I committed mmap()-based changes. See: https://cgit.freebsd.org/src/commit/?id=87ed6840a0a6320295f6abb43a987d20ae126cf7

Although it looks good, replacing read() with mmap() -> memcpy() -> munmap() sequence would make code simpler and cleaner

IMO it would be good to rename driver from ps5dualsense to ps5dsense to be inline with ps4dshock. That is up to you.

Looks mostly good.
There are some missing parts. TODO:

1. Add manpage. You may take share/man/man4/ps4dshock.4 as starting point. Do not forget to add missing touchpad and accelerometer support to BUGS section.
2. Add ps5dualsense.c to sys/conf/files and sys/conf/NOTES
3. Add ps5dualsense subdir to sys/modules/hid/Makefile

It seems that more files should be fixed. See your previous patch: https://reviews.freebsd.org/D56126

In D56126#1283969, @chris.longros_gmail.com wrote:

You're right, the generic Realtek rule already covers this. I missed the USB_VENDOR match at the top of the table. Abandoning this diff.

Good point about the other 0x0bda entries too -- I can submit a cleanup diff to drop those if you'd like.

Other Realtek entries are inherited from Linux. I think we can drop them.

Realtek devices are handled by generic Realtek rule. Does it work for you?

Alder Lake system is not needed

LGTM. Could you test Intel BT identify patch from https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=290639 ?

Radio button can not be correctly handled with hidmap. MS does strange things here assuming every event in report as a keypress. E.g. ASUS TUF always reported radio key as depressed that brokes Linux but works on Windows

It will trigger enumeration value ‘S76_CTRL_MAX’ not specifically handled in switch if -Wswitch option is passed to compiler.
From [1]:

-Wswitch

Warn whenever a switch statement has an index of enumerated type and lacks a case for one or more of the named codes of that enumeration. (The presence of a default label prevents this warning.) case labels that do not correspond to enumerators also provoke warnings when this option is used, unless the enumeration is marked with the flag_enum attribute. This warning is enabled by -Wall.

Obviously case S76_CTRL_MAX is not generally handled in switch() statements.

I dislike this idea.
The sysctl sys.class.drm.card0-DP-1 path matches Linux while sysctl sys.device.drmn1.card0.card0-DP-1 not

In D55694#1274837, @pouria wrote:

The keyboard_color has full RGB coverage and accept 24-bit value as its input, so I don't think registering it in backlight be a good idea.

Adding of EVDEV interface to notyfy userland software would be good idea too. See sys/dev/acpi_support/acpi_asus_wmi.c again

In D55694#1274782, @pouria wrote:

could you take another look at the code?

We have backlight(9) interface since FreeBSD 13

LGTM

Fix braino. size->count

Update page table in chunks having faked page array fit in malloc zone

Not needed after D54225 changes

D54225 uses pmap_qenter() now

Use pmap_qenter() instead homegrown pmap_qmapdev_attr()

s/lkpi_/linuxkpi_/
Add actual arch name to panic message

s/linux_/lkpi_/ in function names

This should be joined with D54223

This should be joined with D54226