Re: [Intel-gfx] [PATCH] drm/i915: Disable D3Cold in s2idle and runtime pm

[Rodrigo Vivi <rodrigo.vivi@intel.com>]

On Mon, Nov 15, 2021 at 09:10:54PM +0530, Tilak Tangudu wrote:
> s2idle and runtime pm puts the pci gfx device in D3Hot, ACPI runtime
> monitors the pci tree,if it sees complete tree as D3Hot,it transitions
> the device to D3Cold.But i915 do not have D3Cold support in S2idle or in
> runtime pm. so disabling D3cold in above flows and its FIXME.
> 
> Added pci D3Cold enable/disable in s2idle and runtime suspend/resume
> flows.
> 
> Signed-off-by: Tilak Tangudu <tilak.tangudu@intel.com>

Just for the clear record, I always preferred the unconditional disallow
of d3cold, but it looks some internal experiments for s3/s4 failed with that
and this approach here was the safest one, so let's move with this and
prevent the d3cold for now and then allow the runtime_pm autosuspend
enabled by default everywhere.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

> ---
>  drivers/gpu/drm/i915/i915_drv.c | 19 +++++++++++++++++++
>  1 file changed, 19 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
> index 46bf3315f616..af6868f12ef0 100644
> --- a/drivers/gpu/drm/i915/i915_drv.c
> +++ b/drivers/gpu/drm/i915/i915_drv.c
> @@ -1194,6 +1194,14 @@ static int i915_drm_suspend_late(struct drm_device *dev, bool hibernation)
>  		goto out;
>  	}
>  
> +	/*
> +	 * FIXME: Temporary hammer to avoid freezing the machine on our DGFX
> +	 * This should be totally removed when we handle the pci states properly
> +	 * on runtime PM and on s2idle cases.
> +	 */
> +	if (suspend_to_idle(dev_priv))
> +		pci_d3cold_disable(pdev);
> +
>  	pci_disable_device(pdev);
>  	/*
>  	 * During hibernation on some platforms the BIOS may try to access
> @@ -1357,6 +1365,8 @@ static int i915_drm_resume_early(struct drm_device *dev)
>  
>  	pci_set_master(pdev);
>  
> +	pci_d3cold_enable(pdev);
> +
>  	disable_rpm_wakeref_asserts(&dev_priv->runtime_pm);
>  
>  	ret = vlv_resume_prepare(dev_priv, false);
> @@ -1533,6 +1543,7 @@ static int intel_runtime_suspend(struct device *kdev)
>  {
>  	struct drm_i915_private *dev_priv = kdev_to_i915(kdev);
>  	struct intel_runtime_pm *rpm = &dev_priv->runtime_pm;
> +	struct pci_dev *pdev = to_pci_dev(dev_priv->drm.dev);
>  	int ret;
>  
>  	if (drm_WARN_ON_ONCE(&dev_priv->drm, !HAS_RUNTIME_PM(dev_priv)))
> @@ -1578,6 +1589,12 @@ static int intel_runtime_suspend(struct device *kdev)
>  		drm_err(&dev_priv->drm,
>  			"Unclaimed access detected prior to suspending\n");
>  
> +	/*
> +	 * FIXME: Temporary hammer to avoid freezing the machine on our DGFX
> +	 * This should be totally removed when we handle the pci states properly
> +	 * on runtime PM and on s2idle cases.
> +	 */
> +	pci_d3cold_disable(pdev);
>  	rpm->suspended = true;
>  
>  	/*
> @@ -1616,6 +1633,7 @@ static int intel_runtime_resume(struct device *kdev)
>  {
>  	struct drm_i915_private *dev_priv = kdev_to_i915(kdev);
>  	struct intel_runtime_pm *rpm = &dev_priv->runtime_pm;
> +	struct pci_dev *pdev = to_pci_dev(dev_priv->drm.dev);
>  	int ret;
>  
>  	if (drm_WARN_ON_ONCE(&dev_priv->drm, !HAS_RUNTIME_PM(dev_priv)))
> @@ -1628,6 +1646,7 @@ static int intel_runtime_resume(struct device *kdev)
>  
>  	intel_opregion_notify_adapter(dev_priv, PCI_D0);
>  	rpm->suspended = false;
> +	pci_d3cold_enable(pdev);
>  	if (intel_uncore_unclaimed_mmio(&dev_priv->uncore))
>  		drm_dbg(&dev_priv->drm,
>  			"Unclaimed access during suspend, bios?\n");
> -- 
> 2.25.1
>