safe/jmp/linux-2.6
16 years agokprobes: update document about batch registration
Masami Hiramatsu [Mon, 28 Apr 2008 09:14:30 +0000 (02:14 -0700)]
kprobes: update document about batch registration

Add the description of batch registration interfaces to
Documentation/kprobes.txt.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: David Miller <davem@davemloft.net>
Cc: "Frank Ch. Eigler" <fche@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agokprobes: add (un)register_jprobes for batch registration
Masami Hiramatsu [Mon, 28 Apr 2008 09:14:29 +0000 (02:14 -0700)]
kprobes: add (un)register_jprobes for batch registration

Introduce unregister_/register_jprobes() for jprobe batch registration.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: David Miller <davem@davemloft.net>
Cc: "Frank Ch. Eigler" <fche@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agokprobes: add (un)register_kretprobes for batch registration
Masami Hiramatsu [Mon, 28 Apr 2008 09:14:29 +0000 (02:14 -0700)]
kprobes: add (un)register_kretprobes for batch registration

Introduce unregister_/register_kretprobes() for kretprobe batch registration.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: David Miller <davem@davemloft.net>
Cc: "Frank Ch. Eigler" <fche@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agokprobes: add (un)register_kprobes for batch registration
Masami Hiramatsu [Mon, 28 Apr 2008 09:14:28 +0000 (02:14 -0700)]
kprobes: add (un)register_kprobes for batch registration

Introduce unregister_/register_kprobes() for kprobe batch registration.  This
can reduce waiting time for synchronized_sched() when a lot of probes have to
be unregistered at once.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: David Miller <davem@davemloft.net>
Cc: "Frank Ch. Eigler" <fche@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agolist.h: add list_is_singular()
Masami Hiramatsu [Mon, 28 Apr 2008 09:14:27 +0000 (02:14 -0700)]
list.h: add list_is_singular()

Add list_is_singular() to check a list has just one entry.

list_is_singular() is useful to check whether a list_head which have been
temporarily allocated for listing objects can be released or not.

Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agokprobes: prevent probing of preempt_schedule()
Srinivasa Ds [Mon, 28 Apr 2008 09:14:26 +0000 (02:14 -0700)]
kprobes: prevent probing of preempt_schedule()

Prohibit users from probing preempt_schedule().  One way of prohibiting the
user from probing functions is by marking such functions with __kprobes.  But
this method doesn't work for those functions, which are already marked to
different section like preempt_schedule() (belongs to __sched section).  So we
use blacklist approach to refuse user from probing these functions.

In blacklist approach we populate the blacklisted function's starting address
and its size in kprobe_blacklist structure.  Then we verify the user specified
address against start and end of the blacklisted function.  So any attempt to
register probe on blacklisted functions will be rejected.

[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Srinivasa DS <srinivasa@in.ibm.com>
Signed-off-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoVT notifier extension for accessibility
Karl Dahlke [Mon, 28 Apr 2008 09:14:25 +0000 (02:14 -0700)]
VT notifier extension for accessibility

Some accessibility modules need to be able to catch the output on the
console before the VT interpretation, and possibly swallow it.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomfd: use shorter set_current_state()
Robert P. J. Day [Mon, 28 Apr 2008 09:14:24 +0000 (02:14 -0700)]
mfd: use shorter set_current_state()

Since this routine declares a separate "tsk" pointer for no other reason than
to call set_task_state(), get rid of it and just invoke set_current_state().

Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: Dmitry Torokhov <dtor@mail.ru>
Cc: Nicolas Pitre <nico@cam.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosm501: add uart support
Magnus Damm [Mon, 28 Apr 2008 09:14:22 +0000 (02:14 -0700)]
sm501: add uart support

This patch extends the sm501 mfd with 8250 uart support. We're currently
doing this in the board specific r2d-1 code already, but it would be nice to
do move things into the mfd since it's more chip specific than board specific.

Signed-off-by: Magnus Damm <damm@igel.co.jp>
Cc: Ben Dooks <ben-linux@fluff.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agospi: spi_imx updates
Andrea Paterniani [Mon, 28 Apr 2008 09:14:21 +0000 (02:14 -0700)]
spi: spi_imx updates

Updates to the i.MX SPI controller driver:

 1) Some comments changed and/or added.

 2) End of transfers is now managed on TXFIFO empty interrupt after the
    last write to TXFIFO.  This speeds interrupt execution by removing
    the wait for TXFIFO to become empty.  On TXFIFO empty interrupt the
    handler needs only to poll for the end of the ongoing transaction
    (SPI_CONTROL_XCH) to close the transfer.
     (2.1) Write only transfers are closed flushing RXFIFO.
     (2.2) Read transfers are closed reading trailing bytes from RXFIFO.
     (2.3) Read transfers where RXFIFO overrun occurred are closed by
           flushing RXFIFO and aborting the message.

 3) Fifos are now flushed via SPI disable after the end of ongoing
    transaction.

Signed-off-by: Andrea Paterniani <a.paterniani@swapp-eng.it>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoatmel_spi: support zero length transfer
Atsushi Nemoto [Mon, 28 Apr 2008 09:14:19 +0000 (02:14 -0700)]
atmel_spi: support zero length transfer

A spi transfer with zero length is not invalid.  For example, such
transfer (len == 0 && delay_usecs != 0) can be used to achieve delay
before first CLK edge after chipselect assertion.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agospi: replace remaining __FUNCTION__ occurrences
Harvey Harrison [Mon, 28 Apr 2008 09:14:19 +0000 (02:14 -0700)]
spi: replace remaining __FUNCTION__ occurrences

__FUNCTION__ is gcc-specific, use __func__

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Acked-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agodocumentation: move spidev_fdx example to its own source file
Randy Dunlap [Mon, 28 Apr 2008 09:14:18 +0000 (02:14 -0700)]
documentation: move spidev_fdx example to its own source file

Move sample source code to its own source file so that it can be used
easier and build-tested/check/maintained by anyone.

(Makefile changes are in a separate patch for all of Documentation/.)

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agospi: pxa2xx_spi "sparse" fixes
David Brownell [Mon, 28 Apr 2008 09:14:17 +0000 (02:14 -0700)]
spi: pxa2xx_spi "sparse" fixes

Various cleanups to pxa2xx_spi suggested by "sparse": make sure that
register addresess are "void __iomem *", and make a few functions properly
static.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Ned Forrester <nforrester@whoi.edu>
Cc: Stephen Street <stephen@streetfiresound.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agospi: use menuconfig for CONFIG_SPI
Alessandro Guido [Mon, 28 Apr 2008 09:14:16 +0000 (02:14 -0700)]
spi: use menuconfig for CONFIG_SPI

Signed-off-by: Alessandro Guido <alessandro.guido@gmail.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agooprofile: change cpu_buffer from array to per_cpu variable
Mike Travis [Mon, 28 Apr 2008 09:14:15 +0000 (02:14 -0700)]
oprofile: change cpu_buffer from array to per_cpu variable

Change cpu_buffer from array to per_cpu variable in oprofile functions.

[akpm@linux-foundation.org: coding-style fixes]
Cc: Philippe Elie <phil.el@wanadoo.fr>
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agox86: configurable DMI scanning code
Thomas Petazzoni [Mon, 28 Apr 2008 09:14:14 +0000 (02:14 -0700)]
x86: configurable DMI scanning code

Turn CONFIG_DMI into a selectable option if EMBEDDED is defined, in
order to be able to remove the DMI table scanning code if it's not
needed, and then reduce the kernel code size.

With CONFIG_DMI (i.e before) :

   text    data     bss     dec     hex filename
1076076  128656   98304 1303036  13e1fc vmlinux

Without CONFIG_DMI (i.e after) :

   text    data     bss     dec     hex filename
1068092  126308   98304 1292704  13b9a0 vmlinux

Result:

   text    data     bss     dec     hex filename
  -7984   -2348       0  -10332   -285c vmlinux

The new option appears in "Processor type and features", only when
CONFIG_EMBEDDED is defined.

This patch is part of the Linux Tiny project, and is based on previous work
done by Matt Mackall <mpm@selenic.com>.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Anvin" <hpa@zytor.com>
Signed-off-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agodrivers/acpi/thermal.c: fix build with CONFIG_DMI=n
Andrew Morton [Mon, 28 Apr 2008 09:14:13 +0000 (02:14 -0700)]
drivers/acpi/thermal.c: fix build with CONFIG_DMI=n

drivers/acpi/thermal.c: In function 'acpi_thermal_init':
drivers/acpi/thermal.c:1794: error: 'thermal_dmi_table' undeclared (first use in this function)
drivers/acpi/thermal.c:1794: error: (Each undeclared identifier is reported only once
drivers/acpi/thermal.c:1794: error: for each function it appears in.)

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoatmel_serial: remove duplicated macro definition
Michael Trimarchi [Mon, 28 Apr 2008 09:14:11 +0000 (02:14 -0700)]
atmel_serial: remove duplicated macro definition

After commit 39d4c922b596633da86878b1a5cc881785b8e5fa (atmel_serial: fix
uart/console concurrent access) the UART_GET_TCR macro got redefined. This
patch removes the duplicated definition.

Signed-off-by: michael trimarchi <trimarchimichael@evidence.eu.com>
Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoserial: use time_before, time_before_eq, etc
Julia Lawall [Mon, 28 Apr 2008 09:14:10 +0000 (02:14 -0700)]
serial: use time_before, time_before_eq, etc

The functions time_before, time_before_eq, time_after, and time_after_eq
are more robust for comparing jiffies against other values.

A simplified version of the semantic patch making this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@ change_compare_np @
expression E;
@@

(
- jiffies <= E
+ time_before_eq(jiffies,E)
|
- jiffies >= E
+ time_after_eq(jiffies,E)
|
- jiffies < E
+ time_before(jiffies,E)
|
- jiffies > E
+ time_after(jiffies,E)
)

@ include depends on change_compare_np @
@@

#include <linux/jiffies.h>

@ no_include depends on !include && change_compare_np @
@@

  #include <linux/...>
+ #include <linux/jiffies.h>
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoserial 8250: tighten test for using backup timer
Alex Williamson [Mon, 28 Apr 2008 09:14:09 +0000 (02:14 -0700)]
serial 8250: tighten test for using backup timer

Thomas Koeller had reported an issue where a device that had been making use
of the UART_BUG_TXEN code in the 8250 driver was mistakenly being caught by
the backup timer test, causing the device to work improperly.

To fix this, tighten the test requirements to enable the backup timer
workaround.

The backup timer is really meant to catch UARTs that don't re-assert the THRE
interrupt.  The expectation is that they do initially assert THRE.  This patch
clarifies the test.

Signed-off-by: Alex Williamson <alex.williamson@hp.com>
Cc: Thomas Koeller <thomas@koeller.dyndns.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoserial: add VR41xx SIU setup for serial console
Yoichi Yuasa [Mon, 28 Apr 2008 09:14:08 +0000 (02:14 -0700)]
serial: add VR41xx SIU setup for serial console

Add VR41xx SIU setup for serial console.

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Cc: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoserial: add vr41xx_siu_early_setup() for serial console
Yoichi Yuasa [Mon, 28 Apr 2008 09:14:08 +0000 (02:14 -0700)]
serial: add vr41xx_siu_early_setup() for serial console

Add vr41xx_siu_early_setup() for serial console.

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Cc: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agouart_get_baud_rate: stop mangling termios
Alan Cox [Mon, 28 Apr 2008 09:14:07 +0000 (02:14 -0700)]
uart_get_baud_rate: stop mangling termios

Russell King noticed this one: We have to avoid replacing B0 when we pick a
baud rate for a "hung up" port.  Ugly but the proper fix is in the tty layer
and means changing the tty<->serial interfaces so we will defer that for now.

[akpm@linux-foundation.org: fix uninitialised var]
Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoFix tty speed handling on 8250
Alan Cox [Mon, 28 Apr 2008 09:14:06 +0000 (02:14 -0700)]
Fix tty speed handling on 8250

We try and write the correct speed back but the serial midlayer already
mangles the speed on us and that means if we request B0 we report back B9600
when we should not.  For now we'll hack around this in the drivers and serial
code, pending a better long term solution.

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agochar: make functions static in synclinkmp.c
Harvey Harrison [Mon, 28 Apr 2008 09:14:05 +0000 (02:14 -0700)]
char: make functions static in synclinkmp.c

All were forward declared with static.

Fixes sparse warnings:
drivers/char/synclinkmp.c:1476:5: warning: symbol 'read_proc' was not declared. Should it be static?
drivers/char/synclinkmp.c:2027:5: warning: symbol 'bh_action' was not declared. Should it be static?
drivers/char/synclinkmp.c:2058:6: warning: symbol 'bh_handler' was not declared. Should it be static?
drivers/char/synclinkmp.c:2103:6: warning: symbol 'bh_receive' was not declared. Should it be static?
drivers/char/synclinkmp.c:2112:6: warning: symbol 'bh_transmit' was not declared. Should it be static?
drivers/char/synclinkmp.c:2124:6: warning: symbol 'bh_status' was not declared. Should it be static?
drivers/char/synclinkmp.c:2136:6: warning: symbol 'isr_timer' was not declared. Should it be static?
drivers/char/synclinkmp.c:2162:6: warning: symbol 'isr_rxint' was not declared. Should it be static?
drivers/char/synclinkmp.c:2221:6: warning: symbol 'isr_rxrdy' was not declared. Should it be static?
drivers/char/synclinkmp.c:2351:6: warning: symbol 'isr_txint' was not declared. Should it be static?
drivers/char/synclinkmp.c:2379:6: warning: symbol 'isr_txrdy' was not declared. Should it be static?
drivers/char/synclinkmp.c:2410:6: warning: symbol 'isr_rxdmaok' was not declared. Should it be static?
drivers/char/synclinkmp.c:2427:6: warning: symbol 'isr_rxdmaerror' was not declared. Should it be static?
drivers/char/synclinkmp.c:2445:6: warning: symbol 'isr_txdmaok' was not declared. Should it be static?
drivers/char/synclinkmp.c:2463:6: warning: symbol 'isr_txdmaerror' was not declared. Should it be static?
drivers/char/synclinkmp.c:2480:6: warning: symbol 'isr_io_pin' was not declared. Should it be static?
drivers/char/synclinkmp.c:3420:5: warning: symbol 'alloc_dma_bufs' was not declared. Should it be static?
drivers/char/synclinkmp.c:3494:5: warning: symbol 'alloc_buf_list' was not declared. Should it be static?
drivers/char/synclinkmp.c:3553:5: warning: symbol 'alloc_frame_bufs' was not declared. Should it be static?
drivers/char/synclinkmp.c:3570:6: warning: symbol 'free_dma_bufs' was not declared. Should it be static?
drivers/char/synclinkmp.c:3580:5: warning: symbol 'alloc_tmp_rx_buf' was not declared. Should it be static?
drivers/char/synclinkmp.c:3588:6: warning: symbol 'free_tmp_rx_buf' was not declared. Should it be static?
drivers/char/synclinkmp.c:3594:5: warning: symbol 'claim_resources' was not declared. Should it be static?
drivers/char/synclinkmp.c:3681:6: warning: symbol 'release_resources' was not declared. Should it be static?
drivers/char/synclinkmp.c:3737:6: warning: symbol 'add_device' was not declared. Should it be static?
drivers/char/synclinkmp.c:3860:6: warning: symbol 'device_init' was not declared. Should it be static?
drivers/char/synclinkmp.c:4054:6: warning: symbol 'enable_loopback' was not declared. Should it be static?
drivers/char/synclinkmp.c:4101:6: warning: symbol 'set_rate' was not declared. Should it be static?
drivers/char/synclinkmp.c:4147:6: warning: symbol 'rx_stop' was not declared. Should it be static?
drivers/char/synclinkmp.c:4168:6: warning: symbol 'rx_start' was not declared. Should it be static?
drivers/char/synclinkmp.c:4225:6: warning: symbol 'tx_start' was not declared. Should it be static?
drivers/char/synclinkmp.c:4295:6: warning: symbol 'tx_stop' was not declared. Should it be static?
drivers/char/synclinkmp.c:4322:6: warning: symbol 'tx_load_fifo' was not declared. Should it be static?
drivers/char/synclinkmp.c:4371:6: warning: symbol 'reset_port' was not declared. Should it be static?
drivers/char/synclinkmp.c:4395:6: warning: symbol 'reset_adapter' was not declared. Should it be static?
drivers/char/synclinkmp.c:4407:6: warning: symbol 'async_mode' was not declared. Should it be static?
drivers/char/synclinkmp.c:4546:6: warning: symbol 'hdlc_mode' was not declared. Should it be static?
drivers/char/synclinkmp.c:4748:6: warning: symbol 'tx_set_idle' was not declared. Should it be static?
drivers/char/synclinkmp.c:4768:6: warning: symbol 'get_signals' was not declared. Should it be static?
drivers/char/synclinkmp.c:4797:6: warning: symbol 'set_signals' was not declared. Should it be static?
drivers/char/synclinkmp.c:4826:6: warning: symbol 'rx_reset_buffers' was not declared. Should it be static?
drivers/char/synclinkmp.c:4837:6: warning: symbol 'rx_free_frame_buffers' was not declared. Should it be static?
drivers/char/synclinkmp.c:4865:5: warning: symbol 'rx_get_frame' was not declared. Should it be static?
drivers/char/synclinkmp.c:5040:6: warning: symbol 'tx_load_dma_buffer' was not declared. Should it be static?
drivers/char/synclinkmp.c:5080:5: warning: symbol 'register_test' was not declared. Should it be static?
drivers/char/synclinkmp.c:5119:5: warning: symbol 'irq_test' was not declared. Should it be static?
drivers/char/synclinkmp.c:5211:5: warning: symbol 'init_adapter' was not declared. Should it be static?
drivers/char/synclinkmp.c:5270:5: warning: symbol 'loopback_test' was not declared. Should it be static?
drivers/char/synclinkmp.c:5335:5: warning: symbol 'adapter_test' was not declared. Should it be static?
drivers/char/synclinkmp.c:5397:5: warning: symbol 'memory_test' was not declared. Should it be static?
drivers/char/synclinkmp.c:5449:6: warning: symbol 'load_pci_memory' was not declared. Should it be static?
drivers/char/synclinkmp.c:5468:6: warning: symbol 'trace_block' was not declared. Should it be static?
drivers/char/synclinkmp.c:5503:6: warning: symbol 'tx_timeout' was not declared. Should it be static?
drivers/char/synclinkmp.c:5530:6: warning: symbol 'status_timeout' was not declared. Should it be static?
drivers/char/synclinkmp.c:5581:15: warning: symbol 'read_reg' was not declared. Should it be static?
drivers/char/synclinkmp.c:5586:6: warning: symbol 'write_reg' was not declared. Should it be static?
drivers/char/synclinkmp.c:5592:5: warning: symbol 'read_reg16' was not declared. Should it be static?
drivers/char/synclinkmp.c:5598:6: warning: symbol 'write_reg16' was not declared. Should it be static?
drivers/char/synclinkmp.c:5604:15: warning: symbol 'read_status_reg' was not declared. Should it be static?
drivers/char/synclinkmp.c:5610:6: warning: symbol 'write_control_reg' was not declared. Should it be static?

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Cc: Paul Fulghum <paulkf@microgate.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosynclink drivers bool conversion
Joe Perches [Mon, 28 Apr 2008 09:14:02 +0000 (02:14 -0700)]
synclink drivers bool conversion

Remove more TRUE/FALSE defines and uses
Remove == TRUE tests
Convert BOOLEAN to bool
Convert int to bool where appropriate

Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Paul Fulghum <paulkf@microgate.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoserial: silence section mismatch warnings in 8250_pci
Sam Ravnborg [Mon, 28 Apr 2008 09:14:02 +0000 (02:14 -0700)]
serial: silence section mismatch warnings in 8250_pci

Fix following warnings:
WARNING: drivers/serial/built-in.o(.data+0x5b8): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:pci_ite887x_exit()
WARNING: drivers/serial/built-in.o(.data+0x5e0): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:pci_plx9050_exit()
WARNING: drivers/serial/built-in.o(.data+0x608): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:pci_plx9050_exit()
WARNING: drivers/serial/built-in.o(.data+0x658): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:pci_plx9050_exit()
WARNING: drivers/serial/built-in.o(.data+0x680): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:pci_plx9050_exit()
WARNING: drivers/serial/built-in.o(.data+0x6a8): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:pci_plx9050_exit()
WARNING: drivers/serial/built-in.o(.data+0x6d0): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:sbs_exit()
WARNING: drivers/serial/built-in.o(.data+0x6f8): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:sbs_exit()
WARNING: drivers/serial/built-in.o(.data+0x720): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:sbs_exit()
WARNING: drivers/serial/built-in.o(.data+0x748): Section mismatch in reference from the variable pci_serial_quirks to the function .devexit.text:sbs_exit()

pci_serial_quirks contains a number of function pointers where the referenced
function is annotated __devexit.  This is OK so we annotate pci_serial_quirks
with __refdata to ignore the __devexit references

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoncpfs: fix sparse warning in ncpsign_kernel.c
Harvey Harrison [Mon, 28 Apr 2008 09:14:01 +0000 (02:14 -0700)]
ncpfs: fix sparse warning in ncpsign_kernel.c

We're casting anyway, might as well cast to the correct sign.
Specific to i386 (ifdef __i386__)

fs/ncpfs/ncpsign_kernel.c:58:23: warning: incorrect type in initializer (different signedness)
fs/ncpfs/ncpsign_kernel.c:58:23:    expected unsigned int *data2
fs/ncpfs/ncpsign_kernel.c:58:23:    got int *<noident>

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Acked-by: Petr Vandrovec <VANDROVE@vc.cvut.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoncpfs: fix sparse warnings in ioctl.c
Harvey Harrison [Mon, 28 Apr 2008 09:14:00 +0000 (02:14 -0700)]
ncpfs: fix sparse warnings in ioctl.c

In both cases, these inode variables arebeing used to test the
server's root inode against NULL.  Change them to s_inode.
fs/ncpfs/ioctl.c:391:18: warning: symbol 'inode' shadows an earlier one
fs/ncpfs/ioctl.c:264:28: originally declared here
fs/ncpfs/ioctl.c:441:17: warning: symbol 'inode' shadows an earlier one
fs/ncpfs/ioctl.c:264:28: originally declared here

In this case, we are about to return anyway, just reuse result.
fs/ncpfs/ioctl.c:521:8: warning: symbol 'result' shadows an earlier one
fs/ncpfs/ioctl.c:268:6: originally declared here

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Acked-by: Petr Vandrovec <VANDROVE@vc.cvut.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoncpfs: add prototypes to ncp_fs.h
Harvey Harrison [Mon, 28 Apr 2008 09:13:59 +0000 (02:13 -0700)]
ncpfs: add prototypes to ncp_fs.h

Removes some externs from C files, noticed from the sparse warnings:
fs/ncpfs/dir.c:90:26: warning: symbol 'ncp_root_dentry_operations' was not declared. Should it be static?
fs/ncpfs/symlink.c:107:5: warning: symbol 'ncp_symlink' was not declared. Should it be static?
fs/ncpfs/symlink.c:101:39: warning: symbol 'ncp_symlink_aops' was not declared. Should it be static?

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Acked-by: Petr Vandrovec <VANDROVE@vc.cvut.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agouml: clean up arch/um/drivers/ubd_kern.c
WANG Cong [Mon, 28 Apr 2008 09:13:58 +0000 (02:13 -0700)]
uml: clean up arch/um/drivers/ubd_kern.c

Make some global functions and variables static.

And remove some useless declarations for local functions, since we just need
to move their definitions ahead.

[jdike@addtoit.com: checkpatch cleanups]
Signed-off-by: WANG Cong <wangcong@zeuux.org>
Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoarch/um/os-Linux/sys-i386/task_size.c: improve a bit
WANG Cong [Mon, 28 Apr 2008 09:13:57 +0000 (02:13 -0700)]
arch/um/os-Linux/sys-i386/task_size.c: improve a bit

Improve this code a bit: check sigaction's return value and remove a useless
fflush().

Acked-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: WANG Cong <wangcong@zeuux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agouml: make several things static
WANG Cong [Mon, 28 Apr 2008 09:13:57 +0000 (02:13 -0700)]
uml: make several things static

Make several things static, because they no longer need to be global.

Acked-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: WANG Cong <wangcong@zeuux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agouml: make three functions static
WANG Cong [Mon, 28 Apr 2008 09:13:56 +0000 (02:13 -0700)]
uml: make three functions static

Make the following three functions static, since they don't need to be global.

arch/um/drivers/mcast_kern.c::mcast_setup()
arch/um/drivers/mconsole_user.c::mconsole_reply_v0()
arch/um/drivers/port_user.c::port_pre_exec()

Acked-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: WANG Cong <wangcong@zeuux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agouml: remove a useless function
WANG Cong [Mon, 28 Apr 2008 09:13:56 +0000 (02:13 -0700)]
uml: remove a useless function

arch/um/drivers/chan_kern.c::chan_out_fd() is not used by anyone.  Remove it.

Acked-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: WANG Cong <wangcong@zeuux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agouml: make a function static
WANG Cong [Mon, 28 Apr 2008 09:13:55 +0000 (02:13 -0700)]
uml: make a function static

arch/um/drivers/chan_kern.c::open_chan() can become static.

Acked-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: WANG Cong <wangcong@zeuux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoarch/um/os-Linux/start_up.c: various improvements.
WANG Cong [Mon, 28 Apr 2008 09:13:53 +0000 (02:13 -0700)]
arch/um/os-Linux/start_up.c: various improvements.

 - lets ptrace_child become void
 - adds checking for the return value of change_sig
 - moves errors info into stderr instead of stdout.

Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: WANG Cong <wangcong@zeuux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agouml: replace remaining __FUNCTION__ occurrences
Harvey Harrison [Mon, 28 Apr 2008 09:13:53 +0000 (02:13 -0700)]
uml: replace remaining __FUNCTION__ occurrences

__FUNCTION__ is gcc-specific, use __func__

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoarch/um/kernel/um_arch.c: some small improvements
WANG Cong [Mon, 28 Apr 2008 09:13:52 +0000 (02:13 -0700)]
arch/um/kernel/um_arch.c: some small improvements

Make some small improvements for arch/um/kernel/um_arch.c.

Signed-off-by: WANG Cong <xiyou.wangcong@gmail.com>
Acked-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agocris: add constfy to pgd_offset()
KOSAKI Motohiro [Mon, 28 Apr 2008 09:13:51 +0000 (02:13 -0700)]
cris: add constfy to pgd_offset()

add constfy to pgd_offset() for avoid following warnings.

  CC      mm/pagewalk.o
mm/pagewalk.c: In function 'walk_page_range':
mm/pagewalk.c:111: warning: passing argument 1 of 'pgd_offset' discards qualifiers from p\
ointer target type

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Matt Mackall <mpm@selenic.com>
Cc: "Vegard Nossum" <vegard.nossum@gmail.com>
Cc: Mikael Starvik <starvik@axis.com>
Cc: Jesper Nilsson <jesper.nilsson@axis.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agocris: remove redundant display of free swap space in show_mem()
Johannes Weiner [Mon, 28 Apr 2008 09:13:51 +0000 (02:13 -0700)]
cris: remove redundant display of free swap space in show_mem()

show_mem() has no need to print the amount of free swap space manually because
show_free_areas() does this already and is called by the former.

The two outputs only differ in text formatting:

  printk("Free swap  = %lukB\n", ...);
  printk("Free swap:       %6ldkB\n", ...);

Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
Cc: Mikael Starvik <mikael.starvik@axis.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agocrisv10: prepare for BKL push down
Alan Cox [Mon, 28 Apr 2008 09:13:50 +0000 (02:13 -0700)]
crisv10: prepare for BKL push down

Just the modem bits this time

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Mikael Starvik <mikael.starvik@axis.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agom68k: replace remaining __FUNCTION__ occurrences
Harvey Harrison [Mon, 28 Apr 2008 09:13:49 +0000 (02:13 -0700)]
m68k: replace remaining __FUNCTION__ occurrences

__FUNCTION__ is gcc-specific, use __func__

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agom68k: remove redundant display of free swap space in show_mem()
Johannes Weiner [Mon, 28 Apr 2008 09:13:48 +0000 (02:13 -0700)]
m68k: remove redundant display of free swap space in show_mem()

show_mem() has no need to print the amount of free swap space manually because
show_free_areas() does this already and is called by the former.

The two outputs only differ in text formatting:

  printk("Free swap  = %lukB\n", ...);
  printk("Free swap:       %6ldkB\n", ...);

Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoalpha: teach the compiler that BUG doesn't return
Andrew Morton [Mon, 28 Apr 2008 09:13:48 +0000 (02:13 -0700)]
alpha: teach the compiler that BUG doesn't return

Fix things like this:

security/selinux/netnode.c: In function 'sel_netnode_find':
security/selinux/netnode.c:126: warning: 'idx' may be used uninitialized in this function
security/selinux/netnode.c: In function 'sel_netnode_sid':
security/selinux/netnode.c:225: warning: 'ret' may be used uninitialized in this function
security/selinux/netnode.c:168: warning: 'idx' may be used uninitialized in this function

due to code correctly not expecting BUG() to return.

For some reason this reduces the object code size for that particular file.

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>
16 years agoarch/alpha/kernel/traps.c: use time_* macros
S.Caglar Onur [Mon, 28 Apr 2008 09:13:47 +0000 (02:13 -0700)]
arch/alpha/kernel/traps.c: use time_* macros

The functions time_before, time_before_eq, time_after, and time_after_eq are
more robust for comparing jiffies against other values.

So implement usage of the time_after() macro, defined in linux/jiffies.h,
which deals with wrapping correctly

[akpm@linux-foundation.org: fix warning]
Signed-off-by: S.Caglar Onur <caglar@pardus.org.tr>
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>
16 years agoalpha: replace __inline with inline
Harvey Harrison [Mon, 28 Apr 2008 09:13:46 +0000 (02:13 -0700)]
alpha: replace __inline with inline

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
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>
16 years agoalpha: remove remaining __FUNCTION__ occurrences
Harvey Harrison [Mon, 28 Apr 2008 09:13:46 +0000 (02:13 -0700)]
alpha: remove remaining __FUNCTION__ occurrences

__FUNCTION__ is gcc-specific, use __func__

The change in pci-iommu,c should be safe as arena has not been assigned
when we get to this point.

Some were within #if 0 blocks, have changed them and left the blocks
as they appear to be debugging infrastructure.

A #define FN __FUNCTION__ was removed and occurances of FN were replaced
with __func__ as well.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
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>
16 years agoalpha: handle kcalloc failure
Jim Meyering [Mon, 28 Apr 2008 09:13:44 +0000 (02:13 -0700)]
alpha: handle kcalloc failure

arch/alpha/kernel/module.c (module_frob_arch_sections): Handle kcalloc failure.

Signed-off-by: Jim Meyering <meyering@redhat.com>
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>
16 years agosmack: use cap_task_prctl
Serge E. Hallyn [Mon, 28 Apr 2008 09:13:43 +0000 (02:13 -0700)]
smack: use cap_task_prctl

With the introduction of per-process securebits, the capabilities-related
prctl callbacks were moved into cap_task_prctl().  Have smack use
cap_task_prctl() so that PR_SET_KEEPCAPS is defined.

Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Acked-by: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosmack: make smk_cipso_doi() and smk_unlbl_ambient()
Casey Schaufler [Mon, 28 Apr 2008 09:13:43 +0000 (02:13 -0700)]
smack: make smk_cipso_doi() and smk_unlbl_ambient()

The functions smk_cipso_doi and smk_unlbl_ambient are not used outside
smackfs.c and should hence be static.

Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoroot_plug: use cap_task_prctl
Serge E. Hallyn [Mon, 28 Apr 2008 09:13:42 +0000 (02:13 -0700)]
root_plug: use cap_task_prctl

With the introduction of per-process securebits, the capabilities-related
prctl callbacks were moved into cap_task_prctl().  Have root_plug use
cap_task_prctl() so that PR_SET_KEEPCAPS is defined.

Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosmack: fix integer as NULL pointer warning in smack_lsm.c
Harvey Harrison [Mon, 28 Apr 2008 09:13:41 +0000 (02:13 -0700)]
smack: fix integer as NULL pointer warning in smack_lsm.c

security/smack/smack_lsm.c:1257:16: warning: Using plain integer as NULL pointer

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Acked-by: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agocapabilities: implement per-process securebits
Andrew G. Morgan [Mon, 28 Apr 2008 09:13:40 +0000 (02:13 -0700)]
capabilities: implement per-process securebits

Filesystem capability support makes it possible to do away with (set)uid-0
based privilege and use capabilities instead.  That is, with filesystem
support for capabilities but without this present patch, it is (conceptually)
possible to manage a system with capabilities alone and never need to obtain
privilege via (set)uid-0.

Of course, conceptually isn't quite the same as currently possible since few
user applications, certainly not enough to run a viable system, are currently
prepared to leverage capabilities to exercise privilege.  Further, many
applications exist that may never get upgraded in this way, and the kernel
will continue to want to support their setuid-0 base privilege needs.

Where pure-capability applications evolve and replace setuid-0 binaries, it is
desirable that there be a mechanisms by which they can contain their
privilege.  In addition to leveraging the per-process bounding and inheritable
sets, this should include suppressing the privilege of the uid-0 superuser
from the process' tree of children.

The feature added by this patch can be leveraged to suppress the privilege
associated with (set)uid-0.  This suppression requires CAP_SETPCAP to
initiate, and only immediately affects the 'current' process (it is inherited
through fork()/exec()).  This reimplementation differs significantly from the
historical support for securebits which was system-wide, unwieldy and which
has ultimately withered to a dead relic in the source of the modern kernel.

With this patch applied a process, that is capable(CAP_SETPCAP), can now drop
all legacy privilege (through uid=0) for itself and all subsequently
fork()'d/exec()'d children with:

  prctl(PR_SET_SECUREBITS, 0x2f);

This patch represents a no-op unless CONFIG_SECURITY_FILE_CAPABILITIES is
enabled at configure time.

[akpm@linux-foundation.org: fix uninitialised var warning]
[serue@us.ibm.com: capabilities: use cap_task_prctl when !CONFIG_SECURITY]
Signed-off-by: Andrew G. Morgan <morgan@kernel.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Reviewed-by: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@tycho.nsa.gov>
Cc: Paul Moore <paul.moore@hp.com>
Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomm/nommu.c: return 0 from kobjsize with invalid objects
Michael Hennerich [Mon, 28 Apr 2008 09:13:38 +0000 (02:13 -0700)]
mm/nommu.c: return 0 from kobjsize with invalid objects

Don't perform kobjsize operations on objects the kernel doesn't manage.

On Blackfin, drivers can get dma coherent memory by calling a function
dma_alloc_coherent(). We do this in nommu by configuring a chunk of uncached
memory at the top of memory.

Since we don't want the kernel to use the uncached memory, we lie to the
kernel, and tell it that it's max memory is between 0, and the start of the
uncached dma coherent section.

this all works well, until this memory gets exposed into userspace (with a
frame buffer), when you look at the process's maps, it shows the framebuf:

root:/proc> cat maps
[snip]
03f0ef00-03f34700 rw-p 00000000 1f:00 192        /dev/fb0
root:/proc>

This is outside the "normal" range for the kernel. When the kernel tries to
find the size of this object (when you run ps), it dies in nommu.c in
kobjsize.

BUG_ON(page->index >= MAX_ORDER);

since the page we are referring to is outside what the kernel thinks is it's
max valid memory.

root:~> while [ 1 ]; ps > /dev/null; done
kernel BUG at mm/nommu.c:119!
Kernel panic - not syncing: BUG!

We fixed this by adding a check to reject out of range object pointers as it
already does that for NULL pointers.

Signed-off-by: Michael Hennerich <Michael.Hennerich@analog.com>
Signed-off-by: Robin Getz <rgetz@blackfin.uclinux.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>
16 years agovmstats: add cond_resched() to refresh_cpu_vm_stats()
Dimitri Sivanich [Mon, 28 Apr 2008 09:13:37 +0000 (02:13 -0700)]
vmstats: add cond_resched() to refresh_cpu_vm_stats()

We've found that it can take quite a bit of time (100's of usec) to get
through the zone loop in refresh_cpu_vm_stats().

Adding a cond_resched() to allow other threads to run in the non-preemptive
case.

Signed-off-by: Dimitri Sivanich <sivanich@sgi.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>
16 years agomm: fix broken gfp_zone with __GFP_THISNODE
KAMEZAWA Hiroyuki [Mon, 28 Apr 2008 09:13:36 +0000 (02:13 -0700)]
mm: fix broken gfp_zone with __GFP_THISNODE

This hack, "base = MAX_NR_ZONES", at __GFP_THISNODE was used for old
zonliests.

Now, new zonelist[] have a list for __GFP_THISNODE and this hack is incorrect.
Should be removed.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomm/page_alloc.c: remove hand-coded get_order()
Pavel Machek [Mon, 28 Apr 2008 09:13:35 +0000 (02:13 -0700)]
mm/page_alloc.c: remove hand-coded get_order()

Remove hand-coded get_order() from page_alloc.c.

Signed-off-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agooom_kill: remove unused parameter in badness()
Li Zefan [Mon, 28 Apr 2008 09:13:35 +0000 (02:13 -0700)]
oom_kill: remove unused parameter in badness()

In commit 4c4a22148909e4c003562ea7ffe0a06e26919e3c, we moved the
memcontroller-related code from badness() to select_bad_process(), so the
parameter 'mem' in badness() is unused now.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomemory hotplug: free memmaps allocated by bootmem
Yasunori Goto [Mon, 28 Apr 2008 09:13:34 +0000 (02:13 -0700)]
memory hotplug: free memmaps allocated by bootmem

This patch is to free memmaps which is allocated by bootmem.

Freeing usemap is not necessary.  The pages of usemap may be necessary for
other sections.

If removing section is last section on the node, its section is the final user
of usemap page.  (usemaps are allocated on its section by previous patch.) But
it shouldn't be freed too, because the section must be logical offline state
which all pages are isolated against page allocater.  If it is freed, page
alloctor may use it which will be removed physically soon.  It will be
disaster.  So, this patch keeps it as it is.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: 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>
16 years agomemory hotplug: allocate usemap on the section with pgdat
Yasunori Goto [Mon, 28 Apr 2008 09:13:33 +0000 (02:13 -0700)]
memory hotplug: allocate usemap on the section with pgdat

Usemaps are allocated on the section which has pgdat by this.

Because usemap size is very small, many other sections usemaps are allocated
on only one page.  If a section has usemap, it can't be removed until removing
other sections.  This dependency is not desirable for memory removing.

Pgdat has similar feature.  When a section has pgdat area, it must be the last
section for removing on the node.  So, if section A has pgdat and section B
has usemap for section A, Both sections can't be removed due to dependency
each other.

To solve this issue, this patch collects usemap on same section with pgdat.
If other sections doesn't have any dependency, this section will be able to be
removed finally.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: 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>
16 years agomemory hotplug: make alloc_bootmem_section()
Yasunori Goto [Mon, 28 Apr 2008 09:13:32 +0000 (02:13 -0700)]
memory hotplug: make alloc_bootmem_section()

alloc_bootmem_section() can allocate specified section's area.  This is used
for usemap to keep same section with pgdat by later patch.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: 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>
16 years agomemory hotplug: align memmap to page size
Yasunori Goto [Mon, 28 Apr 2008 09:13:32 +0000 (02:13 -0700)]
memory hotplug: align memmap to page size

To free memmap easier, this patch aligns it to page size.  Bootmem allocater
may mix some objects in one pages.  It's not good for freeing memmap of memory
hot-remove.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: 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>
16 years agomemory hotplug: register section/node id to free
Yasunori Goto [Mon, 28 Apr 2008 09:13:31 +0000 (02:13 -0700)]
memory hotplug: register section/node id to free

This patch set is to free pages which is allocated by bootmem for
memory-hotremove.  Some structures of memory management are allocated by
bootmem.  ex) memmap, etc.

To remove memory physically, some of them must be freed according to
circumstance.  This patch set makes basis to free those pages, and free
memmaps.

Basic my idea is using remain members of struct page to remember information
of users of bootmem (section number or node id).  When the section is
removing, kernel can confirm it.  By this information, some issues can be
solved.

  1) When the memmap of removing section is allocated on other
     section by bootmem, it should/can be free.
  2) When the memmap of removing section is allocated on the
     same section, it shouldn't be freed. Because the section has to be
     logical memory offlined already and all pages must be isolated against
     page allocater. If it is freed, page allocator may use it which will
     be removed physically soon.
  3) When removing section has other section's memmap,
     kernel will be able to show easily which section should be removed
     before it for user. (Not implemented yet)
  4) When the above case 2), the page isolation will be able to check and skip
     memmap's page when logical memory offline (offline_pages()).
     Current page isolation code fails in this case because this page is
     just reserved page and it can't distinguish this pages can be
     removed or not. But, it will be able to do by this patch.
     (Not implemented yet.)
  5) The node information like pgdat has similar issues. But, this
     will be able to be solved too by this.
     (Not implemented yet, but, remembering node id in the pages.)

Fortunately, current bootmem allocator just keeps PageReserved flags,
and doesn't use any other members of page struct. The users of
bootmem doesn't use them too.

This patch:

This is to register information which is node or section's id.  Kernel can
distinguish which node/section uses the pages allcated by bootmem.  This is
basis for hot-remove sections or nodes.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: 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>
16 years agohugetlbfs: common code update for s390
Gerald Schaefer [Mon, 28 Apr 2008 09:13:29 +0000 (02:13 -0700)]
hugetlbfs: common code update for s390

Huge ptes have a special type on s390 and cannot be handled with the standard
pte functions in certain cases, e.g.  because of a different location of the
invalid bit.  This patch adds some new architecture- specific functions to
hugetlb common code, as a prerequisite for the s390 large page support.

This won't affect other architectures in functionality, but I need to add some
new dummy inline functions to the headers.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agohugetlbfs: add missing TLB flush to hugetlb_cow()
Gerald Schaefer [Mon, 28 Apr 2008 09:13:28 +0000 (02:13 -0700)]
hugetlbfs: add missing TLB flush to hugetlb_cow()

A cow break on a hugetlbfs page with page_count > 1 will set a new pte with
set_huge_pte_at(), w/o any tlb flush operation.  The old pte will remain in
the tlb and subsequent write access to the page will result in a page fault
loop, for as long as it may take until the tlb is flushed from somewhere else.
 This patch introduces an architecture-specific huge_ptep_clear_flush()
function, which is called before the the set_huge_pte_at() in hugetlb_cow().

ATTENTION: This is just a nop on all architectures for now, the s390
implementation will come with our large page patch later.  Other architectures
should define their own huge_ptep_clear_flush() if needed.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agohugetlbfs: architecture header cleanup
Gerald Schaefer [Mon, 28 Apr 2008 09:13:27 +0000 (02:13 -0700)]
hugetlbfs: architecture header cleanup

This patch moves all architecture functions for hugetlb to architecture header
files (include/asm-foo/hugetlb.h) and converts all macros to inline functions.
 It also removes (!) ARCH_HAS_HUGEPAGE_ONLY_RANGE,
ARCH_HAS_HUGETLB_FREE_PGD_RANGE, ARCH_HAS_PREPARE_HUGEPAGE_RANGE,
ARCH_HAS_SETCLEAR_HUGE_PTE and ARCH_HAS_HUGETLB_PREFAULT_HOOK.

Getting rid of the ARCH_HAS_xxx #ifdef and macro fugliness should increase
readability and maintainability, at the price of some code duplication.  An
asm-generic common part would have reduced the loc, but we would end up with
new ARCH_HAS_xxx defines eventually.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: use struct mempolicy pointer in shmem_sb_info
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:26 +0000 (02:13 -0700)]
mempolicy: use struct mempolicy pointer in shmem_sb_info

This patch replaces the mempolicy mode, mode_flags, and nodemask in the
shmem_sb_info struct with a struct mempolicy pointer, initialized to NULL.
This removes dependency on the details of mempolicy from shmem.c and hugetlbfs
inode.c and simplifies the interfaces.

mpol_parse_str() in mempolicy.c is changed to return, via a pointer to a
pointer arg, a struct mempolicy pointer on success.  For MPOL_DEFAULT, the
returned pointer is NULL.  Further, mpol_parse_str() now takes a 'no_context'
argument that causes the input nodemask to be stored in the w.user_nodemask of
the created mempolicy for use when the mempolicy is installed in a tmpfs inode
shared policy tree.  At that time, any cpuset contextualization is applied to
the original input nodemask.  This preserves the previous behavior where the
input nodemask was stored in the superblock.  We can think of the returned
mempolicy as "context free".

Because mpol_parse_str() is now calling mpol_new(), we can remove from
mpol_to_str() the semantic checks that mpol_new() already performs.

Add 'no_context' parameter to mpol_to_str() to specify that it should format
the nodemask in w.user_nodemask for 'bind' and 'interleave' policies.

Change mpol_shared_policy_init() to take a pointer to a "context free" struct
mempolicy and to create a new, "contextualized" mempolicy using the mode,
mode_flags and user_nodemask from the input mempolicy.

  Note: we know that the mempolicy passed to mpol_to_str() or
  mpol_shared_policy_init() from a tmpfs superblock is "context free".  This
  is currently the only instance thereof.  However, if we found more uses for
  this concept, and introduced any ambiguity as to whether a mempolicy was
  context free or not, we could add another internal mode flag to identify
  context free mempolicies.  Then, we could remove the 'no_context' argument
  from mpol_to_str().

Added shmem_get_sbmpol() to return a reference counted superblock mempolicy,
if one exists, to pass to mpol_shared_policy_init().  We must add the
reference under the sb stat_lock to prevent races with replacement of the mpol
by remount.  This reference is removed in mpol_shared_policy_init().

[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: another build fix]
[akpm@linux-foundation.org: yet another build fix]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: support mpol=local tmpfs mount option
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:24 +0000 (02:13 -0700)]
mempolicy: support mpol=local tmpfs mount option

For tmpfs/shmem shared policies, MPOL_DEFAULT is not necessarily equivalent to
"local allocation".  Because shared policies are at the same "scope" level
[see Documentation/vm/numa_memory_policy.txt], as vma policies MPOL_DEFAULT
means "fall back to current task policy".

This patch extends the memory policy string parsing function to display
"local" for MPOL_PREFERRED + MPOL_F_LOCAL.  This allows one to specify local
allocation as the default policy for shared memory areas via the tmpfs mpol
mount option, regardless of the current task's policy.

Also, "local" is now displayed for this policy.  This patch allows us to
accept the same input format as the display.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: rework shmem mpol parsing and display
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:23 +0000 (02:13 -0700)]
mempolicy: rework shmem mpol parsing and display

mm/shmem.c currently contains functions to parse and display memory policy
strings for the tmpfs 'mpol' mount option.  Move this to mm/mempolicy.c with
the rest of the mempolicy support.  With subsequent patches, we'll be able to
remove knowledge of the details [mode, flags, policy, ...] completely from
shmem.c

1) replace shmem_parse_mpol() in mm/shmem.c with mpol_parse_str() in
   mm/mempolicy.c.  Rework to use the policy_types[] array [used by
   mpol_to_str()] to look up mode by name.

2) use mpol_to_str() to format policy for shmem_show_mpol().  mpol_to_str()
   expects a pointer to a struct mempolicy, so temporarily construct one.
   This will be replaced with a reference to a struct mempolicy in the tmpfs
   superblock in a subsequent patch.

   NOTE 1: I changed mpol_to_str() to use a colon ':' rather than an equal
   sign '=' as the nodemask delimiter to match mpol_parse_str() and the
   tmpfs/shmem mpol mount option formatting that now uses mpol_to_str().  This
   is a user visible change to numa_maps, but then the addition of the mode
   flags already changed the display.  It makes sense to me to have the mounts
   and numa_maps display the policy in the same format.  However, if anyone
   objects strongly, I can pass the desired nodemask delimeter as an arg to
   mpol_to_str().

   Note 2: Like show_numa_map(), I don't check the return code from
   mpol_to_str().  I do use a longer buffer than the one provided by
   show_numa_map(), which seems to have sufficed so far.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: clean-up mpol-to-str() mempolicy formatting
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:22 +0000 (02:13 -0700)]
mempolicy: clean-up mpol-to-str() mempolicy formatting

mpol-to-str() formats memory policies into printable strings.  Currently this
is only used to display "numa_maps".  A subsequent patch will use
mpol_to_str() for formatting tmpfs [shmem] mpol mount options, allowing us to
remove essentially duplicate code in mm/shmem.c.  This patch cleans up
mpol_to_str() generally and in preparation for that patch.

1) show_numa_maps() is not checking the return code from mpol_to_str().
   There's not a lot we can do in this context if mpol_to_str() did return the
   error [insufficient space in buffer].  Proposed "solution": just check,
   under DEBUG_VM, that callers are providing sufficient buffer space for the
   policy, flags, and a few nodes.  This way, we'll get some display.
   show_numa_maps() is providing a 50-byte buffer, so it won't trip this
   check.  50-bytes should be sufficient unless one has a large number of
   nodes in a very sparse nodemask.

2) The display of the new mode flags ["static" & "relative"] was set up to
   display multiple flags, separated by a "bar" '|'.  However, this support is
   incomplete--e.g., need_bar was never incremented; and currently, these two
   flags are mutually exclusive.  So remove the "bar" support, for now, and
   only display one flag.

3) Use snprint() to format flags, so as not to overflow the buffer.  Not
   that it's ever happed, AFAIK.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: use MPOL_F_LOCAL to Indicate Preferred Local Policy
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:21 +0000 (02:13 -0700)]
mempolicy: use MPOL_F_LOCAL to Indicate Preferred Local Policy

Now that we're using "preferred local" policy for system default, we need to
make this as fast as possible.  Because of the variable size of the mempolicy
structure [based on size of nodemasks], the preferred_node may be in a
different cacheline from the mode.  This can result in accessing an extra
cacheline in the normal case of system default policy.  Suspect this is the
cause of an observed 2-3% slowdown in page fault testing relative to kernel
without this patch series.

To alleviate this, use an internal mode flag, MPOL_F_LOCAL in the mempolicy
flags member which is guaranteed [?] to be in the same cacheline as the mode
itself.

Verified that reworked mempolicy now performs slightly better on 25-rc8-mm1
for both anon and shmem segments with system default and vma [preferred local]
policy.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: mPOL_PREFERRED cleanups for "local allocation"
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:20 +0000 (02:13 -0700)]
mempolicy: mPOL_PREFERRED cleanups for "local allocation"

Here are a couple of "cleanups" for MPOL_PREFERRED behavior when
v.preferred_node < 0 -- i.e., "local allocation":

1)  [do_]get_mempolicy() calls the now renamed get_policy_nodemask()
    to fetch the nodemask associated with a policy.  Currently,
    get_policy_nodemask() returns the set of nodes with memory, when
    the policy 'mode' is 'PREFERRED, and the preferred_node is < 0.
    Change to return an empty nodemask, as this is what was specified
    to achieve "local allocation".

2)  When a task is moved into a [new] cpuset, mpol_rebind_policy() is
    called to adjust any task and vma policy nodes to be valid in the
    new cpuset.  However, when the policy is MPOL_PREFERRED, and the
    preferred_node is <0, no rebind is necessary.  The "local allocation"
    indication is valid in any cpuset.  Existing code will "do the right
    thing" because node_remap() will just return the argument node when
    it is outside of the valid range of node ids.  However, I think it is
    clearer and cleaner to skip the remap explicitly in this case.

3)  mpol_to_str() produces a printable, "human readable" string from a
    struct mempolicy.  For MPOL_PREFERRED with preferred_node <0,  show
    "local", as this indicates local allocation, as the task migrates
    among nodes.  Note that this matches the usage of "local allocation"
    in libnuma() and numactl.  Without this change, I believe that node_set()
    [via set_bit()] will set bit 31, resulting in a misleading display.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: use MPOL_PREFERRED for system-wide default policy
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:18 +0000 (02:13 -0700)]
mempolicy: use MPOL_PREFERRED for system-wide default policy

Currently, when one specifies MPOL_DEFAULT via a NUMA memory policy API
[set_mempolicy(), mbind() and internal versions], the kernel simply installs a
NULL struct mempolicy pointer in the appropriate context: task policy, vma
policy, or shared policy.  This causes any use of that policy to "fall back"
to the next most specific policy scope.

The only use of MPOL_DEFAULT to mean "local allocation" is in the system
default policy.  This requires extra checks/cases for MPOL_DEFAULT in many
mempolicy.c functions.

There is another, "preferred" way to specify local allocation via the APIs.
That is using the MPOL_PREFERRED policy mode with an empty nodemask.
Internally, the empty nodemask gets converted to a preferred_node id of '-1'.
All internal usage of MPOL_PREFERRED will convert the '-1' to the id of the
node local to the cpu where the allocation occurs.

System default policy, except during boot, is hard-coded to "local
allocation".  By using the MPOL_PREFERRED mode with a negative value of
preferred node for system default policy, MPOL_DEFAULT will never occur in the
'policy' member of a struct mempolicy.  Thus, we can remove all checks for
MPOL_DEFAULT when converting policy to a node id/zonelist in the allocation
paths.

In slab_node() return local node id when policy pointer is NULL.  No need to
set a pol value to take the switch default.  Replace switch default with
BUG()--i.e., shouldn't happen.

With this patch MPOL_DEFAULT is only used in the APIs, including internal
calls to do_set_mempolicy() and in the display of policy in
/proc/<pid>/numa_maps.  It always means "fall back" to the the next most
specific policy scope.  This simplifies the description of memory policies
quite a bit, with no visible change in behavior.

get_mempolicy() continues to return MPOL_DEFAULT and an empty nodemask when
the requested policy [task or vma/shared] is NULL.  These are the values one
would supply via set_mempolicy() or mbind() to achieve that condition--default
behavior.

This patch updates Documentation to reflect this change.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: rework mempolicy Reference Counting [yet again]
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:16 +0000 (02:13 -0700)]
mempolicy: rework mempolicy Reference Counting [yet again]

After further discussion with Christoph Lameter, it has become clear that my
earlier attempts to clean up the mempolicy reference counting were a bit of
overkill in some areas, resulting in superflous ref/unref in what are usually
fast paths.  In other areas, further inspection reveals that I botched the
unref for interleave policies.

A separate patch, suitable for upstream/stable trees, fixes up the known
errors in the previous attempt to fix reference counting.

This patch reworks the memory policy referencing counting and, one hopes,
simplifies the code.  Maybe I'll get it right this time.

See the update to the numa_memory_policy.txt document for a discussion of
memory policy reference counting that motivates this patch.

Summary:

Lookup of mempolicy, based on (vma, address) need only add a reference for
shared policy, and we need only unref the policy when finished for shared
policies.  So, this patch backs out all of the unneeded extra reference
counting added by my previous attempt.  It then unrefs only shared policies
when we're finished with them, using the mpol_cond_put() [conditional put]
helper function introduced by this patch.

Note that shmem_swapin() calls read_swap_cache_async() with a dummy vma
containing just the policy.  read_swap_cache_async() can call alloc_page_vma()
multiple times, so we can't let alloc_page_vma() unref the shared policy in
this case.  To avoid this, we make a copy of any non-null shared policy and
remove the MPOL_F_SHARED flag from the copy.  This copy occurs before reading
a page [or multiple pages] from swap, so the overhead should not be an issue
here.

I introduced a new static inline function "mpol_cond_copy()" to copy the
shared policy to an on-stack policy and remove the flags that would require a
conditional free.  The current implementation of mpol_cond_copy() assumes that
the struct mempolicy contains no pointers to dynamically allocated structures
that must be duplicated or reference counted during copy.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: document {set|get}_policy() vm_ops APIs
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:14 +0000 (02:13 -0700)]
mempolicy: document {set|get}_policy() vm_ops APIs

Document mempolicy return value reference semantics assumed by the rest of the
mempolicy code for the set_ and get_policy vm_ops in <linux/mm.h>--where the
prototypes are defined--to inform any future mempolicy vm_op writers what the
rest of the subsystem expects of them.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: mark shared policies for unref
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:13 +0000 (02:13 -0700)]
mempolicy: mark shared policies for unref

As part of yet another rework of mempolicy reference counting, we want to be
able to identify shared policies efficiently, because they have an extra ref
taken on lookup that needs to be removed when we're finished using the policy.

  Note:  the extra ref is required because the policies are
  shared between tasks/processes and can be changed/freed
  by one task while another task is using them--e.g., for
  page allocation.

Building on David Rientjes mempolicy "mode flags" enhancement, this patch
indicates a "shared" policy by setting a new MPOL_F_SHARED flag in the flags
member of the struct mempolicy added by David.  MPOL_F_SHARED, and any future
"internal mode flags" are reserved from bit zero up, as they will never be
passed in the upper bits of the mode argument of a mempolicy API.

I set the MPOL_F_SHARED flag when the policy is installed in the shared policy
rb-tree.  Don't need/want to clear the flag when removing from the tree as the
mempolicy is freed [unref'd] internally to the sp_delete() function.  However,
a task could hold another reference on this mempolicy from a prior lookup.  We
need the MPOL_F_SHARED flag to stay put so that any tasks holding a ref will
unref, eventually freeing, the mempolicy.

A later patch in this series will introduce a function to conditionally unref
[mpol_free] a policy.  The MPOL_F_SHARED flag is one reason [currently the
only reason] to unref/free a policy via the conditional free.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: rename struct mempolicy 'policy' member to 'mode'
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:12 +0000 (02:13 -0700)]
mempolicy: rename struct mempolicy 'policy' member to 'mode'

The terms 'policy' and 'mode' are both used in various places to describe the
semantics of the value stored in the 'policy' member of struct mempolicy.
Furthermore, the term 'policy' is used to refer to that member, to the entire
struct mempolicy and to the more abstract concept of the tuple consisting of a
"mode" and an optional node or set of nodes.  Recently, we have added "mode
flags" that are passed in the upper bits of the 'mode' [or sometimes,
'policy'] member of the numa APIs.

I'd like to resolve this confusion, which perhaps only exists in my mind, by
renaming the 'policy' member to 'mode' throughout, and fixing up the
Documentation.  Man pages will be updated separately.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: fixup Fallback for Default Shmem Policy
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:11 +0000 (02:13 -0700)]
mempolicy: fixup Fallback for Default Shmem Policy

get_vma_policy() is not handling fallback to task policy correctly when the
get_policy() vm_op returns NULL.  The NULL overwrites the 'pol' variable that
was holding the fallback task mempolicy.  So, it was falling back directly to
system default policy.

Fix get_vma_policy() to use only non-NULL policy returned from the vma
get_policy op.

shm_get_policy() was falling back to current task's mempolicy if the "backing
file system" [tmpfs vs hugetlbfs] does not support the get_policy vm_op and
the vma policy is null.  This is incorrect for show_numa_maps() which is
likely querying the numa_maps of some task other than current.  Remove this
fallback.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: write lock mmap_sem while changing task mempolicy
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:10 +0000 (02:13 -0700)]
mempolicy: write lock mmap_sem while changing task mempolicy

A read of /proc/<pid>/numa_maps holds the target task's mmap_sem for read
while examining each vma's mempolicy.  A vma's mempolicy can fall back to the
task's policy.  However, the task could be changing it's task policy and free
the one that the show_numa_maps() is examining.

To prevent this, grab the mmap_sem for write when updating task mempolicy.
Pointed out to me by Christoph Lameter and extracted and reworked from
Christoph's alternative mempol reference counting patch.

This is analogous to the way that do_mbind() and do_get_mempolicy() prevent
races between task's sharing an mm_struct [a.k.a.  threads] setting and
querying a mempolicy for a particular address.

Note: this is necessary, but not sufficient, to allow us to stop taking an
extra reference on "other task's mempolicy" in get_vma_policy.  Subsequent
patches will complete this update, allowing us to simplify the tests for
whether we need to unref a mempolicy at various points in the code.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: rename mpol_copy to mpol_dup
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:09 +0000 (02:13 -0700)]
mempolicy: rename mpol_copy to mpol_dup

This patch renames mpol_copy() to mpol_dup() because, well, that's what it
does.  Like, e.g., strdup() for strings, mpol_dup() takes a pointer to an
existing mempolicy, allocates a new one and copies the contents.

In a later patch, I want to use the name mpol_copy() to copy the contents from
one mempolicy to another like, e.g., strcpy() does for strings.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomempolicy: rename mpol_free to mpol_put
Lee Schermerhorn [Mon, 28 Apr 2008 09:13:08 +0000 (02:13 -0700)]
mempolicy: rename mpol_free to mpol_put

This is a change that was requested some time ago by Mel Gorman.  Makes sense
to me, so here it is.

Note: I retain the name "mpol_free_shared_policy()" because it actually does
free the shared_policy, which is NOT a reference counted object.  However, ...

The mempolicy object[s] referenced by the shared_policy are reference counted,
so mpol_put() is used to release the reference held by the shared_policy.  The
mempolicy might not be freed at this time, because some task attached to the
shared object associated with the shared policy may be in the process of
allocating a page based on the mempolicy.  In that case, the task performing
the allocation will hold a reference on the mempolicy, obtained via
mpol_shared_policy_lookup().  The mempolicy will be freed when all tasks
holding such a reference have called mpol_put() for the mempolicy.

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoSubject: [PATCH] hugetlb: vmstat events for huge page allocations
Adam Litke [Mon, 28 Apr 2008 09:13:06 +0000 (02:13 -0700)]
Subject: [PATCH] hugetlb: vmstat events for huge page allocations

Allocating huge pages directly from the buddy allocator is not guaranteed to
succeed.  Success depends on several factors (such as the amount of physical
memory available and the level of fragmentation).  With the addition of
dynamic hugetlb pool resizing, allocations can occur much more frequently.
For these reasons it is desirable to keep track of huge page allocation
successes and failures.

Add two new vmstat entries to track huge page allocations that succeed and
fail.  The presence of the two entries is contingent upon CONFIG_HUGETLB_PAGE
being enabled.

[akpm@linux-foundation.org: reduced ifdeffery]
Signed-off-by: Adam Litke <agl@us.ibm.com>
Signed-off-by: Eric Munson <ebmunson@us.ibm.com>
Tested-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Andy Whitcroft <apw@shadowen.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agovmcoreinfo: add page flags values
Ken'ichi Ohmichi [Mon, 28 Apr 2008 09:13:04 +0000 (02:13 -0700)]
vmcoreinfo: add page flags values

Add some values of page flags to the vmcoreinfo data.

The vmcoreinfo data has the minimum debugging information only for dump
filtering.  makedumpfile (dump filtering command) gets it to distinguish
unnecessary pages, and makedumpfile creates a small dumpfile.

An old makedumpfile (v1.2.4 or before) had assumed some values of page flags
internally, and this implementation could not follow the change of these
values.  For example, Christoph Lameter is changing these values by the
follwing patch: http://lkml.org/lkml/2008/2/29/463

So a new makedumpfile (v1.2.5) came to need these values and I created this
patch to let the kernel output them.

Signed-off-by: Ken'ichi Ohmichi <oomichi@mxs.nes.nec.co.jp>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agos390: implement pte special bit
Nick Piggin [Mon, 28 Apr 2008 09:13:03 +0000 (02:13 -0700)]
s390: implement pte special bit

Convert XIP to support non-struct page backed memory, using VM_MIXEDMAP for
the user mappings.

This requires the get_xip_page API to be changed to an address based one.
Improve the API layering a little bit too, while we're here.

This is required in order to support XIP filesystems on memory that isn't
backed with struct page (but memory with struct page is still supported too).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Carsten Otte <cotte@de.ibm.com>
Cc: Jared Hulbert <jaredeh@gmail.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoxip: support non-struct page backed memory
Nick Piggin [Mon, 28 Apr 2008 09:13:02 +0000 (02:13 -0700)]
xip: support non-struct page backed memory

Convert XIP to support non-struct page backed memory, using VM_MIXEDMAP for
the user mappings.

This requires the get_xip_page API to be changed to an address based one.
Improve the API layering a little bit too, while we're here.

This is required in order to support XIP filesystems on memory that isn't
backed with struct page (but memory with struct page is still supported too).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Carsten Otte <cotte@de.ibm.com>
Cc: Jared Hulbert <jaredeh@gmail.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoreturn pfn from direct_access, for XIP
Jared Hulbert [Mon, 28 Apr 2008 09:13:02 +0000 (02:13 -0700)]
return pfn from direct_access, for XIP

Alter the block device ->direct_access() API to work with the new
get_xip_mem() API (that requires both kaddr and pfn are returned).

Some architectures will not do the right thing in their virt_to_page() for use
by XIP (to translate from the kernel virtual address returned by
direct_access(), to a user mappable pfn in XIP's page fault handler.

However, we can't switch it to just return the pfn and not the kaddr, because
we have no good way to get a kva from a pfn, and XIP requires the kva for its
read(2) and write(2) handlers.  So we have to return both.

Signed-off-by: Jared Hulbert <jaredeh@gmail.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Carsten Otte <cotte@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: linux-mm@kvack.org
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomm: add vm_insert_mixed
Nick Piggin [Mon, 28 Apr 2008 09:13:01 +0000 (02:13 -0700)]
mm: add vm_insert_mixed

vm_insert_mixed will insert either a raw pfn or a refcounted struct page into
the page tables, depending on whether vm_normal_page() will return the page or
not.  With the introduction of the new pte bit, this is now a too tricky for
drivers to be doing themselves.

filemap_xip uses this in a subsequent patch.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Jared Hulbert <jaredeh@gmail.com>
Cc: Carsten Otte <cotte@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomm: introduce pte_special pte bit
Nick Piggin [Mon, 28 Apr 2008 09:13:00 +0000 (02:13 -0700)]
mm: introduce pte_special pte bit

s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory
model (which is more dynamic than most).  Instead, they had proposed to
implement it with an additional path through vm_normal_page(), using a bit in
the pte to determine whether or not the page should be refcounted:

vm_normal_page()
{
...
        if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
                if (vma->vm_flags & VM_MIXEDMAP) {
#ifdef s390
if (!mixedmap_refcount_pte(pte))
return NULL;
#else
                        if (!pfn_valid(pfn))
                                return NULL;
#endif
                        goto out;
                }
...
}

This is fine, however if we are allowed to use a bit in the pte to determine
refcountedness, we can use that to _completely_ replace all the vma based
schemes.  So instead of adding more cases to the already complex vma-based
scheme, we can have a clearly seperate and simple pte-based scheme (and get
slightly better code generation in the process):

vm_normal_page()
{
#ifdef s390
if (!mixedmap_refcount_pte(pte))
return NULL;
return pte_page(pte);
#else
...
#endif
}

And finally, we may rather make this concept usable by any architecture rather
than making it s390 only, so implement a new type of pte state for this.
Unfortunately the old vma based code must stay, because some architectures may
not be able to spare pte bits.  This makes vm_normal_page a little bit more
ugly than we would like, but the 2 cases are clearly seperate.

So introduce a pte_special pte state, and use it in mm/memory.c.  It is
currently a noop for all architectures, so this doesn't actually result in any
compiled code changes to mm/memory.o.

BTW:
I haven't put vm_normal_page() into arch code as-per an earlier suggestion.
The reason is that, regardless of where vm_normal_page is actually
implemented, the *abstraction* is still exactly the same. Also, while it
depends on whether the architecture has pte_special or not, that is the
only two possible cases, and it really isn't an arch specific function --
the role of the arch code should be to provide primitive functions and
accessors with which to build the core code; pte_special does that. We do
not want architectures to know or care about vm_normal_page itself, and
we definitely don't want them being able to invent something new there
out of sight of mm/ code. If we made vm_normal_page an arch function, then
we have to make vm_insert_mixed (next patch) an arch function too. So I
don't think moving it to arch code fundamentally improves any abstractions,
while it does practically make the code more difficult to follow, for both
mm and arch developers, and easier to misuse.

[akpm@linux-foundation.org: build fix]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Carsten Otte <cotte@de.ibm.com>
Cc: Jared Hulbert <jaredeh@gmail.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomm: introduce VM_MIXEDMAP
Jared Hulbert [Mon, 28 Apr 2008 09:12:58 +0000 (02:12 -0700)]
mm: introduce VM_MIXEDMAP

This series introduces some important infrastructure work.  The overall result
is that:

1. We now support XIP backed filesystems using memory that have no
   struct page allocated to them. And patches 6 and 7 actually implement
   this for s390.

   This is pretty important in a number of cases. As far as I understand,
   in the case of virtualisation (eg. s390), each guest may mount a
   readonly copy of the same filesystem (eg. the distro). Currently,
   guests need to allocate struct pages for this image. So if you have
   100 guests, you already need to allocate more memory for the struct
   pages than the size of the image. I think. (Carsten?)

   For other (eg. embedded) systems, you may have a very large non-
   volatile filesystem. If you have to have struct pages for this, then
   your RAM consumption will go up proportionally to fs size. Even
   though it is just a small proportion, the RAM can be much more costly
   eg in terms of power, so every KB less that Linux uses makes it more
   attractive to a lot of these guys.

2. VM_MIXEDMAP allows us to support mappings where you actually do want
   to refcount _some_ pages in the mapping, but not others, and support
   COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM
   filesystem in progress. Future iterations of this filesystem will
   most likely want to migrate pages between pagecache and XIP backing,
   which is where the requirement for mixed (some refcounted, some not)
   comes from.

3. pte_special also has a peripheral usage that I need for my lockless
   get_user_pages patch. That was shown to speed up "oltp" on db2 by
   10% on a 2 socket system, which is kind of significant because they
   scrounge for months to try to find 0.1% improvement on these
   workloads. I'm hoping we might finally be faster than AIX on
   pSeries with this :). My reference to lockless get_user_pages is not
   meant to justify this patchset (which doesn't include lockless gup),
   but just to show that pte_special is not some s390 specific thing that
   should be hidden in arch code or xip code: I definitely want to use it
   on at least x86 and powerpc as well.

This patch:

Introduce a new type of mapping, VM_MIXEDMAP.  This is unlike VM_PFNMAP in
that it can support COW mappings of arbitrary ranges including ranges without
struct page *and* ranges with a struct page that we actually want to refcount
(PFNMAP can only support COW in those cases where the un-COW-ed translations
are mapped linearly in the virtual address, and can only support non
refcounted ranges).

VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not
refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it
needs to avoid refcounting pfn_valid pages eg.  for /dev/mem mappings).

Signed-off-by: Jared Hulbert <jaredeh@gmail.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Carsten Otte <cotte@de.ibm.com>
Cc: Jared Hulbert <jaredeh@gmail.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosmaps: account swap entries
Peter Zijlstra [Mon, 28 Apr 2008 09:12:55 +0000 (02:12 -0700)]
smaps: account swap entries

Show the amount of swap for each vma.  This can be used to see where all the
swap goes.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Matt Mackall <mpm@selenic.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoPAGEFLAGS_EXTENDED and separate page flags for Head and Tail
Christoph Lameter [Mon, 28 Apr 2008 09:12:55 +0000 (02:12 -0700)]
PAGEFLAGS_EXTENDED and separate page flags for Head and Tail

Having separate page flags for the head and the tail of a compound page allows
the compiler to use bitops instead of operations on a word to check for a tail
page.  That is f.e.  important for virt_to_head_page() which is used in
various critical code paths (kfree for example):

Code for PageTail(page)

Before:

 mov    (%rdi),%rdx page->flags
 mov    %rdx,%rax 3 bytes
 and    $0x12000,%eax 5 bytes
 cmp    $0x12000,%rax 6 bytes
 je     897 <kfree+0xa7>

After:

 mov    (%rdi),%rax
 test   $0x40,%ah (3 bytes)
 jne    887 <kfree+0x97>

So we go from 14 bytes to 3 bytes and from 3 instructions to one.  From the
use of 2 registers we go to none.

We can only use page flags for this if we have page flags available.  This
patch introduces CONFIG_PAGEFLAGS_EXTENDED that is set if pageflags are not
scarce due to SPARSEMEM using page flags for its sectionid on 32 bit NUMA
platforms.

Additional page flag definitions can be added to the CONFIG_PAGEFLAGS_EXTENDED
section in page-flags.h if the functionality depends on PAGEFLAGS_EXTENDED or
if more page flag overlapping tricks are used for the !PAGEFLAGS_EXTENDED
fallback (the upcoming virtual compound patch may hook in here and Rik's/Lee's
additional page flags to solve the reclaim issues could also be added there
[hint...  hint...  where are these patchsets?]).

Avoiding the overlaying of Pg_reclaim also clears the way for possible use of
compound pages for the pagecache or on the LRU.

Signed-off-by: Christoph Lameter <clameter@sgi.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>
16 years agomm: Get rid of __ZONE_COUNT
Christoph Lameter [Mon, 28 Apr 2008 09:12:54 +0000 (02:12 -0700)]
mm: Get rid of __ZONE_COUNT

It was used to compensate because MAX_NR_ZONES was not available to the
#ifdefs.  Export MAX_NR_ZONES via the new mechanism and get rid of
__ZONE_COUNT.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agopage flags: add PAGEFLAGS_FALSE for flags that are always false
Christoph Lameter [Mon, 28 Apr 2008 09:12:53 +0000 (02:12 -0700)]
page flags: add PAGEFLAGS_FALSE for flags that are always false

Turns out that there are a number of times that a flag is simply always
returning 0.  Define a macro for that.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agopage flags: handle PG_uncached like all other flags
Christoph Lameter [Mon, 28 Apr 2008 09:12:52 +0000 (02:12 -0700)]
page flags: handle PG_uncached like all other flags

Remove the special setup for PG_uncached and simply make it part of the enum.
The page flag will only be allocated when the kernel build includes the
uncached allocator.

Acked-by: Dean Nelson <dcn@sgi.com>
Cc: Jes Sorensen <jes@trained-monkey.org>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agopageflags: eliminate PG_xxx aliases
Christoph Lameter [Mon, 28 Apr 2008 09:12:52 +0000 (02:12 -0700)]
pageflags: eliminate PG_xxx aliases

Remove aliases of PG_xxx.  We can easily drop those now and alias by
specifying the PG_xxx flag in the macro that generates the functions.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agopageflags: use proper page flag functions in Xen
Christoph Lameter [Mon, 28 Apr 2008 09:12:51 +0000 (02:12 -0700)]
pageflags: use proper page flag functions in Xen

Xen uses bitops to manipulate page flags.  Make it use proper page flag
functions.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agopageflags: convert to the use of new macros
Christoph Lameter [Mon, 28 Apr 2008 09:12:50 +0000 (02:12 -0700)]
pageflags: convert to the use of new macros

Replace explicit definitions of page flags through the use of macros.
Significantly reduces the size of the definitions and removes a lot of
opportunity for errors.  Additonal page flags can typically be generated with
a single line.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agopageflags: introduce macros to generate page flag functions
Christoph Lameter [Mon, 28 Apr 2008 09:12:49 +0000 (02:12 -0700)]
pageflags: introduce macros to generate page flag functions

Introduce a set of macros that generate functions to handle page flags.

A page flag function group typically starts with either

SETPAGEFLAG(<part of function name>,<part of PG_ flagname>)

to create a set of page flag operations that are atomic. Or

__SETPAGEFLAG(<part of function name>,<part of PG_ flagname)

to create a set of page flag operations that are not atomic.

Then additional operations can be added using the following macros

TESTSCFLAG Create additional atomic test-and-set and
test-and-clear functions

TESTSETFLAG Create additional test and set function
TESTCLEARFLAG Create additional test and clear function
SETPAGEFLAG Create additional atomic set function
CLEARPAGEFLAG Create additional atomic clear function
__TESTPAGEFLAG Create additional non atomic set function
__SETPAGEFLAG Create additional non atomic clear function

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>