So I've been adding runtime pm to nouveau/radeon, and on X start it does a
lot of pci accesses. Now because the pm on these devices is equivalent
to D3cold, we have to resume them which involves a heavy latency due to
POSTing the cards. The driver configures the autosuspend timeout to 5s for
this reason, and I think the PCI layer config accesses should respect
the autosuspend.
Cc: Huang Ying <[email protected]>
Cc: Bjorn Helgaas <[email protected]>
Cc: Rafael J. Wysocki <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
---
drivers/pci/pci-sysfs.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
index 02d107b..12d3d52 100644
--- a/drivers/pci/pci-sysfs.c
+++ b/drivers/pci/pci-sysfs.c
@@ -487,7 +487,7 @@ pci_config_pm_runtime_put(struct pci_dev *pdev)
struct device *dev = &pdev->dev;
struct device *parent = dev->parent;
- pm_runtime_put(dev);
+ pm_runtime_put_autosuspend(dev);
if (parent)
pm_runtime_put_sync(parent);
}
--
1.7.12.1
On Wed, 2012-11-07 at 15:30 +1000, Dave Airlie wrote:
> So I've been adding runtime pm to nouveau/radeon, and on X start it does a
> lot of pci accesses. Now because the pm on these devices is equivalent
> to D3cold, we have to resume them which involves a heavy latency due to
> POSTing the cards. The driver configures the autosuspend timeout to 5s for
> this reason, and I think the PCI layer config accesses should respect
> the autosuspend.
>
> Cc: Huang Ying <[email protected]>
> Cc: Bjorn Helgaas <[email protected]>
> Cc: Rafael J. Wysocki <[email protected]>
> Signed-off-by: Dave Airlie <[email protected]>
> ---
> drivers/pci/pci-sysfs.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> index 02d107b..12d3d52 100644
> --- a/drivers/pci/pci-sysfs.c
> +++ b/drivers/pci/pci-sysfs.c
> @@ -487,7 +487,7 @@ pci_config_pm_runtime_put(struct pci_dev *pdev)
> struct device *dev = &pdev->dev;
> struct device *parent = dev->parent;
>
> - pm_runtime_put(dev);
> + pm_runtime_put_autosuspend(dev);
> if (parent)
> pm_runtime_put_sync(parent);
> }
I think you do not need that. You can implement timeout
in .runtime_idle callback of the driver.
Best Regards,
Huang Ying
> >
> > Cc: Huang Ying <[email protected]>
> > Cc: Bjorn Helgaas <[email protected]>
> > Cc: Rafael J. Wysocki <[email protected]>
> > Signed-off-by: Dave Airlie <[email protected]>
> > ---
> > drivers/pci/pci-sysfs.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> > index 02d107b..12d3d52 100644
> > --- a/drivers/pci/pci-sysfs.c
> > +++ b/drivers/pci/pci-sysfs.c
> > @@ -487,7 +487,7 @@ pci_config_pm_runtime_put(struct pci_dev *pdev)
> > struct device *dev = &pdev->dev;
> > struct device *parent = dev->parent;
> >
> > - pm_runtime_put(dev);
> > + pm_runtime_put_autosuspend(dev);
> > if (parent)
> > pm_runtime_put_sync(parent);
> > }
>
> I think you do not need that. You can implement timeout
> in .runtime_idle callback of the driver.
If I understand what you are suggesting, I should setup some kinda of timer callback to later call suspend, but that seems pointless for me if we have the autosuspend mechanism in place.
Won't I end up racing my timer against other pm stuff? I'm not really runtime pm expert so maybe I'm just missing something.
Dave.
On Wed, 2012-11-07 at 01:15 -0500, David Airlie wrote:
> > >
> > > Cc: Huang Ying <[email protected]>
> > > Cc: Bjorn Helgaas <[email protected]>
> > > Cc: Rafael J. Wysocki <[email protected]>
> > > Signed-off-by: Dave Airlie <[email protected]>
> > > ---
> > > drivers/pci/pci-sysfs.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> > > index 02d107b..12d3d52 100644
> > > --- a/drivers/pci/pci-sysfs.c
> > > +++ b/drivers/pci/pci-sysfs.c
> > > @@ -487,7 +487,7 @@ pci_config_pm_runtime_put(struct pci_dev *pdev)
> > > struct device *dev = &pdev->dev;
> > > struct device *parent = dev->parent;
> > >
> > > - pm_runtime_put(dev);
> > > + pm_runtime_put_autosuspend(dev);
> > > if (parent)
> > > pm_runtime_put_sync(parent);
> > > }
> >
> > I think you do not need that. You can implement timeout
> > in .runtime_idle callback of the driver.
>
> If I understand what you are suggesting, I should setup some kinda of timer callback to later call suspend, but that seems pointless for me if we have the autosuspend mechanism in place.
>
> Won't I end up racing my timer against other pm stuff? I'm not really runtime pm expert so maybe I'm just missing something.
You can call pm_runtime_autosuspend or pm_runtime_schedule_suspend
in .runtime_idle callback of the driver.
Best Regards,
Huang Ying
----- Original Message -----
> From: "Huang Ying" <[email protected]>
> To: "David Airlie" <[email protected]>
> Cc: [email protected], [email protected], "Bjorn Helgaas" <[email protected]>, "Rafael J.
> Wysocki" <[email protected]>
> Sent: Wednesday, 7 November, 2012 4:26:25 PM
> Subject: Re: [PATCH] pci/runtime-pm: respect devices autosuspend timeout on config access
>
> On Wed, 2012-11-07 at 01:15 -0500, David Airlie wrote:
> > > >
> > > > Cc: Huang Ying <[email protected]>
> > > > Cc: Bjorn Helgaas <[email protected]>
> > > > Cc: Rafael J. Wysocki <[email protected]>
> > > > Signed-off-by: Dave Airlie <[email protected]>
> > > > ---
> > > > drivers/pci/pci-sysfs.c | 2 +-
> > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> > > > index 02d107b..12d3d52 100644
> > > > --- a/drivers/pci/pci-sysfs.c
> > > > +++ b/drivers/pci/pci-sysfs.c
> > > > @@ -487,7 +487,7 @@ pci_config_pm_runtime_put(struct pci_dev
> > > > *pdev)
> > > > struct device *dev = &pdev->dev;
> > > > struct device *parent = dev->parent;
> > > >
> > > > - pm_runtime_put(dev);
> > > > + pm_runtime_put_autosuspend(dev);
> > > > if (parent)
> > > > pm_runtime_put_sync(parent);
> > > > }
> > >
> > > I think you do not need that. You can implement timeout
> > > in .runtime_idle callback of the driver.
> >
> > If I understand what you are suggesting, I should setup some kinda
> > of timer callback to later call suspend, but that seems pointless
> > for me if we have the autosuspend mechanism in place.
> >
> > Won't I end up racing my timer against other pm stuff? I'm not
> > really runtime pm expert so maybe I'm just missing something.
>
> You can call pm_runtime_autosuspend or pm_runtime_schedule_suspend
> in .runtime_idle callback of the driver.
Ah that explains what I was probably missing, I'll go play with that for a while then!
Thanks,
Dave.