serialize assign_irq_vector() use of static variables

Since assign_irq_vector() can be called at runtime, its access of static
variables should be protected by a lock.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Clean and enhance up K8 northbridge access code

 - Factor out the duplicated access/cache code into a single file
   * Shared between i386/x86-64.
 - Share flush code between AGP and IOMMU
   * Fix a bug: AGP didn't wait for end of flush before
 - Drop 8 northbridges limit and allocate dynamically
 - Add lock to serialize AGP and IOMMU GART flushes
 - Add PCI ID for next AMD northbridge
 - Random related cleanups

The old K8 NUMA discovery code is unchanged. New systems
should all use SRAT for this.

Cc: "Navin Boppuri" <navin.boppuri@newisys.com>
Cc: Dave Jones <davej@redhat.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: trivial gart clean-up

A trivial change to have gart_unmap_sg call gart_unmap_single directly,
instead of bouncing through the dma_unmap_single wrapper in
dma-mapping.h.

This change required moving the gart_unmap_single above gart_unmap_sg,
and under gart_map_single (which seems a more logical place that its
current location IMHO).

Signed-off-by: Jon Mason <jdmason@us.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Implement compat functions for PTRACE_SETSIGINFO/GETSIGINFO

Previously we would just silently provide 64 bit services
for this to 32bit processes.

I also added all the other cases explicitely to the ptrace
compat wrapper to make sure this doesn't happen again.

And removed one bogus check in the wrapper.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: iommu_gart_bitmap search to cross next_bit

Allow search for a contiguous block of iommu space to cross the next_bit
marker if we have already committed ourselves to flushing the gart.

There shouldn't be any reason why we'd restrict the search.

Signed-off-by: Mike Waychison <mikew@google.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Remove bogus comment in topology.h

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: x86_64-enable-large-bzImage.patch

enable large bzImages on x86_64. (fix is from x86's build.c) Using this
patch i have successfully built and booted an allyesconfig 13MB+ bzImage
on x86_64 too:

 $ size64 vmlinux
    text    data     bss     dec     hex filename
 23444831        8202642 3439360 35086833        21761f1 vmlinux

 -rw-rw-r--  1 mingo mingo 13121740 Apr 19 09:32 arch/x86_64/boot/bzImage

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Remove most of ia32_unistd.h

It's only needed for three system calls, no need to maintain
a full list forever.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: pci-dma.c clean-up - trivial

Replace hard coded DMA masks with #defines from
include/linux/dma-mapping.h

Signed-off-by: Jon Mason <jdmason@us.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: x86_64 version of the smp alternative patch.

Changes are largely identical to the i386 version:

 * alternative #define are moved to the new alternative.h file.
 * one new elf section with pointers to the lock prefixes which can be
   nop'ed out for non-smp.
 * two new elf sections simliar to the "classic" alternatives to
   replace SMP code with simpler UP code.
 * fixup headers to use alternative.h instead of defining their own
   LOCK / LOCK_PREFIX macros.

The patch reuses the i386 version of the alternatives code to avoid code
duplication.  The code in alternatives.c was shuffled around a bit to
reduce the number of #ifdefs needed.  It also got some tweaks needed for
x86_64 (vsyscall page handling) and new features (noreplacement option
which was x86_64 only up to now).  Debug printk's are changed from
compile-time to runtime.

Loosely based on a early version from Bastian Blank <waldi@debian.org>

Signed-off-by: Gerd Hoffmann <kraxel@suse.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386/x86-64: Emulate CPUID4 on AMD

Intel systems report the cache level data from CPUID 4 in sysfs.
Add a CPUID 4 emulation for AMD CPUs to report the same
information for them. This allows programs to read this
information in a uniform way.

The AMD way to report this is less flexible so some assumptions
are hardcoded (e.g. no L3)

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] i386/x86-64: Use new official CPUID to get APICID/core split on AMD platforms

Previously the apicid<->coreid split was computed based on the max
number of cores. Now use a new CPUID AMD defined for that. On most
systems right now it should be 0 and the old method will be used.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Use __always_inline for __inline_memcpy

Inspired from i386 changes

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Use local APIC ID from local APIC instead of CPUID

vSMPowered systems use apic_cluster too.  Forcing apic_physflat works
on these systems too, but only if we change phys_pkg_id to use
hard_smp_prcoessor_id() instead of cpuid_ebx.  I am guessing other
multichassi cluster systems would need this too.

Signed-off-by: ravikiran thirumalai <kiran@scalex86.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] x86_64: Update defconfig

Enable some hwmon drivers as modules and tulip and stack unwinding

Kernel image should be somewhat bigger now because of the unwind
information being included, but you'll get exact backtraces for that.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Merge /pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SPARC]: Add iomap interfaces.
  [OPENPROM]: Rewrite driver to use in-kernel device tree.
  [OPENPROMFS]: Rewrite using in-kernel device tree and seq_file.
  [SPARC]: Add unique device_node IDs and a ".node" property.
  [SPARC]: Add of_set_property() interface.
  [SPARC64]: Export auxio_register to modules.
  [SPARC64]: Add missing interfaces to dma-mapping.h
  [SPARC64]: Export _PAGE_IE to modules.
  [SPARC64]: Allow floppy driver to build modular.
  [SPARC]: Export x_bus_type to modules.
  [RIOWATCHDOG]: Fix the build.
  [CPWATCHDOG]: Fix the build.
  [PARPORT] sunbpp: Fix typo.
  [MTD] sun_uflash: Port to new EBUS device layer.

Merge /pub/scm/linux/kernel/git/davem/net-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6: (42 commits)
  [IOAT]: Do not dereference THIS_MODULE directly to set unsafe.
  [NETROM]: Fix possible null pointer dereference.
  [NET] netpoll: break recursive loop in netpoll rx path
  [NET] netpoll: don't spin forever sending to stopped queues
  [IRDA]: add some IBM think pads
  [ATM]: atm/mpc.c warning fix
  [NET]: skb_find_text ignores to argument
  [NET]: make net/core/dev.c:netdev_nit static
  [NET]: Fix GSO problems in dev_hard_start_xmit()
  [NET]: Fix CHECKSUM_HW GSO problems.
  [TIPC]: Fix incorrect correction to discovery timer frequency computation.
  [TIPC]: Get rid of dynamically allocated arrays in broadcast code.
  [TIPC]: Fixed link switchover bugs
  [TIPC]: Enhanced & cleaned up system messages; fixed 2 obscure memory leaks.
  [TIPC]: First phase of assert() cleanup
  [TIPC]: Disallow config operations that aren't supported in certain modes.
  [TIPC]: Fixed memory leak in tipc_link_send() when destination is unreachable
  [TIPC]: Added missing warning for out-of-memory condition
  [TIPC]: Withdrawing all names from nameless port now returns success, not error
  [TIPC]: Optimized argument validation done by connect().
  ...

typo fixes

Signed-off-by: Adrian Bunk <bunk@stusta.de>

[PATCH] sched: fix SCHED_FIFO bug in sys_sched_rr_get_interval()

The introduction of SCHED_BATCH scheduling class with a value of 3 means
that the expression (p->policy & SCHED_FIFO) will return true if policy
is SCHED_BATCH or SCHED_FIFO.

Unfortunately, this expression is used in sys_sched_rr_get_interval()
and in the absence of a comment to say that this is intentional I
presume that it is unintentional and erroneous.

The fix is to change the expression to (p->policy == SCHED_FIFO).

Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Clean up 'inline is not at beginning' warnings for usb storage

Usually we don't care much about 'gcc -W' warnings, but some of us do build
kernels that way to look for problems, and then the fewer warnings we have
to wade through the better. Especially when they are very easy and
non-intrusive to clean up. Which is the case for the following warnings
spewed by drivers/usb/storage/usb.h :

  drivers/usb/storage/usb.h:163: warning: `inline' is not at beginning of
+declaration
  drivers/usb/storage/usb.h:166: warning: `inline' is not at beginning of
+declaration

There's also some precedence for cleaning up these warnings. I've had
a few patches merged in the past that remove exactly this class of
warnings.

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Adrian Bunk <bunk@stusta.de>

[PATCH] initramfs overwrite fix

This patch ensures that initramfs overwrites work correctly, even when dealing
with device nodes of different types.  Furthermore, when replacing a file
which already exists, we must make very certain that we truncate the existing
file.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Cc: Michael Neuling <mikey@neuling.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] drivers/md/md.c: make code static

Make needlessly global code static.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Allow the write_mostly flag to be set via sysfs

It appears in /sys/mdX/md/dev-YYY/state
and can be set or cleared by writing 'writemostly' or '-writemostly'
respectively.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Allow resync_start to be set and queried via sysfs

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Allow raid 'layout' to be read and set via sysfs

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Allow rdev state to be set via sysfs

The md/dev-XXX/state file can now be written:

 "faulty" simulates an error on the device
 "remove" removes the device from the array (if it is not busy)

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Set/get state of array via sysfs

This allows the state of an md/array to be directly controlled via sysfs and
adds the ability to stop and array without tearing it down.

Array states/settings:

 clear
     No devices, no size, no level
     Equivalent to STOP_ARRAY ioctl
 inactive
     May have some settings, but array is not active
        all IO results in error
     When written, doesn't tear down array, but just stops it
 suspended (not supported yet)
     All IO requests will block. The array can be reconfigured.
     Writing this, if accepted, will block until array is quiescent
 readonly
     no resync can happen.  no superblocks get written.
     write requests fail
 read-auto
     like readonly, but behaves like 'clean' on a write request.

 clean - no pending writes, but otherwise active.
     When written to inactive array, starts without resync
     If a write request arrives then
       if metadata is known, mark 'dirty' and switch to 'active'.
       if not known, block and switch to write-pending
     If written to an active array that has pending writes, then fails.
 active
     fully active: IO and resync can be happening.
     When written to inactive array, starts with resync

 write-pending (not supported yet)
     clean, but writes are blocked waiting for 'active' to be written.

 active-idle
     like active, but no writes have been seen for a while (100msec).

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Don't write dirty/clean update to spares - leave them alone

- record the 'event' count on each individual device (they
  might sometimes be slightly different now)
- add a new value for 'sb_dirty': '3' means that the super
  block only needs to be updated to record a clean<->dirty
  transition.
- Prefer odd event numbers for dirty states and even numbers
  for clean states
- Using all the above, don't update the superblock on
  a spare device if the update is just doing a clean-dirty
  transition.  To accomodate this, a transition from
  dirty back to clean might now decrement the events counter
  if nothing else has changed.

The net effect of this is that spare drives will not see any IO requests
during normal running of the array, so they can go to sleep if that is what
they want to do.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Allow re-add to work on array without bitmaps

When an array has a bitmap, a device can be removed and re-added and only
blocks changes since the removal (as recorded in the bitmap) will be resynced.

It should be possible to do a similar thing to arrays without bitmaps.  i.e.
if a device is removed and re-added and *no* changes have been made in the
interim, then the add should not require a resync.

This patch allows that option.  This means that when assembling an array one
device at a time (e.g.  during device discovery) the array can be enabled
read-only as soon as enough devices are available, but extra devices can still
be added without causing a resync.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Fix bug that stops raid5 resync from happening

As data_disks is *less* than raid_disks, the current test here is obviously
wrong.  And as the difference is already available in conf->max_degraded, it
makes much more sense to use that.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Fix Kconfig error

RAID5 recently changed to RAID456

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: md Kconfig speeling feex

I was experimenting with Linux SW raid today and found a spelling error when
reading the help menus...  (and fly spell found more).

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Calculate correct array size for raid10 in new offset mode

The size calculation made assumtion which the new offset mode didn't
follow.  This gets the size right in all cases.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: Change md/bitmap file handling to use bmap to file blocks-fix

Fix problems with new bmap based access to bitmap files.

1/ When not using a file based bitmap, attach a NULL list of buffers
   to each page so the common free_buffer routine can cope.
2/ Use submit_bh to read as well as write, rather than vfs_read.
   This makes read and write more symetric.
3/ sync the file before reading, to ensure that the page cache has no
   dirty pages that might get written out later.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md/bitmap: change md/bitmap file handling to use bmap to file blocks

If md is asked to store a bitmap in a file, it tries to hold onto the page
cache pages for that file, manipulate them directly, and call a cocktail of
operations to write the file out.  I don't believe this is a supportable
approach.

This patch changes the approach to use the same approach as swap files.  i.e.
bmap is used to enumerate all the block address of parts of the file and we
write directly to those blocks of the device.

swapfile only uses parts of the file that provide a full pages at contiguous
addresses.  We don't have that luxury so we have to cope with pages that are
non-contiguous in storage.  To handle this we attach buffers to each page, and
store the addresses in those buffers.

With this approach the pagecache may contain data which is inconsistent with
what is on disk.  To alleviate the problems this can cause, md invalidates the
pagecache when releasing the file.  If the file is to be examined while the
array is active (a non-critical but occasionally useful function), O_DIRECT io
must be used.  And new version of mdadm will have support for this.

This approach simplifies a lot of code:
 - we no longer need to keep a list of pages which we need to wait for,
   as the b_endio function can keep track of how many outstanding
   writes there are.  This saves a mempool.
 - -EAGAIN returns from write_page are no longer possible (not sure if
    they ever were actually).

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md/bitmap: tidy up i_writecount handling in md/bitmap

md/bitmap modifies i_writecount of a bitmap file to make sure that no-one else
writes to it.  The reverting of the change is sometimes done twice, and there
is one error path where it is omitted.

This patch tidies that up.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md/bitmap: remove dead code from md/bitmap

bitmap_active is never called, and the BITMAP_ACTIVE flag is never users or
tested, so discard them both.

Also remove some out-of-date 'todo' comments.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md/bitmap: remove unnecessary page reference manipulations from md/bitmap code

md/bitmap gets a collection of pages representing the bitmap when it
initialises the bitmap, and puts all the references when discarding the
bitmap.

It also occasionally takes extra references without any good reason, and
sometimes drops them ...  though it doesn't always drop them, which can result
in a memory leak.

This patch removes the unnecessary 'get_page' calls, and the corresponding
'put_page' calls.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md/bitmap: use set_bit etc for bitmap page attributes

In particular, this means that we use 4 bits per page instead of a whole
unsigned long.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md/bitmap: cleaner separation of page attribute handlers in md/bitmap

md/bitmap has some attributes per-page.  Handling of these attributes in
largely abstracted in set_page_attr and clear_page_attr.  However
get_page_attr exposes the format used to store them.  So prior to changing
that format, introduce test_page_attr instead of get_page_attr, and make
appropriate usage changes.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md/bitmap: remove bitmap writeback daemon

md/bitmap currently has a separate thread to wait for writes to the bitmap
file to complete (as we cannot get a callback on that action).

However this isn't needed as bitmap_unplug is called from process context and
waits for the writeback thread to do it's work.  The same result can be
achieved by doing the waiting directly in bitmap_unplug.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md/bitmap: fix online removal of file-backed bitmaps

When "mdadm --grow /dev/mdX --bitmap=none" is used to remove a filebacked
bitmap, the bitmap was disconnected from the array, but the file wasn't closed
(until the array was stopped).

The file also wasn't closed if adding the bitmap file failed.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: split reshape portion of raid5 sync_request into a separate function

... as raid5 sync_request is WAY too big.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: make md_print_devices() static

This patch makes the needlessly global md_print_devices() static.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: support stripe/offset mode in raid10

The "industry standard" DDF format allows for a stripe/offset layout where
data is duplicated on different stripes.  e.g.

  A  B  C  D
  D  A  B  C
  E  F  G  H
  H  E  F  G

(columns are drives, rows are stripes, LETTERS are chunks of data).

This is similar to raid10's 'far' mode, but not quite the same.  So enhance
'far' mode with a 'far/offset' option which follows the layout of DDFs
stripe/offset.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: allow a linear array to have drives added while active

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] md: allow checkpoint of recovery with version-1 superblock

For a while we have had checkpointing of resync.  The version-1 superblock
allows recovery to be checkpointed as well, and this patch implements that.

Due to early carelessness we need to add a feature flag to signal that the
recovery_offset field is in use, otherwise older kernels would assume that a
partially recovered array is in fact fully recovered.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>