lenovo-ymc: Yoga 7 14IRL8 keyboard backlight stays on in tablet mode; EC expects VPC 0x2A with posture data

[Mahir Faisal]

Hello,

On a Lenovo Yoga 7 14IRL8 the keyboard backlight stays on when the
laptop is folded into tablet/tent/stand mode under Linux, while on
Windows it turns off. I traced the difference to a VPC command that
the Windows YMC service sends on every posture change and that Linux
does not send. Details below; I'd appreciate your view on whether
lenovo-ymc/ideapad-laptop should send it.

System
------
  Model:  Lenovo Yoga 7 14IRL8 (DMI product name: 82YL)
  BIOS:   LHCN30WW (latest)
  Kernel: 7.2.11-300.fc45.x86_64 (Fedora)

Symptom
-------
When folded, the EC disables the keyboard (no input reaches a VT) and
SW_TABLET_MODE is reported correctly by lenovo-ymc. The keyboard
backlight, however, stays at its previous level. Writes through
KBLC (via the ideapad-laptop kbd_backlight LED / UPower) are stored in
the EC (KBLS reads back the new value) but are not applied while
folded, so there is no way to turn the backlight off without
unfolding.

The BIOS changelog suggests this is meant to be EC-driven. LHCN10WW,
"EC Notification":
  "Fix the issue of keyboard backlight is on when turn on in notebook
   mode and change NoteBook mode to Yoga mode."

The DSDT contains no tablet-mode handling in KBLC, and _Q1B only does
Notify (WMIY, 0xD0). WMIY.WMAB just returns PCMD.

What Windows does
-----------------
The "Lenovo Yoga Mode Control" package (YMC.exe 3.0.1.5, bound to
ACPI\YMC2017) sends IOCTL 0x831020C0 to \\.\EnergyDrv with a 12-byte
buffer {0x0C, 1, X}. EnergyDrv is AcpiVpc.sys from "Lenovo Energy
Management". Its handler for that IOCTL maps function 0x0C to VPC
command 0x2A (VPCCMD_W_YMC) and issues it through VPCW with X as the
data byte, using the same VDAT/VCMD sequence as write_ec_cmd().

Values of X observed in YMC.exe:
  0x0A  service start
  0x08  posture != laptop (tent/stand/tablet)
  0x09  posture == laptop
  0x0B  service stop

Current Linux behaviour
-----------------------
ideapad_laptop_trigger_ec() sends VPCCMD_W_YMC with data 1, and only
for models in ymc_ec_trigger_quirk_dmi_table. Booting with
ideapad_laptop.ymc_ec_trigger=1 on this machine has no effect on the
backlight.

Test
----
Using acpi_call against \_SB.PC00.LPCB.H_EC.VPC0, with VPCW 0 <data>
followed by VPCW 1 0x2a, then polling VPCR 1 until 0:

  data 0x0A, then fold      -> no change, backlight stays on
  fold, then data 0x08      -> EC turns the keyboard backlight off
  unfold, then data 0x09    -> keyboard and backlight work normally

I'm currently running a small userspace daemon that sends 0x0A at
start, then 0x08/0x09 on each SW_TABLET_MODE change. It has worked
reliably so far.

Suggestion
----------
On a YMC event, send VPCCMD_W_YMC with 0x08 when the reported posture
is anything other than laptop and 0x09 for laptop, instead of 1. This
might also be worth doing outside the quirk table, since it matches
what the Windows driver does on every model it supports. I can't tell
whether data 1 is still needed on the models already in the quirk
table (Yoga 7 14ARB7 / 14ACN6) for event re-arming, so someone with
that hardware would need to check.

I can provide an acpidump, the disassembly excerpts, or test patches
on this machine.

Disclosure: AI assistance was used to troubleshoot the issue and draft
this email.

Thanks,
Mahir