The .remove() callback for a platform driver returns an int which makes
many driver authors wrongly assume it's possible to do error handling by
returning an error code. However the value returned is ignored (apart from
emitting a warning).
To improve here there is a quest to make the remove callback return void.
In the first step of this quest all drivers are converted to .remove_new()
which already returns void. Eventually after all drivers are converted,
.remove_new() is renamed to .remove().
Before this driver might have returned an error. In this case emit a
warning that tells more about the problem than the generic warning by
the core, and instead of making the remove callback return zero
unconditionally, convert to .remove_new() which is equivalent.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Markus Schneider-Pargmann <msp@baylibre.com>
Link: https://lore.kernel.org/r/20230605161458.117361-1-u.kleine-koenig@pengutronix.de
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
The Linux kernel is notorious for misspelling X-Box, X-box, XBox or XBOX;
the official spelling is actually just Xbox. Plain and simple.
Tried to respect the existing notes but still following the style guide.
No functional changes intended. This only affects ancillary parts.
Signed-off-by: Ismael Ferreras Morezuelas <swyterzone@gmail.com>
Link: https://lore.kernel.org/r/401b1d94-1348-15fd-b48f-a80e8885c7a4@gmail.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Upon receiving a PS/2 command the device and controller are supposed to
stop sending normal data (scancodes or movement packets) and instead
immediately start delivering ACK/NAK and command response. Unfortunately
often EC has an output buffer which may contain latched data by the time
the EC receives a command from the host. The kernel used to ignore such
data, but that may cause "stuck" keys if the data dropped happens to be a
break code or a part of a break code. This occasionally happens, for
example, on Chromebooks when the kernel tries to toggle CapsLock LED on
a keyboard while user releases Alt+Search keyboard shortcut.
Fix this by passing the first non-ACK byte to the normal handler for a
handful of PS/2 commands that are expected to be used during normal device
operation (as opposed to probe/configuration time).
Reviewed-by: Raul E Rangel <rrangel@chromium.org>
Link: https://lore.kernel.org/r/20230511185252.386941-8-dmitry.torokhov@gmail.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Instead of exposing inner workings of libps2 to drivers such as atkbd and
psmouse, have them define pre-receive and receive callbacks, and provide a
common handler that can be used with underlying serio port.
While at this add kerneldoc to the module.
Link: https://lore.kernel.org/r/ZGK81cxqjr/KS1kA@google.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
When getting unexpected data while waiting for an acknowledgement it does
not matter what command phase is currently executed, and ps2_handle_ack()
should indicate that no further processing is needed for the received data
byte. Remove PS2_FLAG_ACK_CMD and associated handling.
Note that while it is possible to make ps2_handle_ack (and
ps2_handle_repsonse) return void, it will be done when the code will be
converted to common PS/2 interrupt handler later.
Reviewed-by: Raul E Rangel <rrangel@chromium.org>
Link: https://lore.kernel.org/r/20230511185252.386941-3-dmitry.torokhov@gmail.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
The touch controller needs some time to wake-up after setting the wake-up
gpio. Without having a delay after wake-up probing regularly fails in
edt_ft5x06_ts_identify() with an error (i.e. EREMOTEIO) that was caused
by a failed i2c transfer.
The datasheet sets the wake-up time to 5 ms, although it is not entirely
clear.
Signed-off-by: Philipp Puschmann <p.puschmann@pironex.com>
Link: https://lore.kernel.org/r/20230505115823.545803-1-p.puschmann@pironex.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
The touchscreen can be put into a deep sleep state that prevents it from
emitting touch irqs. Put the touchscreen into deep sleep during suspend
if it is not marked as a wakeup source.
This also fixes a problem with the touchscreen getting unresponsive after
system resume when a falling edge trigger is used for the interrupt.
When left on during suspend, the touchscreen would pull the interrupt
line down in response to touch events, leaving the interrupt effectively
disabled after resume.
Signed-off-by: Maximilian Weigand <mweigand@mweigand.net>
Reviewed-by: Alistair Francis <alistair@alistair23.me>
Link: https://lore.kernel.org/r/20230504120316.408687-2-mweigand2017@gmail.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
input_abs_set_val() can nominally call input_alloc_absinfo() which may
allocate memory with GFP_KERNEL flag. This does not happen when
input_abs_set_val() is called by the input core to set current MT slot when
handling a new input event, but it trips certain static analyzers.
Rearrange the code to access the relevant structures directly.
Reported-by: Teng Qi <starmiku1207184332@gmail.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Link: https://lore.kernel.org/r/ZFBg379uuHjf+YEM@google.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
With CONFIG_DEBUG_SLAB=y:
# Subtest: input_core
1..3
input: Test input device as /devices/virtual/input/input1
8<--- cut here ---
Unable to handle kernel paging request at virtual address 6b6b6dd7 when read
...
__lock_acquire from lock_acquire+0x26c/0x300
lock_acquire from _raw_spin_lock_irqsave+0x50/0x64
_raw_spin_lock_irqsave from devres_remove+0x20/0x7c
devres_remove from devres_destroy+0x8/0x24
devres_destroy from input_free_device+0x2c/0x60
input_free_device from kunit_try_run_case+0x70/0x94 [kunit]
Without CONFIG_DEBUG_SLAB=y:
KTAP version 1
# Subtest: input_core
1..3
input: Test input device as /devices/virtual/input/input1
------------[ cut here ]------------
WARNING: CPU: 0 PID: 694 at lib/refcount.c:28 refcount_warn_saturate+0x54/0x100
refcount_t: underflow; use-after-free.
...
Call Trace: [<0037cad4>] dump_stack+0xc/0x10
[<00377614>] __warn+0x7e/0xb4
[<0037768c>] warn_slowpath_fmt+0x42/0x62
[<001eee1c>] refcount_warn_saturate+0x54/0x100
[<000b1d34>] kfree_const+0x0/0x20
[<0036290a>] __kobject_del+0x0/0x6e
[<001eee1c>] refcount_warn_saturate+0x54/0x100
[<00362a1a>] kobject_put+0xa2/0xb6
[<11965770>] kunit_generic_run_threadfn_adapter+0x0/0x1c [kunit]
As per the comments for input_allocate_device() and
input_register_device(), input_free_device() must be called only to free
devices that have not been registered. input_unregister_device()
already calls input_put_device(), thus leading to a use-after-free.
Moreover, the kunit_suite.exit() method is called after every test case,
even on failures. As the test itself already does cleanups in its
failure paths, this may lead to a second use-after-free.
Fix the first issue by dropping the call to input_allocate_device() from
input_test_exit().
Fix the second issue by making the cleanup code conditional on a
successful test.
Fixes: fdefcbdd6f ("Input: Add KUnit tests for some of the input core helper functions")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://lore.kernel.org/r/957b3b309a44d39fb6e38b2a526b250f69ea3d2c.1683022164.git.geert+renesas@glider.be
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Currently, uinput doesn't use the input_set_timestamp API, so any
event injected using uinput is not accurately timestamped in terms of
measuring when the actual event happened. Hence, call the
input_set_timestamp API from uinput in order to provide a more
accurate sense of time for the event. Propagate only the timestamps
which are a) positive, b) within a pre-defined offset (10 secs) from
the current time, and c) not in the future.
Signed-off-by: Biswarup Pal <biswarupp@google.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Siarhei Vishniakou <svv@google.com>
Link: https://lore.kernel.org/r/20230427000152.1407471-1-biswarupp@google.com
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
The watchdog_timer can schedule tx_timeout_task and watchdog_work
can also arm watchdog_timer. The process is shown below:
----------- timer schedules work ------------
cyttsp4_watchdog_timer() //timer handler
schedule_work(&cd->watchdog_work)
----------- work arms timer ------------
cyttsp4_watchdog_work() //workqueue callback function
cyttsp4_start_wd_timer()
mod_timer(&cd->watchdog_timer, ...)
Although del_timer_sync() and cancel_work_sync() are called in
cyttsp4_remove(), the timer and workqueue could still be rearmed.
As a result, the possible use after free bugs could happen. The
process is shown below:
(cleanup routine) | (timer and workqueue routine)
cyttsp4_remove() | cyttsp4_watchdog_timer() //timer
cyttsp4_stop_wd_timer() | schedule_work()
del_timer_sync() |
| cyttsp4_watchdog_work() //worker
| cyttsp4_start_wd_timer()
| mod_timer()
cancel_work_sync() |
| cyttsp4_watchdog_timer() //timer
| schedule_work()
del_timer_sync() |
kfree(cd) //FREE |
| cyttsp4_watchdog_work() // reschedule!
| cd-> //USE
This patch changes del_timer_sync() to timer_shutdown_sync(),
which could prevent rearming of the timer from the workqueue.
Fixes: 17fb1563d6 ("Input: cyttsp4 - add core driver for Cypress TMA4XX touchscreen devices")
Signed-off-by: Duoming Zhou <duoming@zju.edu.cn>
Link: https://lore.kernel.org/r/20230421082919.8471-1-duoming@zju.edu.cn
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
First of all, previously the 16-bit magnitude was written as-is to the
device which actually discarded the upper 8 bits since the device has
8-bit registers only. This meant that a strong_magnitude of 0xFF00 would
result in 0. To correct this shift the strong_magnitude / weak_magnitude
input values so we discard the lower 8 bits and keep the upper bits
instead.
Secondly the RTP mode that is used by default interprets the values as
signed (2s complement), so 0x81 = 0%, 0x00 = 50%, 0x7F = 100%. This
doesn't match the FF_RUMBLE interface at all, so let's tell the device
to interpret the data as unsigned instead which gets us 0x00 = 0% and
0xFF = 100%.
As last change switch ERM to using "Closed-Loop Mode, Unidirectional"
instead of "Open-Loop Mode" since it's recommended by the datasheet
compared to open loop and better matches our use case of 0% - 100%
vibration.
Signed-off-by: Luca Weiss <luca@z3ntu.xyz>
Link: https://lore.kernel.org/r/20230430-drv260x-improvements-v1-4-1fb28b4cc698@z3ntu.xyz
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
After starting to use regmap API to access registers the edt-ft5x06
driver depends on symbols provided by REGMAP_I2C:
edt-ft5x06.o: in function `edt_ft5x06_ts_probe':
edt-ft5x06.c:1154: undefined reference to `__regmap_init_i2c'
edt-ft5x06.o: in function `edt_ft5x06_ts_identify':
edt-ft5x06.c:897: undefined reference to `__regmap_init_i2c'
Make sure support for I2C regmap is actually selected by adding this
dependency to Kconfig.
Fixes: 9dfd9708ff ("Input: edt-ft5x06 - convert to use regmap API")
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Link: https://lore.kernel.org/r/ZDRBExF1xmxalMZc@makrotopia.org
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
The driver had the code to differentiate between finger and palm touches,
but did not use this information when reporting contacts. Change it so that
proper "tool" type is assigned to reported contacts.
Signed-off-by: JungHoon Hyun <hyunjunghoon@melfas.com>
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
