Quoting Janusz Krzysztofik (2019-05-30 10:24:26)
> In order to support driver hot unbind, some cleanup operations, now
> performed on PCI driver remove, must be called later, after all device
> file descriptors are closed.
>
> Split out those operations from the tail of pci_driver.remove()
> callback and put them into drm_driver.release() which is called as soon
> as all references to the driver are put. As a result, those cleanups
> will be now run on last drm_dev_put(), either still called from
> pci_driver.remove() if all device file descriptors are already closed,
> or on last drm_release() file operation.
>
> Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@xxxxxxxxxxxxxxx>
> ---
> drivers/gpu/drm/i915/i915_drv.c | 17 +++++++++++++----
> drivers/gpu/drm/i915/i915_drv.h | 1 +
> drivers/gpu/drm/i915/i915_gem.c | 10 +++++++++-
> 3 files changed, 23 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
> index 83d2eb9e74cb..8be69f84eb6d 100644
> --- a/drivers/gpu/drm/i915/i915_drv.c
> +++ b/drivers/gpu/drm/i915/i915_drv.c
> @@ -738,6 +738,7 @@ static int i915_load_modeset_init(struct drm_device *dev)
>
> cleanup_gem:
> i915_gem_suspend(dev_priv);
> + i915_gem_fini_hw(dev_priv);
> i915_gem_fini(dev_priv);
> cleanup_modeset:
> intel_modeset_cleanup(dev);
> @@ -1685,7 +1686,6 @@ static void i915_driver_cleanup_hw(struct drm_i915_private *dev_priv)
> pci_disable_msi(pdev);
>
> pm_qos_remove_request(&dev_priv->pm_qos);
> - i915_ggtt_cleanup_hw(dev_priv);
> }
>
> /**
> @@ -1909,6 +1909,7 @@ int i915_driver_load(struct pci_dev *pdev, const struct pci_device_id *ent)
Would it make sense to rename load/unload from the legacy drm stubs over
to match the pci entry points?
> out_cleanup_hw:
> i915_driver_cleanup_hw(dev_priv);
> + i915_ggtt_cleanup_hw(dev_priv);
> out_cleanup_mmio:
> i915_driver_cleanup_mmio(dev_priv);
> out_runtime_pm_put:
> @@ -1960,21 +1961,29 @@ void i915_driver_unload(struct drm_device *dev)
> cancel_delayed_work_sync(&dev_priv->gpu_error.hangcheck_work);
> i915_reset_error_state(dev_priv);
>
> - i915_gem_fini(dev_priv);
> + i915_gem_fini_hw(dev_priv);
>
> intel_power_domains_fini_hw(dev_priv);
>
> i915_driver_cleanup_hw(dev_priv);
> - i915_driver_cleanup_mmio(dev_priv);
>
> enable_rpm_wakeref_asserts(dev_priv);
> - intel_runtime_pm_cleanup(dev_priv);
> }
>
> static void i915_driver_release(struct drm_device *dev)
> {
> struct drm_i915_private *dev_priv = to_i915(dev);
>
> + disable_rpm_wakeref_asserts(dev_priv);
> +
> + i915_gem_fini(dev_priv);
> +
> + i915_ggtt_cleanup_hw(dev_priv);
> + i915_driver_cleanup_mmio(dev_priv);
> +
> + enable_rpm_wakeref_asserts(dev_priv);
> + intel_runtime_pm_cleanup(dev_priv);
We should really propagate the release nomenclature down and replace our
mixed fini/cleanup. Consistency is helpful when trying to work out which
phase the code is in.
> i915_driver_cleanup_early(dev_priv);
> i915_driver_destroy(dev_priv);
> }
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index a2664ea1395b..d08e7bd83544 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -3047,6 +3047,7 @@ void i915_gem_init_mmio(struct drm_i915_private *i915);
> int __must_check i915_gem_init(struct drm_i915_private *dev_priv);
> int __must_check i915_gem_init_hw(struct drm_i915_private *dev_priv);
> void i915_gem_init_swizzling(struct drm_i915_private *dev_priv);
> +void i915_gem_fini_hw(struct drm_i915_private *dev_priv);
> void i915_gem_fini(struct drm_i915_private *dev_priv);
> int i915_gem_wait_for_idle(struct drm_i915_private *dev_priv,
> unsigned int flags, long timeout);
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 7cafd5612f71..c6a8e665a6ba 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -4667,7 +4667,7 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
> return ret;
> }
>
> -void i915_gem_fini(struct drm_i915_private *dev_priv)
> +void i915_gem_fini_hw(struct drm_i915_private *dev_priv)
> {
> GEM_BUG_ON(dev_priv->gt.awake);
>
> @@ -4681,6 +4681,14 @@ void i915_gem_fini(struct drm_i915_private *dev_priv)
> intel_uc_fini_hw(dev_priv);
> intel_uc_fini(dev_priv);
> intel_engines_cleanup(dev_priv);
intel_engines_cleanup -> i915_gem_fini -- that is in principle just
freeing structs. One side effect it does have is to make all engines
unavailable (but it doesn't update the engine_mask so the inconsistency
might catch us out if it is not one of the last cleanup actions).
intel_uc_fini() is a bit of a mixed bag. It looks like it flushes
runtime state, so preferrably that flush should be moved to the
_fini_hw so that _fini is pure cleanup. So for the time being, best to
leave intel_uc_fini() here.
> + mutex_unlock(&dev_priv->drm.struct_mutex);
> +
> + i915_gem_drain_freed_objects(dev_priv);
> +}
> +
> +void i915_gem_fini(struct drm_i915_private *dev_priv)
> +{
> + mutex_lock(&dev_priv->drm.struct_mutex);
> i915_gem_contexts_fini(dev_priv);
> i915_gem_fini_scratch(dev_priv);
> mutex_unlock(&dev_priv->drm.struct_mutex);
That split looks sensible to me, with the consideration as to whether
defer intel_engines_cleanup() as well,
Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
