Re: [PCI] 3233e41d3e: WARNING:at_drivers/pci/pci.c:#pci_reset_hotplug_slot

[Lukas Wunner <lukas@xxxxxxxxx>]

On Thu, Jul 23, 2020 at 05:13:06PM +0800, kernel test robot wrote:
> FYI, we noticed the following commit (built with gcc-9):
[...]
> commit: 3233e41d3e8ebcd44e92da47ffed97fd49b84278 ("[PATCH] PCI: pciehp: Fix 
> AB-BA deadlock between reset_lock and device_lock")
[...]
> caused below changes (please refer to attached dmesg/kmsg for entire 
> log/backtrace):
> [    0.971752] WARNING: CPU: 0 PID: 1 at drivers/pci/pci.c:4905 
> pci_reset_hotplug_slot+0x70/0x80

Thank you, trusty robot.

I botched the call to lockdep_assert_held_write(), it should have been
conditional on "if (probe)".

Happy to respin the patch, but I'd like to hear opinions on the locking
issues surrounding xen and octeon (and the patch in general).

In particular, would a solution be entertained wherein the pci_dev is
reset by the PCI core after driver unbinding, contingent on a flag which
is set by a PCI driver to indicate that the pci_dev is returned to the
core in an unclean state?

Also, why does xen require a device reset on bind?

Thanks!

Lukas