GitHub/mt8127/android_kernel_alcatel_ttab.git
11 years agommc: use regulator_can_change_voltage() instead of regulator_count_voltages
Marek Szyprowski [Tue, 4 Dec 2012 14:01:02 +0000 (15:01 +0100)]
mmc: use regulator_can_change_voltage() instead of regulator_count_voltages

mmc_regulator_set_ocr() depends on the ability of regulator to change the
voltage value. When regulator cannot change its voltage output, some code
is skipped to avoid reporting false errors on some boards, which use MMC
hosts with fixed regulators (e.g. Samsung Goni and UniversalC210 boards).

This patch replaces a hacky workaround based on regulator_count_voltages()
value with the correct call to recently introduced
regulator_can_change_voltage() function in regulators core.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-pxav3: add pm runtime support
Kevin Liu [Fri, 1 Feb 2013 09:48:30 +0000 (17:48 +0800)]
mmc: sdhci-pxav3: add pm runtime support

Signed-off-by: Kevin Liu <kliu5@marvell.com>
Signed-off-by: Jialing Fu <jlfu@marvell.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: core: fix indentation
Jaehoon Chung [Fri, 1 Feb 2013 05:32:22 +0000 (14:32 +0900)]
mmc: core: fix indentation

This patch fixes incorrect indentation.  (Just code cleanup)

Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: block: don't start new request when the card is removed
Seungwon Jeon [Tue, 22 Jan 2013 10:48:07 +0000 (19:48 +0900)]
mmc: block: don't start new request when the card is removed

It's not necessary to start a new request while error handling if
the card was removed.

Signed-off-by: Seungwon Jeon <tgih.jun@samsung.com>
Acked-by: Jaehoon Chung <jh80.chung@samsung.com>
Tested-by: Konstantin Dorfman <kdorfman@codeaurora.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: core: fix permanent sleep of mmcqd during card removal
Seungwon Jeon [Tue, 22 Jan 2013 10:48:03 +0000 (19:48 +0900)]
mmc: core: fix permanent sleep of mmcqd during card removal

This patch is derived from:
"mmc: fix async request mechanism for sequential read scenarios".

According as async transfer, a request is handled with twice mmc_start_req.
When the card is removed, the request is actually not issued in the first
mmc_start_req [__mmc_start_data_req]. And then mmc_wait_for_data_req_done
will come in the next mmc_start_req. But there is no event for completions.
wake_up_interruptible is needed in __mmc_start_data_req for the case of
removed card.

Signed-off-by: Seungwon Jeon <tgih.jun@samsung.com>
Acked-by: Jaehoon Chung <jh80.chung@samsung.com>
Tested-by: Konstantin Dorfman <kdorfman@codeaurora.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci: enhance preset value function
Kevin Liu [Thu, 31 Jan 2013 03:31:37 +0000 (11:31 +0800)]
mmc: sdhci: enhance preset value function

4d55c5a1 ("mmc: sdhci: enable preset value after uhs initialization")
added preset value support and enabled it by default during sd card init.

Below are the enhancements introduced by this patch:

1. In current code, preset value is enabled after setting clock finished,
which means the clock is manually set by driver firstly and then suddenly
switched to preset value at this point. So the first setting is useless
and unnecessary. What's more, the first clock setting may differ from the
preset one.  The better way is enable preset value just after switch to
UHS mode so the preset value can take effect immediately. So move preset
value enable from mmc_sd_init_card to sdhci_set_ios which will be called
during set timing.

2. In current code, preset value is disabled at the beginning of
mmc_attach_sd.  It's too late since low freq (400khz) should be set in
mmc_power_up.  So move preset value disable to sdhci_set_ios which will
be called during power up.

3. host->clock and ios->drv_type should also be updated according to the
preset value if it's enabled. Current code missed this.

4. This patch also introduce a quirk to disable preset value in case
preset value doesn't work.

This patch has been verified on sdhci-pxav3 platform with both preset
enabled and disabled.

Signed-off-by: Kevin Liu <kliu5@marvell.com>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: mmc_spi: Fix return value evaluation of irq_of_parse_and_map()
Roland Stigge [Wed, 30 Jan 2013 17:05:08 +0000 (18:05 +0100)]
mmc: mmc_spi: Fix return value evaluation of irq_of_parse_and_map()

When irq_of_parse_and_map() returns an error, it does as zero. But in
mmc_spi_get_pdata(), the error return case is compared against NO_IRQ.
This might work where NO_IRQ is zero (defaults to zero when undefined,
as on MIPS) but not where NO_IRQ is different, e.g. on ARM where it's -1.

This patch changes to comparison with 0 which is the error return value
of irq_of_parse_and_map().

Tested on ARM that mmc_spi is working now.

Signed-off-by: Roland Stigge <stigge@antcom.de>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: MAINTAINERS update for omap_hsmmc
Balaji T K [Wed, 6 Feb 2013 14:34:43 +0000 (20:04 +0530)]
mmc: MAINTAINERS update for omap_hsmmc

Update Maintainer email for omap_hsmmc, as Venkatraman will no longer
be able to maintain omap_hsmmc driver.

Signed-off-by: Balaji T K <balajitk@ti.com>
Acked-by: Venkatraman S <svenkatr@gmail.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-pltfm: Add a common clk API implementation of get_timeout_clock
Lars-Peter Clausen [Mon, 28 Jan 2013 18:27:12 +0000 (19:27 +0100)]
mmc: sdhci-pltfm: Add a common clk API implementation of get_timeout_clock

Quite a few drivers have a implementation of the get_timeout_clock
callback which simply returns the result of clk_get_rate on the device's
clock. This patch adds a common implementation of this to the sdhci-pltfm
module and replaces all custom implementations with the common one.

Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Tested-by: Stephen Warren <swarren@wwwdotorg.org>
Acked-by: Shawn Guo <shawn.guo@linaro.org>
Tested-by: Kevin Liu <kliu5@marvell.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci: update signal voltage switch code
Kevin Liu [Mon, 17 Dec 2012 11:29:26 +0000 (19:29 +0800)]
mmc: sdhci: update signal voltage switch code

The protocol related code is moved to core stack. So update the host
driver accordingly.

Signed-off-by: Kevin Liu <kliu5@marvell.com>
Tested-by: Tim Wang <wangtt@marvell.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: core: Fixup signal voltage switch
Johan Rudholm [Mon, 28 Jan 2013 14:08:28 +0000 (15:08 +0100)]
mmc: core: Fixup signal voltage switch

When switching SD and SDIO cards from 3.3V to 1.8V signal levels, the
clock should be gated for 5 ms during the step. After enabling the
clock, the host should wait for at least 1 ms before checking for
failure. Failure by the card to switch is indicated by dat[0:3] being
pulled low. The host should check for this condition and power-cycle
the card if failure is indicated.

Add a retry mechanism for the SDIO case.

If the voltage switch fails repeatedly, give up and continue the
initialization using the original voltage.

This patch places a couple of requirements on the host driver:

 1) mmc_set_ios with ios.clock = 0 must gate the clock
 2) mmc_power_off must actually cut the power to the card
 3) The card_busy host_ops member must be implemented

if these requirements are not fulfilled, the 1.8V signal voltage switch
will still be attempted but may not be successful.

Signed-off-by: Johan Rudholm <johan.rudholm@stericsson.com>
Signed-off-by: Kevin Liu <kliu5@marvell.com>
Acked-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Wei WANG <wei_wang@realsil.com.cn>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: core: Break out start_signal_voltage_switch
Johan Rudholm [Mon, 28 Jan 2013 14:08:27 +0000 (15:08 +0100)]
mmc: core: Break out start_signal_voltage_switch

Allow callers to access the start_signal_voltage_switch host_ops
member without going through any cmd11 logic. This is mostly a
preparation for the following signal voltage switch patch.

Also, reset ios.signal_voltage to its original value if
start_signal_voltage_switch fails.

Signed-off-by: Johan Rudholm <johan.rudholm@stericsson.com>
Acked-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Wei WANG <wei_wang@realsil.com.cn>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: core: Add card_busy to host_ops
Johan Rudholm [Mon, 28 Jan 2013 14:08:26 +0000 (15:08 +0100)]
mmc: core: Add card_busy to host_ops

This host_ops member is used to test if the card is signaling busy by
pulling dat[0:3] low.

Signed-off-by: Johan Rudholm <johan.rudholm@stericsson.com>
Acked-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Wei WANG <wei_wang@realsil.com.cn>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: core: Add mmc_power_cycle
Johan Rudholm [Mon, 28 Jan 2013 14:08:25 +0000 (15:08 +0100)]
mmc: core: Add mmc_power_cycle

Add mmc_power_cycle which can be used to power cycle for instance
SD-cards.

Signed-off-by: Johan Rudholm <johan.rudholm@stericsson.com>
Acked-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Wei WANG <wei_wang@realsil.com.cn>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sd: Simplify by using mmc_host_uhs
Johan Rudholm [Mon, 28 Jan 2013 14:08:24 +0000 (15:08 +0100)]
mmc: sd: Simplify by using mmc_host_uhs

Signed-off-by: Johan Rudholm <johan.rudholm@stericsson.com>
Acked-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: core: expose RPMB partition only for CMD23 capable hosts
Balaji T K [Fri, 25 Jan 2013 11:30:30 +0000 (17:00 +0530)]
mmc: core: expose RPMB partition only for CMD23 capable hosts

SET_BLOCK_COUNT CMD23 is needed for all access to RPMB partition.  If
block count is not set by CMD23, all subsequent read/write commands fail
as per eMMC specification. So, If the host does not support CMD23, do not
expose RPMB partition.

Accessing RPMB partition can cause hang / huge delay for hosts which do
not support CMD23.

Signed-off-by: Balaji T K <balajitk@ti.com>
Reported-and-Tested-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: goldfish: emulated MMC device
Mike Lockwood [Mon, 21 Jan 2013 23:43:46 +0000 (23:43 +0000)]
mmc: goldfish: emulated MMC device

This driver handles the virtual MMC device present in the Goldfish emulator.
The patch folds together initial work from Mike Lockwood and patches by
San Mehat, Jun Nakajima and Tom Keel <thomas.keel@intel.com> plus cleanups
by Alan Cox to get it all into 3.6 shape.

Signed-off-by: Mike A. Chan <mikechan@google.com>
[cleaned up and x86 support added]
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Yunhong Jiang <yunhong.jiang@intel.com>
Signed-off-by: Xiaohui Xin <xiaohui.xin@intel.com>
Signed-off-by: Jun Nakajima <jun.nakajima@intel.com>
Signed-off-by: Bruce Beare <bruce.j.beare@intel.com>
[Moved to 3.4]
Signed-off-by: Tom Keel <thomas.keel@intel.com>
[Moved to 3.7]
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: dt: bus-width can be an optional property
Shawn Guo [Mon, 28 Jan 2013 21:49:11 +0000 (16:49 -0500)]
mmc: dt: bus-width can be an optional property

None of mmc drivers implements bus-width as a required device tree
property.  Instead, some drivers like atmel-mci, dw_mmc, sdhci-s3c
implement it as an optional one, and will force bus width to be 1
when the property is absent.  Let's change the common binding to
reflect what the drivers are usually doing.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-esdhc-imx: support 8bit mode
Sascha Hauer [Mon, 21 Jan 2013 11:02:28 +0000 (19:02 +0800)]
mmc: sdhci-esdhc-imx: support 8bit mode

The i.MX esdhc has a nonstandard bit layout for the SDHCI_HOST_CONTROL
register. To support 8bit bus width on i.MX populate the platform_bus_width
callback. This is tested on an i.MX25, but should according to the datasheets
work on the other i.MX using this hardware aswell. The i.MX6, while having
a SDHCI_SPEC_300 controller, still uses the same nonstandard register layout.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Tested-by: Dirk Behme <dirk.behme@de.bosch.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci: rename platform_8bit_width to platform_bus_width
Sascha Hauer [Mon, 21 Jan 2013 11:02:27 +0000 (19:02 +0800)]
mmc: sdhci: rename platform_8bit_width to platform_bus_width

The 8bit in the function name is misleading. When set, it will be
used to set the bus width, regardless of whether 8bit or another
bus width is requested, so change the function name to
platform_bus_width.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Tested-by: Dirk Behme <dirk.behme@de.bosch.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-esdhc-imx: Auto CMD23 support for usdhc
Shawn Guo [Mon, 21 Jan 2013 11:02:26 +0000 (19:02 +0800)]
mmc: sdhci-esdhc-imx: Auto CMD23 support for usdhc

SDHCI core will try to use Auto CMD23 for mmc card.  Currently, we will
see the following message with mmc card on usdhc due to the lacking of
Auto CMD23 support in the driver.

$ mmc0: new high speed MMC card at address 0001
mmcblk1: mmc0:0001 MMC02G 1.87 GiB
mmcblk1: error -84 transferring data, sector 0, nr 8, cmd response 0x900, card status 0xb00
mmcblk1: retrying using single block read
 mmcblk1:

Enable Auto CMD23 support for usdhc so that mmc card can work in
multiple block mode.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Tested-by: Dirk Behme <dirk.behme@de.bosch.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-esdhc-imx: manually reset MIX_CTRL for usdhc
Shawn Guo [Mon, 21 Jan 2013 11:02:25 +0000 (19:02 +0800)]
mmc: sdhci-esdhc-imx: manually reset MIX_CTRL for usdhc

It's another violation to SDHC spec that software reset on usdhc
does not reset MIX_CTRL register.  Have to do it manually, otherwise
the preserving of the register bits (e.g. AC23EN) may cause mmc card
fail to be initialized.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Tested-by: Dirk Behme <dirk.behme@de.bosch.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-esdhc-imx: separate transfer mode from command write for usdhc
Shawn Guo [Mon, 21 Jan 2013 11:02:24 +0000 (19:02 +0800)]
mmc: sdhci-esdhc-imx: separate transfer mode from command write for usdhc

The combining of SDHCI_TRANSFER_MODE and SDHCI_COMMAND writes is only
required for esdhc, but not necessarily for usdhc.  Different from
esdhc where the bits for transfer mode and command are all in the same
register CMD_XFR_TYP, usdhc has a newly introduced register MIX_CTRL
to hold transfer mode bits.  So it makes more sense to separate transfer
mode from command write for usdhc.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Tested-by: Dirk Behme <dirk.behme@de.bosch.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: core: move the cache disabling operation to mmc_suspend
Maya Erez [Mon, 28 Jan 2013 21:44:22 +0000 (16:44 -0500)]
mmc: core: move the cache disabling operation to mmc_suspend

Cache control is an eMMC feature and in therefore should be
part of MMC's bus resume operations, performed in mmc_suspend,
rather than in the generic mmc_suspend_host().

Signed-off-by: Maya Erez <merez@codeaurora.org>
Acked-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agoMAINTAINERS: mmc: add maintainer entry for dw_mmc driver
Seungwon Jeon [Thu, 17 Jan 2013 12:33:23 +0000 (21:33 +0900)]
MAINTAINERS: mmc: add maintainer entry for dw_mmc driver

Add maintainer entry for the Synopsys DW host driver which is used in
various SOC including EXYNOS series.  As Will Newton will no longer be
able to take care of dw_mmc*, I and Jaehoon Chung are willing to maintain
it.

Signed-off-by: Seungwon Jeon <tgih.jun@samsung.com>
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Acked-by: Will Newton <will.newton@imgtec.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-esdhc-imx: Remove unused variables
Fabio Estevam [Thu, 17 Jan 2013 02:32:52 +0000 (00:32 -0200)]
mmc: sdhci-esdhc-imx: Remove unused variables

3f175a6e5 (mmc: sdhci-esdhc-imx: remove ESDHC_CD_GPIO handling from IO
accessory) introduced the following build warnings:

drivers/mmc/host/sdhci-esdhc-imx.c:149:30: warning: unused variable 'boarddata' [-Wunused-variable]
drivers/mmc/host/sdhci-esdhc-imx.c:181:30: warning: unused variable 'boarddata' [-Wunused-variable]

Remove the unused variables.

Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Acked-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: fix DT binding documentation SDHCI left-over
Guennadi Liakhovetski [Wed, 16 Jan 2013 16:20:41 +0000 (17:20 +0100)]
mmc: fix DT binding documentation SDHCI left-over

The file Documentation/devicetree/bindings/mmc/mmc.txt is common for all
MMC host drivers. Use a generic MMC host reference instead of an SDHCI
left-over.

Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-esdhc-imx: name esdhc specific definitions with ESDHC_ prefix
Shawn Guo [Tue, 15 Jan 2013 15:36:53 +0000 (23:36 +0800)]
mmc: sdhci-esdhc-imx: name esdhc specific definitions with ESDHC_ prefix

Rename esdhc local definitions with ESDHC_ rather than SDHCI_ prefix,
so that we can distinguish them from SDHCI core definitions from name.

A couple of bit fields are also changed use shift for consistency and
better readability.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Tested-by: Dirk Behme <dirk.behme@de.bosch.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-esdhc-imx: remove D3CD check from SDHCI_HOST_CONTROL write
Shawn Guo [Tue, 15 Jan 2013 15:36:52 +0000 (23:36 +0800)]
mmc: sdhci-esdhc-imx: remove D3CD check from SDHCI_HOST_CONTROL write

SDHCI_CTRL_D3CD is not a standard SDHCI_HOST_CONTROL, so there is no
need to check it in SDHCI_HOST_CONTROL write at all.  Remove it.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Tested-by: Dirk Behme <dirk.behme@de.bosch.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: sdhci-esdhc-imx: fix host version read
Shawn Guo [Tue, 15 Jan 2013 15:30:27 +0000 (23:30 +0800)]
mmc: sdhci-esdhc-imx: fix host version read

When commit 95a2482 (mmc: sdhci-esdhc-imx: add basic imx6q usdhc
support) works around host version issue on imx6q, it gets the
register address fixup "reg ^= 2" lost for imx25/35/51/53 esdhc.
Thus, the controller version on these SoCs is wrongly identified
as v1 while it's actually v2.

Add the address fixup back and take a different approach to correct
imx6q host version, so that the host version read gets back to work
for all SoCs.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: vt8500: Remove erroneous __exitp in wmt_mci_driver
Tony Prisk [Sun, 13 Jan 2013 06:19:20 +0000 (19:19 +1300)]
mmc: vt8500: Remove erroneous __exitp in wmt_mci_driver

With the __devinit/__devexit attributes having been removed, this
__exitp attribute causes an unused function warning and should be
removed as well.

Signed-off-by: Tony Prisk <linux@prisktech.co.nz>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: dw_mmc: Remove DW_MCI_QUIRK_NO_WRITE_PROTECT
Doug Anderson [Fri, 11 Jan 2013 17:03:54 +0000 (17:03 +0000)]
mmc: dw_mmc: Remove DW_MCI_QUIRK_NO_WRITE_PROTECT

The original quirk was added in the change 'mmc: dw_mmc: add quirk to
indicate missing write protect line'.  The original quirk was added at
a controller level even though each slot has its own write protect (so
the quirk should be at the slot level).  A recent change (mmc: dw_mmc:
Add "disable-wp" device tree property) added a slot-level quirk and
support for the quirk directly to dw_mmc.

Signed-off-by: Doug Anderson <dianders@chromium.org>
Acked-by: Will Newton <will.newton@imgtec.com>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: dw_mmc: Handle wp-gpios from device tree
Doug Anderson [Fri, 11 Jan 2013 17:03:53 +0000 (17:03 +0000)]
mmc: dw_mmc: Handle wp-gpios from device tree

On some SoCs (like exynos5250) you need to use an external GPIO for
write protect.  Add support for wp-gpios to the core dw_mmc driver
since it could be useful across multiple SoCs.

With this change I am able to make use of the write protect for the
external SD slot on exynos5250-snow.

Signed-off-by: Doug Anderson <dianders@chromium.org>
Acked-by: Seungwon Jeon <tgih.jun@samsung.com>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: dw_mmc: exynos: Remove code for wp-gpios
Doug Anderson [Fri, 11 Jan 2013 17:03:52 +0000 (17:03 +0000)]
mmc: dw_mmc: exynos: Remove code for wp-gpios

The exynos code claimed the write protect with devm_gpio_request() but
never did anything with it.  That meant that anyone using a write
protect GPIO would effectively be write protected all the time.

The handling for wp-gpios belongs in the main dw_mmc driver and has
been moved there.

Signed-off-by: Doug Anderson <dianders@chromium.org>
Acked-by: Seungwon Jeon <tgih.jun@samsung.com>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agoARM: dts: Add disable-wp for sd card slot on smdk5250
Doug Anderson [Fri, 11 Jan 2013 17:03:51 +0000 (17:03 +0000)]
ARM: dts: Add disable-wp for sd card slot on smdk5250

The next change will remove the code from the dw_mmc-exynos that added
the DW_MCI_QUIRK_NO_WRITE_PROTECT.  Keep existing functionality of
having no write protect pin on smdk5250 by adding the disable-wp
property.

Signed-off-by: Doug Anderson <dianders@chromium.org>
Acked-by: Seungwon Jeon <tgih.jun@samsung.com>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: dw_mmc: Add "disable-wp" device tree property
Doug Anderson [Fri, 11 Jan 2013 17:03:50 +0000 (17:03 +0000)]
mmc: dw_mmc: Add "disable-wp" device tree property

The "disable-wp" property is used to specify that a given SD card slot
doesn't have a concept of write protect.  This eliminates the need for
special case code for SD slots that should never be write protected
(like a micro SD slot or a dev board).

The dw_mmc driver is special in needing to specify "disable-wp"
because the lack of a "wp-gpios" property means to use the special
purpose write protect line.  On some other mmc devices the lack of
"wp-gpios" means that write protect should be disabled.

Signed-off-by: Doug Anderson <dianders@chromium.org>
Acked-by: Seungwon Jeon <tgih.jun@samsung.com>
Acked-by: Will Newton <will.newton@imgtec.com>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: correct the EXCEPTION_EVENTS_STATUS value in comment
Zhang, YiX X [Tue, 8 Jan 2013 06:07:39 +0000 (06:07 +0000)]
mmc: correct the EXCEPTION_EVENTS_STATUS value in comment

The right value is 54 according to eMMC 4.5 specification.

Signed-off-by: ZhangYi <yix.x.zhang@intel.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: mxs-mmc: Fix warning due to incorrect type
Fabio Estevam [Tue, 8 Jan 2013 00:42:35 +0000 (22:42 -0200)]
mmc: mxs-mmc: Fix warning due to incorrect type

Fixes the following warning when building with W=1 option:

drivers/mmc/host/mxs-mmc.c: In function 'mxs_mmc_adtc':
drivers/mmc/host/mxs-mmc.c:401:2: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]

The warning happens because 'i' is used in 'for_each_sg(sgl, sg, sg_len, i)' and should be made unsigned.

Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Acked-by: Marek Vasut <marex@denx.de>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: mxs-mmc: Add MODULE_ALIAS()
Fabio Estevam [Mon, 7 Jan 2013 22:41:52 +0000 (20:41 -0200)]
mmc: mxs-mmc: Add MODULE_ALIAS()

Add an entry for MODULE_ALIAS().

Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Acked-by: Marek Vasut <marex@denx.de>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agommc: mvsdio: add pinctrl integration
Thomas Petazzoni [Mon, 28 Jan 2013 11:26:53 +0000 (06:26 -0500)]
mmc: mvsdio: add pinctrl integration

On many Marvell SoCs, the pins used for the SDIO interface are part of
the MPP pins, that are muxable pins. In order to get the muxing of
those pins correct, this commit integrates the mvsdio driver with the
pinctrl infrastructure by calling devm_pinctrl_get_select_default()
during ->probe().

Note that we permit this function to fail because not all Marvell
platforms have yet been fully converted to using the pinctrl
infrastructure.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Stefan Peter <s.peter@mpl.ch>
Tested-by: Florian Fainelli <florian@openwrt.org>
Signed-off-by: Jason Cooper <jason@lakedaemon.net>
Signed-off-by: Chris Ball <cjb@laptop.org>
11 years agogianfar: fix compile fail for NET_POLL=y due to struct packing
Paul Gortmaker [Sun, 24 Feb 2013 05:38:31 +0000 (05:38 +0000)]
gianfar: fix compile fail for NET_POLL=y due to struct packing

Commit ee873fda3bec7c668407b837fc5519eb961fcd37 ("gianfar: Pack struct
gfar_priv_grp into three cachelines") moved the irq number and names
off into a separate struct and created accessors for them.  However
it was never tested with NET_POLL enabled, and so some conversions
that were simply overlooked went undetected until now.

Make the netpoll ones also use the gfar_irq() accessors.

Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Claudiu Manoil <claudiu.manoil@freescale.com>
Cc: Jianhua Xie <jianhua.xie@freescale.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agodrivers/vfio: remove depends on CONFIG_EXPERIMENTAL
Kees Cook [Sun, 24 Feb 2013 16:59:44 +0000 (09:59 -0700)]
drivers/vfio: remove depends on CONFIG_EXPERIMENTAL

The CONFIG_EXPERIMENTAL config item has not carried much meaning for a
while now and is almost always enabled by default. As agreed during the
Linux kernel summit, remove it from any "depends on" lines in Kconfigs.

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
11 years agocputime: Use local_clock() for full dynticks cputime accounting
Frederic Weisbecker [Sat, 23 Feb 2013 16:28:45 +0000 (17:28 +0100)]
cputime: Use local_clock() for full dynticks cputime accounting

Running the full dynticks cputime accounting with preemptible
kernel debugging trigger the following warning:

[    4.488303] BUG: using smp_processor_id() in preemptible [00000000] code: init/1
[    4.490971] caller is native_sched_clock+0x22/0x80
[    4.493663] Pid: 1, comm: init Not tainted 3.8.0+ #13
[    4.496376] Call Trace:
[    4.498996]  [<ffffffff813410eb>] debug_smp_processor_id+0xdb/0xf0
[    4.501716]  [<ffffffff8101e642>] native_sched_clock+0x22/0x80
[    4.504434]  [<ffffffff8101db99>] sched_clock+0x9/0x10
[    4.507185]  [<ffffffff81096ccd>] fetch_task_cputime+0xad/0x120
[    4.509916]  [<ffffffff81096dd5>] task_cputime+0x35/0x60
[    4.512622]  [<ffffffff810f146e>] acct_update_integrals+0x1e/0x40
[    4.515372]  [<ffffffff8117d2cf>] do_execve_common+0x4ff/0x5c0
[    4.518117]  [<ffffffff8117cf14>] ? do_execve_common+0x144/0x5c0
[    4.520844]  [<ffffffff81867a10>] ? rest_init+0x160/0x160
[    4.523554]  [<ffffffff8117d457>] do_execve+0x37/0x40
[    4.526276]  [<ffffffff810021a3>] run_init_process+0x23/0x30
[    4.528953]  [<ffffffff81867aac>] kernel_init+0x9c/0xf0
[    4.531608]  [<ffffffff8188356c>] ret_from_fork+0x7c/0xb0

We use sched_clock() to perform and fixup the cputime
accounting. However we are calling it with preemption enabled
from the read side, which trigger the bug above.

To fix this up, use local_clock() instead. It takes care of
preemption and also provide a more reliable clock source. This
is welcome for this kind of statistic that is widely relied on
in userspace.

Reported-by: Thomas Gleixner <tglx@linutronix.de>
Reported-by: Ingo Molnar <mingo@kernel.org>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kevin Hilman <khilman@linaro.org>
Link: http://lkml.kernel.org/r/1361636925-22288-3-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
11 years agocputime: Constify timeval_to_cputime(timeval) argument
Li Zhong [Sat, 23 Feb 2013 16:28:44 +0000 (17:28 +0100)]
cputime: Constify timeval_to_cputime(timeval) argument

Saw the following compiler warning on the linux-next tree:

  kernel/itimer.c: In function 'set_cpu_itimer':
  kernel/itimer.c:152:2: warning: passing argument 1 of 'timeval_to_cputime' discards 'const' qualifier from pointer target type [enabled by default]
  ...

timeval_to_cputime() is always passed a constant timeval in
argument, we need to teach the nsecs based cputime
implementation about that.

Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kevin Hilman <khilman@linaro.org>
Link: http://lkml.kernel.org/r/1361636925-22288-2-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kevin Hilman <khilman@linaro.org>
11 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal
Linus Torvalds [Sun, 24 Feb 2013 02:50:11 +0000 (18:50 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/signal

Pull signal handling cleanups from Al Viro:
 "This is the first pile; another one will come a bit later and will
  contain SYSCALL_DEFINE-related patches.

   - a bunch of signal-related syscalls (both native and compat)
     unified.

   - a bunch of compat syscalls switched to COMPAT_SYSCALL_DEFINE
     (fixing several potential problems with missing argument
     validation, while we are at it)

   - a lot of now-pointless wrappers killed

   - a couple of architectures (cris and hexagon) forgot to save
     altstack settings into sigframe, even though they used the
     (uninitialized) values in sigreturn; fixed.

   - microblaze fixes for delivery of multiple signals arriving at once

   - saner set of helpers for signal delivery introduced, several
     architectures switched to using those."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal: (143 commits)
  x86: convert to ksignal
  sparc: convert to ksignal
  arm: switch to struct ksignal * passing
  alpha: pass k_sigaction and siginfo_t using ksignal pointer
  burying unused conditionals
  make do_sigaltstack() static
  arm64: switch to generic old sigaction() (compat-only)
  arm64: switch to generic compat rt_sigaction()
  arm64: switch compat to generic old sigsuspend
  arm64: switch to generic compat rt_sigqueueinfo()
  arm64: switch to generic compat rt_sigpending()
  arm64: switch to generic compat rt_sigprocmask()
  arm64: switch to generic sigaltstack
  sparc: switch to generic old sigsuspend
  sparc: COMPAT_SYSCALL_DEFINE does all sign-extension as well as SYSCALL_DEFINE
  sparc: kill sign-extending wrappers for native syscalls
  kill sparc32_open()
  sparc: switch to use of generic old sigaction
  sparc: switch sys_compat_rt_sigaction() to COMPAT_SYSCALL_DEFINE
  mips: switch to generic sys_fork() and sys_clone()
  ...

11 years agoMerge branch 'drm/hdmi-for-3.9' of git://anongit.freedesktop.org/tegra/linux into...
Dave Airlie [Sun, 24 Feb 2013 02:39:42 +0000 (12:39 +1000)]
Merge branch 'drm/hdmi-for-3.9' of git://anongit.freedesktop.org/tegra/linux into drm-next

Thierry writes:
"Remove a duplicate implementation of the CEA VIC lookup and move the CEA
and other mode tables to drm_edid.c to make it more difficult to create
duplicates of the tables.

Add some helpers to pack CEA-861/HDMI AVI, audio and SPD infoframes into
binary buffers that can easily be written into hardware registers. A new
helper function makes it easy construct an AVI infoframe from a DRM
display mode.

Convert the Tegra and Radeon drivers to use the new HDMI helpers."
* 'drm/hdmi-for-3.9' of git://anongit.freedesktop.org/tegra/linux:
  drm/radeon: Use generic HDMI infoframe helpers
  drm/tegra: Use generic HDMI infoframe helpers
  drm: Add EDID helper documentation
  drm: Add HDMI infoframe helpers
  video: Add generic HDMI infoframe helpers
  drm: Add some missing forward declarations
  drm: Move mode tables to drm_edid.c
  drm: Remove duplicate drm_mode_cea_vic()

11 years agoMerge branch 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel...
Dave Airlie [Sun, 24 Feb 2013 02:39:02 +0000 (12:39 +1000)]
Merge branch 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel into drm-next

Two regressions fixes from snowboarding land

* 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel:
  drm/i915: Revert hdmi HDP pin checks
  drm/i915: Handle untiled planes when computing their offsets

11 years agoMerge branch 'drm/tegra-for-3.9' of git://anongit.freedesktop.org/tegra/linux into...
Dave Airlie [Sun, 24 Feb 2013 02:38:22 +0000 (12:38 +1000)]
Merge branch 'drm/tegra-for-3.9' of git://anongit.freedesktop.org/tegra/linux into drm-next

Thierry writes:
"Add support for 2 hardware overlays found on Tegra. These support YUV
pixel formats and can be used as video overlays. .mode_set_base() is
implemented and support for VBLANK and page-flipping is added.

A few minor bug fixes are also included and a new debugfs file allows
to inspect the framebuffers attached to the Tegra DRM device."

* 'drm/tegra-for-3.9' of git://anongit.freedesktop.org/tegra/linux:
  drm/tegra: Add list of framebuffers to debugfs
  drm/tegra: Fix color expansion
  drm/tegra: Split DC_CMD_STATE_CONTROL register write
  drm/tegra: Implement page-flipping support
  drm/tegra: Implement VBLANK support
  drm/tegra: Implement .mode_set_base()
  drm/tegra: Add plane support
  drm/tegra: Remove bogus tegra_framebuffer structure
  drm: Add consistency check for page-flipping

11 years agovlan: adjust vlan_set_encap_proto() for its callers
Cong Wang [Thu, 21 Feb 2013 23:32:27 +0000 (23:32 +0000)]
vlan: adjust vlan_set_encap_proto() for its callers

There are two places to call vlan_set_encap_proto():
vlan_untag() and __pop_vlan_tci().

vlan_untag() assumes skb->data points after mac addr, otherwise
the following code

        vhdr = (struct vlan_hdr *) skb->data;
        vlan_tci = ntohs(vhdr->h_vlan_TCI);
        __vlan_hwaccel_put_tag(skb, vlan_tci);

        skb_pull_rcsum(skb, VLAN_HLEN);

won't be correct. But __pop_vlan_tci() assumes points _before_
mac addr.

In vlan_set_encap_proto(), it looks for some magic L2 value
after mac addr:

        rawp = skb->data;
        if (*(unsigned short *) rawp == 0xFFFF)
...

Therefore __pop_vlan_tci() is obviously wrong.

A quick fix is avoiding using skb->data in vlan_set_encap_proto(),
use 'vhdr+1' is always correct in both cases.

Cc: David S. Miller <davem@davemloft.net>
Cc: Jesse Gross <jesse@nicira.com>
Signed-off-by: Cong Wang <amwang@redhat.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'akpm' (more incoming from Andrew)
Linus Torvalds [Sun, 24 Feb 2013 01:50:35 +0000 (17:50 -0800)]
Merge branch 'akpm' (more incoming from Andrew)

Merge second patch-bomb from Andrew Morton:

 - A little DM fix

 - the MM queue

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (154 commits)
  ksm: allocate roots when needed
  mm: cleanup "swapcache" in do_swap_page
  mm,ksm: swapoff might need to copy
  mm,ksm: FOLL_MIGRATION do migration_entry_wait
  ksm: shrink 32-bit rmap_item back to 32 bytes
  ksm: treat unstable nid like in stable tree
  ksm: add some comments
  tmpfs: fix mempolicy object leaks
  tmpfs: fix use-after-free of mempolicy object
  mm/fadvise.c: drain all pagevecs if POSIX_FADV_DONTNEED fails to discard all pages
  mm: export mmu notifier invalidates
  mm: accelerate mm_populate() treatment of THP pages
  mm: use long type for page counts in mm_populate() and get_user_pages()
  mm: accurately document nr_free_*_pages functions with code comments
  HWPOISON: change order of error_states[]'s elements
  HWPOISON: fix misjudgement of page_action() for errors on mlocked pages
  memcg: stop warning on memcg_propagate_kmem
  net: change type of virtio_chan->p9_max_pages
  vmscan: change type of vm_total_pages to unsigned long
  fs/nfsd: change type of max_delegations, nfsd_drc_max_mem and nfsd_drc_mem_used
  ...

11 years agoksm: allocate roots when needed
Hugh Dickins [Sat, 23 Feb 2013 00:36:12 +0000 (16:36 -0800)]
ksm: allocate roots when needed

It is a pity to have MAX_NUMNODES+MAX_NUMNODES tree roots statically
allocated, particularly when very few users will ever actually tune
merge_across_nodes 0 to use more than 1+1 of those trees.  Not a big
deal (only 16kB wasted on each machine with CONFIG_MAXSMP), but a pity.

Start off with 1+1 statically allocated, then if merge_across_nodes is
ever tuned, allocate for nr_node_ids+nr_node_ids.  Do not attempt to
free up the extra if it's tuned back, that would be a waste of effort.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: cleanup "swapcache" in do_swap_page
Hugh Dickins [Sat, 23 Feb 2013 00:36:10 +0000 (16:36 -0800)]
mm: cleanup "swapcache" in do_swap_page

I dislike the way in which "swapcache" gets used in do_swap_page():
there is always a page from swapcache there (even if maybe uncached by
the time we lock it), but tests are made according to "swapcache".
Rework that with "page != swapcache", as has been done in unuse_pte().

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm,ksm: swapoff might need to copy
Hugh Dickins [Sat, 23 Feb 2013 00:36:09 +0000 (16:36 -0800)]
mm,ksm: swapoff might need to copy

Before establishing that KSM page migration was the cause of my
WARN_ON_ONCE(page_mapped(page))s, I suspected that they came from the
lack of a ksm_might_need_to_copy() in swapoff's unuse_pte() - which in
many respects is equivalent to faulting in a page.

In fact I've never caught that as the cause: but in theory it does at
least need the KSM_RUN_UNMERGE check in ksm_might_need_to_copy(), to
avoid bringing a KSM page back in when it's not supposed to be.

I intended to copy how it's done in do_swap_page(), but have a strong
aversion to how "swapcache" ends up being used there: rework it with
"page != swapcache".

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm,ksm: FOLL_MIGRATION do migration_entry_wait
Hugh Dickins [Sat, 23 Feb 2013 00:36:07 +0000 (16:36 -0800)]
mm,ksm: FOLL_MIGRATION do migration_entry_wait

In "ksm: remove old stable nodes more thoroughly" I said that I'd never
seen its WARN_ON_ONCE(page_mapped(page)).  True at the time of writing,
but it soon appeared once I tried fuller tests on the whole series.

It turned out to be due to the KSM page migration itself: unmerge_and_
remove_all_rmap_items() failed to locate and replace all the KSM pages,
because of that hiatus in page migration when old pte has been replaced
by migration entry, but not yet by new pte.  follow_page() finds no page
at that instant, but a KSM page reappears shortly after, without a
fault.

Add FOLL_MIGRATION flag, so follow_page() can do migration_entry_wait()
for KSM's break_cow().  I'd have preferred to avoid another flag, and do
it every time, in case someone else makes the same easy mistake; but did
not find another transgressor (the common get_user_pages() is of course
safe), and cannot be sure that every follow_page() caller is prepared to
sleep - ia64's xencomm_vtop()? Now, THP's wait_split_huge_page() can
already sleep there, since anon_vma locking was changed to mutex, but
maybe that's somehow excluded.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: shrink 32-bit rmap_item back to 32 bytes
Hugh Dickins [Sat, 23 Feb 2013 00:36:06 +0000 (16:36 -0800)]
ksm: shrink 32-bit rmap_item back to 32 bytes

Think of struct rmap_item as an extension of struct page (restricted to
MADV_MERGEABLE areas): there may be a lot of them, we need to keep them
small, especially on 32-bit architectures of limited lowmem.

Siting "int nid" after "unsigned int checksum" works nicely on 64-bit,
making no change to its 64-byte struct rmap_item; but bloats the 32-bit
struct rmap_item from (nicely cache-aligned) 32 bytes to 36 bytes, which
rounds up to 40 bytes once allocated from slab.  We'd better avoid that.

Hey, I only just remembered that the anon_vma pointer in struct
rmap_item has no purpose until the rmap_item is hung from a stable tree
node (which has its own nid field); and rmap_item's nid field no purpose
than to say which tree root to tell rb_erase() when unlinking from an
unstable tree.

Double them up in a union.  There's just one place where we set anon_vma
early (when we already hold mmap_sem): now we must remove tree_rmap_item
from its unstable tree there, before overwriting nid.  No need to
spatter BUG()s around: we'd be seeing oopses if this were wrong.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: treat unstable nid like in stable tree
Hugh Dickins [Sat, 23 Feb 2013 00:36:05 +0000 (16:36 -0800)]
ksm: treat unstable nid like in stable tree

An inconsistency emerged in reviewing the NUMA node changes to KSM: when
meeting a page from the wrong NUMA node in a stable tree, we say that
it's okay for comparisons, but not as a leaf for merging; whereas when
meeting a page from the wrong NUMA node in an unstable tree, we bail out
immediately.

Now, it might be that a wrong NUMA node in an unstable tree is more
likely to correlate with instablility (different content, with rbnode
now misplaced) than page migration; but even so, we are accustomed to
instablility in the unstable tree.

Without strong evidence for which strategy is generally better, I'd
rather be consistent with what's done in the stable tree: accept a page
from the wrong NUMA node for comparison, but not as a leaf for merging.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: add some comments
Hugh Dickins [Sat, 23 Feb 2013 00:36:03 +0000 (16:36 -0800)]
ksm: add some comments

Added slightly more detail to the Documentation of merge_across_nodes, a
few comments in areas indicated by review, and renamed get_ksm_page()'s
argument from "locked" to "lock_it".  No functional change.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agotmpfs: fix mempolicy object leaks
Greg Thelen [Sat, 23 Feb 2013 00:36:02 +0000 (16:36 -0800)]
tmpfs: fix mempolicy object leaks

Fix several mempolicy leaks in the tmpfs mount logic.  These leaks are
slow - on the order of one object leaked per mount attempt.

Leak 1 (umount doesn't free mpol allocated in mount):
    while true; do
        mount -t tmpfs -o mpol=interleave,size=100M nodev /mnt
        umount /mnt
    done

Leak 2 (errors parsing remount options will leak mpol):
    mount -t tmpfs -o size=100M nodev /mnt
    while true; do
        mount -o remount,mpol=interleave,size=x /mnt 2> /dev/null
    done
    umount /mnt

Leak 3 (multiple mpol per mount leak mpol):
    while true; do
        mount -t tmpfs -o mpol=interleave,mpol=interleave,size=100M nodev /mnt
        umount /mnt
    done

This patch fixes all of the above.  I could have broken the patch into
three pieces but is seemed easier to review as one.

[akpm@linux-foundation.org: fix handling of mpol_parse_str() errors, per Hugh]
Signed-off-by: Greg Thelen <gthelen@google.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agotmpfs: fix use-after-free of mempolicy object
Greg Thelen [Sat, 23 Feb 2013 00:36:01 +0000 (16:36 -0800)]
tmpfs: fix use-after-free of mempolicy object

The tmpfs remount logic preserves filesystem mempolicy if the mpol=M
option is not specified in the remount request.  A new policy can be
specified if mpol=M is given.

Before this patch remounting an mpol bound tmpfs without specifying
mpol= mount option in the remount request would set the filesystem's
mempolicy object to a freed mempolicy object.

To reproduce the problem boot a DEBUG_PAGEALLOC kernel and run:
    # mkdir /tmp/x

    # mount -t tmpfs -o size=100M,mpol=interleave nodev /tmp/x

    # grep /tmp/x /proc/mounts
    nodev /tmp/x tmpfs rw,relatime,size=102400k,mpol=interleave:0-3 0 0

    # mount -o remount,size=200M nodev /tmp/x

    # grep /tmp/x /proc/mounts
    nodev /tmp/x tmpfs rw,relatime,size=204800k,mpol=??? 0 0
        # note ? garbage in mpol=... output above

    # dd if=/dev/zero of=/tmp/x/f count=1
        # panic here

Panic:
    BUG: unable to handle kernel NULL pointer dereference at           (null)
    IP: [<          (null)>]           (null)
    [...]
    Oops: 0010 [#1] SMP DEBUG_PAGEALLOC
    Call Trace:
      mpol_shared_policy_init+0xa5/0x160
      shmem_get_inode+0x209/0x270
      shmem_mknod+0x3e/0xf0
      shmem_create+0x18/0x20
      vfs_create+0xb5/0x130
      do_last+0x9a1/0xea0
      path_openat+0xb3/0x4d0
      do_filp_open+0x42/0xa0
      do_sys_open+0xfe/0x1e0
      compat_sys_open+0x1b/0x20
      cstar_dispatch+0x7/0x1f

Non-debug kernels will not crash immediately because referencing the
dangling mpol will not cause a fault.  Instead the filesystem will
reference a freed mempolicy object, which will cause unpredictable
behavior.

The problem boils down to a dropped mpol reference below if
shmem_parse_options() does not allocate a new mpol:

    config = *sbinfo
    shmem_parse_options(data, &config, true)
    mpol_put(sbinfo->mpol)
    sbinfo->mpol = config.mpol  /* BUG: saves unreferenced mpol */

This patch avoids the crash by not releasing the mempolicy if
shmem_parse_options() doesn't create a new mpol.

How far back does this issue go? I see it in both 2.6.36 and 3.3.  I did
not look back further.

Signed-off-by: Greg Thelen <gthelen@google.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/fadvise.c: drain all pagevecs if POSIX_FADV_DONTNEED fails to discard all pages
Mel Gorman [Sat, 23 Feb 2013 00:35:59 +0000 (16:35 -0800)]
mm/fadvise.c: drain all pagevecs if POSIX_FADV_DONTNEED fails to discard all pages

Rob van der Heij reported the following (paraphrased) on private mail.

The scenario is that I want to avoid backups to fill up the page
cache and purge stuff that is more likely to be used again (this is
with s390x Linux on z/VM, so I don't give it as much memory that
we don't care anymore). So I have something with LD_PRELOAD that
intercepts the close() call (from tar, in this case) and issues
a posix_fadvise() just before closing the file.

This mostly works, except for small files (less than 14 pages)
that remains in page cache after the face.

Unfortunately Rob has not had a chance to test this exact patch but the
test program below should be reproducing the problem he described.

The issue is the per-cpu pagevecs for LRU additions.  If the pages are
added by one CPU but fadvise() is called on another then the pages
remain resident as the invalidate_mapping_pages() only drains the local
pagevecs via its call to pagevec_release().  The user-visible effect is
that a program that uses fadvise() properly is not obeyed.

A possible fix for this is to put the necessary smarts into
invalidate_mapping_pages() to globally drain the LRU pagevecs if a
pagevec page could not be discarded.  The downside with this is that an
inode cache shrink would send a global IPI and memory pressure
potentially causing global IPI storms is very undesirable.

Instead, this patch adds a check during fadvise(POSIX_FADV_DONTNEED) to
check if invalidate_mapping_pages() discarded all the requested pages.
If a subset of pages are discarded it drains the LRU pagevecs and tries
again.  If the second attempt fails, it assumes it is due to the pages
being mapped, locked or dirty and does not care.  With this patch, an
application using fadvise() correctly will be obeyed but there is a
downside that a malicious application can force the kernel to send
global IPIs and increase overhead.

If accepted, I would like this to be considered as a -stable candidate.
It's not an urgent issue but it's a system call that is not working as
advertised which is weak.

The following test program demonstrates the problem.  It should never
report that pages are still resident but will without this patch.  It
assumes that CPU 0 and 1 exist.

int main() {
int fd;
int pagesize = getpagesize();
ssize_t written = 0, expected;
char *buf;
unsigned char *vec;
int resident, i;
cpu_set_t set;

/* Prepare a buffer for writing */
expected = FILESIZE_PAGES * pagesize;
buf = malloc(expected + 1);
if (buf == NULL) {
printf("ENOMEM\n");
exit(EXIT_FAILURE);
}
buf[expected] = 0;
memset(buf, 'a', expected);

/* Prepare the mincore vec */
vec = malloc(FILESIZE_PAGES);
if (vec == NULL) {
printf("ENOMEM\n");
exit(EXIT_FAILURE);
}

/* Bind ourselves to CPU 0 */
CPU_ZERO(&set);
CPU_SET(0, &set);
if (sched_setaffinity(getpid(), sizeof(set), &set) == -1) {
perror("sched_setaffinity");
exit(EXIT_FAILURE);
}

/* open file, unlink and write buffer */
fd = open("fadvise-test-file", O_CREAT|O_EXCL|O_RDWR);
if (fd == -1) {
perror("open");
exit(EXIT_FAILURE);
}
unlink("fadvise-test-file");
while (written < expected) {
ssize_t this_write;
this_write = write(fd, buf + written, expected - written);

if (this_write == -1) {
perror("write");
exit(EXIT_FAILURE);
}

written += this_write;
}
free(buf);

/*
 * Force ourselves to another CPU. If fadvise only flushes the local
 * CPUs pagevecs then the fadvise will fail to discard all file pages
 */
CPU_ZERO(&set);
CPU_SET(1, &set);
if (sched_setaffinity(getpid(), sizeof(set), &set) == -1) {
perror("sched_setaffinity");
exit(EXIT_FAILURE);
}

/* sync and fadvise to discard the page cache */
fsync(fd);
if (posix_fadvise(fd, 0, expected, POSIX_FADV_DONTNEED) == -1) {
perror("posix_fadvise");
exit(EXIT_FAILURE);
}

/* map the file and use mincore to see which parts of it are resident */
buf = mmap(NULL, expected, PROT_READ, MAP_SHARED, fd, 0);
if (buf == NULL) {
perror("mmap");
exit(EXIT_FAILURE);
}
if (mincore(buf, expected, vec) == -1) {
perror("mincore");
exit(EXIT_FAILURE);
}

/* Check residency */
for (i = 0, resident = 0; i < FILESIZE_PAGES; i++) {
if (vec[i])
resident++;
}
if (resident != 0) {
printf("Nr unexpected pages resident: %d\n", resident);
exit(EXIT_FAILURE);
}

munmap(buf, expected);
close(fd);
free(vec);
exit(EXIT_SUCCESS);
}

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Rob van der Heij <rvdheij@gmail.com>
Tested-by: Rob van der Heij <rvdheij@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: export mmu notifier invalidates
Cliff Wickman [Sat, 23 Feb 2013 00:35:58 +0000 (16:35 -0800)]
mm: export mmu notifier invalidates

We at SGI have a need to address some very high physical address ranges
with our GRU (global reference unit), sometimes across partitioned
machine boundaries and sometimes with larger addresses than the cpu
supports.  We do this with the aid of our own 'extended vma' module
which mimics the vma.  When something (either unmap or exit) frees an
'extended vma' we use the mmu notifiers to clean them up.

We had been able to mimic the functions
__mmu_notifier_invalidate_range_start() and
__mmu_notifier_invalidate_range_end() by locking the per-mm lock and
walking the per-mm notifier list.  But with the change to a global srcu
lock (static in mmu_notifier.c) we can no longer do that.  Our module has
no access to that lock.

So we request that these two functions be exported.

Signed-off-by: Cliff Wickman <cpw@sgi.com>
Acked-by: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: accelerate mm_populate() treatment of THP pages
Michel Lespinasse [Sat, 23 Feb 2013 00:35:56 +0000 (16:35 -0800)]
mm: accelerate mm_populate() treatment of THP pages

This change adds a follow_page_mask function which is equivalent to
follow_page, but with an extra page_mask argument.

follow_page_mask sets *page_mask to HPAGE_PMD_NR - 1 when it encounters
a THP page, and to 0 in other cases.

__get_user_pages() makes use of this in order to accelerate populating
THP ranges - that is, when both the pages and vmas arrays are NULL, we
don't need to iterate HPAGE_PMD_NR times to cover a single THP page (and
we also avoid taking mm->page_table_lock that many times).

Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: use long type for page counts in mm_populate() and get_user_pages()
Michel Lespinasse [Sat, 23 Feb 2013 00:35:55 +0000 (16:35 -0800)]
mm: use long type for page counts in mm_populate() and get_user_pages()

Use long type for page counts in mm_populate() so as to avoid integer
overflow when running the following test code:

int main(void) {
  void *p = mmap(NULL, 0x100000000000, PROT_READ,
                 MAP_PRIVATE | MAP_ANON, -1, 0);
  printf("p: %p\n", p);
  mlockall(MCL_CURRENT);
  printf("done\n");
  return 0;
}

Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: accurately document nr_free_*_pages functions with code comments
Zhang Yanfei [Sat, 23 Feb 2013 00:35:54 +0000 (16:35 -0800)]
mm: accurately document nr_free_*_pages functions with code comments

nr_free_zone_pages(), nr_free_buffer_pages() and nr_free_pagecache_pages()
are horribly badly named, so accurately document them with code comments
in case of the misuse of them.

[akpm@linux-foundation.org: tweak comments]
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoHWPOISON: change order of error_states[]'s elements
Naoya Horiguchi [Sat, 23 Feb 2013 00:35:53 +0000 (16:35 -0800)]
HWPOISON: change order of error_states[]'s elements

error_states[] has two separate states "unevictable LRU page" and
"mlocked LRU page", and the former one has the higher priority now.  But
because of that the latter one is rarely chosen because pages with
PageMlocked highly likely have PG_unevictable set.  On the other hand,
PG_unevictable without PageMlocked is common for ramfs or SHM_LOCKed
shared memory, so reversing the priority of these two states helps us
clearly distinguish them.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Chen Gong <gong.chen@linux.intel.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoHWPOISON: fix misjudgement of page_action() for errors on mlocked pages
Naoya Horiguchi [Sat, 23 Feb 2013 00:35:51 +0000 (16:35 -0800)]
HWPOISON: fix misjudgement of page_action() for errors on mlocked pages

memory_failure() can't handle memory errors on mlocked pages correctly,
because page_action() judges such errors as ones on "unknown pages"
instead of ones on "unevictable LRU page" or "mlocked LRU page".  In
order to determine page_state page_action() checks page flags at the
timing of the judgement, but such page flags are not the same with those
just after memory_failure() is called, because memory_failure() does
unmapping of the error pages before doing page_action().  This unmapping
changes the page state, especially page_remove_rmap() (called from
try_to_unmap_one()) clears PG_mlocked, so page_action() can't catch
mlocked pages after that.

With this patch, we store the page flag of the error page before doing
unmap, and (only) if the first check with page flags at the time decided
the error page is unknown, we do the second check with the stored page
flag.  This implementation doesn't change error handling for the page
types for which the first check can determine the page state correctly.

[akpm@linux-foundation.org: tweak comments]
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Chen Gong <gong.chen@linux.intel.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: stop warning on memcg_propagate_kmem
Hugh Dickins [Sat, 23 Feb 2013 00:35:50 +0000 (16:35 -0800)]
memcg: stop warning on memcg_propagate_kmem

Whilst I run the risk of a flogging for disloyalty to the Lord of Sealand,
I do have CONFIG_MEMCG=y CONFIG_MEMCG_KMEM not set, and grow tired of the
"mm/memcontrol.c:4972:12: warning: `memcg_propagate_kmem' defined but not
used [-Wunused-function]" seen in 3.8-rc: move the #ifdef outwards.

Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Glauber Costa <glommer@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agonet: change type of virtio_chan->p9_max_pages
Zhang Yanfei [Sat, 23 Feb 2013 00:35:49 +0000 (16:35 -0800)]
net: change type of virtio_chan->p9_max_pages

This member of struct virtio_chan is calculated from nr_free_buffer_pages
so change its type to unsigned long in case of overflow.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: David Miller <davem@davemloft.net>
Cc: Eric Van Hensbergen <ericvh@gmail.com>
Cc: Ron Minnich <rminnich@sandia.gov>
Cc: Latchesar Ionkov <lucho@ionkov.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agovmscan: change type of vm_total_pages to unsigned long
Zhang Yanfei [Sat, 23 Feb 2013 00:35:48 +0000 (16:35 -0800)]
vmscan: change type of vm_total_pages to unsigned long

This variable is calculated from nr_free_pagecache_pages so
change its type to unsigned long.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofs/nfsd: change type of max_delegations, nfsd_drc_max_mem and nfsd_drc_mem_used
Zhang Yanfei [Sat, 23 Feb 2013 00:35:47 +0000 (16:35 -0800)]
fs/nfsd: change type of max_delegations, nfsd_drc_max_mem and nfsd_drc_mem_used

The three variables are calculated from nr_free_buffer_pages so change
their types to unsigned long in case of overflow.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofs/buffer.c: change type of max_buffer_heads to unsigned long
Zhang Yanfei [Sat, 23 Feb 2013 00:35:46 +0000 (16:35 -0800)]
fs/buffer.c: change type of max_buffer_heads to unsigned long

max_buffer_heads is calculated from nr_free_buffer_pages(), so change
its type to unsigned long in case of overflow.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoia64: use %ld to print pages calculated in nr_free_buffer_pages
Zhang Yanfei [Sat, 23 Feb 2013 00:35:45 +0000 (16:35 -0800)]
ia64: use %ld to print pages calculated in nr_free_buffer_pages

Now the function nr_free_buffer_pages returns unsigned long, so use %ld
to print its return value.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: fix return type for functions nr_free_*_pages
Zhang Yanfei [Sat, 23 Feb 2013 00:35:43 +0000 (16:35 -0800)]
mm: fix return type for functions nr_free_*_pages

Currently, the amount of RAM that functions nr_free_*_pages return is
held in unsigned int.  But in machines with big memory (exceeding 16TB),
the amount may be incorrect because of overflow, so fix it.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Simon Horman <horms@verge.net.au>
Cc: Julian Anastasov <ja@ssi.bg>
Cc: David Miller <davem@davemloft.net>
Cc: Eric Van Hensbergen <ericvh@gmail.com>
Cc: Ron Minnich <rminnich@sandia.gov>
Cc: Latchesar Ionkov <lucho@ionkov.net>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: cleanup mem_cgroup_init comment
Michal Hocko [Sat, 23 Feb 2013 00:35:41 +0000 (16:35 -0800)]
memcg: cleanup mem_cgroup_init comment

We should encourage all memcg controller initialization independent on a
specific mem_cgroup to be done here rather than exploit css_alloc
callback and assume that nothing happens before root cgroup is created.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <htejun@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: move memcg_stock initialization to mem_cgroup_init
Michal Hocko [Sat, 23 Feb 2013 00:35:40 +0000 (16:35 -0800)]
memcg: move memcg_stock initialization to mem_cgroup_init

memcg_stock are currently initialized during the root cgroup allocation
which is OK but it pointlessly pollutes memcg allocation code with
something that can be called when the memcg subsystem is initialized by
mem_cgroup_init along with other controller specific parts.

This patch wraps the current memcg_stock initialization code into a
helper calls it from the controller subsystem initialization code.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <htejun@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: move mem_cgroup_soft_limit_tree_init to mem_cgroup_init
Michal Hocko [Sat, 23 Feb 2013 00:35:39 +0000 (16:35 -0800)]
memcg: move mem_cgroup_soft_limit_tree_init to mem_cgroup_init

Per-node-zone soft limit tree is currently initialized when the root
cgroup is created which is OK but it pointlessly pollutes memcg
allocation code with something that can be called when the memcg
subsystem is initialized by mem_cgroup_init along with other controller
specific parts.

While we are at it let's make mem_cgroup_soft_limit_tree_init void
because it doesn't make much sense to report memory failure because if
we fail to allocate memory that early during the boot then we are
screwed anyway (this saves some code).

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <htejun@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: use up free swap space before reaching OOM kill
Minchan Kim [Sat, 23 Feb 2013 00:35:37 +0000 (16:35 -0800)]
mm: use up free swap space before reaching OOM kill

Recently, Luigi reported there are lots of free swap space when OOM
happens.  It's easily reproduced on zram-over-swap, where many instance
of memory hogs are running and laptop_mode is enabled.  He said there
was no problem when he disabled laptop_mode.  The problem when I
investigate problem is following as.

Assumption for easy explanation: There are no page cache page in system
because they all are already reclaimed.

1. try_to_free_pages disable may_writepage when laptop_mode is enabled.
2. shrink_inactive_list isolates victim pages from inactive anon lru list.
3. shrink_page_list adds them to swapcache via add_to_swap but it doesn't
   pageout because sc->may_writepage is 0 so the page is rotated back into
   inactive anon lru list. The add_to_swap made the page Dirty by SetPageDirty.
4. 3 couldn't reclaim any pages so do_try_to_free_pages increase priority and
   retry reclaim with higher priority.
5. shrink_inactlive_list try to isolate victim pages from inactive anon lru list
   but got failed because it try to isolate pages with ISOLATE_CLEAN mode but
   inactive anon lru list is full of dirty pages by 3 so it just returns
   without  any reclaim progress.
6. do_try_to_free_pages doesn't set may_writepage due to zero total_scanned.
   Because sc->nr_scanned is increased by shrink_page_list but we don't call
   shrink_page_list in 5 due to short of isolated pages.

Above loop is continued until OOM happens.

The problem didn't happen before [1] was merged because old logic's
isolatation in shrink_inactive_list was successful and tried to call
shrink_page_list to pageout them but it still ends up failed to page out
by may_writepage.  But important point is that sc->nr_scanned was
increased although we couldn't swap out them so do_try_to_free_pages
could set may_writepages.

Since commit f80c0673610e ("mm: zone_reclaim: make isolate_lru_page()
filter-aware") was introduced, it's not a good idea any more to depends
on only the number of scanned pages for setting may_writepage.  So this
patch adds new trigger point of setting may_writepage as below
DEF_PRIOIRTY - 2 which is used to show the significant memory pressure
in VM so it's good fit for our purpose which would be better to lose
power saving or clickety rather than OOM killing.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Reported-by: Luigi Semenzato <semenzato@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: use NUMA_NO_NODE
David Rientjes [Sat, 23 Feb 2013 00:35:36 +0000 (16:35 -0800)]
mm: use NUMA_NO_NODE

Make a sweep through mm/ and convert code that uses -1 directly to using
the more appropriate NUMA_NO_NODE.

Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agommu_notifier_unregister NULL Pointer deref and multiple ->release() callouts
Robin Holt [Sat, 23 Feb 2013 00:35:34 +0000 (16:35 -0800)]
mmu_notifier_unregister NULL Pointer deref and multiple ->release() callouts

There is a race condition between mmu_notifier_unregister() and
__mmu_notifier_release().

Assume two tasks, one calling mmu_notifier_unregister() as a result of a
filp_close() ->flush() callout (task A), and the other calling
mmu_notifier_release() from an mmput() (task B).

                A                               B
t1                                              srcu_read_lock()
t2              if (!hlist_unhashed())
t3                                              srcu_read_unlock()
t4              srcu_read_lock()
t5                                              hlist_del_init_rcu()
t6                                              synchronize_srcu()
t7              srcu_read_unlock()
t8              hlist_del_rcu()  <--- NULL pointer deref.

Additionally, the list traversal in __mmu_notifier_release() is not
protected by the by the mmu_notifier_mm->hlist_lock which can result in
callouts to the ->release() notifier from both mmu_notifier_unregister()
and __mmu_notifier_release().

-stable suggestions:

The stable trees prior to 3.7.y need commits 21a92735f660 and
70400303ce0c cherry-picked in that order prior to cherry-picking this
commit.  The 3.7.y tree already has those two commits.

Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Sagi Grimberg <sagig@mellanox.co.il>
Cc: Haggai Eran <haggaie@mellanox.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/memory_hotplug: use pgdat_end_pfn() instead of open coding the same.
Cody P Schafer [Sat, 23 Feb 2013 00:35:32 +0000 (16:35 -0800)]
mm/memory_hotplug: use pgdat_end_pfn() instead of open coding the same.

Replace open coded pgdat_end_pfn() with helper function.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/memory_hotplug: use ensure_zone_is_initialized()
Cody P Schafer [Sat, 23 Feb 2013 00:35:31 +0000 (16:35 -0800)]
mm/memory_hotplug: use ensure_zone_is_initialized()

Remove open coding of ensure_zone_is_initialzied().

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: add helper ensure_zone_is_initialized()
Cody P Schafer [Sat, 23 Feb 2013 00:35:30 +0000 (16:35 -0800)]
mm: add helper ensure_zone_is_initialized()

ensure_zone_is_initialized() checks if a zone is in a empty & not
initialized state (typically occuring after it is created in memory
hotplugging), and, if so, calls init_currently_empty_zone() to
initialize the zone.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/page_alloc: add informative debugging message in page_outside_zone_boundaries()
Cody P Schafer [Sat, 23 Feb 2013 00:35:28 +0000 (16:35 -0800)]
mm/page_alloc: add informative debugging message in page_outside_zone_boundaries()

Add a debug message which prints when a page is found outside of the
boundaries of the zone it should belong to. Format is:
"page $pfn outside zone [ $start_pfn - $end_pfn ]"

[akpm@linux-foundation.org: s/pr_debug/pr_err/]
Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agommzone: add pgdat_{end_pfn,is_empty}() helpers & consolidate.
Cody P Schafer [Sat, 23 Feb 2013 00:35:27 +0000 (16:35 -0800)]
mmzone: add pgdat_{end_pfn,is_empty}() helpers & consolidate.

Add pgdat_end_pfn() and pgdat_is_empty() helpers which match the similar
zone_*() functions.

Change node_end_pfn() to be a wrapper of pgdat_end_pfn().

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/page_alloc: add a VM_BUG in __free_one_page() if the zone is uninitialized.
Cody P Schafer [Sat, 23 Feb 2013 00:35:25 +0000 (16:35 -0800)]
mm/page_alloc: add a VM_BUG in __free_one_page() if the zone is uninitialized.

Freeing pages to uninitialized zones is not handled by
__free_one_page(), and should never happen when the code is correct.

Ran into this while writing some code that dynamically onlines extra
zones.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: add zone_is_empty() and zone_is_initialized()
Cody P Schafer [Sat, 23 Feb 2013 00:35:24 +0000 (16:35 -0800)]
mm: add zone_is_empty() and zone_is_initialized()

Factoring out these 2 checks makes it more clear what we are actually
checking for.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: add & use zone_end_pfn() and zone_spans_pfn()
Cody P Schafer [Sat, 23 Feb 2013 00:35:23 +0000 (16:35 -0800)]
mm: add & use zone_end_pfn() and zone_spans_pfn()

Add 2 helpers (zone_end_pfn() and zone_spans_pfn()) to reduce code
duplication.

This also switches to using them in compaction (where an additional
variable needed to be renamed), page_alloc, vmstat, memory_hotplug, and
kmemleak.

Note that in compaction.c I avoid calling zone_end_pfn() repeatedly
because I expect at some point the sycronization issues with start_pfn &
spanned_pages will need fixing, either by actually using the seqlock or
clever memory barrier usage.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: add SECTION_IN_PAGE_FLAGS
Cody P Schafer [Sat, 23 Feb 2013 00:35:21 +0000 (16:35 -0800)]
mm: add SECTION_IN_PAGE_FLAGS

Instead of directly utilizing a combination of config options to determine
this, add a macro to specifically address it.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Cc: David Hansen <dave@linux.vnet.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/mlock.c: document scary-looking stack expansion mlock chain
Johannes Weiner [Sat, 23 Feb 2013 00:35:20 +0000 (16:35 -0800)]
mm/mlock.c: document scary-looking stack expansion mlock chain

The fact that mlock calls get_user_pages, and get_user_pages might call
mlock when expanding a stack looks like a potential recursion.

However, mlock makes sure the requested range is already contained
within a vma, so no stack expansion will actually happen from mlock.

Should this ever change: the stack expansion mlocks only the newly
expanded range and so will not result in recursive expansion.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Hugh Dickins <hughd@google.com>
Acked-by: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: refactor inactive_file_is_low() to use get_lru_size()
Johannes Weiner [Sat, 23 Feb 2013 00:35:19 +0000 (16:35 -0800)]
mm: refactor inactive_file_is_low() to use get_lru_size()

An inactive file list is considered low when its active counterpart is
bigger, regardless of whether it is a global zone LRU list or a memcg
zone LRU list.  The only difference is in how the LRU size is assessed.

get_lru_size() does the right thing for both global and memcg reclaim
situations.

Get rid of inactive_file_is_low_global() and
mem_cgroup_inactive_file_is_low() by using get_lru_size() and compare
the numbers in common code.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: shmem: use new radix tree iterator
Johannes Weiner [Sat, 23 Feb 2013 00:35:17 +0000 (16:35 -0800)]
mm: shmem: use new radix tree iterator

In shmem_find_get_pages_and_swap(), use the faster radix tree iterator
construct from commit 78c1d78488a3 ("radix-tree: introduce bit-optimized
iterator").

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Konstantin Khlebnikov <khlebnikov@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: stop hotremove lockdep warning
Hugh Dickins [Sat, 23 Feb 2013 00:35:16 +0000 (16:35 -0800)]
ksm: stop hotremove lockdep warning

Complaints are rare, but lockdep still does not understand the way
ksm_memory_callback(MEM_GOING_OFFLINE) takes ksm_thread_mutex, and holds
it until the ksm_memory_callback(MEM_OFFLINE): that appears to be a
problem because notifier callbacks are made under down_read of
blocking_notifier_head->rwsem (so first the mutex is taken while holding
the rwsem, then later the rwsem is taken while still holding the mutex);
but is not in fact a problem because mem_hotplug_mutex is held
throughout the dance.

There was an attempt to fix this with mutex_lock_nested(); but if that
happened to fool lockdep two years ago, apparently it does so no longer.

I had hoped to eradicate this issue in extending KSM page migration not
to need the ksm_thread_mutex.  But then realized that although the page
migration itself is safe, we do still need to lock out ksmd and other
users of get_ksm_page() while offlining memory - at some point between
MEM_GOING_OFFLINE and MEM_OFFLINE, the struct pages themselves may
vanish, and get_ksm_page()'s accesses to them become a violation.

So, give up on holding ksm_thread_mutex itself from MEM_GOING_OFFLINE to
MEM_OFFLINE, and add a KSM_RUN_OFFLINE flag, and wait_while_offlining()
checks, to achieve the same lockout without being caught by lockdep.
This is less elegant for KSM, but it's more important to keep lockdep
useful to other users - and I apologize for how long it took to fix.

Signed-off-by: Hugh Dickins <hughd@google.com>
Reported-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Tested-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm: remove offlining arg to migrate_pages
Hugh Dickins [Sat, 23 Feb 2013 00:35:14 +0000 (16:35 -0800)]
mm: remove offlining arg to migrate_pages

No functional change, but the only purpose of the offlining argument to
migrate_pages() etc, was to ensure that __unmap_and_move() could migrate a
KSM page for memory hotremove (which took ksm_thread_mutex) but not for
other callers.  Now all cases are safe, remove the arg.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: enable KSM page migration
Hugh Dickins [Sat, 23 Feb 2013 00:35:13 +0000 (16:35 -0800)]
ksm: enable KSM page migration

Migration of KSM pages is now safe: remove the PageKsm restrictions from
mempolicy.c and migrate.c.

But keep PageKsm out of __unmap_and_move()'s anon_vma contortions, which
are irrelevant to KSM: it looks as if that code was preventing hotremove
migration of KSM pages, unless they happened to be in swapcache.

There is some question as to whether enforcing a NUMA mempolicy migration
ought to migrate KSM pages, mapped into entirely unrelated processes; but
moving page_mapcount > 1 is only permitted with MPOL_MF_MOVE_ALL anyway,
and it seems reasonable to assume that you wouldn't set MADV_MERGEABLE on
any area where this is a worry.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: make !merge_across_nodes migration safe
Hugh Dickins [Sat, 23 Feb 2013 00:35:11 +0000 (16:35 -0800)]
ksm: make !merge_across_nodes migration safe

The new KSM NUMA merge_across_nodes knob introduces a problem, when it's
set to non-default 0: if a KSM page is migrated to a different NUMA node,
how do we migrate its stable node to the right tree?  And what if that
collides with an existing stable node?

ksm_migrate_page() can do no more than it's already doing, updating
stable_node->kpfn: the stable tree itself cannot be manipulated without
holding ksm_thread_mutex.  So accept that a stable tree may temporarily
indicate a page belonging to the wrong NUMA node, leave updating until the
next pass of ksmd, just be careful not to merge other pages on to a
misplaced page.  Note nid of holding tree in stable_node, and recognize
that it will not always match nid of kpfn.

A misplaced KSM page is discovered, either when ksm_do_scan() next comes
around to one of its rmap_items (we now have to go to cmp_and_merge_page
even on pages in a stable tree), or when stable_tree_search() arrives at a
matching node for another page, and this node page is found misplaced.

In each case, move the misplaced stable_node to a list of migrate_nodes
(and use the address of migrate_nodes as magic by which to identify them):
we don't need them in a tree.  If stable_tree_search() finds no match for
a page, but it's currently exiled to this list, then slot its stable_node
right there into the tree, bringing all of its mappings with it; otherwise
they get migrated one by one to the original page of the colliding node.
stable_tree_search() is now modelled more like stable_tree_insert(), in
order to handle these insertions of migrated nodes.

remove_node_from_stable_tree(), remove_all_stable_nodes() and
ksm_check_stable_tree() have to handle the migrate_nodes list as well as
the stable tree itself.  Less obviously, we do need to prune the list of
stale entries from time to time (scan_get_next_rmap_item() does it once
each full scan): whereas stale nodes in the stable tree get naturally
pruned as searches try to brush past them, these migrate_nodes may get
forgotten and accumulate.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: make KSM page migration possible
Hugh Dickins [Sat, 23 Feb 2013 00:35:10 +0000 (16:35 -0800)]
ksm: make KSM page migration possible

KSM page migration is already supported in the case of memory hotremove,
which takes the ksm_thread_mutex across all its migrations to keep life
simple.

But the new KSM NUMA merge_across_nodes knob introduces a problem, when
it's set to non-default 0: if a KSM page is migrated to a different NUMA
node, how do we migrate its stable node to the right tree?  And what if
that collides with an existing stable node?

So far there's no provision for that, and this patch does not attempt to
deal with it either.  But how will I test a solution, when I don't know
how to hotremove memory?  The best answer is to enable KSM page migration
in all cases now, and test more common cases.  With THP and compaction
added since KSM came in, page migration is now mainstream, and it's a
shame that a KSM page can frustrate freeing a page block.

Without worrying about merge_across_nodes 0 for now, this patch gets KSM
page migration working reliably for default merge_across_nodes 1 (but
leave the patch enabling it until near the end of the series).

It's much simpler than I'd originally imagined, and does not require an
additional tier of locking: page migration relies on the page lock, KSM
page reclaim relies on the page lock, the page lock is enough for KSM page
migration too.

Almost all the care has to be in get_ksm_page(): that's the function which
worries about when a stable node is stale and should be freed, now it also
has to worry about the KSM page being migrated.

The only new overhead is an additional put/get/lock/unlock_page when
stable_tree_search() arrives at a matching node: to make sure migration
respects the raised page count, and so does not migrate the page while
we're busy with it here.  That's probably avoidable, either by changing
internal interfaces from using kpage to stable_node, or by moving the
ksm_migrate_page() callsite into a page_freeze_refs() section (even if not
swapcache); but this works well, I've no urge to pull it apart now.

(Descents of the stable tree may pass through nodes whose KSM pages are
under migration: being unlocked, the raised page count does not prevent
that, nor need it: it's safe to memcmp against either old or new page.)

You might worry about mremap, and whether page migration's rmap_walk to
remove migration entries will find all the KSM locations where it inserted
earlier: that should already be handled, by the satisfyingly heavy hammer
of move_vma()'s call to ksm_madvise(,,,MADV_UNMERGEABLE,).

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: remove old stable nodes more thoroughly
Hugh Dickins [Sat, 23 Feb 2013 00:35:08 +0000 (16:35 -0800)]
ksm: remove old stable nodes more thoroughly

Switching merge_across_nodes after running KSM is liable to oops on stale
nodes still left over from the previous stable tree.  It's not something
that people will often want to do, but it would be lame to demand a reboot
when they're trying to determine which merge_across_nodes setting is best.

How can this happen?  We only permit switching merge_across_nodes when
pages_shared is 0, and usually set run 2 to force that beforehand, which
ought to unmerge everything: yet oopses still occur when you then run 1.

Three causes:

1. The old stable tree (built according to the inverse
   merge_across_nodes) has not been fully torn down.  A stable node
   lingers until get_ksm_page() notices that the page it references no
   longer references it: but the page is not necessarily freed as soon as
   expected, particularly when swapcache.

   Fix this with a pass through the old stable tree, applying
   get_ksm_page() to each of the remaining nodes (most found stale and
   removed immediately), with forced removal of any left over.  Unless the
   page is still mapped: I've not seen that case, it shouldn't occur, but
   better to WARN_ON_ONCE and EBUSY than BUG.

2. __ksm_enter() has a nice little optimization, to insert the new mm
   just behind ksmd's cursor, so there's a full pass for it to stabilize
   (or be removed) before ksmd addresses it.  Nice when ksmd is running,
   but not so nice when we're trying to unmerge all mms: we were missing
   those mms forked and inserted behind the unmerge cursor.  Easily fixed
   by inserting at the end when KSM_RUN_UNMERGE.

3.  It is possible for a KSM page to be faulted back from swapcache
   into an mm, just after unmerge_and_remove_all_rmap_items() scanned past
   it.  Fix this by copying on fault when KSM_RUN_UNMERGE: but that is
   private to ksm.c, so dissolve the distinction between
   ksm_might_need_to_copy() and ksm_does_need_to_copy(), doing it all in
   the one call into ksm.c.

A long outstanding, unrelated bugfix sneaks in with that third fix:
ksm_does_need_to_copy() would copy from a !PageUptodate page (implying I/O
error when read in from swap) to a page which it then marks Uptodate.  Fix
this case by not copying, letting do_swap_page() discover the error.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: get_ksm_page locked
Hugh Dickins [Sat, 23 Feb 2013 00:35:06 +0000 (16:35 -0800)]
ksm: get_ksm_page locked

In some places where get_ksm_page() is used, we need the page to be locked.

When KSM migration is fully enabled, we shall want that to make sure that
the page just acquired cannot be migrated beneath us (raised page count is
only effective when there is serialization to make sure migration
notices).  Whereas when navigating through the stable tree, we certainly
do not want to lock each node (raised page count is enough to guarantee
the memcmps, even if page is migrated to another node).

Since we're about to add another use case, add the locked argument to
get_ksm_page() now.

Hmm, what's that rcu_read_lock() about?  Complete misunderstanding, I
really got the wrong end of the stick on that!  There's a configuration in
which page_cache_get_speculative() can do something cheaper than
get_page_unless_zero(), relying on its caller's rcu_read_lock() to have
disabled preemption for it.  There's no need for rcu_read_lock() around
get_page_unless_zero() (and mapping checks) here.  Cut out that silliness
before making this any harder to understand.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: reorganize ksm_check_stable_tree
Hugh Dickins [Sat, 23 Feb 2013 00:35:05 +0000 (16:35 -0800)]
ksm: reorganize ksm_check_stable_tree

Memory hotremove's ksm_check_stable_tree() is pitifully inefficient
(restarting whenever it finds a stale node to remove), but rearrange so
that at least it does not needlessly restart from nid 0 each time.  And
add a couple of comments: here is why we keep pfn instead of page.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoksm: trivial tidyups
Hugh Dickins [Sat, 23 Feb 2013 00:35:03 +0000 (16:35 -0800)]
ksm: trivial tidyups

Add NUMA() and DO_NUMA() macros to minimize blight of #ifdef
CONFIG_NUMAs (but indeed we don't want to expand struct rmap_item by nid
when not NUMA).  Add comment, remove "unsigned" from rmap_item->nid, as
"int nid" elsewhere.  Define ksm_merge_across_nodes 1U when #ifndef NUMA
to help optimizing out.  Use ?: in get_kpfn_nid().  Adjust a few
comments noticed in ongoing work.

Leave stable_tree_insert()'s rb_linkage until after the node has been
set up, as unstable_tree_search_insert() does: ksm_thread_mutex and page
lock make either way safe, but we're going to copy and I prefer this
precedent.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Izik Eidus <izik.eidus@ravellosystems.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>