cpuset: remove sched domain hooks from cpusets

Remove the cpuset hooks that defined sched domains depending on the setting
of the 'cpu_exclusive' flag.

The cpu_exclusive flag can only be set on a child if it is set on the
parent.

This made that flag painfully unsuitable for use as a flag defining a
partitioning of a system.

It was entirely unobvious to a cpuset user what partitioning of sched
domains they would be causing when they set that one cpu_exclusive bit on
one cpuset, because it depended on what CPUs were in the remainder of that
cpusets siblings and child cpusets, after subtracting out other
cpu_exclusive cpusets.

Furthermore, there was no way on production systems to query the
result.

Using the cpu_exclusive flag for this was simply wrong from the get go.

Fortunately, it was sufficiently borked that so far as I know, almost no
successful use has been made of this.  One real time group did use it to
affectively isolate CPUs from any load balancing efforts.  They are willing
to adapt to alternative mechanisms for this, such as someway to manipulate
the list of isolated CPUs on a running system.  They can do without this
present cpu_exclusive based mechanism while we develop an alternative.

There is a real risk, to the best of my understanding, of users
accidentally setting up a partitioned scheduler domains, inhibiting desired
load balancing across all their CPUs, due to the nonobvious (from the
cpuset perspective) side affects of the cpu_exclusive flag.

Furthermore, since there was no way on a running system to see what one was
doing with sched domains, this change will be invisible to any using code.
Unless they have real insight to the scheduler load balancing choices, they
will be unable to detect that this change has been made in the kernel's
behaviour.

Initial discussion on lkml of this patch has generated much comment.  My
(probably controversial) take on that discussion is that it has reached a
rough concensus that the current cpuset cpu_exclusive mechanism for
defining sched domains is borked.  There is no concensus on the
replacement.  But since we can remove this mechanism, and since its
continued presence risks causing unwanted partitioning of the schedulers
load balancing, we should remove it while we can, as we proceed to work the
replacement scheduler domain mechanisms.

Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Dinakar Guniguntala <dino@in.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

I/OAT: Add DCA services

Add code to connect to the DCA driver and provide cpu tags for use by
drivers that would like to use Direct Cache Access hints.

    [Adrian Bunk]                Several Kconfig cleanup items
    [Andrew Morten, Chris Leech] Fix for using cpu_physical_id() even when
         built for uni-processor

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

DCA: Add Direct Cache Access driver

Direct Cache Access (DCA) is a method for warming the CPU cache before data
is used, with the intent of lessening the impact of cache misses.  This
patch adds a manager and interface for matching up client requests for DCA
services with devices that offer DCA services.

In order to use DCA, a module must do bus writes with the appropriate tag
bits set to trigger a cache read for a specific CPU.  However, different
CPUs and chipsets can require different sets of tag bits, and the methods
for determining the correct bits may be simple hardcoding or may be a
hardware specific magic incantation.  This interface is a way for DCA
clients to find the correct tag bits for the targeted CPU without needing
to know the specifics.

    [Dave Miller] use DEFINE_SPINLOCK()

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

I/OAT: Add support for MSI and MSI-X

Add support for MSI and MSI-X interrupt handling, including the ability
to choose the desired interrupt method.

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Acked-by: David S. Miller <davem@davemloft.net>
[bunk@kernel.org: drivers/dma/ioat_dma.c: make 3 functions static]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

I/OAT: Split PCI startup from DMA handling code

Split the general PCI startup from the DMA handling code in order to
prepare for adding support for DCA services and future versions of the
ioatdma device.

    [Rusty Russell] Removal of __unsafe() usage.

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

I/OAT: code cleanup from checkpatch output

Take care of a bunch of little code nits in ioatdma files

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

I/OAT: Rename the source file

Rename the ioatdma.c file in preparation for splitting into multiple files,
which will allow for easier adding new functionality.

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

I/OAT: New device ids

Add device ids for new revs of the Intel I/OAT DMA engine

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

V850: cleanup struct irqaction initializers

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Miles Bader <miles@gnu.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: definitively kill subprocesses on panic

In a stock 2.6.22.6 kernel, poweroff a user mode linux guest (2.6.22.6 running
in skas0 mode) will halt the host linux.  I think the reason is the kernel
thread abort because of a bug.  Then the sys_reboot in process of user mode
linux guest is not trapped by the user mode linux kernel and is executed by
host.  I think it is better to make sure all of our children process to quit
when user mode linux kernel abort.

[ jdike - the kernel process needs to ignore SIGTERM, plus the waitpid/kill
loop is needed to make sure that all of our children are dead before the
kernel exits ]

Signed-off-by: Lepton Wu <ytht.net@gmail.com>
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: arch/um/drivers formatting

Style fixes for the rest of the drivers.  arch/um/drivers should be pretty
CodingStyle-compliant now.

Except for the ubd driver, which will have to be treated separately.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix stub address calculations

The calculation of CONFIG_STUB_CODE and CONFIG_STUB_DATA didn't take into
account anything but 3G/1G and 2G/2G, leaving the other vmsplits out in the
cold.

I'd rather not duplicate the four known host vmsplit cases for each of these
symbols.  I'd also like to calculate them based on the highest userspace
address.

The Kconfig language seems not to allow calculation of hex constants, so I
moved this to as-layout.h.  CONFIG_STUB_CODE, CONFIG_STUB_DATA, and
CONFIG_STUB_START are now gone.  In their place are STUB_CODE, STUB_DATA, and
STUB_START in as-layout.h.

i386 and x86_64 seem to differ as to whether an unadorned constant is an int
or a long, so I cast them to unsigned long so they can be printed
consistently.  However, they are also used in stub.S, where C types don't work
so well.  So, there are ASM_ versions of these constants for use in stub.S.  I
also ifdef-ed the non-asm-friendly portion of as-layout.h.

With this in place, most of the rest of this patch is changing CONFIG_STUB_*
to STUB_*, except in stub.S, where they are changed to ASM_STUB_*.

defconfig has the old symbols deleted.

I also print these addresses out in case there is any problem mapping them on
the host.

The two stub.S files had some trailing whitespace, so that is cleaned up here.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: correctly handle skb allocation failures

Handle memory allocation failures when reading packets.

We have to read something from the host, even if we can't allocate any
memory.  If we don't, the host side of the device may fill up and stop
delivering interrupts because no new packets can be queued.

A single sk_buff is allocated whenever an MTU is seen which is larger
than any seen earlier.  This is used to read packets if there is a
memory allocation failure.

The large MTU check is done from eth_configure, which is called when a
interface is added to the system.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: network driver MTU cleanups

A bunch of MTU-related cleanups in the network code.

First, there is the addition of the notion of a maximally-sized packet, which
is the MTU plus headers.  This is used to size the skb that will receive a
packet.  This allows ether_adjust_skb to go away, as it was used to resize the
skb after it was allocated.

Since the skb passed into the low-level read routine is no longer resized, and
possibly reallocated, there, they (and the write routines) don't need to get
an sk_buff **.  They just need the sk_buff * now.  The callers of
ether_adjust_skb still need to do the skb_put, so that's now inlined.

The MAX_PACKET definitions in most of the drivers are gone.

The set_mtu methods were all the same and did nothing, so they can be
removed.

The ethertap driver had a typo which doubled the size of the packet rather
than adding two bytes to it.  It also wasn't defining its setup_size, causing
a zero-byte kmalloc and crash when the invalid pointer returned from kmalloc
was dereferenced.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: network formatting

Style and other non-functional changes in the UML networking code, including
include tidying
style violations
copyright updates
printks getting severities
userspace code calling libc directly rather than using the os_*
wrappers

There's also a exit path cleanup in the pcap driver.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: use *SEC_PER_*SEC constants

There are various uses of powers of 1000, plus the odd BILLION constant in the
time code.  However, there are perfectly good definitions of *SEC_PER_*SEC in
linux/time.h which can be used instaed.

These are replaced directly in kernel code.  Userspace code imports those
constants as UM_*SEC_PER_*SEC and uses these.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: eliminate SIGALRM

Now that ITIMER_REAL is no longer used, there is no need for any use of
SIGALRM whatsoever.  This patch removes all mention of it.

In addition, real_alarm_handler took a signal argument which is now always
SIGVTALRM.  So, that is gone.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: time build fix

Put back an implementation of timeval_to_ns in arch/um/os-Linux/time.c.
tglx pointed out in his review of tickless support that there was a
perfectly good implementation of it in linux/time.h.  The problem is that
this is userspace code which can't pull in kernel headers and there doesn't
seem to be a libc version.

So, I'm copying the version from linux/time.h rather than resurrecting my
version.  This causes some declaration changes as it now returns a signed
value rather than an unsigned value.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: eliminate interrupts in the idle loop

Now, the idle loop now longer needs SIGALRM firing - it can just sleep for the
requisite amount of time and fake a timer interrupt when it finishes.

Any use of ITIMER_REAL now goes away.  disable_timer only turns off
ITIMER_VIRTUAL.  switch_timers is no longer needed, so it, and all calls, goes
away.

disable_timer now returns the amount of time remaining on the timer.
default_idle uses this to tell idle_sleep how long to sleep.  idle_sleep will
call alarm_handler if nanosleep returns 0, which is the case if it didn't
return early due to an interrupt.  Otherwise, it just returns.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: tickless support

Enable tickless support.

CONFIG_TICK_ONESHOT and CONFIG_NO_HZ are enabled.

itimer_clockevent gets CLOCK_EVT_FEAT_ONESHOT and an implementation of
.set_next_event.

CONFIG_UML_REAL_TIME_CLOCK goes away because it only makes sense when there is
a clock ticking away all the time.  timer_handler now just calls do_IRQ once
without trying to figure out how many ticks to emulate.

The idle loop now needs to turn ticking on and off.

Userspace ticks keep happening as usual.  However, the userspace loop keep
track of when the next wakeup should happen and suppresses process ticks until
that happens.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: clocksource support

Add clocksource support.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: GENERIC_CLOCKEVENTS support

Enable CONFIG_GENERIC_CLOCKEVENTS.

timer_irq gets its name changed to timer_handler, and becomes the recipient of
timer signals.

The clock_event_device is set up to imitate the current ticking clock, i.e.
CLOCK_EVT_FEAT_ONESHOT is not enabled yet.

disable_timer now doesn't ignore SIGALRM and SIGVTALRM because that breaks
delay calibration.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: GENERIC_TIME support

Enable CONFIG_GENERIC_TIME.

As a side-effect of this, the UML implementations of do_gettimeofday and
do_settimeofday go away, as these are provided by generic code.  set_time also
goes away since it was only used by do_settimeofday.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: separate timer initialization

Move timer signal initialization from init_irq_signals to a new function,
timer_init.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: simplify interval setting

set_interval took a timer type as an argument, but it always specified a
virtual timer.  So, it is not needed, and it is gone, and set_interval is
simplified appropriately.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix timer switching

Fix up the switching between virtual and real timers.  The idle loop sleeps,
so the timer at that point must be real time.  At all other times, the timer
must be virtual.  Even when userspace is running, and the kernel is asleep,
the virtual timer is correct because the process timer will be running and the
process timer will be firing.

The timer switch used to be in the context switch and timer handler code.
This is moved to the idle loop and the signal handler, making it much more
clear why it is happening.

switch_timers now returns the old timer type so that it may be restored.  The
signal handler uses this in order to restore the previous timer type when it
returns.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: eliminate hz()

Eliminate hz() since its only purpose was to provide a kernel-space constant
to userspace code.  This can be done instead by providing the constant
directly through kernel_constants.h.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: more idiomatic parameter parsing

Make mconsole parameter parsing slightly more idiomatic.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: remove unused file

arch/um/os-Linux/tt.c is no longer used.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: remove unneeded void * cast

vmalloc() returns a void pointer, so casting to (void *) is pretty pointless.

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: eliminate floating point state from register file

The floating point fields in the pt_regs register file aren't used, so they
are deleted.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: style fixes in FP code

Tidy the code affected by the floating point fixes.

A bunch of unused stuff is gone, including two sigcontext.c files,
which turned out to be entirely unneeded.

There are the usual fixes -
whitespace and style cleanups
copyright updates
emacs formatting comments gone
include cleanups
adding severities to printks

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: sysrq and mconsole fixes

Fix the passing of printk output back to the mconsole client.  The existing
code was somewhat confused, accumulating output in a buffer, but writing it
out entirely whenever a new chunk was added.  This is fixed.

The earlier include cleanups caused linux/sysrq.h to not be included - this is
fixed by adding the include back, under CONFIG_MAGIC_SYSRQ.

CONFIG_MAGIC_SYSRQ is also defaulted to on in defconfig.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: coredumping floating point fixes

Fix core dumping of floating point state.  ELF_CORE_COPY_FPREGS gets a
definitions, and as a result, dump_fpu no longer needs to exist.  Also,
elf_fpregset_t needed a real definition.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: ptrace floating point fixes

Handle floating point state better in ptrace.  The code now correctly
distinguishes between PTRACE_[GS]ETFPREGS and PTRACE_[GS]ETFPXREGS.  The FPX
requests get handed off to arch-specific code because that's not generic.

get_fpregs, set_fpregs, set_fpregs, and set_fpxregs needed real
implementations.

Something here exposed a missing include in asm/page.h, which needed
linux/types.h in order to get gfp_t, so that's fixed here.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: floating point signal delivery fixes

Handle floating point state in across signals correctly.  UML/i386 needs to
know whether the host does PTRACE_[GS]ETFPXREGS, so an arch_init_registers
hook is added, which on x86_64 does nothing.

UML doesn't save and restore floating point registers on kernel entry and
exit, so they need to be copied between the host process and the sigcontext.
save_fpx_registers and restore_fpx_registers are added for this purpose.
save_fp_registers and restore_fp_registers already exist.

There was a bunch of floating point state conversion code in
arch/um/sys-i386/ptrace.c which isn't needed there, but is needed in signal.c,
so it is moved over.

The i386 code now distinguishes between fp and fpx state and handles them
correctly.  The x86_64 code just needs to copy state as-is between the host
process and the stack.  There are also some fixes there to pass the correct
address of the floating point state around.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: don't use glibc asm/user.h

Stop including asm/user.h from libc - it seems to be disappearing from
distros.  It's replaced with sys/user.h which defines user_fpregs_struct and
user_fpxregs_struct instead of user_i387_struct and struct user_fxsr_struct on
i386.

As a bonus, on x86_64, I get to dump some stupid typedefs which were needed in
order to get asm/user.h to compile.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix hostfs style

Style fixes in hostfs.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: remove unneeded if from hostfs

Get rid of an empty if statement which might look like a bug to a
casual reader.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: clean up tlb flush path

Tidy the tlb flushing code.

With tt mode gone, there is no reason to have the capability to have
called directly from do_mmap, do_mprotect, and do_munmap, rather than
calling a function pointer that it is given.

There was a large amount of data that was passed from function to
function, being used at the lowest level, without being changed.  This
stuff is now encapsulated in a structure which is initialized at the
top layer and passed down.  This simplifies the code, reduces the
amount of code needed to pass the parameters around, and saves on
stack space.

A somewhat more subtle change is the meaning of the current operation
index.  It used to start at -1, being pre-incremented when adding an
operation.  It now starts at 0, being post-incremented, with
associated adjustments of +/- 1 on comparisons.

In addition, tlb.h contained a couple of declarations which had no
users outside of tlb.c, so they could be moved or deleted.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: userspace files should call libc directly

A number of files that were changed in the recent removal of tt mode
are userspace files which call the os_* wrappers instead of calling
libc directly.  A few other files were affected by this, through

This patch makes these call glibc directly.

There are also style fixes in the affected areas.

os_print_error has no remaining callers, so it is deleted.

There is a interface change to os_set_exec_close, eliminating a
parameter which was always the same.  The callers are fixed as well.

os_process_pc got its error path cleaned up.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix inlines

"extern inline" will have different semantics with gcc 4.3.

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: replace clone with fork

Convert the boot-time host ptrace testing from clone to fork.  They were
essentially doing fork anyway.  This cleans up the code a bit, and makes
valgrind a bit happier about grinding it.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: remove os_* usage from userspace files

This patch fixes some userspace files which were calling libc through the os_*
wrappers.

It turns out that there was only one user of os_new_tty_pgrp, so it can be
deleted.

There are also some style and whitespace fixes in here.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: free LDT state on process exit

The space allocated for a process LDT wasn't being freed when the process
exited.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: rename pt_regs general-purpose register file

Before the removal of tt mode, access to a register on the skas-mode side of a
pt_regs struct looked like pt_regs.regs.skas.regs.regs[FOO].  This was bad
enough, but it became pt_regs.regs.regs.regs[FOO] with the removal of the
union from the middle.  To get rid of the run of three "regs", the last field
is renamed to "gp".

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fold mmu_context_skas into mm_context

This patch folds mmu_context_skas into struct mm_context, changing all users
of these structures as needed.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: get rid of do_longjmp

do_longjmp used to be needed when UML didn't have its own implementation of
setjmp and longjmp.  They came from libc, and couldn't be called directly from
kernel code, as the libc jmp_buf couldn't be imported there.  do_longjmp was a
userspace function which served to provide longjmp access to kernel code.

This is gone, and a number of void * pointers can now be jmp_buf *.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: remove __u64 usage from physical memory subsystem

Eliminate some uses of __u64 in the physical memory support.  It's hard to get
a definition of __u64 in both kernel and userspace code on x86_64, so this
changes them to unsigned long long.

There are also a copyright update and formatting comment removal from the
affected header.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: style fixes pass 3

Formatting changes in the files which have been changed in the course
of folding foo_skas functions into their callers.  These include:
copyright updates
header file trimming
style fixes
adding severity to printks

These changes should be entirely non-functional.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: remove code made redundant by CHOOSE_MODE removal

This patch makes a number of simplifications enabled by the removal of
CHOOSE_MODE.  There were lots of functions that looked like

int foo(args){
foo_skas(args);
}

The bodies of foo_skas are now folded into foo, and their declarations (and
sometimes entire header files) are deleted.

In addition, the union uml_pt_regs, which was a union between the tt and skas
register formats, is now a struct, with the tt-mode arm of the union being
removed.

It turns out that usr2_handler was unused, so it is gone.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: style fixes pass 2

Formatting changes in the files which have been changed in the course
of removing CHOOSE_MODE.  These include:
copyright updates
header file trimming
style fixes
adding severity to printks

These changes should be entirely non-functional.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: throw out CHOOSE_MODE

The next stage after removing code which depends on CONFIG_MODE_TT is removing
the CHOOSE_MODE abstraction, which provided both compile-time and run-time
branching to either tt-mode or skas-mode code.

This patch removes choose-mode.h and all inclusions of it, and replaces all
CHOOSE_MODE invocations with the skas branch.  This leaves a number of trivial
functions which will be dealt with in a later patch.

There are some changes in the uaccess and tls support which go somewhat beyond
this and eliminate some of the now-redundant functions.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: style fixes pass 1

Formatting changes in the files which have been changed in the
tt-removal patchset so far.  These include:
copyright updates
header file trimming
style fixes
adding severity to printks
indenting Kconfig help according to the predominant kernel style

These changes should be entirely non-functional.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: remove sysdep/thread.h

This patch removes thread.h, which turns out not to be needed any more.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: throw out CONFIG_MODE_TT

This patchset throws out tt mode, which has been non-functional for a while.

This is done in phases, interspersed with code cleanups on the affected files.

The removal is done as follows:
remove all code, config options, and files which depend on
CONFIG_MODE_TT
get rid of the CHOOSE_MODE macro, which decided whether to
call tt-mode or skas-mode code, and replace invocations with their
skas portions
replace all now-trivial procedures with their skas equivalents

There are now a bunch of now-redundant pieces of data structures, including
mode-specific pieces of the thread structure, pt_regs, and mm_context.  These
are all replaced with their skas-specific contents.

As part of the ongoing style compliance project, I made a style pass over all
files that were changed.  There are three such patches, one for each phase,
covering the files affected by that phase but no later ones.

I noticed that we weren't freeing the LDT state associated with a process when
it exited, so that's fixed in one of the later patches.

The last patch is a tidying patch which I've had for a while, but which caused
inexplicable crashes under tt mode.  Since that is no longer a problem, this
can now go in.

This patch:

Start getting rid of tt mode support.

This patch throws out CONFIG_MODE_TT and all config options, code, and files
which depend on it.

CONFIG_MODE_SKAS is gone and everything that depends on it is included
unconditionally.

The few changed lines are in re-written Kconfig help, lines which needed
something skas-related removed from them, and a few more which weren't
strictly deletions.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

UML: remove unnecessary hostfs_getattr()

Currently hostfs_getattr() just defines the default behavior.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: add VDE networking support

Added vde network backend in uml to introduce native Virtual Distributed
Ethernet support (using libvdeplug).

Signed-off-by: Luca Bigliardi <shammash@artha.org>
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: physmem code tidying

Tidying of the UML physical memory system.  These are mostly style fixes,
however the includes were cleaned as well.  This uncovered a need for
mem_user.h to be included in mode_kern_skas.h.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: stop saving process FP state

Throw out a lot of code dealing with saving and restoring floating-point
state.  In skas mode, where processes run in a restoring floating-point state
on kernel entry and exit is pointless.

This eliminates most of arch/um/os-Linux/sys-{i386,x86_64}/registers.c.  Most
of what remained is now arch-indpendent, and can be moved up to
arch/um/os-Linux/registers.c.  Both arches need the jmp_buf accessor
get_thread_reg, and i386 needs {save,restore}_fp_regs because it cheats during
sigreturn by getting the fp state using ptrace rather than copying it out of
the process sigcontext.

After this, it turns out that arch/um/include/skas/mode-skas.h is almost
completely unneeded.  The declarations in it are variables which either don't
exist or which don't have global scope.  The one exception is
kill_off_processes_skas.  If that's removed, this header can be deleted.

This uncovered a bug in user.h, which wasn't correctly making sure that a
size_t definition was available to both userspace and kernelspace files.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: stop specially protecting kernel stacks

Map all of physical memory as executable to avoid having to change stack
protections during fork and exit.

unprotect_stack is now called only from MODE_TT code, so it is marked as such.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix nonremovability of watchdog

The UML watchdog driver was using the wrong config variable to control whether
it can be unloaded once active.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix an IPV6 libc vs kernel symbol clash

On some systems, with IPV6 configured, there is a clash between the kernel's
in6addr_any and the one in libc.

This is handled in the usual (gross) way of defining the kernel symbol out of
the way on the gcc command line.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: stop using libc asm/page.h

Remove includes of asm/page.h from libc code.  This header seems to be
disappearing, and UML doesn't make much use of it anyway.

The one use, PAGE_SHIFT in stub.h, is handled by copying the constant from the
kernel side of the house in common_offsets.h.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: console tidying

Tidy line.c:
The includes are more minimal
Lots of style fixes
All the printks have severities
Removed some commented-out code
Deleted a useless printk when ioctl is called
Fixed some whitespace damage

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix console writing bugs

The previous console cleanup patch switched generic_read and generic_write
from calling os_{read,write}_file to calling read and write directly.  Because
the calling convention is different, they now need to get any error from errno
rather than the return value.  I did this for generic_read, but forgot about
generic_write.

While chasing some output corruption, I noticed that line_write was
unnecessarily calling flush_buffer, and deleted it.  I don't understand why,
but the corruption disappeared.  This is unneeded because there already is a
perfectly good mechanism for finding out when the host output device has some
room to write data - there is an interrupt that comes in when writes can
happen again.  line_write calling flush_buffer seemed to just be an attempt to
opportunistically get some data out to the host.

I also made write_chan short-circuit calling into the host-level code for
zero-length writes.  Calling libc write with a length of zero conflated write
not being able to write anything with asking it not to write anything.  Better
to just cut it off as soon as possible.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: console subsystem tidying

This does a lot of cleanup on the UML console system.  This patch should be
entirely non-functional.

The tidying is as follows:
header cleanups - the includes should be closer to minimal and complete
all printks now have a severity
lots of style fixes
fd_close is restructured a little in order to reduce the nesting
some functions were calling the os_* wrappers when they can
call libc directly
port_accept had a unnecessary variable
it also tested a pid unecessarily before killing it
some functions were made static
xterm_free is gone, as it was identical to generic_free

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: fix error cleanup ordering

I messed up the error cleanup ordering in the console port driver.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: tidy recently-moved code

Now that the generic console operations are in a userspace file, we
can do the following:
directly call into libc instead of through the os_* wrappers
eliminate os_window_size since it has only one user

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

uml: move userspace code to userspace file

Move some code from a kernelspace file to a userspace file where it fits
better.  This enables some tidying which is the subject of a later patch.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tty: bring the old cris driver back somewhere into the realm of new tty buffering

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Mikael Starvik <starvik@axis.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

CRIS: cleanup struct irqaction initializers

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Mikael Starvik <starvik@axis.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m32r: convert to generic sys_ptrace

Convert m32r to the generic sys_ptrace.  The conversion requires an
architecture hook after ptrace_attach which this patch adds.  The hook
will also be needed for a conersion of ia64 to the generic ptrace code.

Thanks to Hirokazu Takata for fixing a bug in the first version of this
code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m32r: serial: remove M32R_SIO_SHARE_IRQS

Remove an unused symbol M32R_SIO_SHARE_IRQS from drivers/serial/m32r_sio.h.

Signed-off-by: Hirokazu Takata <takata@linux-m32r.org>
Cc: "Robert P. J. Day" <rpjday@mindspring.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

M32R: cleanup struct irqaction initializers

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/asm-m32r/thread_info.h: kmalloc + memset conversion to kzalloc

Signed-off-by: Mariusz Kozlowski <m.kozlowski@tuxland.pl>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: beautify vmlinux.lds

Introduced a consistent style in vmlinux.lds and it now matches the
soon-to-be common style for all arch's vmlinux.lds files.

In addition:
- Replaced hardcoded constant with PAGE_SIZE
- Fix page.h so PAGE_SIZE can be used from assembler and in lds files
- Move a few labels inside brackets so linker alignment will not
  make label point ot a too low address
- Replaced DWARF and STABS sections with definitions from asm-generic

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: convert to generic sys_ptrace

This patch converts alpha to the generic sys_ptrace.  We use
force_successful_syscall_return to avoid having to pass the pt_regs pointer
down to the function.  I think the removal of the assemly stub is correct,
but I could only compile-test this patch, so please give it a spin before
commiting :)

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cleanup arch/alpha/Makefile

- binutils 2.7 is far below the current minimum supported version,
  and there's therefore no longer a need for an extra test

- since even gcc 3.2 already supports all options used we can use them
  unconditionally

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

M68KNOMMU: remove unused config symbol CONFIG_DISKtel

Remove unused config symbol CONFIG_DISKtel.
Pointed out by Robert P. J. Day <rpjday@mindspring.com>.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

FRV: cleanup struct irqaction initializers

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/asm-frv/thread_info.h: kmalloc + memset conversion to kzalloc

Signed-off-by: Mariusz Kozlowski <m.kozlowski@tuxland.pl>
Acked-By: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

remove frv usage of flush_tlb_pgtables()

frv is the last user in the tree of that dubious hook, and it's my
understanding that it's not even needed.  It's only called by memory.c
free_pgd_range() which is always called within an mmu_gather, and
tlb_flush() on frv will do a flush_tlb_mm(), which from my reading of the
code, seems to do what flush_tlb_ptables() does, which is to clear the
cached PGE.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-By: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add node states sysfs class attributeS

Add a per node state sysfs class attribute file to /sys/devices/system/node
to display node state masks.

E.g., on a 4-cell HP ia64 NUMA platform, we have 5 nodes: 4 representing
the actual hardware cells and one memory-only pseudo-node representing a
small amount [512MB] of "hardware interleaved" memory.  With this patch, in
/sys/devices/system/node we see:

#ls -1F /sys/devices/system/node
has_cpu
has_normal_memory
node0/
node1/
node2/
node3/
node4/
online
possible
#cat /sys/devices/system/node/possible
0-255
#cat /sys/devices/system/node/online
0-4
#cat /sys/devices/system/node/has_normal_memory
0-4
#cat /sys/devices/system/node/has_cpu
0-3

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmstat.c: cleanups

This patch contains the following cleanups:
- make the needlessly global setup_vmstat() static
- remove the unused refresh_vm_stats()

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mempolicy.c: cleanups

This patch contains the following cleanups:
- every file should include the headers containing the prototypes for
  its global functions
- make the follosing needlessly global functions static:
  - migrate_to_node()
  - do_mbind()
  - sp_alloc()
  - mpol_rebind_policy()

[akpm@linux-foundation.org: fix uninitialised var warning]
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/shmem.c: make 3 functions static

This patch makes three needlessly global functions static.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: fix dynamic pool resize failure case

When gather_surplus_pages() fails to allocate enough huge pages to satisfy
the requested reservation, it frees what it did allocate back to the buddy
allocator.  put_page() should be called instead of update_and_free_page()
to ensure that pool counters are updated as appropriate and the page's
refcount is decremented.

Signed-off-by: Adam Litke <agl@us.ibm.com>
Acked-by: Dave Hansen <haveblue@us.ibm.com>
Cc: David Gibson <hermes@gibson.dropbear.id.au>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Ken Chen <kenchen@google.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: fix hugepage allocation with memoryless nodes

Anton found a problem with the hugetlb pool allocation when some nodes have
no memory (http://marc.info/?l=linux-mm&m=118133042025995&w=2).  Lee worked
on versions that tried to fix it, but none were accepted.  Christoph has
created a set of patches which allow for GFP_THISNODE allocations to fail
if the node has no memory.

Currently, alloc_fresh_huge_page() returns NULL when it is not able to
allocate a huge page on the current node, as specified by its custom
interleave variable.  The callers of this function, though, assume that a
failure in alloc_fresh_huge_page() indicates no hugepages can be allocated
on the system period.  This might not be the case, for instance, if we have
an uneven NUMA system, and we happen to try to allocate a hugepage on a
node with less memory and fail, while there is still plenty of free memory
on the other nodes.

To correct this, make alloc_fresh_huge_page() search through all online
nodes before deciding no hugepages can be allocated.  Add a helper function
for actually allocating the hugepage.  Use a new global nid iterator to
control which nid to allocate on.

Note: we expect particular semantics for __GFP_THISNODE, which are now
enforced even for memoryless nodes.  That is, there is should be no
fallback to other nodes.  Therefore, we rely on the nid passed into
alloc_pages_node() to be the nid the page comes from.  If this is
incorrect, accounting will break.

Tested on x86 !NUMA, x86 NUMA, x86_64 NUMA and ppc64 NUMA (with 2
memoryless nodes).

Before on the ppc64 box:
Trying to clear the hugetlb pool
Done.       0 free
Trying to resize the pool to 100
Node 0 HugePages_Free:     25
Node 1 HugePages_Free:     75
Node 2 HugePages_Free:      0
Node 3 HugePages_Free:      0
Done. Initially     100 free
Trying to resize the pool to 200
Node 0 HugePages_Free:     50
Node 1 HugePages_Free:    150
Node 2 HugePages_Free:      0
Node 3 HugePages_Free:      0
Done.     200 free

After:
Trying to clear the hugetlb pool
Done.       0 free
Trying to resize the pool to 100
Node 0 HugePages_Free:     50
Node 1 HugePages_Free:     50
Node 2 HugePages_Free:      0
Node 3 HugePages_Free:      0
Done. Initially     100 free
Trying to resize the pool to 200
Node 0 HugePages_Free:    100
Node 1 HugePages_Free:    100
Node 2 HugePages_Free:      0
Node 3 HugePages_Free:      0
Done.     200 free

Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Acked-by: Christoph Lameter <clameter@sgi.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <hermes@gibson.dropbear.id.au>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Ken Chen <kenchen@google.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: fix pool resizing corner case

When shrinking the size of the hugetlb pool via the nr_hugepages sysctl, we
are careful to keep enough pages around to satisfy reservations.  But the
calculation is flawed for the following scenario:

Action                          Pool Counters (Total, Free, Resv)
======                          =============
Set pool to 1 page              1 1 0
Map 1 page MAP_PRIVATE          1 1 0
Touch the page to fault it in   1 0 0
Set pool to 3 pages             3 2 0
Map 2 pages MAP_SHARED          3 2 2
Set pool to 2 pages             2 1 2 <-- Mistake, should be 3 2 2
Touch the 2 shared pages        2 0 1 <-- Program crashes here

The last touch above will terminate the process due to lack of huge pages.

This patch corrects the calculation so that it factors in pages being used
for private mappings.  Andrew, this is a standalone fix suitable for
mainline.  It is also now corrected in my latest dynamic pool resizing
patchset which I will send out soon.

Signed-off-by: Adam Litke <agl@us.ibm.com>
Acked-by: Ken Chen <kenchen@google.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlbfs read() support

Support for reading from hugetlbfs files.  libhugetlbfs lets application
text/data to be placed in large pages.  When we do that, oprofile doesn't
work - since libbfd tries to read from it.

This code is very similar to what do_generic_mapping_read() does, but I
can't use it since it has PAGE_CACHE_SIZE assumptions.

[akpm@linux-foundation.org: cleanups, fix leak]
[bunk@stusta.de: make hugetlbfs_read() static]
Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com>
Acked-by: William Irwin <bill.irwin@oracle.com>
Tested-by: Nishanth Aravamudan <nacc@us.ibm.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: allow extending ftruncate on hugetlbfs

For historical reason, expanding ftruncate that increases file size on
hugetlbfs is not allowed due to pages were pre-faulted and lack of fault
handler.  Now that we have demand faulting on hugetlb since 2.6.15, there
is no reason to hold back that limitation.

This will make hugetlbfs behave more like a normal fs.  I'm writing a user
level code that uses hugetlbfs but will fall back to tmpfs if there are no
hugetlb page available in the system.  Having hugetlbfs specific ftruncate
behavior is a bit quirky and I would like to remove that artificial
limitation.

Signed-off-by: <kenchen@google.com>
Acked-by: Wiliam Irwin <wli@holomorphy.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: Add hugetlb_dynamic_pool sysctl

The maximum size of the huge page pool can be controlled using the overall
size of the hugetlb filesystem (via its 'size' mount option).  However in the
common case the this will not be set as the pool is traditionally fixed in
size at boot time.  In order to maintain the expected semantics, we need to
prevent the pool expanding by default.

This patch introduces a new sysctl controlling dynamic pool resizing.  When
this is enabled the pool will expand beyond its base size up to the size of
the hugetlb filesystem.  It is disabled by default.

Signed-off-by: Adam Litke <agl@us.ibm.com>
Acked-by: Andy Whitcroft <apw@shadowen.org>
Acked-by: Dave McCracken <dave.mccracken@oracle.com>
Cc: William Irwin <bill.irwin@oracle.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Ken Chen <kenchen@google.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: Try to grow hugetlb pool for MAP_SHARED mappings

Shared mappings require special handling because the huge pages needed to
fully populate the VMA must be reserved at mmap time.  If not enough pages are
available when making the reservation, allocate all of the shortfall at once
from the buddy allocator and add the pages directly to the hugetlb pool.  If
they cannot be allocated, then fail the mapping.  The page surplus is
accounted for in the same way as for private mappings; faulted surplus pages
will be freed at unmap time.  Reserved, surplus pages that have not been used
must be freed separately when their reservation has been released.

Signed-off-by: Adam Litke <agl@us.ibm.com>
Acked-by: Andy Whitcroft <apw@shadowen.org>
Acked-by: Dave McCracken <dave.mccracken@oracle.com>
Cc: William Irwin <bill.irwin@oracle.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Ken Chen <kenchen@google.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: Try to grow hugetlb pool for MAP_PRIVATE mappings

Because we overcommit hugepages for MAP_PRIVATE mappings, it is possible that
the hugetlb pool will be exhausted or completely reserved when a hugepage is
needed to satisfy a page fault.  Before killing the process in this situation,
try to allocate a hugepage directly from the buddy allocator.

The explicitly configured pool size becomes a low watermark.  When dynamically
grown, the allocated huge pages are accounted as a surplus over the watermark.
 As huge pages are freed on a node, surplus pages are released to the buddy
allocator so that the pool will shrink back to the watermark.

Surplus accounting also allows for friendlier explicit pool resizing.  When
shrinking a pool that is fully in-use, increase the surplus so pages will be
returned to the buddy allocator as soon as they are freed.  When growing a
pool that has a surplus, consume the surplus first and then allocate new
pages.

Signed-off-by: Adam Litke <agl@us.ibm.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Andy Whitcroft <apw@shadowen.org>
Acked-by: Dave McCracken <dave.mccracken@oracle.com>
Cc: William Irwin <bill.irwin@oracle.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Ken Chen <kenchen@google.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: Move update_and_free_page

Dynamic huge page pool resizing.

In most real-world scenarios, configuring the size of the hugetlb pool
correctly is a difficult task.  If too few pages are allocated to the pool,
applications using MAP_SHARED may fail to mmap() a hugepage region and
applications using MAP_PRIVATE may receive SIGBUS.  Isolating too much memory
in the hugetlb pool means it is not available for other uses, especially those
programs not using huge pages.

The obvious answer is to let the hugetlb pool grow and shrink in response to
the runtime demand for huge pages.  The work Mel Gorman has been doing to
establish a memory zone for movable memory allocations makes dynamically
resizing the hugetlb pool reliable within the limits of that zone.  This patch
series implements dynamic pool resizing for private and shared mappings while
being careful to maintain existing semantics.  Please reply with your comments
and feedback; even just to say whether it would be a useful feature to you.
Thanks.

How it works
============

Upon depletion of the hugetlb pool, rather than reporting an error immediately,
first try and allocate the needed huge pages directly from the buddy allocator.
Care must be taken to avoid unbounded growth of the hugetlb pool, so the
hugetlb filesystem quota is used to limit overall pool size.

The real work begins when we decide there is a shortage of huge pages.  What
happens next depends on whether the pages are for a private or shared mapping.
Private mappings are straightforward.  At fault time, if alloc_huge_page()
fails, we allocate a page from the buddy allocator and increment the source
node's surplus_huge_pages counter.  When free_huge_page() is called for a page
on a node with a surplus, the page is freed directly to the buddy allocator
instead of the hugetlb pool.

Because shared mappings require all of the pages to be reserved up front, some
additional work must be done at mmap() to support them.  We determine the
reservation shortage and allocate the required number of pages all at once.
These pages are then added to the hugetlb pool and marked reserved.  Where that
is not possible the mmap() will fail.  As with private mappings, the
appropriate surplus counters are updated.  Since reserved huge pages won't
necessarily be used by the process, we can't be sure that free_huge_page() will
always be called to return surplus pages to the buddy allocator.  To prevent
the huge page pool from bloating, we must free unused surplus pages when their
reservation has ended.

Controlling it
==============

With the entire patch series applied, pool resizing is off by default so unless
specific action is taken, the semantics are unchanged.

To take advantage of the flexibility afforded by this patch series one must
tolerate a change in semantics.  To control hugetlb pool growth, the following
techniques can be employed:

 * A sysctl tunable to enable/disable the feature entirely
 * The size= mount option for hugetlbfs filesystems to limit pool size

Performance
===========

When contiguous memory is readily available, it is expected that the cost of
dynamicly resizing the pool will be small.  This series has been performance
tested with 'stream' to measure this cost.

Stream (http://www.cs.virginia.edu/stream/) was linked with libhugetlbfs to
enable remapping of the text and data/bss segments into huge pages.

Stream with small array
-----------------------
Baseline:  nr_hugepages = 0, No libhugetlbfs segment remapping
Preallocated: nr_hugepages = 5, Text and data/bss remapping
Dynamic: nr_hugepages = 0, Text and data/bss remapping

Rate (MB/s)
Function Baseline Preallocated Dynamic
Copy: 4695.6266 5942.8371 5982.2287
Scale: 4451.5776 5017.1419 5658.7843
Add: 5815.8849 7927.7827 8119.3552
Triad: 5949.4144 8527.6492 8110.6903

Stream with large array
-----------------------
Baseline:  nr_hugepages =  0, No libhugetlbfs segment remapping
Preallocated: nr_hugepages = 67, Text and data/bss remapping
Dynamic: nr_hugepages =  0, Text and data/bss remapping

Rate (MB/s)
Function Baseline Preallocated Dynamic
Copy: 2227.8281 2544.2732 2546.4947
Scale: 2136.3208 2430.7294 2421.2074
Add: 2773.1449 4004.0021 3999.4331
Triad: 2748.4502 3777.0109 3773.4970

* All numbers are averages taken from 10 consecutive runs with a maximum
  standard deviation of 1.3 percent noted.

This patch:

Simply move update_and_free_page() so that it can be reused later in this
patch series.  The implementation is not changed.

Signed-off-by: Adam Litke <agl@us.ibm.com>
Acked-by: Andy Whitcroft <apw@shadowen.org>
Acked-by: Dave McCracken <dave.mccracken@oracle.com>
Acked-by: William Irwin <bill.irwin@oracle.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Ken Chen <kenchen@google.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory hotplug: Hot-add with sparsemem-vmemmap

This patch is to avoid panic when memory hot-add is executed with
sparsemem-vmemmap.  Current vmemmap-sparsemem code doesn't support memory
hot-add.  Vmemmap must be populated when hot-add.  This is for
2.6.23-rc2-mm2.

Todo: # Even if this patch is applied, the message "[xxxx-xxxx] potential
        offnode page_structs" is displayed. To allocate memmap on its node,
        memmap (and pgdat) must be initialized itself like chicken and
        egg relationship.

      # vmemmap_unpopulate will be necessary for followings.
         - For cancel hot-add due to error.
         - For unplug.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fix memory hot remove not configured case.

Now, arch dependent code around CONFIG_MEMORY_HOTREMOVE is a mess.
This patch cleans up them. This is against 2.6.23-rc6-mm1.

 - fix compile failure on ia64/ CONFIG_MEMORY_HOTPLUG && !CONFIG_MEMORY_HOTREMOVE case.
 - For !CONFIG_MEMORY_HOTREMOVE, add generic no-op remove_memory(),
   which returns -EINVAL.
 - removed remove_pages() only used in powerpc.
 - removed no-op remove_memory() in i386, sh, sparc64, x86_64.

 - only powerpc returns -ENOSYS at memory hot remove(no-op). changes it
   to return -EINVAL.

Note:
Currently, only ia64 supports CONFIG_MEMORY_HOTREMOVE. I welcome other
archs if there are requirements and testers.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory unplug: ia64 interface

IA64 memory unplug interface.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory unplug: page offline

Logic.
 - set all pages in  [start,end)  as isolated migration-type.
   by this, all free pages in the range will be not-for-use.
 - Migrate all LRU pages in the range.
 - Test all pages in the range's refcnt is zero or not.

Todo:
 - allocate migration destination page from better area.
 - confirm page_count(page)== 0 && PageReserved(page) page is safe to be freed..
 (I don't like this kind of page but..
 - Find out pages which cannot be migrated.
 - more running tests.
 - Use reclaim for unplugging other memory type area.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>