Re: ERROR: Writing to dgpu_disable cause Input/Output error

[Armin Wolf <W_Armin@xxxxxx>]

Am 09.01.24 um 01:00 schrieb Bjorn Helgaas:

> On Sat, Jan 06, 2024 at 11:33:35PM +0100, Armin Wolf wrote:
> > Am 04.01.24 um 03:50 schrieb Athul Krishna:
> >
> > > Sent with Proton Mail secure email.
> > >
> > > On Thursday, January 4th, 2024 at 1:05 AM, Armin Wolf <W_Armin@xxxxxx> wrote:
> > >
> > >
> > > > Am 03.01.24 um 19:51 schrieb Athul Krishna:
> > > >
> > > > > Hello,
> > > > > This is my first time reporting an issue in the kernel.
> > > > >
> > > > > Device Details:
> > > > >
> > > > > * Asus Zephyrus G14 (||||||GA402RJ)
> > > > > * Latest BIOS
> > > > > * Arch_x86_64
> > > > > * Kernel: 6.6.9
> > > > > * Minimal install using archinstall
> > > > >
> > > > > ISSUE: Using /dgpu_disable /provided by _asus-nb-wmi _to disable and
> > > > > enable dedicated gpu,
> > > > > causes system crash and reboots, randomly.
> > > > > 9/10 times writing 0 to dgpu_disable will produce an Input/Output
> > > > > error, but the value will be changed to 0, half the time system will
> > > > > crash and reboot. While writing 1 to it doesn't produce an error, I
> > > > > have observed system crash twice just after that.
> > > > >
> > > > > Steps to Reproduce:
> > > > >
> > > > > * Remove dpgu: echo 1 | sudo tee ../remove (dgpu path)
> > > > > * echo 1 | sudo tee /sys/devices/platform/asus-nb-wmi/dgpu_disable
> > > > > * echo 0 | sudo tee /sys/devices/platform/asus-nb-wmi/dgpu_disable
> > > > >
> > > > > * echo 1 | sudo tee /sys/bus/pci/rescan
> > > > >
> > > > > After writing 0 to dgpu_disable, there's an entry in journal about an
> > > > > ACPI bug.
> > > > > Output of 'journalctl -p 3' and lspci is attached.
> > > > Hi,
> > > >
> > > > Can you share the output of "acpidump" and the content of "/sys/bus/wmi/devices/05901221-D566-11D1-B2F0-00A0C9062910[-X]/bmof"?
> > > > The bmof files contain a description of the WMI interfaces of your machine, which might be important for diagnosing the error.
> > > >
> > > > Thanks,
> > > > Armin Wolf
> > > Here's the output of 'acpidump > acpidump.out' and 'cat /sys/bus/wmi/devices/05901221-D566-11D1-B2F0-00A0C9062910[-X]/bmof'
> > Ok, it seems the ACPI code tries to access an object ("GC00") which does not exist.
> > This is the reason why disabling the dGPU fails with -EIO.
> >
> > I am unfortunately not that knowledgeable when it comes to PCI problems, i CCed the linux-pci mailing list in hope that
> > they can better help you in this regard.
> FWIW, I don't know enough about what's going on here to see a PCI
> connection.  I do see a bunch of PCI-related stuff around rfkill, but
> I don't think that's involved here.
>
> I think the path here is this, which doesn't seem to touch anything in
> PCI:
>
>    dgpu_disable_store
>      asus_wmi_set_devstate(ASUS_WMI_DEVID_DGPU, ..., &result)
>        asus_wmi_evaluate_method(ASUS_WMI_METHODID_DEVS, ...)
>          asus_wmi_evaluate_method3
>            wmi_evaluate_method(ASUS_WMI_MGMT_GUID, ...)
>      if (result > 1)
>        return -EIO
>
> But if I missed it, let me know and I'll be happy to take another
> look.
>
> Bjorn

The issue happens when a PCI bus rescan is done after writing to "dgpu_disable".
As a side note a bugzilla bugreport for this issue was recently created:

https://bugzilla.kernel.org/show_bug.cgi?id=218354

Thanks,
Armin Wolf