Input: wacom_serial4 - add support for Wacom ArtPad II tablet

Tablet initially begins communicating at 9600 baud, so this command
should be used to connect to the device:

    $ inputattach --daemon --baud 9600 --wacom_iv /dev/ttyS0

https://github.com/linuxwacom/xf86-input-wacom/issues/40

Signed-off-by: Jason Gerecke <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: elan_i2c - add id for touchpad found in Lenovo s21e-20

Lenovo s21e-20 uses ELAN0601 in its ACPI tables for the Elan touchpad.

Signed-off-by: Vincent Batts <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: raspberrypi-ts - select CONFIG_INPUT_POLLDEV

When CONFIG_INPUT_POLLDEV is disabled, we get a link error:

drivers/input/touchscreen/raspberrypi-ts.o: In function `rpi_ts_probe':
raspberrypi-ts.c:(.text+0xec): undefined reference to `devm_input_allocate_polled_device'
raspberrypi-ts.c:(.text+0xec): relocation truncated to fit: R_AARCH64_CALL26 against undefined symbol `devm_input_allocate_polled_device'
raspberrypi-ts.c:(.text+0x19c): undefined reference to `input_register_polled_device'
raspberrypi-ts.c:(.text+0x19c): relocation truncated to fit: R_AARCH64_CALL26 against undefined symbol `input_register_polled_device'

Select that symbol like we do from the other similar drivers.

Fixes: 0b9f28fed3f7 ("Input: add official Raspberry Pi's touchscreen driver")
Signed-off-by: Arnd Bergmann <[email protected]>
Reviewed-by: Nicolas Saenz Julienne <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: msm-vibrator - use correct gpio header

When CONFIG_GPIOLIB is not set, we get a couple of build
errors during test building:

drivers/input/misc/msm-vibrator.c: In function 'msm_vibrator_start':
drivers/input/misc/msm-vibrator.c:79:3: error: implicit declaration of function 'gpiod_set_value_cansleep'; did you mean 'gpio_set_value_cansleep'? [-Werror=implicit-function-declaration]
   gpiod_set_value_cansleep(vibrator->enable_gpio, 1);
   ^~~~~~~~~~~~~~~~~~~~~~~~
   gpio_set_value_cansleep
drivers/input/misc/msm-vibrator.c: In function 'msm_vibrator_probe':
drivers/input/misc/msm-vibrator.c:176:26: error: implicit declaration of function 'devm_gpiod_get'; did you mean 'devm_gpio_free'? [-Werror=implicit-function-declaration]
  vibrator->enable_gpio = devm_gpiod_get(&pdev->dev, "enable",
                          ^~~~~~~~~~~~~~
                          devm_gpio_free
drivers/input/misc/msm-vibrator.c:177:13: error: 'GPIOD_OUT_LOW' undeclared (first use in this function); did you mean 'GPIOF_INIT_LOW'?
             GPIOD_OUT_LOW);
             ^~~~~~~~~~~~~
             GPIOF_INIT_LOW
drivers/input/misc/msm-vibrator.c:177:13: note: each undeclared identifier is reported only once for each function it appears in

This is easy to avoid when we use gpio/consumer.h as the documented interface.

Fixes: 0f681d09e66e ("Input: add new vibrator driver for various MSM SOCs")
Signed-off-by: Arnd Bergmann <[email protected]>
Reviewed-by: Linus Walleij <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Merge branch 'next' into for-linus

Prepare input updates for 5.1 merge window.

Input: ti_am335x_tsc - remove set but not used variable 'tscadc_dev'

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/input/touchscreen/ti_am335x_tsc.c: In function 'titsc_suspend':
drivers/input/touchscreen/ti_am335x_tsc.c:510:24: warning:
 variable 'tscadc_dev' set but not used [-Wunused-but-set-variable]

drivers/input/touchscreen/ti_am335x_tsc.c: In function 'titsc_resume':
drivers/input/touchscreen/ti_am335x_tsc.c:527:24: warning:
 variable 'tscadc_dev' set but not used [-Wunused-but-set-variable]

It's not used any more after 333e07ec4b33 ("Input: ti_am335x_tsc: Mark TSC
device as wakeup source")

Signed-off-by: YueHaibing <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: i8042 - rework DT node name comparisons

Convert string compares of DT node names to use of_node_name_eq helper
instead. For the root node on SUN DT, we need to retrieve the 'name'
property as it is the rare case where the 'name' property and node name
differ. With both changes, it removes direct access to the node name
pointer.

While at it, convert the open coded loop to use for_each_child_of_node().

Signed-off-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: goodix - print values in case of inconsistencies

"Invalid config" gives little idea what's wrong. Print the values that
must not be 0 so we know which ones are off.

Signed-off-by: Guido Günther <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: goodix - refer to touchscreen.txt in device tree bindings

Refer to touchscreen.txt for generic touch properties. This avoids
duplication and we're using the generic code to parse these in the
driver. While at that add touchscreen-size-{x,y} which are respected by
the driver as well.

Signed-off-by: Guido Günther <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: goodix - support Goodix gt5688

From what I've seen in vendor trees it's fine to treat this as gt1x¹.
Tested on the Purism Librem 5 Devkit (Rocktech JH057N00900 panel).

[1]: https://github.com/TadiT7/android_kernel_mtk-4.4/tree/master/drivers/input/touchscreen/mediatek/GT5688

Signed-off-by: Guido Günther <[email protected]>
Reviewed-by: Bastien Nocera <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: synaptics_i2c - remove redundant spinlock

Remove a leftover spinlock.

This was required back when mod_delayed_work() did not exist, and had to
be implemented with a cancel + queue. See commit e7c2f967445d
("workqueue: use mod_delayed_work() instead of __cancel + queue")

schedule_delayed_work() and mod_delayed_work() can now be used
concurrently. So the spinlock is no longer needed.

Signed-off-by: Sven Van Asbroeck <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: elan_i2c - add ACPI ID for touchpad in Lenovo V330-15ISK

This adds ELAN0617 to the ACPI table to support Elan touchpad found in
Lenovo V330-15ISK.

Signed-off-by: Mauro Ciancio <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: st-keyscan - fix potential zalloc NULL dereference

This patch fixes the following static checker warning:

drivers/input/keyboard/st-keyscan.c:156 keyscan_probe()
error: potential zalloc NULL dereference: 'keypad_data->input_dev'

Reported-by: Dan Carpenter <[email protected]>
Signed-off-by: Gabriel Fernandez <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: apanel - switch to using brightness_set_blocking()

Now that LEDs core allows "blocking" flavor of "set brightness" method we
can use it and get rid of private work item. As a bonus, we are no longer
forgetting to cancel it when we unbind the driver.

Reviewed-by: Sven Van Asbroeck <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: db9 - mark expected switch fall-through

In preparation to enabling -Wimplicit-fallthrough, mark switch
cases where we are expecting to fall through.

This patch fixes the following warning:

drivers/input/joystick/db9.c: In function ‘db9_saturn_read_packet’:
drivers/input/joystick/db9.c:256:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
   if (tmp == 0xff) {
      ^
drivers/input/joystick/db9.c:263:2: note: here
  default:
  ^~~~~~~

Notice that, in this particular case, the code comment is modified
in accordance with what GCC is expecting to find.

This patch is part of the ongoing efforts to enable
-Wimplicit-fallthrough.

Signed-off-by: Gustavo A. R. Silva <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: qt2160 - remove redundant spinlock

Remove a spinlock which prevents schedule_delayed_work() and
mod_delayed_work() from executing concurrently.

This was required back when mod_delayed_work() did not exist,
and had to be implemented with a cancel + schedule. See
commit e7c2f967445d ("workqueue: use mod_delayed_work() instead of
__cancel + queue")

schedule_delayed_work() and mod_delayed_work() can now be used
concurrently. So the spinlock is no longer needed.

Signed-off-by: Sven Van Asbroeck <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: st1232 - handle common DT bindings

This is required to specify generic touchscreen properties via DT.

Signed-off-by: Matthias Fend <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ims-pcu - switch to using brightness_set_blocking()

Now that LEDs core allows "blocking" flavor of "set brightness" method we
can use it and get rid of private work item.

Signed-off-by: Dmitry Torokhov <[email protected]>

Input: st1232 - switch to gpiod API

Use devm_gpiod_get_optional() and gpiod_set_value_cansleep() instead
of the old API. The st1232_ts_power() now passes on the inverted "poweron"
value to reflect the correct logical value.

Signed-off-by: Martin Kepplinger <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Revert "Input: elan_i2c - add ACPI ID for touchpad in ASUS Aspire F5-573G"

This reverts commit 7db54c89f0b30a101584e09d3729144e6170059d as it
breaks Acer Aspire V-371 and other devices. According to Elan:

"Acer Aspire F5-573G is MS Precision touchpad which should use hid
 multitouch driver. ELAN0501 should not be added in elan_i2c."

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=202503
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: qt2160 - switch to using brightness_set_blocking()

Now that LEDs core allows "blocking" flavor of "set brightness" method we
can use it and get rid of private work items.

Reviewed-by: Sven Van Asbroeck <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - fetch touchscreen geometry from DT

Fetching the geometry from the ILI251x registers seems unreliable and
sometimes returns all zeroes. Add support for fetching the geometry and
axis inversion from DT instead.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: msm-vibrator - tweak an error message

The PTR_ERR(NULL) value is zero and it's not useful to print that.

Signed-off-by: Dan Carpenter <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: matrix_keypad - use flush_delayed_work()

We should be using flush_delayed_work() instead of flush_work() in
matrix_keypad_stop() to ensure that we are not missing work that is
scheduled but not yet put in the workqueue (i.e. its delay timer has not
expired yet).

Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ps2-gpio - flush TX work when closing port

To ensure that TX work is not running after serio port has been torn down,
let's flush it when closing the port.

Reported-by: Sven Van Asbroeck <[email protected]>
Acked-by: Danilo Krummrich <[email protected]>
Reviewed-by: Sven Van Asbroeck <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: cap11xx - switch to using set_brightness_blocking()

Updating LED state requires access to regmap and therefore we may sleep,
so we could not do that directly form set_brightness() method.
Historically we used private work to adjust the brightness, but with the
introduction of set_brightness_blocking() we no longer need it.

As a bonus, not having our own work item means we do not have
use-after-free issue as we neglected to cancel outstanding work on
driver unbind.

Reported-by: Sven Van Asbroeck <[email protected]>
Reviewed-by: Sven Van Asbroeck <[email protected]>
Acked-by: Jacek Anaszewski <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: elantech - enable 3rd button support on Fujitsu CELSIUS H780

Like Fujitsu CELSIUS H760, the H780 also has a three-button Elantech
touchpad, but the driver needs to be told so to enable the middle touchpad
button.

The elantech_dmi_force_crc_enabled quirk was not necessary with the H780.

Also document the fw_version and caps values detected for both H760 and
H780 models.

Signed-off-by: Matti Kurkela <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: tm2-touchkey - acknowledge that setting brightness is a blocking call

We need to access I2C bus when switching brightness, and that may block,
therefore we have to set stmfts_brightness_set() as LED's
brightness_set_blocking() method.

Fixes: 72d1f2346ded ("Input: tm2-touchkey - add touchkey driver support for TM2")
Acked-by: Andi Shyti <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: stmfts - acknowledge that setting brightness is a blocking call

We need to turn regulators on and off when switching brightness, and
that may block, therefore we have to set stmfts_brightness_set() as
LED's brightness_set_blocking() method.

Fixes: 78bcac7b2ae1 ("Input: add support for the STMicroelectronics FingerTip touchscreen")
Acked-by: Andi Shyti <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - switch to using devm_device_add_group()

By switching to devm_device_add_group() we can complete driver conversion
to using managed resources and get rid of ili210x_i2c_remove().

Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - add ILI251X support

Add support for ILI251x touch controller. This controller is similar
to the ILI210x, except for the following differences:
- Does not support I2C R-W transfer, Read must be followed by an
  obscenely long delay, and then followed by Write
- Does support 10 simultaneous touch inputs.
- Touch data format is slightly different, pressure reporting does not
  work although the touch data contain such information.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - add OF match table

Add OF match table for the ili210x touchscreen.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - reorder probe

Perform the register access only after the I2C client data are set,
this is only done in preparation for the subsequent patch which
uses the I2C client data in the register IO function.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - rework the touchscreen sample processing

Get rid of the packed structures for representing data as that does not
apply to other similar Ilitek touchscreens. Instead, implement a function
which parses the data and reports touch events and coordinates.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - convert to devm IRQ

Convert the driver to devm_request_irq(), drop the related unmanaged
deregistration code and add ili210x_irq_teardown() to tear the IRQ
down and cancel possible touchscreen pending work.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - add reset GPIO support

The touchscreen can have a reset GPIO connected to it, add support
for such an arrangement.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - convert to devm_ functions

Convert the driver to dev-managed allocations.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - drop get_pendown_state

The .get_pendown_state callback is set only by the platform data code,
which was just removed. Thus, get_pendown_state() always returns false,
so drop that altogether.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - drop platform data support

There is not a single user of the ili210x platform data in the kernel,
just drop it.

Signed-off-by: Marek Vasut <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: ili210x - add DT binding document

Add DT binding document for the Ilitek ILI210x and ILI251x
touchscreen controllers.

Signed-off-by: Marek Vasut <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: bma150 - register input device after setting private data

Otherwise we introduce a race condition where userspace can request input
before we're ready leading to null pointer dereference such as

input: bma150 as /devices/platform/i2c-gpio-2/i2c-5/5-0038/input/input3
Unable to handle kernel NULL pointer dereference at virtual address 00000018
pgd = (ptrval)
[00000018] *pgd=55dac831, *pte=00000000, *ppte=00000000
Internal error: Oops: 17 [#1] PREEMPT ARM
Modules linked in: bma150 input_polldev [last unloaded: bma150]
CPU: 0 PID: 2870 Comm: accelerometer Not tainted 5.0.0-rc3-dirty #46
Hardware name: Samsung S5PC110/S5PV210-based board
PC is at input_event+0x8/0x60
LR is at bma150_report_xyz+0x9c/0xe0 [bma150]
pc : [<80450f70>]    lr : [<7f0a614c>]    psr: 800d0013
sp : a4c1fd78  ip : 00000081  fp : 00020000
r10: 00000000  r9 : a5e2944c  r8 : a7455000
r7 : 00000016  r6 : 00000101  r5 : a7617940  r4 : 80909048
r3 : fffffff2  r2 : 00000000  r1 : 00000003  r0 : 00000000
Flags: Nzcv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
Control: 10c5387d  Table: 54e34019  DAC: 00000051
Process accelerometer (pid: 2870, stack limit = 0x(ptrval))
Stackck: (0xa4c1fd78 to 0xa4c20000)
fd60:                                                       fffffff3 fc813f6c
fd80: 40410581 d7530ce3 a5e2817c a7617f00 a5e29404 a5e2817c 00000000 7f008324
fda0: a5e28000 8044f59c a5fdd9d0 a5e2945c a46a4a00 a5e29668 a7455000 80454f10
fdc0: 80909048 a5e29668 a5fdd9d0 a46a4a00 806316d0 00000000 a46a4a00 801df5f0
fde0: 00000000 d7530ce3 a4c1fec0 a46a4a00 00000000 a5fdd9d0 a46a4a08 801df53c
fe00: 00000000 801d74bc a4c1fec0 00000000 a4c1ff70 00000000 a7038da8 00000000
fe20: a46a4a00 801e91fc a411bbe0 801f2e88 00000004 00000000 80909048 00000041
fe40: 00000000 00020000 00000000 dead4ead a6a88da0 00000000 ffffe000 806fcae8
fe60: a4c1fec8 00000000 80909048 00000002 a5fdd9d0 a7660110 a411bab0 00000001
fe80: dead4ead ffffffff ffffffff a4c1fe8c a4c1fe8c d7530ce3 20000013 80909048
fea0: 80909048 a4c1ff70 00000001 fffff000 a4c1e000 00000005 00026038 801eabd8
fec0: a7660110 a411bab0 b9394901 00000006 a696201b 76fb3000 00000000 a7039720
fee0: a5fdd9d0 00000101 00000002 00000096 00000000 00000000 00000000 a4c1ff00
ff00: a6b310f4 805cb174 a6b310f4 00000010 00000fe0 00000010 a4c1e000 d7530ce3
ff20: 00000003 a5f41400 a5f41424 00000000 a6962000 00000000 00000003 00000002
ff40: ffffff9c 000a0000 80909048 d7530ce3 a6962000 00000003 80909048 ffffff9c
ff60: a6962000 801d890c 00000000 00000000 00020000 a7590000 00000004 00000100
ff80: 00000001 d7530ce3 000288b8 00026320 000288b8 00000005 80101204 a4c1e000
ffa0: 00000005 80101000 000288b8 00026320 000288b8 000a0000 00000000 00000000
ffc0: 000288b8 00026320 000288b8 00000005 7eef3bac 000264e8 00028ad8 00026038
ffe0: 00000005 7eef3300 76f76e91 76f78546 800d0030 000288b8 00000000 00000000
[<80450f70>] (input_event) from [<a5e2817c>] (0xa5e2817c)
Code: e1a08148 eaffffa8 e351001f 812fff1e (e590c018)
---[ end trace 1c691ee85f2ff243 ]---

Signed-off-by: Jonathan Bakker <[email protected]>
Signed-off-by: Paweł Chmiel <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: add new vibrator driver for various MSM SOCs

This patch adds a new vibrator driver that supports various Qualcomm
MSM SOCs. Driver was tested on a LG Nexus 5 (hammerhead) phone.

Signed-off-by: Brian Masney <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: gpio-keys - add shutdown callback

On some platforms (e.g.: ARCH_BRCMSTB) it is possible to enter
"poweroff" while leaving some wake-up sources enabled such as key
presses in order to allow for the system to wake-up.

Wire up a .shutdown() callback which calls into the existing
gpio_keys_suspend() since the logic is essentially the same.

Signed-off-by: Florian Fainelli <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: sx8654 - do not override interrupt trigger

We should rely on the interrupt trigger (level vs edge) set up by the
firmware or board code instead of forcing what we consider appropriate.

Signed-off-by: Dmitry Torokhov <[email protected]>

Input: sx8654 - convert #defined flags to BIT(x)

Some of the #defined register values are one-bit flags. Convert them to
use the BIT(x) macro instead of 1 byte hexadecimal values. This improves
readability and clarifies the intent.

Signed-off-by: Richard Leitner <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: sx8654 - use common of_touchscreen functions

of_touchscreen.c provides a common interface for a axis inversion and
swapping of touchscreens. Therefore use it in the sx8654 driver.

Signed-off-by: Richard Leitner <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: sx8654 - add sx8650 support

The sx8654 and sx8650 are quite similar, therefore add support for the
sx8650 within the sx8654 driver.

Signed-off-by: Richard Leitner <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
[dtor: use __be16 in sx8650_irq, add missing del_timer_sync]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: sx8654 - add sx8655 and sx8656 to compatibles

As the sx865[456] share the same datasheet and differ only in the
presence of a "capacitive proximity detection circuit" and a "haptics
motor driver for LRA/ERM" add them to the compatbiles. As the driver
doesn't implement these features it should be no problem.

Signed-off-by: Richard Leitner <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: sx8654 - add reset-gpio support

The sx8654 features a NRST input which may be connected to a GPIO.
Therefore add support for hard-resetting the sx8654 via this NRST.

If the reset-gpio property is provided the sx8654 is resetted via NRST
instead of the soft-reset via I2C.

Signed-off-by: Richard Leitner <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: st1232 - add Martin as module author

This adds myself as an author of the st1232 driver module as Tony's
email address doesn't seem to work anymore.

Signed-off-by: Martin Kepplinger <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: st1232 - add support for st1633

Add support for the Sitronix ST1633 touchscreen controller to the st1232
driver. A protocol spec can be found here:
www.ampdisplay.com/documents/pdf/AM-320480B6TZQW-TC0H.pdf

Signed-off-by: Martin Kepplinger <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: pwm-vibra - stop regulator after disabling pwm, not before

This patch fixes order of disable calls in pwm_vibrator_stop.
Currently when starting device, we first enable vcc regulator and then
setup and enable pwm. When stopping, we should do this in oposite order,
so first disable pwm and then disable regulator.
Previously order was the same as in start.

Signed-off-by: Paweł Chmiel <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: pwm-vibra - prevent unbalanced regulator

pwm_vibrator_stop disables the regulator, but it can be called from
multiple places, even when the regulator is already disabled. Fix this
by using regulator_is_enabled check when starting and stopping device.

Signed-off-by: Jonathan Bakker <[email protected]>
Signed-off-by: Paweł Chmiel <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: snvs_pwrkey - allow selecting driver for i.MX 7D

The i.MX SNVS Power Key driver supports the i.MX 7D SoC family too.
Allow to enable the i.MX SNVS Power Key driver even if only i.MX 7D
SoC is selected.

Signed-off-by: Stefan Agner <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: input_event - fix the CONFIG_SPARC64 mixup

Arnd Bergmann pointed out that CONFIG_* cannot be used in a uapi header.
Override with an equivalent conditional.

Fixes: 2e746942ebac ("Input: input_event - provide override for sparc64")
Fixes: 152194fe9c3f ("Input: extend usable life of event timestamps to 2106 on 32 bit systems")
Signed-off-by: Deepa Dinamani <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: olpc_apsp - assign priv->dev earlier

The dev field needs to be set when serio_register_port() is called,
because the open callback may use it (in the error handling path).

Fixes: commit af518342effd ("Input: olpc_apsp - check FIFO status on open(), not probe()")
Signed-off-by: Lubomir Rintel <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: uinput - fix undefined behavior in uinput_validate_absinfo()

An integer overflow may arise in uinput_validate_absinfo() if "max - min"
can't be represented by an "int". We should check for overflow before
trying to use the result.

Reported-by: Kyungtae Kim <[email protected]>
Reviewed-by: Peter Hutterer <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Merge tag 'v4.20' into for-linus

Sync with mainline to get linux/overflow.h among other things.

Input: raspberrypi-ts - fix link error

Fix link error when TOUCHSCREEN_RASPBERRYPI_FW is enabled as a module and
the dependent module is built-in.  The 'depends on RASPBERRYPI_FIRMWARE' by
itself prevents the touchscreen driver from being built-in when the
firmware is configured as a module.  However, the '|| COMPILE_TEST' still
allows it unless we explicitly prevent that configuration with
'|| (RASPBERRYPI_FIRMWARE=n && COMPILE_TEST)'.

ld: drivers/input/touchscreen/raspberrypi-ts.o: in function `rpi_ts_probe':
raspberrypi-ts.c:(.text+0x3a8): undefined reference to `rpi_firmware_get'
ld: raspberrypi-ts.c:(.text+0x3a8): relocation truncated to fit: R_AARCH64_CALL26 against undefined symbol `rpi_firmware_get'
ld: raspberrypi-ts.c:(.text+0x4c8): undefined reference to `rpi_firmware_property'
ld: raspberrypi-ts.c:(.text+0x4c8): relocation truncated to fit: R_AARCH64_CALL26 against undefined symbol `rpi_firmware_property'

Rework so that TOUCHSCREEN_RASPBERRYPI_FW depends on
RASPBERRYPI_FIRMWARE=n if COMPILE_TEST is enabled.

Fixes: 0b9f28fed3f7 ("Input: add official Raspberry Pi's touchscreen driver")
Suggested-by: Arnd Bergmann <[email protected]>
Signed-off-by: Anders Roxell <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: xpad - add support for SteelSeries Stratus Duo

Add support for the SteelSeries Stratus Duo, a wireless Xbox 360
controller. The Stratus Duo ships with a USB dongle to enable wireless
connectivity, but it can also function as a wired controller by connecting
it directly to a PC via USB, hence the need for two USD PIDs. 0x1430 is the
dongle, and 0x1431 is the controller.

Signed-off-by: Tom Panfil <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: edt-ft5x06 - add offset support for ev-ft5726

Unfortunately the evervision focaltech implementation uses two offset
registers, one for the x coordinate and one for y.

This patch extends the driver to handle those offset registers only for
devices that support these.

Signed-off-by: Marco Felsch <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: edt-ft5x06 - add support to update ev-ft5726 registers

Currently only the threshold and gain parameters can be read.

Signed-off-by: Marco Felsch <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: edt-ft5x06 - add support for Evervision FT5726

Evervision displays are using different Focaltech touchscreen
controllers. This commit adds the initial support for the ones using the
FT5726 controller. Receiving the touch data is the same as for the
GENERIC_FT but the x and y cooridnates are swapped. The main differences
are the register addresses where the GAIN and THRESHOLD parameters are
stored.

Signed-off-by: Marco Felsch <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: input_event - provide override for sparc64

The usec part of the timeval is defined as
__kernel_suseconds_t tv_usec; /* microseconds */

Arnd noticed that sparc64 is the only architecture that defines
__kernel_suseconds_t as int rather than long.

This breaks the current y2038 fix for kernel as we only access and define
the timeval struct for non-kernel use cases.  But, this was hidden by an
another typo in the use of __KERNEL__ qualifier.

Fix the typo, and provide an override for sparc64.

Fixes: 152194fe9c3f ("Input: extend usable life of event timestamps to 2106 on 32 bit systems")
Reported-by: Arnd Bergmann <[email protected]>
Signed-off-by: Deepa Dinamani <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: mtk-pmic-keys - remove duplicated include from mtk-pmic-keys.c

iSOrt includes in alphabetical order and remove duplicated include file
linux/kernel.h

Signed-off-by: YueHaibing <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: mcs_touchkey - use struct_size() in kzalloc()

One of the more common cases of allocation size calculations is finding the
size of a structure that has a zero-sized array at the end, along with memory
for some number of elements for that array. For example:

struct foo {
    int stuff;
    void *entry[];
};

instance = kzalloc(sizeof(struct foo) + sizeof(void *) * count, GFP_KERNEL);

Instead of leaving these open-coded and prone to type mistakes, we can now
use the new struct_size() helper:

instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL);

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: tca6416-keypad - use struct_size() in kzalloc()

One of the more common cases of allocation size calculations is finding the
size of a structure that has a zero-sized array at the end, along with
memory for some number of elements for that array. For example:

struct foo {
    int stuff;
    void *entry[];
};

instance = kzalloc(sizeof(struct foo) + sizeof(void *) * count, GFP_KERNEL);

Instead of leaving these open-coded and prone to type mistakes, we can now
use the new struct_size() helper:

instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL);

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Merge tag 'v4.20' into next

Merge with mainline to bring in the new APIs.

Input: ad7879 - add check for read errors in interrupt

regmap_bulk_read() can return a non zero value on failure. The fix checks
if the function call succeeded before calling mod_timer. The issue was
identified by a static analysis tool.

Signed-off-by: Aditya Pakki <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: tm2-touchkey - add support for aries touchkey variant

The touchkey variant found on aries board is slighty different,
it uses a fixed regulator and writes/read to the same place

Signed-off-by: Jonathan Bakker <[email protected]>
Signed-off-by: Paweł Chmiel <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: tm2-touchkey - allow specifying custom keycodes

Not all devices use the same keycodes in the same order,
so add possibility to define keycodes for buttons present
on actual hardware.

If keycodes property is not present, we assume that device has
at least MENU and BACK keys.

Signed-off-by: Jonathan Bakker <[email protected]>
Signed-off-by: Paweł Chmiel <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: tm2-touchkey - correct initial brightness

tm2-touchkey doesn't have brightness levels, but only on/off states,
so replace LED_FULL with LED_ON.

Signed-off-by: Jonathan Bakker <[email protected]>
Signed-off-by: Paweł Chmiel <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: tm2-touchkey - add support for midas touchkey

The touchkey on midas boards is almost identical.
The only real difference is that it uses the same register for both
keycode and base.

Signed-off-by: Simon Shields <[email protected]>
Signed-off-by: Paweł Chmiel <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: soc_button_array - fix mapping of the 5th GPIO in a PNP0C40 device

The Microsoft documenation for the PNP0C40 device aka the
"Windows-compatible button array" describes the 5th GpioInt listed in
the resources as: '5. Interrupt corresponding to the "Rotation Lock"
button, if supported'.

Notice this describes the 5th entry as a button while we sofar have been
mapping it to EV_SW, SW_ROTATE_LOCK. On my Point of View TAB P1006W-232
which actually comes with a rotation-lock button, the button indeed is a
button and not a slider/switch. An image search for other Windows tablets
has found 2 more models with a rotation-lock button and on both of those
it too is a push-button and not a slider/switch.

Further evidence can be found in the HUT extension HUTRR52 from Microsoft
which adds rotation lock support to the HUT, which describes 2 different
usages: "0xC9 System Display Rotation Lock Button" and
"0xCA System Display Rotation Lock Slider Switch" note that switch is seen
as a separate thing here and the non switch wording is an exact match for
the "Windows-compatible button array" spec wording.

TL;DR: our current mapping of the 5th GPIO to SW_ROTATE_LOCK is wrong
because the 5th GPIO is for a push-button not a switch.

This commit fixes this by maping the 5th GPIO to KEY_ROTATE_LOCK_TOGGLE.

Signed-off-by: Hans de Goede <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: soc_button_array - add usage-page 0x01 usage-id 0xca mapping

The ACPI0011 _DSD button descriptor on a CHT based Intel Compute Sticks
contains a mapping for usage-page 0x01 usage-id 0xca.

As described in hutrr52_system_display_rotation_lock_controls_0.pdf this
should be mapped as a "System Display Rotation Lock Slider Switch", this
commit adds support for this, silencing the following warning:

soc_button_array ACPI0011:00: Unknown button index 4 upage 01 usage ca,
ignoring

Signed-off-by: Hans de Goede <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Merge branch 'next' into for-linus

Prepare input updates for 4.21 merge window.

Input: elan_i2c - add ACPI ID for touchpad in ASUS Aspire F5-573G

This adds ELAN0501 to the ACPI table to support Elan touchpad found in ASUS
Aspire F5-573G.

Signed-off-by: Patrick Dreyer <[email protected]>
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: atmel_mxt_ts - don't try to free unallocated kernel memory

If the user attempts to update Atmel device with an invalid configuration
cfg file, error handling code is trying to free cfg file memory which is
not allocated yet hence results into kernel crash.

This patch fixes the order of memory free operations.

Signed-off-by: Sanjeev Chugh <[email protected]>
Fixes: a4891f105837 ("Input: atmel_mxt_ts - zero terminate config firmware file")
Cc: [email protected]
Signed-off-by: Dmitry Torokhov <[email protected]>

Linux 4.20

Merge branch 'fixes' of git://git./linux/kernel/git/viro/vfs

Pull vfs fixes from Al Viro:
 "A couple of fixes - no common topic ;-)"

[ The aio spectre patch also came in from Jens, so now we have that
  doubly fixed .. ]

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  proc/sysctl: don't return ENOMEM on lookup when a table is unregistering
  aio: fix spectre gadget in lookup_ioctx

Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "This is two simple target fixes and one discard related I/O starvation
  problem in sd.

  The discard problem occurs because the discard page doesn't have a
  mempool backing so if the allocation fails due to memory pressure, we
  then lose the forward progress we require if the writeout is on the
  same device. The fix is to back it with a mempool"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: sd: use mempool for discard special page
  scsi: target: iscsi: cxgbit: add missing spin_lock_init()
  scsi: target: iscsi: cxgbit: fix csk leak

Merge tag 'compiler-attributes-for-linus-v4.20' of https://github.com/ojeda/linux

Pull compiler_types.h fix from Miguel Ojeda:
 "A cleanup for userspace in compiler_types.h: don't pollute userspace
  with macro definitions (Xiaozhou Liu)

  This is harmless for the kernel, but v4.19 was released with a few
  macros exposed to userspace as the patch explains; which this removes,
  so it *could* happen that we break something for someone (although
  leaving inline redefined is probably worse)"

* tag 'compiler-attributes-for-linus-v4.20' of https://github.com/ojeda/linux:
  include/linux/compiler_types.h: don't pollute userspace with macro definitions

Merge tag 'auxdisplay-for-linus-v4.20' of https://github.com/ojeda/linux

Pull auxdisplay fix from Miguel Ojeda:
 "charlcd: fix x/y command parsing (Mans Rullgard)"

* tag 'auxdisplay-for-linus-v4.20' of https://github.com/ojeda/linux:
  auxdisplay: charlcd: fix x/y command parsing

Revert "vfs: Allow userns root to call mknod on owned filesystems."

This reverts commit 55956b59df336f6738da916dbb520b6e37df9fbd.

commit 55956b59df33 ("vfs: Allow userns root to call mknod on owned filesystems.")
enabled mknod() in user namespaces for userns root if CAP_MKNOD is
available. However, these device nodes are useless since any filesystem
mounted from a non-initial user namespace will set the SB_I_NODEV flag on
the filesystem. Now, when a device node s created in a non-initial user
namespace a call to open() on said device node will fail due to:

bool may_open_dev(const struct path *path)
{
        return !(path->mnt->mnt_flags & MNT_NODEV) &&
                !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
}

The problem with this is that as of the aforementioned commit mknod()
creates partially functional device nodes in non-initial user namespaces.
In particular, it has the consequence that as of the aforementioned commit
open() will be more privileged with respect to device nodes than mknod().
Before it was the other way around. Specifically, if mknod() succeeded
then it was transparent for any userspace application that a fatal error
must have occured when open() failed.

All of this breaks multiple userspace workloads and a widespread assumption
about how to handle mknod(). Basically, all container runtimes and systemd
live by the slogan "ask for forgiveness not permission" when running user
namespace workloads. For mknod() the assumption is that if the syscall
succeeds the device nodes are useable irrespective of whether it succeeds
in a non-initial user namespace or not. This logic was chosen explicitly
to allow for the glorious day when mknod() will actually be able to create
fully functional device nodes in user namespaces.
A specific problem people are already running into when running 4.18 rc
kernels are failing systemd services. For any distro that is run in a
container systemd services started with the PrivateDevices= property set
will fail to start since the device nodes in question cannot be
opened (cf. the arguments in [1]).

Full disclosure, Seth made the very sound argument that it is already
possible to end up with partially functional device nodes. Any filesystem
mounted with MS_NODEV set will allow mknod() to succeed but will not allow
open() to succeed. The difference to the case here is that the MS_NODEV
case is transparent to userspace since it is an explicitly set mount option
while the SB_I_NODEV case is an implicit property enforced by the kernel
and hence opaque to userspace.

[1]: https://github.com/systemd/systemd/pull/9483

Signed-off-by: Christian Brauner <[email protected]>
Cc: "Eric W. Biederman" <[email protected]>
Cc: Seth Forshee <[email protected]>
Cc: Serge Hallyn <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

dma-mapping: fix flags in dma_alloc_wc

We really need the writecombine flag in dma_alloc_wc, fix a stupid
oversight.

Fixes: 7ed1d91a9e ("dma-mapping: translate __GFP_NOFAIL to DMA_ATTR_NO_WARN")
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Input: drv2667 - fix indentation issues

There are some statements that are indented incorrectly, fix this by
removing the extra tabs.

Signed-off-by: Colin Ian King <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "4 fixes"

* emailed patches from Andrew Morton <[email protected]>:
  mm, page_alloc: fix has_unmovable_pages for HugePages
  fork,memcg: fix crash in free_thread_stack on memcg charge fail
  mm: thp: fix flags for pmd migration when split
  mm, memory_hotplug: initialize struct pages for the full memory section

mm, page_alloc: fix has_unmovable_pages for HugePages

While playing with gigantic hugepages and memory_hotplug, I triggered
the following #PF when "cat memoryX/removable":

  BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
  #PF error: [normal kernel read fault]
  PGD 0 P4D 0
  Oops: 0000 [#1] SMP PTI
  CPU: 1 PID: 1481 Comm: cat Tainted: G            E     4.20.0-rc6-mm1-1-default+ #18
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
  RIP: 0010:has_unmovable_pages+0x154/0x210
  Call Trace:
   is_mem_section_removable+0x7d/0x100
   removable_show+0x90/0xb0
   dev_attr_show+0x1c/0x50
   sysfs_kf_seq_show+0xca/0x1b0
   seq_read+0x133/0x380
   __vfs_read+0x26/0x180
   vfs_read+0x89/0x140
   ksys_read+0x42/0x90
   do_syscall_64+0x5b/0x180
   entry_SYSCALL_64_after_hwframe+0x44/0xa9

The reason is we do not pass the Head to page_hstate(), and so, the call
to compound_order() in page_hstate() returns 0, so we end up checking
all hstates's size to match PAGE_SIZE.

Obviously, we do not find any hstate matching that size, and we return
NULL.  Then, we dereference that NULL pointer in
hugepage_migration_supported() and we got the #PF from above.

Fix that by getting the head page before calling page_hstate().

Also, since gigantic pages span several pageblocks, re-adjust the logic
for skipping pages.  While are it, we can also get rid of the
round_up().

[[email protected]: remove round_up(), adjust skip pages logic per Michal]
Link: http://lkml.kernel.org/r/[email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Oscar Salvador <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Cc: Pavel Tatashin <[email protected]>
Cc: Mike Rapoport <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fork,memcg: fix crash in free_thread_stack on memcg charge fail

Commit 9b6f7e163cd0 ("mm: rework memcg kernel stack accounting") will
result in fork failing if allocating a kernel stack for a task in
dup_task_struct exceeds the kernel memory allowance for that cgroup.

Unfortunately, it also results in a crash.

This is due to the code jumping to free_stack and calling
free_thread_stack when the memcg kernel stack charge fails, but without
tsk->stack pointing at the freshly allocated stack.

This in turn results in the vfree_atomic in free_thread_stack oopsing
with a backtrace like this:

#5 [ffffc900244efc88] die at ffffffff8101f0ab
 #6 [ffffc900244efcb8] do_general_protection at ffffffff8101cb86
 #7 [ffffc900244efce0] general_protection at ffffffff818ff082
    [exception RIP: llist_add_batch+7]
    RIP: ffffffff8150d487  RSP: ffffc900244efd98  RFLAGS: 00010282
    RAX: 0000000000000000  RBX: ffff88085ef55980  RCX: 0000000000000000
    RDX: ffff88085ef55980  RSI: 343834343531203a  RDI: 343834343531203a
    RBP: ffffc900244efd98   R8: 0000000000000001   R9: ffff8808578c3600
    R10: 0000000000000000  R11: 0000000000000001  R12: ffff88029f6c21c0
    R13: 0000000000000286  R14: ffff880147759b00  R15: 0000000000000000
    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
 #8 [ffffc900244efda0] vfree_atomic at ffffffff811df2c7
 #9 [ffffc900244efdb8] copy_process at ffffffff81086e37
#10 [ffffc900244efe98] _do_fork at ffffffff810884e0
#11 [ffffc900244eff10] sys_vfork at ffffffff810887ff
#12 [ffffc900244eff20] do_syscall_64 at ffffffff81002a43
    RIP: 000000000049b948  RSP: 00007ffcdb307830  RFLAGS: 00000246
    RAX: ffffffffffffffda  RBX: 0000000000896030  RCX: 000000000049b948
    RDX: 0000000000000000  RSI: 00007ffcdb307790  RDI: 00000000005d7421
    RBP: 000000000067370f   R8: 00007ffcdb3077b0   R9: 000000000001ed00
    R10: 0000000000000008  R11: 0000000000000246  R12: 0000000000000040
    R13: 000000000000000f  R14: 0000000000000000  R15: 000000000088d018
    ORIG_RAX: 000000000000003a  CS: 0033  SS: 002b

The simplest fix is to assign tsk->stack right where it is allocated.

Link: http://lkml.kernel.org/r/[email protected]
Fixes: 9b6f7e163cd0 ("mm: rework memcg kernel stack accounting")
Signed-off-by: Rik van Riel <[email protected]>
Acked-by: Roman Gushchin <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Cc: Shakeel Butt <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Tejun Heo <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm: thp: fix flags for pmd migration when split

When splitting a huge migrating PMD, we'll transfer all the existing PMD
bits and apply them again onto the small PTEs.  However we are fetching
the bits unconditionally via pmd_soft_dirty(), pmd_write() or
pmd_yound() while actually they don't make sense at all when it's a
migration entry.  Fix them up.  Since at it, drop the ifdef together as
not needed.

Note that if my understanding is correct about the problem then if
without the patch there is chance to lose some of the dirty bits in the
migrating pmd pages (on x86_64 we're fetching bit 11 which is part of
swap offset instead of bit 2) and it could potentially corrupt the
memory of an userspace program which depends on the dirty bit.

Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Peter Xu <[email protected]>
Reviewed-by: Konstantin Khlebnikov <[email protected]>
Reviewed-by: William Kucharski <[email protected]>
Acked-by: Kirill A. Shutemov <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Dave Jiang <[email protected]>
Cc: "Aneesh Kumar K.V" <[email protected]>
Cc: Souptick Joarder <[email protected]>
Cc: Konstantin Khlebnikov <[email protected]>
Cc: Zi Yan <[email protected]>
Cc: <[email protected]> [4.14+]
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm, memory_hotplug: initialize struct pages for the full memory section

If memory end is not aligned with the sparse memory section boundary,
the mapping of such a section is only partly initialized.  This may lead
to VM_BUG_ON due to uninitialized struct page access from
is_mem_section_removable() or test_pages_in_a_zone() function triggered
by memory_hotplug sysfs handlers:

Here are the the panic examples:
 CONFIG_DEBUG_VM=y
 CONFIG_DEBUG_VM_PGFLAGS=y

 kernel parameter mem=2050M
 --------------------------
 page:000003d082008000 is uninitialized and poisoned
 page dumped because: VM_BUG_ON_PAGE(PagePoisoned(p))
 Call Trace:
 ( test_pages_in_a_zone+0xde/0x160)
   show_valid_zones+0x5c/0x190
   dev_attr_show+0x34/0x70
   sysfs_kf_seq_show+0xc8/0x148
   seq_read+0x204/0x480
   __vfs_read+0x32/0x178
   vfs_read+0x82/0x138
   ksys_read+0x5a/0xb0
   system_call+0xdc/0x2d8
 Last Breaking-Event-Address:
   test_pages_in_a_zone+0xde/0x160
 Kernel panic - not syncing: Fatal exception: panic_on_oops

 kernel parameter mem=3075M
 --------------------------
 page:000003d08300c000 is uninitialized and poisoned
 page dumped because: VM_BUG_ON_PAGE(PagePoisoned(p))
 Call Trace:
 ( is_mem_section_removable+0xb4/0x190)
   show_mem_removable+0x9a/0xd8
   dev_attr_show+0x34/0x70
   sysfs_kf_seq_show+0xc8/0x148
   seq_read+0x204/0x480
   __vfs_read+0x32/0x178
   vfs_read+0x82/0x138
   ksys_read+0x5a/0xb0
   system_call+0xdc/0x2d8
 Last Breaking-Event-Address:
   is_mem_section_removable+0xb4/0x190
 Kernel panic - not syncing: Fatal exception: panic_on_oops

Fix the problem by initializing the last memory section of each zone in
memmap_init_zone() till the very end, even if it goes beyond the zone end.

Michal said:

: This has alwways been problem AFAIU.  It just went unnoticed because we
: have zeroed memmaps during allocation before f7f99100d8d9 ("mm: stop
: zeroing memory during allocation in vmemmap") and so the above test
: would simply skip these ranges as belonging to zone 0 or provided a
: garbage.
:
: So I guess we do care for post f7f99100d8d9 kernels mostly and
: therefore Fixes: f7f99100d8d9 ("mm: stop zeroing memory during
: allocation in vmemmap")

Link: http://lkml.kernel.org/r/[email protected]
Fixes: f7f99100d8d9 ("mm: stop zeroing memory during allocation in vmemmap")
Signed-off-by: Mikhail Zaslonko <[email protected]>
Reviewed-by: Gerald Schaefer <[email protected]>
Suggested-by: Michal Hocko <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Reported-by: Mikhail Gavrilov <[email protected]>
Tested-by: Mikhail Gavrilov <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Alexander Duyck <[email protected]>
Cc: Pasha Tatashin <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Cc: Heiko Carstens <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Merge git://git./linux/kernel/git/davem/sparc

Pull sparc fixes from David Miller:
 "Just some small fixes here and there, and a refcount leak in a serial
  driver, nothing serious"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc:
  serial/sunsu: fix refcount leak
  sparc: Set "ARCH: sunxx" information on the same line
  sparc: vdso: Drop implicit common-page-size linker flag

Merge git://git./linux/kernel/git/davem/net

Pull more networking fixes from David Miller:
 "Some more bug fixes have trickled in, we have:

  1) Local MAC entries properly in mscc driver, from Allan W. Nielsen.

  2) Eric Dumazet found some more of the typical "pskb_may_pull() -->
     oops forgot to reload the header pointer" bugs in ipv6 tunnel
     handling.

  3) Bad SKB socket pointer in ipv6 fragmentation handling, from Herbert
     Xu.

  4) Overflow fix in sk_msg_clone(), from Vakul Garg.

  5) Validate address lengths in AF_PACKET, from Willem de Bruijn"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  qmi_wwan: Fix qmap header retrieval in qmimux_rx_fixup
  qmi_wwan: Add support for Fibocom NL678 series
  tls: Do not call sk_memcopy_from_iter with zero length
  ipv6: tunnels: fix two use-after-free
  Prevent overflow of sk_msg in sk_msg_clone()
  packet: validate address length
  net: netxen: fix a missing check and an uninitialized use
  tcp: fix a race in inet_diag_dump_icsk()
  MAINTAINERS: update cxgb4 and cxgb3 maintainer
  ipv6: frags: Fix bogus skb->sk in reassembled packets
  mscc: Configured MAC entries should be locked.

auxdisplay: charlcd: fix x/y command parsing

The x/y command parsing has been broken since commit 129957069e6a
("staging: panel: Fixed checkpatch warning about simple_strtoul()").

Commit b34050fadb86 ("auxdisplay: charlcd: Fix and clean up handling of
x/y commands") fixed some problems by rewriting the parsing code,
but also broke things further by removing the check for a complete
command before attempting to parse it.  As a result, parsing is
terminated at the first x or y character.

This reinstates the check for a final semicolon.  Whereas the original
code use strchr(), this is wasteful seeing as the semicolon is always
at the end of the buffer.  Thus check this character directly instead.

Signed-off-by: Mans Rullgard <[email protected]>
Signed-off-by: Miguel Ojeda <[email protected]>

serial/sunsu: fix refcount leak

The function of_find_node_by_path() acquires a reference to the node
returned by it and that reference needs to be dropped by its caller.

su_get_type() doesn't do that. The match node are used as an identifier
to compare against the current node, so we can directly drop the refcount
after getting the node from the path as it is not used as pointer.

Fix this by use a single variable and drop the refcount right after
of_find_node_by_path().

Signed-off-by: Yangtao Li <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

sparc: Set "ARCH: sunxx" information on the same line

While checking boot log from SPARC qemu, I saw that the "ARCH: sunxx"
information was split on two different line.
This patchs merge both line together.
In the meantime, thoses information need to be printed via pr_info
since printk print them by default via the warning loglevel.

Signed-off-by: Corentin Labbe <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

sparc: vdso: Drop implicit common-page-size linker flag

GNU linker's -z common-page-size's default value is based on the target
architecture. arch/sparc/vdso/Makefile sets it to the architecture
default, which is implicit and redundant. Drop it.

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Nick Desaulniers <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull kvm fix from Paolo Bonzini:
 "A simple patch for a pretty bad bug: Unbreak AMD nested
  virtualization."

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: x86: nSVM: fix switch to guest mmu

qmi_wwan: Fix qmap header retrieval in qmimux_rx_fixup

This patch fixes qmap header retrieval when modem is configured for
dl data aggregation.

Signed-off-by: Daniele Palmas <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'timers-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull timer fix from Ingo Molnar:
 "Fix a division by zero crash in the posix-timers code"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  posix-timers: Fix division by zero bug