On Wed, Jan 13, 2021 at 5:05 PM Doug Anderson [off-list ref] wrote:

Hi,

On Wed, Jan 13, 2021 at 7:09 AM Benjamin Tissoires
[off-list ref] wrote:

quoted

quoted

I wanted to apply the series yesterday, but for these kinds of changes
I like giving it a spin on actual hardware. Turns out that my XPS-13
can not boot to v5.11-rc2, which makes testing the new branch slightly
more difficult.

I'll give it a spin next week, but I think I should be able to land it for 5.12.

Regarding the defconfig conflict, no worries, we can handle it with
Stephen and Linus.

After 2 full kernel bisects (I messed up the first because I am an
idiot and inverted good and bad after the first reboot), I found my
culprit, and I was able to test the series today.

The series works fine regarding enumeration and removing of devices,
but it prevents my system from being suspended. If I rmmod
i2c-hid-acpi, suspend works fine, but if it is present, it immediately
comes back, which makes me think that something must be wrong.

I also just reverted the series and confirmed that suspend/resume now
works, meaning that patch 1/4 needs to be checked.

Can you give me any hints about what type of failure you're seeing?
Any logs?  I don't have an ACPI system to test with...

I don't have any logs, just that the system comes back up. There is a
chance we are not powering the device down correctly, which triggers
an IRQ and which puts the system back on.

Is there any chance that some type of userspace / udev rule is getting
tripped up by the driver being renamed?  We ran into something like
this recently on Chrome OS where we had a tool that was hardcoded to
look for "i2c-hid" and needed to be adapted to account for the new
driver name.  Often userspace tweaks with wakeup rules based on driver
name...

I don't think there is anything like that on a regular desktop.

I'll go stare at the code now and see if anything jumps out.

Thanks, but don't spend too much time on it, unless something really
jumps out. I'll debug that tomorrow. It's much easier with an actual
device than by just looking at the code.

Cheers,
Benjamin

-Doug