]> err.no Git - linux-2.6/log
linux-2.6
16 years agox86: provide paravirtualized hook for rdtscp
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:32:05 +0000 (13:32 +0100)]
x86: provide paravirtualized hook for rdtscp

This patch adds a field in pv_cpu_ops for a paravirtualized hook
for rdtscp, needed for x86_64.

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: cleanup write_tsc
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:32:05 +0000 (13:32 +0100)]
x86: cleanup write_tsc

write_tsc() does not need to be enclosed in any paravirt closure,
as it uses wrmsr(). So we rip off the duplicate in msr.h
and the definition from paravirt.h

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: adjust PVOP_CALL/VCALL macros for 64-bit
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:32:05 +0000 (13:32 +0100)]
x86: adjust PVOP_CALL/VCALL macros for 64-bit

This patch adjust the PVOP_VCALL and PVOP_CALL macros to
work with x86_64. It has a different calling convention, and
we use auxiliary macros to account for both calling conventions
as cleanly as possible

Comments are adjusted accordingly.

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: change paravirt_32.c name
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:32:04 +0000 (13:32 +0100)]
x86: change paravirt_32.c name

This patch changes paravirt_32.c to paravirt.c. The goal
is to have paravirt support in x86_64, so we do it in a common file

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, ptrace: overflow signal API
Markus Metzger [Wed, 30 Jan 2008 12:32:04 +0000 (13:32 +0100)]
x86, ptrace: overflow signal API

Establish the user API for sending a user-defined signal to the traced task on a BTS buffer overflow.

This should complete the user API for the BTS ptrace extension.
The patches so far implement wrap-around overflow handling as is needed for debugging.

The remaining open is another overflow handling mechanism that sends a signal to the traced task on a buffer overflow.
This will take some more time from my side.

Since, from a user perspective, this occurs behind the scenes, the patch set should already be useful. More features may/will be added on top of it (overflow signal, pageable back-up buffers, kernel tracing, core file support, profiling, ...).

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, ptrace: add buffer size checks
Markus Metzger [Wed, 30 Jan 2008 12:32:03 +0000 (13:32 +0100)]
x86, ptrace: add buffer size checks

Pass the buffer size for (most) ptrace commands that pass user-allocated buffers and check that size before accessing the buffer. Unfortunately, PTRACE_BTS_GET already uses all 4 parameters.
Commands that access user buffers return the number of bytes or records read or written.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, ptrace: support 32bit-cross-64bit BTS recording
Markus Metzger [Wed, 30 Jan 2008 12:32:03 +0000 (13:32 +0100)]
x86, ptrace: support 32bit-cross-64bit BTS recording

Support BTS recording of 32bit and 64bit tasks from 32bit or 64bit tasks.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, ptrace: rlimit BTS buffer allocation
Markus Metzger [Wed, 30 Jan 2008 12:32:03 +0000 (13:32 +0100)]
x86, ptrace: rlimit BTS buffer allocation

Check the rlimit of the tracing task for total and locked memory when allocating the BTS buffer.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: move deeply indented code to reenter_kprobe
Masami Hiramatsu [Wed, 30 Jan 2008 12:32:02 +0000 (13:32 +0100)]
x86: move deeply indented code to reenter_kprobe

Move some deeply indented code related to re-entrance processing
from kprobe_handler() to reenter_kprobe().

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add reenter_kprobe helper
Harvey Harrison [Wed, 30 Jan 2008 12:32:02 +0000 (13:32 +0100)]
x86: add reenter_kprobe helper

[ mhiramat@redhat.com: updated it to latest x86.git ]

Factor common X86_32, X86_64 kprobe reenter logic from deeply
indented section to helper function.

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>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
16 years agox86: fix kprobe_handler reenable preemption
Masami Hiramatsu [Wed, 30 Jan 2008 12:32:01 +0000 (13:32 +0100)]
x86: fix kprobe_handler reenable preemption

Fix a preemption bug in kprobe_handler(). It has to call preempt_enable()
before returning.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: voluntary leave_mm before entering ACPI C3
Venki Pallipadi [Wed, 30 Jan 2008 12:32:01 +0000 (13:32 +0100)]
x86: voluntary leave_mm before entering ACPI C3

Aviod TLB flush IPIs during C3 states by voluntary leave_mm()
before entering C3.

The performance impact of TLB flush on C3 should not be significant with
respect to C3 wakeup latency. Also, CPUs tend to flush TLB in hardware while in
C3 anyways.

On a 8 logical CPU system, running make -j2, the number of tlbflush IPIs goes
down from 40 per second to ~ 0. Total number of interrupts during the run
of this workload was ~1200 per second, which makes it ~3% savings in wakeups.

There was no measurable performance or power impact however.

[ akpm@linux-foundation.org: symbol export fixes. ]

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: add some pirq debugging
Ingo Molnar [Wed, 30 Jan 2008 12:32:00 +0000 (13:32 +0100)]
x86: add some pirq debugging

we use a few static mapping rules in our pirq routing functions,
and for example regression f3ac84324fd94 was due to the pirq
being out of range of the remapping array. Put in a few
WARN_ON_ONCE() lines so that we get notified about any such
out-of-bound incidents.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoPCI: remove default PCI expansion ROM memory allocation
Gary Hade [Wed, 30 Jan 2008 12:31:59 +0000 (13:31 +0100)]
PCI: remove default PCI expansion ROM memory allocation

increasing number of PCI slots in large multi-node systems.  The kernel
currently attempts by default to allocate memory for all PCI expansion
ROMs so there has also been an increasing number of PCI memory
allocation failures seen on these systems.  This occurs because the BIOS
either (1) provides insufficient PCI memory resource for all the
expansion ROMs or (2) provides adequate PCI memory resource for
expansion ROMs but provides the space in kernel unexpected BIOS assigned
P2P non-prefetch windows.

The resulting PCI memory allocation failures may be benign when related
to memory requests for expansion ROMs themselves but in some cases they
can occur when attempting to allocate space for more critical BARs.
This can happen when a successful expansion ROM allocation request
consumes memory resource that was intended for a non-ROM BAR.  We have
seen this happen during PCI hotplug of an adapter that contains a P2P
bridge where successful memory allocation for an expansion ROM BAR on
device behind the bridge consumed memory that was intended for a non-ROM
BAR on the P2P bridge.  In all cases the allocation failure messages can
be very confusing for users.

This patch addresses the issue by changing the kernel default behavior
so that expansion ROM memory allocations are no longer attempted by
default when the BIOS has not assigned a specific address range to the
expansion ROM BAR.  This was done by changing the 'pci=rom' boot option
behavior for BIOS unassigned expansion ROMs to actually match it's
current kernel-parameters.txt description which already implies "off" by
default. Behavior for BIOS assigned expansion ROMs implemented in
pcibios_assign_resources() [arch/x86/pci/i386.c] is unchanged.

Signed-off-by: Gary Hade <garyhade@us.ibm.com>
Cc: Greg KH <greg@kroah.com>
Cc: Jan Beulich <jbeulich@novell.com>
Acked-by: "Jun'ichi Nomura" <j-nomura@ce.jp.nec.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 dmi_alloc() to not advance alloc index in case of
Parag Warudkar [Wed, 30 Jan 2008 12:31:59 +0000 (13:31 +0100)]
x86: fix dmi_alloc() to not advance alloc index in case of

dmi_alloc() for CONFIG_X86_64 is defined to allocate from a static array
and it maintains a allocation index which is advanced each time allocation
is attempted - it gets incremented even if an allocation fails thereby
depriving any future request that may be small enough to be satisfied from
the array.

Fix this by first testing if allocation is going to be possible and
incrementing alloc index only then.

Signed-off-by: Parag Warudkar <parag.warudkar@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: fix DMI out of memory problems
Parag Warudkar [Wed, 30 Jan 2008 12:31:59 +0000 (13:31 +0100)]
x86: fix DMI out of memory problems

People with HP Desktops (including me) encounter couple of DMI errors
during boot - dmi_save_oem_strings_devices: out of memory and
dmi_string: out of memory.

On some HP desktops the DMI data include OEM strings (type 11) out of
which only few are meaningful and most other are empty. DMI code
religiously creates copies of these 27 strings (65 bytes each in my
case) and goes OOM in dmi_string().

If DMI_MAX_DATA is bumped up a little then it goes and fails in
dmi_save_oem_strings while allocating dmi_devices of sizeof(struct
dmi_device) corresponding to these strings.

On x86_64 since we cannot use alloc_bootmem this early, the code uses a
static array of 2048 bytes (DMI_MAX_DATA) for allocating the memory DMI
needs. It does not survive the creation of empty strings and devices.

Fix this by detecting and not newly allocating empty strings and instead
using a one statically defined dmi_empty_string.

Also do not create a new struct dmi_device for each empty string - use
one statically define dmi_device with .name=dmi_empty_string and add
that to the dmi_devices list.

On x64 this should stop the OOM with same current size of DMI_MAX_DATA
and on x86 this should save a good amount of (27*65 bytes +
27*sizeof(struct dmi_device) bootmem.

Compile and boot tested on both 32-bit and 64-bit x86.

Signed-off-by: Parag Warudkar <parag.warudkar@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: remove unused include/asm-x86/processor_32/64.h
Ingo Molnar [Wed, 30 Jan 2008 12:31:58 +0000 (13:31 +0100)]
x86: remove unused include/asm-x86/processor_32/64.h

remove unused include/asm-x86/processor_32/64.h.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: finish processor.h integration
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:57 +0000 (13:31 +0100)]
x86: finish processor.h integration

What's left in processor_32.h and processor_64.h cannot be cleanly
integrated. However, it's just a couple of definitions. They are moved
to processor.h around ifdefs, and the original files are deleted. Note that
there's much less headers included in the final 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: remove __init modifier from header declaration
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:57 +0000 (13:31 +0100)]
x86: remove __init modifier from header declaration

This patch removes the __init modifier from an extern function
declaration in acpi.h.

Besides not being strictly needed, it requires the inclusion of
linux/init.h, which is usually not even included directly, increasing
header mess by a lot.

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: x86 ptrace generic requests
Roland McGrath [Wed, 30 Jan 2008 12:31:56 +0000 (13:31 +0100)]
x86: x86 ptrace generic requests

This removes duplicated code by calling the generic ptrace_request and
compat_ptrace_request functions for the things they already handle.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 core dump TLS
Roland McGrath [Wed, 30 Jan 2008 12:31:56 +0000 (13:31 +0100)]
x86: x86 core dump TLS

This makes ELF core dumps of 32-bit processes include a new
note type NT_386_TLS (0x200) giving the contents of the TLS
slots in struct user_desc format.  This lets post mortem
examination figure out what the segment registers mean like
the debugger does with get_thread_area on a live process.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ia32_binfmt removal
Roland McGrath [Wed, 30 Jan 2008 12:31:56 +0000 (13:31 +0100)]
x86: x86 ia32_binfmt removal

Remove the old ia32_binfmt.c file, which is no longer used.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: compat_binfmt_elf
Roland McGrath [Wed, 30 Jan 2008 12:31:55 +0000 (13:31 +0100)]
x86: compat_binfmt_elf

This switches x86-64's 32-bit ELF support to use the shared
fs/compat_binfmt_elf.c code instead of our own ia32_binfmt.c.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 user_regset cleanup
Roland McGrath [Wed, 30 Jan 2008 12:31:55 +0000 (13:31 +0100)]
x86: x86 user_regset cleanup

This removes a bunch of dead code that is no longer needed now
that the user_regset interfaces are being used for all these jobs.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 ptrace user_regset
Roland McGrath [Wed, 30 Jan 2008 12:31:54 +0000 (13:31 +0100)]
x86: x86 ptrace user_regset

This cleans up the PTRACE_*REGS* request code so each one is just a
simple call to copy_regset_to_user or copy_regset_from_user.  The
ptrace layouts already match the user_regset formats (core dump formats).

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 CORE_DUMP_USE_REGSET
Roland McGrath [Wed, 30 Jan 2008 12:31:54 +0000 (13:31 +0100)]
x86: x86 CORE_DUMP_USE_REGSET

This switches x86 to the user_regset-based code for ELF core dumps.
The core dumps come out exactly the same as before.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 user_regset_view
Roland McGrath [Wed, 30 Jan 2008 12:31:53 +0000 (13:31 +0100)]
x86: x86 user_regset_view

This defines task_user_regset_view and the tables
describing the x86 user_regset layouts for 32 and 64.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 user_regset general regs
Roland McGrath [Wed, 30 Jan 2008 12:31:52 +0000 (13:31 +0100)]
x86: x86 user_regset general regs

This adds accessor functions in the user_regset style for
the general registers (struct user_regs_struct).

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 user_regset TLS
Roland McGrath [Wed, 30 Jan 2008 12:31:52 +0000 (13:31 +0100)]
x86: x86 user_regset TLS

This adds accessor functions in the user_regset style for the TLS data.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 TLS desc_struct cleanup
Roland McGrath [Wed, 30 Jan 2008 12:31:51 +0000 (13:31 +0100)]
x86: x86 TLS desc_struct cleanup

This cleans up the TLS code to use struct desc_struct and to separate the
encoding and installation magic from the interface wrappers.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 i387 cleanup
Roland McGrath [Wed, 30 Jan 2008 12:31:51 +0000 (13:31 +0100)]
x86: x86 i387 cleanup

This removes all the old code that is no longer used after
the i387 unification and cleanup.  The i387_64.h is renamed
to i387.h with no changes, but since it replaces the nonempty
one-line stub i387.h it looks like a big diff and not a rename.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 i387 user_regset
Roland McGrath [Wed, 30 Jan 2008 12:31:50 +0000 (13:31 +0100)]
x86: x86 i387 user_regset

This revamps the i387 code to be shared across 32-bit, 64-bit,
and 32-on-64.  It does so by consolidating the code in one place
based on the user_regset accessor interfaces.  This switches
32-bit to using the i387_64.h header and 64-bit to using the
i387.c that was previously i387_32.c, but that's what took the
least cleanup in each file.  Here i387.h is stubbed to always
include i387_64.h rather than renaming the file, to keep this
diff smaller and easier to read.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: i387 renaming
Roland McGrath [Wed, 30 Jan 2008 12:31:50 +0000 (13:31 +0100)]
x86: i387 renaming

This renames arch/x86/kernel/{i387_32.c => i387.c}.
This is a pure renaming, but paves the way for merging
the 32-bit and 64-bit versions of this code.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 user_regset math_emu
Roland McGrath [Wed, 30 Jan 2008 12:31:49 +0000 (13:31 +0100)]
x86: x86 user_regset math_emu

This converts the ptrace/signal accessors for i387 math_emu
state to the user_regset interface style, and calls these
from the old interfaces.

It also cleans up math_emulate's ptrace check to be a
single-step check, which is what it really wants.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 i387 header cleanup
Roland McGrath [Wed, 30 Jan 2008 12:31:49 +0000 (13:31 +0100)]
x86: x86 i387 header cleanup

This moves some code into asm-x86/i387_64.h in preparation for
unifying this code between 32 and 64.  The 32-bit versions of
some things are copied in some existing names changed to match
32-bit names and share code.  For 64, save_i387 is moved into
an inline from i387_64.c; this matches restore_i387, which is
already an inline, and makes sense since there is exactly one
caller (in signal_64.c).  The save_i387 function could use more
cosmetic cleanup, but it is just moved verbatim in this patch.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86 i387 unify structs
Roland McGrath [Wed, 30 Jan 2008 12:31:48 +0000 (13:31 +0100)]
x86: x86 i387 unify structs

The i387_fxsave_struct formats really have the same layout
on 32 and 64, with only some slightly different use of a few
fields.  The i387_fsave_struct and i387_soft_struct formats
are never used by 64-bit kernels, but it doesn't hurt to
have the unused types in the union and cuts down on the
amount of #ifdef hair required throughout the i387 code.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: x86-64 wired cpu_has macros
Roland McGrath [Wed, 30 Jan 2008 12:31:48 +0000 (13:31 +0100)]
x86: x86-64 wired cpu_has macros

This adds hard-wired definitions for the remaining cpu_has_* macros
that correspond to flags required-features.h demands are set for
64-bit.  Using these can efficiently avoid some #ifdef's when
merging 32-bit and 64-bit code together.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: compat_sys_ptrace
Roland McGrath [Wed, 30 Jan 2008 12:31:48 +0000 (13:31 +0100)]
x86: compat_sys_ptrace

This adds a generic definition of compat_sys_ptrace that calls
compat_arch_ptrace, parallel to sys_ptrace/arch_ptrace.  Some
machines needing this already define a function by that name.
The new generic function is defined only on machines that
put #define __ARCH_WANT_COMPAT_SYS_PTRACE into asm/ptrace.h.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: compat_ptrace_request
Roland McGrath [Wed, 30 Jan 2008 12:31:47 +0000 (13:31 +0100)]
x86: compat_ptrace_request

This adds a compat_ptrace_request that is the analogue of ptrace_request
for the things that 32-on-64 ptrace implementations can share in common.
So far there are just a couple of requests handled generically.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: ptrace_request peekdata/pokedata
Roland McGrath [Wed, 30 Jan 2008 12:31:47 +0000 (13:31 +0100)]
x86: ptrace_request peekdata/pokedata

This makes ptrace_request handle {PEEK,POKE}{TEXT,DATA} directly.
Every arch_ptrace that could call generic_ptrace_peekdata already
has a default case calling ptrace_request, so this keeps things
simpler for the arch code.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: user_regset user-copy helpers
Roland McGrath [Wed, 30 Jan 2008 12:31:47 +0000 (13:31 +0100)]
x86: user_regset user-copy helpers

This defines two new inlines in linux/regset.h, for use in arch_ptrace
implementations and the like.  These provide simplified wrappers for using
the user_regset interfaces to copy thread regset data into the caller's
user-space memory.  The inlines are trivial, but make the common uses in
places such as ptrace implementation much more concise, easier to read, and
less prone to code-copying errors.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: compat_binfmt_elf Kconfig
Roland McGrath [Wed, 30 Jan 2008 12:31:46 +0000 (13:31 +0100)]
x86: compat_binfmt_elf Kconfig

This adds Kconfig and Makefile bits to build fs/compat_binfmt_elf.c,
just added.  Each arch that wants to use this file needs to add a
"select COMPAT_BINFMT_ELF" line in its Kconfig bits that enable COMPAT.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: compat_binfmt_elf
Roland McGrath [Wed, 30 Jan 2008 12:31:46 +0000 (13:31 +0100)]
x86: compat_binfmt_elf

This adds fs/compat_binfmt_elf.c, a wrapper around fs/binfmt_elf.c for
32-bit ELF support on 64-bit kernels.  It can replace all the hand-rolled
versions of this that each 32/64 arch has, which are all about the same.

To use this, an arch's asm/elf.h has to define at least a few compat_*
macros that parallel the various macros that fs/binfmt_elf.c uses for
native support.

There is no attempt to deal with compat macros for the core dump format
support.  To use this file, the arch has to define compat_gregset_t for
linux/elfcore-compat.h and #define CORE_DUMP_USE_REGSET.  The 32-bit
compatible formats should come automatically from task_user_regset_view
called on a 32-bit task.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: user_regset helpers
Roland McGrath [Wed, 30 Jan 2008 12:31:45 +0000 (13:31 +0100)]
x86: user_regset helpers

This adds some inlines to linux/regset.h intended for arch code to use in
its user_regset get and set functions.  These make it pretty easy to deal
with the interface's optional kernel-space or user-space pointers and its
generalized access to a part of the register data at a time.

In simple cases where the internal data structure matches the exported
layout (core dump format), a get function can be nothing but a call to
user_regset_copyout, and a set function a call to user_regset_copyin.

In other cases the exported layout is usually made up of a few pieces each
stored contiguously in a different internal data structure.  These helpers
make it straightforward to write a get or set function by processing each
contiguous chunk of the data in order.  The start_pos and end_pos arguments
are always constants, so these inlines collapse to a small amount of code.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoelf core dump: notes user_regset
Roland McGrath [Wed, 30 Jan 2008 12:31:45 +0000 (13:31 +0100)]
elf core dump: notes user_regset

This modifies the ELF core dump code under #ifdef CORE_DUMP_USE_REGSET.
It changes nothing when this macro is not defined.  When it's #define'd
by some arch header (e.g. asm/elf.h), the arch must support the
user_regset (linux/regset.h) interface for reading thread state.

This provides an alternate version of note segment writing that is based
purely on the user_regset interfaces.  When CORE_DUMP_USE_REGSET is set,
the arch need not define macros such as ELF_CORE_COPY_REGS and ELF_ARCH.
All that information is taken from the user_regset data structures.
The core dumps come out exactly the same if arch's definitions for its
user_regset details are correct.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoelf core dump: notes reorg
Roland McGrath [Wed, 30 Jan 2008 12:31:44 +0000 (13:31 +0100)]
elf core dump: notes reorg

This pulls out the code for writing the notes segment of an ELF core dump
into separate functions.  This cleanly isolates into one cluster of
functions everything that deals with the note formats and the hooks into
arch code to fill them.  The top-level elf_core_dump function itself now
deals purely with the generic ELF format and the memory segments.

This only moves code around into functions that can be inlined away.
It should not change any behavior at all.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: user_regset header
Roland McGrath [Wed, 30 Jan 2008 12:31:44 +0000 (13:31 +0100)]
x86: user_regset header

The new header <linux/regset.h> defines the types struct user_regset and
struct user_regset_view, with some associated declarations.  This new set
of interfaces will become the standard way for arch code to expose
user-mode machine-specific state.  A single set of entry points into arch
code can do all the low-level work in one place to fill the needs of core
dumps, ptrace, and any other user-mode debugging facilities that might come
along in the future.

For existing arch code to adapt to the user_regset interfaces, each arch
can work from the code it already has to support core files and ptrace.
The formats you want for user_regset are the core file formats.  The only
wrinkle in adapting old ptrace implementation code as user_regset get and
set functions is that these functions can be called on current as well as
on another task_struct that is stopped and switched out as for ptrace.
For some kinds of machine state, you may have to load it directly from CPU
registers or otherwise differently for current than for another thread.
(Your core dump support already handles this in elf_core_copy_regs for
current and elf_core_copy_task_regs for other tasks, so just check there.)
The set function should also be made to work on current in case that
entails some special cases, though this was never required before for
ptrace.  Adding this flexibility covers the arch needs to open the door to
more sophisticated new debugging facilities that don't always need to
context-switch to do every little thing.

The copyin/copyout helper functions (in a later patch) relieve the arch
code of most of the cumbersome details of the flexible get/set interfaces.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify module_{32|64}.h
Harvey Harrison [Tue, 18 Dec 2007 11:58:10 +0000 (03:58 -0800)]
x86: unify module_{32|64}.h

This adds one case to the MODULE_PROC_FAMILY block testing
for X86_64.  There are no new things defined on X86_64 than
there were before.

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: kprobes leftover cleanups
Harvey Harrison [Wed, 30 Jan 2008 12:31:43 +0000 (13:31 +0100)]
x86: kprobes leftover cleanups

Eliminate __always_inline, all of these static functions are
only called once.  Minor whitespace cleanup.  Eliminate one
supefluous return at end of void function.  Change the one
#ifndef to #ifdef to match the sense of the rest of the config
tests.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Acked-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoarch/x86/: spelling fixes
Joe Perches [Wed, 30 Jan 2008 12:31:42 +0000 (13:31 +0100)]
arch/x86/: spelling fixes

Spelling fixes.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: use helper in fault_64.c
Harvey Harrison [Wed, 30 Jan 2008 12:31:42 +0000 (13:31 +0100)]
x86: use helper in fault_64.c

Use the fixup_exception() helper in 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: unify extable_{32|64}.c
Harvey Harrison [Wed, 30 Jan 2008 12:31:41 +0000 (13:31 +0100)]
x86: unify extable_{32|64}.c

Introduce fixup_exception() on 64-bit and use it in kprobes to
eliminate an #ifdef.

Only 64-bit needs search_extable() due to a stepping bug.

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: move i387 definitions to processor.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:41 +0000 (13:31 +0100)]
x86: move i387 definitions to processor.h

This patch moves i387 definitions from processor_32.h and processor_64.h
to processor.h. They are different. Very different. And there's appearently
nothing we can do about it, so they're enclosed inside ifdefs.

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: unify asm nops
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:40 +0000 (13:31 +0100)]
x86: unify asm nops

There's only one difference between the NOPs used in asm code for i386 and x86_64:
i386 has a lot more variants. The code is moved to processor.h, and adjusted
accordingly.

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: unify prefetch operations
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:40 +0000 (13:31 +0100)]
x86: unify prefetch operations

This patch moves the prefetch[w]? functions to processor.h

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: move definitions to processor.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:39 +0000 (13:31 +0100)]
x86: move definitions to processor.h

This patch moves definitions that are present in only one of the files
(between processor_32.h and processor_64.h), to processor.h. They're mostly
structures and function definitions.

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: unify mm_segment_t definition
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:38 +0000 (13:31 +0100)]
x86: unify mm_segment_t definition

This patch moves the mm_segment_t structure definition to processor.h
This makes mmsegment.h file useless, and it is deleted.

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: remove legacy stuff from processor_64.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:37 +0000 (13:31 +0100)]
x86: remove legacy stuff from processor_64.h

This patch removes definitions and macros that are not used anymore
from processor_64.h

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: unify x86_cpuinfo struct.
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:33 +0000 (13:31 +0100)]
x86: unify x86_cpuinfo struct.

x86_cpuinfo is one more to the family of "not fundamentally different"
structs. It's unified in processor.h, with very specific fields enclosed
around ifdefs.

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: change bitwise operations to get a void parameter.
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:31 +0000 (13:31 +0100)]
x86: change bitwise operations to get a void parameter.

This patch changes the bitwise operations in bitops.h to get
a void pointers as a parameter. Before this patch, a lot of warnings
can be seen. They're gone after 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: unify TASK_ALIGN definitions
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:31 +0000 (13:31 +0100)]
x86: unify TASK_ALIGN definitions

This patch moves the TASK_ALIGN constraints to common header.
The base of it is the same for x86_64 and i386. The only difference
is the presence of vSMP in x86_64. As it's not a worry in i386, we can
safely use the same code for both.

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: unify thread struct.
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:31 +0000 (13:31 +0100)]
x86: unify thread struct.

The thread_struct is not fundamentally different between architectures, and
this patch puts it in the common header. What's really unique for each of
them is enclosed in ifdefs.

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 64-bit with a load_sp0 function.
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:31 +0000 (13:31 +0100)]
x86: provide 64-bit with a load_sp0 function.

Paravirt guests need to inform the underlying hypervisor whenever the sp0
tss field changes. i386 already has such a function, and we use it for
x86_64 too. There's an unnecessary (for 64-bit) msr handling part in the original
version, and it is placed around an ifdef. Making no more sense in
processor_32.h, it is moved to the common 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: unify tss_struct
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:31 +0000 (13:31 +0100)]
x86: unify tss_struct

Although slighly different, the tss_struct is very similar in x86_64 and
i386. The really different part, which matchs the hardware vision of it, is
now called x86_hw_tss, and each of the architectures provides yours.
It's then used as a field in the outter tss_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: unify current_text_addr
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:27 +0000 (13:31 +0100)]
x86: unify current_text_addr

current_text_addr() has a different implementation in x86_64 and
i386, but it is not fundamentally different. I stick to the i386
implementation, that seem to be a common base, and move it to processor.h

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: unify common parts of processor.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:27 +0000 (13:31 +0100)]
x86: unify common parts of processor.h

This patch moves the pieces of processor_32.h and processor_64 that are
equal to processor.h. Only what's exactly the same is moved around, the rest
not being touched.

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: move the definition of set_iopl_mask to common header
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:27 +0000 (13:31 +0100)]
x86: move the definition of set_iopl_mask to common header

This patch moves the definition of set_iopl_mask to processor.h,
instead of letting it at processor_32.h.
For x86_64, nothing is done, as we don't really need such a function.
However, having it on both arches saves us from putting an ifdef in the
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: unify paravirt pieces of processor.h
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:27 +0000 (13:31 +0100)]
x86: unify paravirt pieces of processor.h

This patch unifies the paravirt pieces of processor.h
The functionality present in 32 bit, but not (yet) in 64-bit,
like load_sp0 is _not_ done here, and let to a different patch.

With this unification, we get paravirt for free in x86_64 processor.h

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: move load_cr3 to a common place.
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:27 +0000 (13:31 +0100)]
x86: move load_cr3 to a common place.

There are currently two definitions of load_cr3, that essentially do the
same thing. This patch moves them all to processor.h.

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: move desc_empty to where they belong
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:27 +0000 (13:31 +0100)]
x86: move desc_empty to where they belong

This patch moves the (duplicated) desc_empty implementation to desc.h,
where the descriptor things belong.

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: get rid of _MASK flags
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:27 +0000 (13:31 +0100)]
x86: get rid of _MASK flags

There's no need for the *_MASK flags (TF_MASK, IF_MASK, etc), found in
processor.h (both _32 and _64). They have a one-to-one mapping with the
EFLAGS value. This patch removes the definitions, and use the already
existent X86_EFLAGS_ version when applicable.

[ roland@redhat.com: KVM build 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: move tsc definitions to were they belong
Glauber de Oliveira Costa [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: move tsc definitions to were they belong

This patch wipes out the definitions of tsc_disable from processor_32.h
and move it to tsc.h, were it belongs

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: clean up arch/x86/kernel/early-quirks.c
Neil Horman [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: clean up arch/x86/kernel/early-quirks.c

clean up checkpatch errors. No code changed.

      text    data     bss     dec     hex filename
       705     120       0     825     339 early-quirks.o.before
       705     120       0     825     339 early-quirks.o.after

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: cleanup i387_32.c according to checkpatch
Cyrill Gorcunov [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: cleanup i387_32.c according to checkpatch

clean up checkpatch warnings/errors on i387_32.c

The old and new i387_32.s (asm listings) were checked with diff to
be identical so it's safe to apply this patch.

Signed-off-by: Cyrill Gorcunov <gorunov@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: unify kexec_{32|64}.h
Harvey Harrison [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: unify kexec_{32|64}.h

One section collecting all constant defines.  Ifdef the asm
blocks for X86_32/64.

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>
Reviewed-by: Vivek Goyal <vgoyal@redhat.com>
16 years agox86: local.h fix checkpatch warnings
Harvey Harrison [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: local.h fix checkpatch warnings

Mostly space after comma, one space after if.

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 local_{32|64}.h
Harvey Harrison [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: unify local_{32|64}.h

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: introduce asm helpers in local_{32|64}.h
Harvey Harrison [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: introduce asm helpers in local_{32|64}.h

Handle the use of long on X86_32 and quad on X86_64

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agox86: fix asm memory constraints in local_64.h
Harvey Harrison [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: fix asm memory constraints in local_64.h

Use the shorter +m form rather than =m and m.

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 local_{32|64}.h
Harvey Harrison [Wed, 30 Jan 2008 12:31:26 +0000 (13:31 +0100)]
x86: clean up local_{32|64}.h

Common prefix from both files moved to local.h

Change __inline__ to inline

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 include/asm-x86/pda.h
Cyrill Gorcunov [Wed, 30 Jan 2008 12:31:25 +0000 (13:31 +0100)]
x86: clean up include/asm-x86/pda.h

clean up include/asm-x86/pda.h, as suggested by checkpatch.pl.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, kexec: force x86 arches to boot kdump kernels on boot cpu
Neil Horman [Wed, 30 Jan 2008 12:31:25 +0000 (13:31 +0100)]
x86, kexec: force x86 arches to boot kdump kernels on boot cpu

Recently a kdump bug was discovered in which a system would hang inside
calibrate_delay during the booting of the kdump kernel.  This was caused
by the fact that the jiffies counter was not being incremented during
timer calibration.  The root cause of this problem was found to be a
bios misconfiguration of the hypertransport bus.  On system affected by
this hang, the bios had assigned APIC ids which used extended apic bits
(more than the nominal 4 bit ids's), but failed to configure bit 17 of
the hypertransport transaction config register, which indicated that the
mask for the destination field of interrupt packets accross the ht bus
(see section 3.3.9 of
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF).
If a crash occurs on a cpu with an APIC id that extends beyond 4 bits,
it will not recieve interrupts during the kdump kernel boot, and this
hang will be the result.  The fix is to add this patch, whcih add an
early pci quirk check, to forcibly enable this bit in the httcfg
register.  This enables all cpus on a system to receive interrupts, and
allows kdump kernel bootup to procede normally.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: replace outb_p() with udelay(2) in drivers/input/mouse/pc110pad.c
Ingo Molnar [Wed, 30 Jan 2008 12:31:24 +0000 (13:31 +0100)]
x86: replace outb_p() with udelay(2) in drivers/input/mouse/pc110pad.c

replace outb_p() with udelay(2). This is a real ISA device so it likely
needs this particular delay.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: also define AT_VECTOR_SIZE_ARCH
Jan Beulich [Wed, 30 Jan 2008 12:31:24 +0000 (13:31 +0100)]
x86: also define AT_VECTOR_SIZE_ARCH

The patch introducing this left out x86-64, despite it also having
extra entries.

[ mingo@elte.hu: re-merged this to after the unification patches. ]

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: adjust enable_NMI_through_LVT0()
Jan Beulich [Wed, 30 Jan 2008 12:31:24 +0000 (13:31 +0100)]
x86: adjust enable_NMI_through_LVT0()

Its previous use in a call to on_each_cpu() was pointless, as at the
time that code gets executed only one CPU is online. Further, the
function can be __cpuinit, and for this to work without
CONFIG_HOTPLUG_CPU setup_nmi() must also get an attribute (this one
can even be __init; on 64-bits check_timer() also was lacking that
attribute).

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: make __{save,restore}_processor_state static
Jan Beulich [Wed, 30 Jan 2008 12:31:23 +0000 (13:31 +0100)]
x86: make __{save,restore}_processor_state static

.. allowing to remove their declarations from a global include file
(the symbols don't exist for anything but x86).

Likewise for 64-bits' fix_processor_context(), just that that one was
properly declared in an arch-specific header.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: move to .rodata/.init.data
Jan Beulich [Wed, 30 Jan 2008 12:31:23 +0000 (13:31 +0100)]
x86: move to .rodata/.init.data

The array is never written, and on 64-bits it's not even being used
past initial boot.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86-64: honor notify_die() returning NOTIFY_STOP
Jan Beulich [Wed, 30 Jan 2008 12:31:23 +0000 (13:31 +0100)]
x86-64: honor notify_die() returning NOTIFY_STOP

This requires making die() return a value, making its callers honor
this (and be prepared that it may return), and making oops_end() have
two additional parameters.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86-64: make pda's cpunumber and nodenumber unsigned
Jan Beulich [Wed, 30 Jan 2008 12:31:21 +0000 (13:31 +0100)]
x86-64: make pda's cpunumber and nodenumber unsigned

This generally allows better code to be generated, since the zero-
extension during 32-bit operations comes for free (needed when the
result is used as array index or similar), whereas sign extension must
be done explicitly and frequently requires a one byte larger
instruction due to the necessary rex64 prefix.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoi386: hard_{en,dis}able_TSC can be static
Jan Beulich [Wed, 30 Jan 2008 12:31:21 +0000 (13:31 +0100)]
i386: hard_{en,dis}able_TSC can be static

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: kprobes code for x86 unification
Masami Hiramatsu [Wed, 30 Jan 2008 12:31:21 +0000 (13:31 +0100)]
x86: kprobes code for x86 unification

This patch unifies kprobes code.

- Unify kprobes_*.h to kprobes.h
- Unify kprobes_*.c to kprobes.c
  (Differences are separated by ifdefs)
 - Most differences are related to REX prefix and rip relatives.
 - Two inline assembly code are different.
 - One difference in kprobe_handlre()
 - One fixup exception code is different, but it will be unified
   if mm/extable_*.c are unified.
- Merge history logs into arch/x86/kernel/kprobes.c.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: prepare kprobes code for x86 unification
Masami Hiramatsu [Wed, 30 Jan 2008 12:31:21 +0000 (13:31 +0100)]
x86: prepare kprobes code for x86 unification

This patch cleanup kprobes code on x86 for unification.
This patch is based on Arjan's previous work.

- Remove spurious whitespace changes
- Add harmless includes
- Make the 32/64 files more identical
 - Generalize structure fields' and local variable name.
 - Wrap accessing to stack address by macros.
 - Modify bitmap making macro.
 - Merge fixup code into is_riprel() and change its name to fix_riprel().
 - Set MAX_INSN_SIZE to 16 on both arch.
 - Use u32 for bitmaps on both architectures.
 - Clarify some comments.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Arjan van de Ven <arjan@infradead.org>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: return probe-booster for x86-64
Masami Hiramatsu [Wed, 30 Jan 2008 12:31:21 +0000 (13:31 +0100)]
x86: return probe-booster for x86-64

This patch adds kretprobe-booster to kprobes_64.c.

- Changes are based on x86-32.
- Rewrite register saving/restoring code

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: kprobe-booster for x86-64
Masami Hiramatsu [Wed, 30 Jan 2008 12:31:21 +0000 (13:31 +0100)]
x86: kprobe-booster for x86-64

This patch adds kprobe-booster to kprobes_64.c.

- Changes are based on x86-32.
- Add REX prefix checking code.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86: FIFO ticket spinlocks
Nick Piggin [Wed, 30 Jan 2008 12:31:21 +0000 (13:31 +0100)]
x86: FIFO ticket spinlocks

Introduce ticket lock spinlocks for x86 which are FIFO. The implementation
is described in the comments. The straight-line lock/unlock instruction
sequence is slightly slower than the dec based locks on modern x86 CPUs,
however the difference is quite small on Core2 and Opteron when working out of
cache, and becomes almost insignificant even on P4 when the lock misses cache.
trylock is more significantly slower, but they are relatively rare.

On an 8 core (2 socket) Opteron, spinlock unfairness is extremely noticable,
with a userspace test having a difference of up to 2x runtime per thread, and
some threads are starved or "unfairly" granted the lock up to 1 000 000 (!)
times. After this patch, all threads appear to finish at exactly the same
time.

The memory ordering of the lock does conform to x86 standards, and the
implementation has been reviewed by Intel and AMD engineers.

The algorithm also tells us how many CPUs are contending the lock, so
lockbreak becomes trivial and we no longer have to waste 4 bytes per
spinlock for it.

After this, we can no longer spin on any locks with preempt enabled
and cannot reenable interrupts when spinning on an irq safe lock, because
at that point we have already taken a ticket and the would deadlock if
the same CPU tries to take the lock again.  These are questionable anyway:
if the lock happens to be called under a preempt or interrupt disabled section,
then it will just have the same latency problems. The real fix is to keep
critical sections short, and ensure locks are reasonably fair (which this
patch does).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agospinlock: lockbreak cleanup
Nick Piggin [Wed, 30 Jan 2008 12:31:20 +0000 (13:31 +0100)]
spinlock: lockbreak cleanup

The break_lock data structure and code for spinlocks is quite nasty.
Not only does it double the size of a spinlock but it changes locking to
a potentially less optimal trylock.

Put all of that under CONFIG_GENERIC_LOCKBREAK, and introduce a
__raw_spin_is_contended that uses the lock data itself to determine whether
there are waiters on the lock, to be used if CONFIG_GENERIC_LOCKBREAK is
not set.

Rename need_lockbreak to spin_needbreak, make it use spin_is_contended to
decouple it from the spinlock implementation, and make it typesafe (rwlocks
do not have any need_lockbreak sites -- why do they even get bloated up
with that break_lock then?).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, ptrace: new ptrace BTS API
Markus Metzger [Wed, 30 Jan 2008 12:31:20 +0000 (13:31 +0100)]
x86, ptrace: new ptrace BTS API

Here's the new ptrace BTS API that supports two different overflow handling mechanisms (wrap-around and buffer-full-signal) to support two different use cases (debugging and profiling).

It further combines buffer allocation and configuration.

Opens:
- memory rlimit
- overflow signal

What would be the right signal to use?

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, ptrace: change BTS GET ptrace interface
Markus Metzger [Wed, 30 Jan 2008 12:31:20 +0000 (13:31 +0100)]
x86, ptrace: change BTS GET ptrace interface

Change the ptrace interface to mimick an array from newst to oldest.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, ptrace: use jiffies for BTS timestamps
Markus Metzger [Wed, 30 Jan 2008 12:31:20 +0000 (13:31 +0100)]
x86, ptrace: use jiffies for BTS timestamps

Replace sched_clock() with jiffies for BTS timestamps.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agox86, ptrace: remove bad comment
Markus Metzger [Wed, 30 Jan 2008 12:31:20 +0000 (13:31 +0100)]
x86, ptrace: remove bad comment

Remove no longer correct comment.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>