Re: [PATCH v4 0/7] Support inhibiting input devices

From: Hans de Goede
Date: Wed Jun 10 2020 - 09:52:29 EST


Hi,

On 6/10/20 12:38 PM, Rafael J. Wysocki wrote:
On Wed, Jun 10, 2020 at 11:50 AM Hans de Goede <hdegoede@xxxxxxxxxx> wrote:

<snip>

A different, but related issue is how to make devices actually use the
new inhibit support on the builtin keyboard + touchpad when say the lid
is closed. Arguably this is an userspace problem, but it is a tricky
one. Currently on most modern Linux distributions suspend-on-lid-close
is handled by systemd-logind and most modern desktop-environments are
happy to have logind handle this for them.

But most knowledge about input devices and e.g. heurisitics to decide
if a touchpad is internal or external are part of libinput. Now we could
have libinput use the new inhibit support (1), but then when the lid
closes we get race between whatever process is using libinput trying
to inhibit the touchpad (which must be done before to suspend to disable
it as wakeup source) and logind trying to suspend the system.

One solution here would be to move the setting of the inhibit sysfs
attr into logind, but that requires adding a whole bunch of extra
knowledge to logind which does not really belong there IMHO.

I've been thinking a bit about this and to me it seems that the kernel
is in the ideal position to automatically inhibit some devices when
some EV_SW transitions from 0->1 (and uninhibit again on 1->0). The
issue here is to chose on which devices to enable this. I believe
that the auto inhibit on some switches mechanism is best done inside
the kernel (disabled by default) and then we can have a sysfs
attr called auto_inhibit_ev_sw_mask which can be set to e.g.
(1 << SW_LID) to make the kernel auto-inhibit the input-device whenever
the lid is closed, or to ((1 << SW_LID) | (1 << SW_TABLET_MODE)) to
inhibit both when the lid is closed or when switched to tablet mode.

I agree that the kernel is the right place to handle this, but it
requires some extra knowledge about dependencies between devices.

It'd be kind of like power resources in ACPI, so for each state of a
"master" device (in principle, there may be more states of it than
just two) there would be a list of "dependent" intput devices that
need to be inhibited when the "master" device goes into that state.

So a big part of the reason to punt the decision on which input
devices to enable this auto-inhibit is that we don't really have
information about those relationsships / device-links you are
suggesting here. libinput is already doing inhibiting inside
userspace for e.g. the tablet-mode switch but it relies on heuristics
+ quirk tables to decide which keyboards should be inhibited and which
not.

E.g. for a 360 degree hinges 2-in-1 we want to disable the builtin
keyboard, when folded into in tablet mode, but not any external ones.

Mostly the builtin kbd will be PS2 but I have one such 2-in-1 here
in my home office with a USB kbd ...

In general of the master devices there will be only 1, there will be
only 1 lid switch and only 1 tablet-mode switch. So my idea with the
auto_inhibit_ev_sw_mask, is for it to be a per input-device setting.

So using your terms, all input devices with the (1 << SW_LID) bit
set in their auto_inhibit_ev_sw_mask will be dependents of the
(master) device which actually is reporting the SW_LID bit.

The idea here is for this to work the same as how the rfkill code
from net/rfkill/input.c works, except instead of binding e.g.
KEY_WLAN to toggling the sw-state of rfkill devices with a type
of RFKILL_TYPE_WLAN. This will bind SW_LID to inhibiting input
devices with the SW_LID bit set in their auto_inhibit_ev_sw_mask.

Regards,

Hans