[Intel-gfx] [RFC PATCH 1/1] drm/i915: Split off pci_driver.remove() tail to drm_driver.release()

Daniel Vetter daniel at ffwll.ch
Mon Jun 3 07:28:18 UTC 2019


On Thu, May 30, 2019 at 10:40:09AM +0100, Chris Wilson wrote:
> 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 at linux.intel.com>
> > ---
> >  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?

+1 on that rename, load/unload is really terribly confusing and has
horrible semantics in the dri1 shadow attach world ...
-Daniel

> 
> >  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 at chris-wilson.co.uk>
> -Chris

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch


More information about the Intel-gfx mailing list