On Wed, Jan 15, 2020 at 11:12:23PM -0500, Martin K. Petersen wrote:

Guenter,

quoted

The hwmon-next branch is based on v5.5-rc1. It might be better to
either merge hwmon-next into mainline, or to apply the drivetemp patch
to mainline, and test the result. I have seen some (unrelated) weird
tracebacks in the driver core with v5.5-rc1, so that may not be the
best baseline for a test.

I'm afraid the warnings still happen with hwmon-next on top of
linus/master.

Can you by any chance provide a full traceback ?

The warning you reported suggests that a devm_ function was called on a
device pointer prior to a device registration. I don't immediately see
how that happens. A full traceback might give us an idea.

I suspect that the underlying problem is in hwmon_device_register_with_info(),
which uses devm_ functions to allocate memory associated with the device
pointer passed to it, in this case the SCSI device. This is inherently
wrong (independent of this driver), since the lifetime of the hardware
monitoring device does not necessarily match the lifetime of its parent.
The impact here is that we may get a memory leak under some circumstances.
I'll have to fix that in the hardware monitoring core.

Either case, I would like to track down how the warning happens, so any
information you can provide that lets me reproduce the problem would be
very helpful.

Thanks,
Guenter