Re: Issue with 395f121e6199 ("PCI/PM: Wait for device to become ready after power-on")

Re: Issue with 395f121e6199 ("PCI/PM: Wait for device to become ready after power-on")

[Mika Westerberg]

Hi,

For some reason I don't see this in linux-pci archives. Maybe it was not
accepted because of the attachment? In any case the full dmesg can be
found here as well:

  https://gist.github.com/westeri/ca2574a4e48dc6e3eb76184922e103e5

On Wed, Nov 27, 2019 at 05:34:37PM +0200, Mika Westerberg wrote:
> Hi,
> 
> I noticed that latest Bjorn's pci/pm branch breaks native PCIe hotplug
> so that when I unplug a device (connected over TBT) the pciehp notices
> it but then the core PCI code starts waiting for all the devices that
> are now gone:
> 
> [  160.344476] pcieport 0000:04:04.0: pciehp: pciehp_check_link_active: lnk_status = 5041
> [  160.344531] pcieport 0000:04:04.0: pciehp: Slot(4): Card not present
> [  160.346004] pcieport 0000:04:04.0: pciehp: pciehp_unconfigure_device: domain:bus:dev = 0000:3a:00
> [  161.421569] pcieport 0000:3a:00.0: not ready 1023ms after Switch to D0; waiting
> [  162.509610] pcieport 0000:3a:00.0: not ready 2047ms after Switch to D0; waiting
> [  164.621620] pcieport 0000:3a:00.0: not ready 4095ms after Switch to D0; waiting
> [  169.165529] pcieport 0000:3a:00.0: not ready 8191ms after Switch to D0; waiting
> [  177.869521] pcieport 0000:3a:00.0: not ready 16383ms after Switch to D0; waiting
> [  194.765581] pcieport 0000:3a:00.0: not ready 32767ms after Switch to D0; waiting
> ...
> 
> It seems to continue forever so the devices are not removed from the
> system.
> 
> The test system used here is Dell XPS 9380 but it happens also on other
> systems that support native PCIe hotplug.
> 
> Reverting 395f121e6199 ("PCI/PM: Wait for device to become ready after
> power-on") makes the problem go away.
> 
> I've attached full dmesg from the system. The way I reproduce this is
> that I boot the system up (no device connected yet), the I connect TBT
> dock that includes the PCIe switch etc. Then I unplug the dock. The
> unplug happens around 160.335355 in the dmesg.

Re: Issue with 395f121e6199 ("PCI/PM: Wait for device to become ready after power-on")

[Bjorn Helgaas]

On Thu, Nov 28, 2019 at 10:23:14AM +0200, Mika Westerberg wrote:
> Hi,
> 
> For some reason I don't see this in linux-pci archives. Maybe it was not
> accepted because of the attachment? In any case the full dmesg can be
> found here as well:
> 
>   https://gist.github.com/westeri/ca2574a4e48dc6e3eb76184922e103e5
> 
> On Wed, Nov 27, 2019 at 05:34:37PM +0200, Mika Westerberg wrote:
> > Hi,
> > 
> > I noticed that latest Bjorn's pci/pm branch breaks native PCIe hotplug
> > so that when I unplug a device (connected over TBT) the pciehp notices
> > it but then the core PCI code starts waiting for all the devices that
> > are now gone:
> > 
> > [  160.344476] pcieport 0000:04:04.0: pciehp: pciehp_check_link_active: lnk_status = 5041
> > [  160.344531] pcieport 0000:04:04.0: pciehp: Slot(4): Card not present
> > [  160.346004] pcieport 0000:04:04.0: pciehp: pciehp_unconfigure_device: domain:bus:dev = 0000:3a:00
> > [  161.421569] pcieport 0000:3a:00.0: not ready 1023ms after Switch to D0; waiting
> > [  162.509610] pcieport 0000:3a:00.0: not ready 2047ms after Switch to D0; waiting
> > [  164.621620] pcieport 0000:3a:00.0: not ready 4095ms after Switch to D0; waiting
> > [  169.165529] pcieport 0000:3a:00.0: not ready 8191ms after Switch to D0; waiting
> > [  177.869521] pcieport 0000:3a:00.0: not ready 16383ms after Switch to D0; waiting
> > [  194.765581] pcieport 0000:3a:00.0: not ready 32767ms after Switch to D0; waiting
> > ...
> > 
> > It seems to continue forever so the devices are not removed from the
> > system.
> > 
> > The test system used here is Dell XPS 9380 but it happens also on other
> > systems that support native PCIe hotplug.
> > 
> > Reverting 395f121e6199 ("PCI/PM: Wait for device to become ready after
> > power-on") makes the problem go away.
> > 
> > I've attached full dmesg from the system. The way I reproduce this is
> > that I boot the system up (no device connected yet), the I connect TBT
> > dock that includes the PCIe switch etc. Then I unplug the dock. The
> > unplug happens around 160.335355 in the dmesg.

Thanks for testing this.  I dropped 395f121e6199 ("PCI/PM: Wait for
device to become ready after power-on") for now.  We can sort this out
for v5.6.

Re: Issue with 395f121e6199 ("PCI/PM: Wait for device to become ready after power-on")

[Mika Westerberg]

On Thu, Nov 28, 2019 at 08:59:21AM -0600, Bjorn Helgaas wrote:
> On Thu, Nov 28, 2019 at 10:23:14AM +0200, Mika Westerberg wrote:
> > Hi,
> > 
> > For some reason I don't see this in linux-pci archives. Maybe it was not
> > accepted because of the attachment? In any case the full dmesg can be
> > found here as well:
> > 
> >   https://gist.github.com/westeri/ca2574a4e48dc6e3eb76184922e103e5
> > 
> > On Wed, Nov 27, 2019 at 05:34:37PM +0200, Mika Westerberg wrote:
> > > Hi,
> > > 
> > > I noticed that latest Bjorn's pci/pm branch breaks native PCIe hotplug
> > > so that when I unplug a device (connected over TBT) the pciehp notices
> > > it but then the core PCI code starts waiting for all the devices that
> > > are now gone:
> > > 
> > > [  160.344476] pcieport 0000:04:04.0: pciehp: pciehp_check_link_active: lnk_status = 5041
> > > [  160.344531] pcieport 0000:04:04.0: pciehp: Slot(4): Card not present
> > > [  160.346004] pcieport 0000:04:04.0: pciehp: pciehp_unconfigure_device: domain:bus:dev = 0000:3a:00
> > > [  161.421569] pcieport 0000:3a:00.0: not ready 1023ms after Switch to D0; waiting
> > > [  162.509610] pcieport 0000:3a:00.0: not ready 2047ms after Switch to D0; waiting
> > > [  164.621620] pcieport 0000:3a:00.0: not ready 4095ms after Switch to D0; waiting
> > > [  169.165529] pcieport 0000:3a:00.0: not ready 8191ms after Switch to D0; waiting
> > > [  177.869521] pcieport 0000:3a:00.0: not ready 16383ms after Switch to D0; waiting
> > > [  194.765581] pcieport 0000:3a:00.0: not ready 32767ms after Switch to D0; waiting
> > > ...
> > > 
> > > It seems to continue forever so the devices are not removed from the
> > > system.
> > > 
> > > The test system used here is Dell XPS 9380 but it happens also on other
> > > systems that support native PCIe hotplug.
> > > 
> > > Reverting 395f121e6199 ("PCI/PM: Wait for device to become ready after
> > > power-on") makes the problem go away.
> > > 
> > > I've attached full dmesg from the system. The way I reproduce this is
> > > that I boot the system up (no device connected yet), the I connect TBT
> > > dock that includes the PCIe switch etc. Then I unplug the dock. The
> > > unplug happens around 160.335355 in the dmesg.
> 
> Thanks for testing this.  I dropped 395f121e6199 ("PCI/PM: Wait for
> device to become ready after power-on") for now.  We can sort this out
> for v5.6.

Thanks!