[PATCH 0/2] PCI: pci_unignore_hotplug()

[PATCH 0/2] PCI: pci_unignore_hotplug()

[Alex Williamson]

I'd kinda like to use pci_ignore_hotplug() for devices in use by a
user via vfio-pci, but the interface seems only partially implemented
since we can only set ignore_hotplug and not clear it.  This adds the
capability to clear the flag and applies it as seems to be appropriate
to the existing users.  I don't have a system with power control for
a radeon or nouveau device, so I'd certainly appreciate testing.  It
seems though that once the device is in D0, we've passed the critical
point where the device could be accidentally unplugged.  Thanks,

Alex

---

Alex Williamson (2):
      pci: Add reciprocal function for pci_ignore_hotplug()
      nouveau/radeon: Un-ignore hotplug on resume


 drivers/gpu/drm/nouveau/nouveau_drm.c |    1 +
 drivers/gpu/drm/radeon/radeon_drv.c   |    1 +
 include/linux/pci.h                   |    5 +++++
 3 files changed, 7 insertions(+)

[PATCH 1/2] pci: Add reciprocal function for pci_ignore_hotplug()

[Alex Williamson]

We have pci_ignore_hotplug(), but there's no way to undo this
without manipulating the device itself.  It seems like the original
intention of the function is to provide the driver with a way to
ignore hotplug over a specific time window, ex. the time while the
device is powered off.  Once the device is powered on again, the
flag should be cleared to avoid leaving driver cruft on the device.
Add a trivial helper function to enable this.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
---
 include/linux/pci.h |    5 +++++
 1 file changed, 5 insertions(+)

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 211e9da..26f9779 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -1044,6 +1044,11 @@ static inline void pci_ignore_hotplug(struct pci_dev *dev)
 	dev->ignore_hotplug = 1;
 }
 
+static inline void pci_unignore_hotplug(struct pci_dev *dev)
+{
+	dev->ignore_hotplug = 0;
+}
+
 static inline int pci_enable_wake(struct pci_dev *dev, pci_power_t state,
 				  bool enable)
 {

[PATCH 2/2] nouveau/radeon: Un-ignore hotplug on resume

[Alex Williamson]

Commit b440bde74f04 enabled drivers to be able to set an ignore
hotplug flag on the device and used it in the suspend routine for
nouveau and radeon to prevent hotplug when the device is powered off.
However, there was no undo function provided, so once set, the ignore
hotplug flag lives on for the life of the pci_dev.  I think the
intention was only to ignore hotplug across the power event for the
device, so we should be able to clear the flag on resume.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
---
 drivers/gpu/drm/nouveau/nouveau_drm.c |    1 +
 drivers/gpu/drm/radeon/radeon_drv.c   |    1 +
 2 files changed, 2 insertions(+)

diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index 8763deb..b168c3a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -749,6 +749,7 @@ nouveau_pmops_runtime_resume(struct device *dev)
 		return -EINVAL;
 
 	pci_set_power_state(pdev, PCI_D0);
+	pci_unignore_hotplug(pdev);
 	pci_restore_state(pdev);
 	ret = pci_enable_device(pdev);
 	if (ret)
diff --git a/drivers/gpu/drm/radeon/radeon_drv.c b/drivers/gpu/drm/radeon/radeon_drv.c
index 5d684be..1c7c79b 100644
--- a/drivers/gpu/drm/radeon/radeon_drv.c
+++ b/drivers/gpu/drm/radeon/radeon_drv.c
@@ -472,6 +472,7 @@ static int radeon_pmops_runtime_resume(struct device *dev)
 	drm_dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
 
 	pci_set_power_state(pdev, PCI_D0);
+	pci_unignore_hotplug(pdev);
 	pci_restore_state(pdev);
 	ret = pci_enable_device(pdev);
 	if (ret)

Re: [PATCH 0/2] PCI: pci_unignore_hotplug()

[Alex Deucher]

On Tue, Feb 24, 2015 at 3:12 PM, Alex Williamson
<alex.williamson@redhat.com> wrote:
> I'd kinda like to use pci_ignore_hotplug() for devices in use by a
> user via vfio-pci, but the interface seems only partially implemented
> since we can only set ignore_hotplug and not clear it.  This adds the
> capability to clear the flag and applies it as seems to be appropriate
> to the existing users.  I don't have a system with power control for
> a radeon or nouveau device, so I'd certainly appreciate testing.  It
> seems though that once the device is in D0, we've passed the critical
> point where the device could be accidentally unplugged.  Thanks,

Something re-broke this recently.  See:
https://bugzilla.kernel.org/show_bug.cgi?id=61891
https://bugzilla.kernel.org/show_bug.cgi?id=89731

Alex

>
> Alex
>
> ---
>
> Alex Williamson (2):
>       pci: Add reciprocal function for pci_ignore_hotplug()
>       nouveau/radeon: Un-ignore hotplug on resume
>
>
>  drivers/gpu/drm/nouveau/nouveau_drm.c |    1 +
>  drivers/gpu/drm/radeon/radeon_drv.c   |    1 +
>  include/linux/pci.h                   |    5 +++++
>  3 files changed, 7 insertions(+)
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 0/2] PCI: pci_unignore_hotplug()

[Bjorn Helgaas]

On Tue, Feb 24, 2015 at 03:23:27PM -0500, Alex Deucher wrote:
> On Tue, Feb 24, 2015 at 3:12 PM, Alex Williamson
> <alex.williamson@redhat.com> wrote:
> > I'd kinda like to use pci_ignore_hotplug() for devices in use by a
> > user via vfio-pci, but the interface seems only partially implemented
> > since we can only set ignore_hotplug and not clear it.  This adds the
> > capability to clear the flag and applies it as seems to be appropriate
> > to the existing users.  I don't have a system with power control for
> > a radeon or nouveau device, so I'd certainly appreciate testing.  It
> > seems though that once the device is in D0, we've passed the critical
> > point where the device could be accidentally unplugged.  Thanks,
> 
> Something re-broke this recently.  See:
> https://bugzilla.kernel.org/show_bug.cgi?id=61891
> https://bugzilla.kernel.org/show_bug.cgi?id=89731

I'm not opposed to adding pci_unignore_hotplug() when we have users for it.
I don't really want to throw in the nouveau and radeon changes while there
are open issues in the same area, though.

I haven't looked at the bugzillas above.  Let me know if I should.

Bjorn

Re: [PATCH 0/2] PCI: pci_unignore_hotplug()

[Alex Deucher]

On Fri, Mar 6, 2015 at 1:49 PM, Bjorn Helgaas <bhelgaas@google.com> wrote:
> On Tue, Feb 24, 2015 at 03:23:27PM -0500, Alex Deucher wrote:
>> On Tue, Feb 24, 2015 at 3:12 PM, Alex Williamson
>> <alex.williamson@redhat.com> wrote:
>> > I'd kinda like to use pci_ignore_hotplug() for devices in use by a
>> > user via vfio-pci, but the interface seems only partially implemented
>> > since we can only set ignore_hotplug and not clear it.  This adds the
>> > capability to clear the flag and applies it as seems to be appropriate
>> > to the existing users.  I don't have a system with power control for
>> > a radeon or nouveau device, so I'd certainly appreciate testing.  It
>> > seems though that once the device is in D0, we've passed the critical
>> > point where the device could be accidentally unplugged.  Thanks,
>>
>> Something re-broke this recently.  See:
>> https://bugzilla.kernel.org/show_bug.cgi?id=61891
>> https://bugzilla.kernel.org/show_bug.cgi?id=89731
>
> I'm not opposed to adding pci_unignore_hotplug() when we have users for it.
> I don't really want to throw in the nouveau and radeon changes while there
> are open issues in the same area, though.
>
> I haven't looked at the bugzillas above.  Let me know if I should.

Please do.  My comments were not targeted at this patch set, but
rather at the fact that pci hotplug (acpiphp in this case) is again
trying to remove the driver even though it's flagged to ignore hotplug
in the hybrid laptop case.

Alex