Hi, On 11/24/21 15:31, Hans de Goede wrote: > Hi, > > On 11/24/21 05:13, Lukas Wunner wrote: >> On Mon, Nov 22, 2021 at 03:29:43PM -0600, Bjorn Helgaas wrote: >>> On Mon, Nov 22, 2021 at 05:45:32PM +0100, Hans de Goede wrote: >>>> With 5.16-rc2 I'm getting the following lockdep warning when unplugging >>>> a Lenovo X1C8 from a Lenovo 2nd gen TB3 dock: >> >> Thanks for the report. I'm aware of this issue, it's still on my todo >> list. Theodore already came across it a while ago: >> >> https://lore.kernel.org/linux-pci/20190402021933.GA2966@xxxxxxx/ >> >> It's a false positive, we need to use a separate lockdep class either >> for each hotplug port or for each level in the PCI hierarchy. > > Can we easily determine what the level in the PCI hierarchy is ? > > If yes; and if having a separate lock class per level is enough, > then the code could simply switch to down_read_nested > (and other xxx_nested) functions passing the level as > "subclass" parameter. > > If no, maybe we should add an "int level" member to > struct controller ? > > And then make the switch to the foo_nested locking functions > based on that ? So I've given the above idea a shot and fixes the false-positive lockdep warning for me. I've posted a 2 patch series fixing this upstream: https://lore.kernel.org/linux-pci/20211129121934.4963-1-hdegoede@xxxxxxxxxx/T/#t Note the 2nd patch can probably use a Fixes: tag but I had no idea which commit to put there. Or maybe add a Cc: stable to both patches? Regards, Hans
