Input events not working on ARC HS58

[jzbydniewski]

HS58, toolchain 2023.09 using glibc, buildroot arc-2023.09, Linux 5.15.127

Input events (mouse, keboard events) are not visible in applications

• Raspberry platform

pi@raspberrypi4:~ $ evemu-record /dev/input/event1 &
[1] 169299
pi@raspberrypi4:~ $ # EVEMU 1.3
# Kernel: 6.6.51+rpt-rpi-2712
# Input device name: "vc4-hdmi-0"
# Input device ID: bus 0x1e vendor 0000 product 0000 version 0x01
# Supported events:
[..]
#   Event type 2 (EV_REL)
#     Event code 0 (REL_X)
#     Event code 1 (REL_Y)
#   Event type 4 (EV_MSC)
#     Event code 4 (MSC_SCAN)
#   Event type 20 (EV_REP)
#     Event code 0 (REP_DELAY)
#     Event code 1 (REP_PERIOD)
[..]
################################
#      Waiting for events      #
################################

pi@raspberrypi4:~ $ evemu-event /dev/input/event1 --type EV_REL --code REL_X --value 3 --sync
E: 0.000001 0002 0000 0003      # EV_REL / REL_X                3
E: 0.000001 0000 0000 0000      # ------------ SYN_REPORT (0) ---------- +0ms

• HS58 platform

# evemu-record /dev/input/event0 &
# # EVEMU 1.3
# Kernel: 5.15.127
# Input device name: "Logitech USB Optical Mouse"
# Input device ID: bus 0x03 vendor 0x46d product 0xc077 version 0x111
# Supported events:
[..]
#   Event type 2 (EV_REL)
#     Event code 0 (REL_X)
#     Event code 1 (REL_Y)
#     Event code 8 (REL_WHEEL)
#     Event code 11 (REL_WHEEL_HI_RES)
#   Event type 4 (EV_MSC)
#     Event code 4 (MSC_SCAN)
[..]
################################
#      Waiting for events      #
################################

# evemu-event /dev/input/event0 --type EV_REL --code REL_X --value 3 --sync
error: could not play event

My investigation shows that this is due to difference in size of the structure "input_event" between user space side and kernel. This is due to the fact that glibc is using 64-bit members in timeval struct, while kernel requires 32-bit members.

I attach a patch that works for me - although a bit hacky I hope this help you in preparing a proper fix.

patch-input-event.patch

[abrodkin]

@jzbydniewski nice catch, thanks! I'm not sure though the patch is correct. See, it's not a unique situation of having 64-bit time on 32-bit system, in fact IIRC it's now a standard for glibc for any 32-bit system. That said, the fix for the Linux kernel needs to be in place already, possibly in a bit later kernel version. I'll let our glibc maintainer @pavelvkozlov to look into it.
As usual a question to you: how soon do you need that fixed and how important is that one for you?

[jzbydniewski]

Updated patch, the previous one was incomplete:
patch-input-event.patch

In terms of priority I would say "medium", I think I can get away with workaround for some time

[pavelvkozlov]

Hi @jzbydniewski,

On the kernel side, nothing has changed in the input event data structure since torvalds@152194f which specified using unsigned 32-bit sec/usec timestamp fields in the input event structure.

The issue was in glibc and was fixed by the commit: bminor/glibc@a4ed047

The correct behavior of glibc is to provide 32-bit timestamp fields to the kernel.

So, I would recommend using toolchain 2024.06 or late and buildroot arc-2024.06. Toolchain 2024.06 uses glibc 2.39 which has mentioned commit.