On Sat, 16 Jun 2018 17:43:09 +0200,
Pali Rohár wrote:
>
> On Saturday 16 June 2018 09:05:41 Takashi Iwai wrote:
> > On Fri, 15 Jun 2018 21:09:59 +0200,
> > Pali Rohár wrote:
> > >
> > > On Friday 15 June 2018 14:51:47 Takashi Iwai wrote:
> > > > On Fri, 08 Jun 2018 13:10:57 +0200,
> > > > Pali Rohár wrote:
> > > > >
> > > > > Hi! With up-to-date thinkpad_acpi.ko driver on ThinkPad T480s I'm seeing
> > > > > a strange behavior of LEDs which are integrated into mic mute (Fn+F4)
> > > > > and mute (Fn+F1) keys.
> > > > >
> > > > > When thinkpad_acpi.ko is not loaded, then mute key is working fine. When
> > > > > pressed, it correctly generates KEY_MUTE on AT Translated Set 2 keyboard
> > > > > input device and also turn on/of mute led. But when micmute key is
> > > > > pressed then, nothing happen. No key event is reported and also led is
> > > > > not turned on/off.
> > > > >
> > > > > On the other hand, when thinkpad_acpi.ko is loaded, then both buttons
> > > > > mute and micmute correctly generates input events; mute via AT keyboard
> > > > > and micmute via ThinkPad Extra Buttons. But led is not changed. When
> > > > > thinkpad_acpi.ko is loaded it turn off both leds (mute and micmute) and
> > > > > leds after pressing any of those buttons, leds are not turned on again.
> > > > >
> > > > > When thinkpad_acpi.ko is unloaded, then pressing mute button again start
> > > > > switching led on/off.
> > > > >
> > > > > So it seems that some init sequence of thinkpad_acpi.ko breaks mute led.
> > > > > And fini sequence of thinkpad_acpi.ko makes mute led working again.
> > > >
> > > > Usually the mute LED on Thinkpad is triggered from HD-audio driver
> > > > (sound/pci/hda/thinkpad_helper.c), and it's a soft-bound via
> > > > symbol_request(tpacpi_led_set). I thought thinkpad_acpi is
> > > > auto-loaded when the module gets bound.
> > > >
> > > > A possible explanation would be that TPT480s has neither IBM0068,
> > > > LEN0068 nor LEN0268 ACPI HIDs, hence the driver is not auto-loaded.
> > >
> > > I have Debian Stretch kernel (4.9) which does not have LEN0268 alias for
> > > thinkpad_acpi.ko. So thinkpad_acpi.ko is not loaded automatically. But I
> > > have put thinkpad_acpi into /etc/modules and it is now automatically
> > > loaded at boot.
> >
> > That's odd. It's exposed via
> > MODULE_DEVICE_TABLE(acpi, ibm_htk_device_ids);
> >
> > It's been already in 4.9. At this point, something is fishy.
>
> $ /sbin/modinfo thinkpad_acpi | grep alias
> alias: dmi:bvnIBM:bvrI[MU]ET??WW*
> alias: tpacpi
> alias: acpi*:LEN0068:*
> alias: acpi*:IBM0068:*
>
> No there is no LEN0268 on 4.9.
OK, that's the cause. It's really old.
The commit a3c42a467a25 ("platform/x86: thinkpad_acpi: Adding new
hotkey ID for Lenovo thinkpad") has to be backported.
Also, in the HD-audio side, the commit 2ecb704a1290 ("ALSA: hda - add
a new condition to check if it is thinkpad") is needed, too.
> > > I also compiled upstream version of thinkpad_acpi.ko, loaded it in
> > > Stretch kernel, but it behaves in same way.
> > >
> > > Maybe... there could be a problem that thinkpad_acpi.ko must be already
> > > loaded when sound subsystem is doing initialization? If yes, this could
> > > explain it as /etc/modules is loaded at later stage and manually loading
> > > of new version of thinkpad_acpi.ko at runtime does not help when sound
> > > subsystem is already running.
> >
> > Not really. The HD-audio driver tries to bind with tpacpi_led_set()
> > via symbol_request(). i.e. if it's not present, it tries to load a
> > module.
> >
> > Check whether hda_fixup_thinkpad_acpi() is called and the symbol gets
> > loaded or not.
> >
> > But, I don't think it's worth to debug such an old kernel primarily.
>
> It is default one used by the last released Debian stable version.
Heh, that explains :)
And there was a recent regression in HD-audio that was addressed in
4.9.104. If you're using some earlier 4.9.x, you might hit the
problem regarding HD-audio thinkpad_acpi binding.
(But I guess it doesn't work in anyway without the backport of the
commit above.)
> > Could you test the latest Linus tree or 4.17.x at least as a test
> > basis?
>
> Ok, will do that later.
If my analysis above is correct, everything should work with the
recent upstream kernel as is.
Once after you confirm it, I can cook a patch to add the mixer enum to
change LED behavior as you wanted.
Takashi
