| Message ID | 20190614061723.1173-1-daniel.vetter@ffwll.ch (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/kms: Catch mode_object lifetime errors | expand |
On Fri, Jun 14, 2019 at 08:17:23AM +0200, Daniel Vetter wrote: > Only dynamic mode objects, i.e. those which are refcounted and have a free > callback, can be added while the overall drm_device is visible to > userspace. All others must be added before drm_dev_register and > removed after drm_dev_unregister. > > Small issue around drivers still using the load/unload callbacks, we > need to make sure we set dev->registered so that load/unload code in > these callbacks doesn't trigger false warnings. Only a small > adjustement in drm_dev_register was needed. > > Motivated by some irc discussions about object ids of dynamic objects > like blobs become invalid, and me going on a bit an audit spree. > Seems like a very worthwhile change, any idea how many drivers are going to be sad after this change? > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > --- > drivers/gpu/drm/drm_drv.c | 4 ++-- > drivers/gpu/drm/drm_mode_object.c | 4 ++++ > 2 files changed, 6 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c > index cb6f0245de7c..48c84e3e1931 100644 > --- a/drivers/gpu/drm/drm_drv.c > +++ b/drivers/gpu/drm/drm_drv.c > @@ -997,14 +997,14 @@ int drm_dev_register(struct drm_device *dev, unsigned long flags) > if (ret) > goto err_minors; > > - dev->registered = true; > - > if (dev->driver->load) { > ret = dev->driver->load(dev, flags); > if (ret) > goto err_minors; > } > > + dev->registered = true; > + > if (drm_core_check_feature(dev, DRIVER_MODESET)) > drm_modeset_register_all(dev); > > diff --git a/drivers/gpu/drm/drm_mode_object.c b/drivers/gpu/drm/drm_mode_object.c > index 1c6e51135962..c355ba8e6d5d 100644 > --- a/drivers/gpu/drm/drm_mode_object.c > +++ b/drivers/gpu/drm/drm_mode_object.c > @@ -42,6 +42,8 @@ int __drm_mode_object_add(struct drm_device *dev, struct drm_mode_object *obj, > { > int ret; > > + WARN_ON(dev->registered && !obj_free_cb); These should probably have a comment above them giving some guidance to the driver developer. With some comments, this is: Reviewed-by: Sean Paul <sean@poorly.run> > + > mutex_lock(&dev->mode_config.idr_mutex); > ret = idr_alloc(&dev->mode_config.object_idr, register_obj ? obj : NULL, > 1, 0, GFP_KERNEL); > @@ -102,6 +104,8 @@ void drm_mode_object_register(struct drm_device *dev, > void drm_mode_object_unregister(struct drm_device *dev, > struct drm_mode_object *object) > { > + WARN_ON(dev->registered && !object->free_cb); > + > mutex_lock(&dev->mode_config.idr_mutex); > if (object->id) { > idr_remove(&dev->mode_config.object_idr, object->id); > -- > 2.20.1 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
On Fri, Jun 28, 2019 at 7:24 PM Sean Paul <sean@poorly.run> wrote: > > On Fri, Jun 14, 2019 at 08:17:23AM +0200, Daniel Vetter wrote: > > Only dynamic mode objects, i.e. those which are refcounted and have a free > > callback, can be added while the overall drm_device is visible to > > userspace. All others must be added before drm_dev_register and > > removed after drm_dev_unregister. > > > > Small issue around drivers still using the load/unload callbacks, we > > need to make sure we set dev->registered so that load/unload code in > > these callbacks doesn't trigger false warnings. Only a small > > adjustement in drm_dev_register was needed. > > > > Motivated by some irc discussions about object ids of dynamic objects > > like blobs become invalid, and me going on a bit an audit spree. > > > > Seems like a very worthwhile change, any idea how many drivers are going > to be sad after this change? None I think/hope, really just defense WARN_ON just in case. The main ones that would be sad are all the ones that have a ->load callback, but I'm taking care of them. Everyone else does things correctly and calls drm_dev_register last in their probe function (or around where they set up fbdev, which is also register the driver at least to the fbdev world, so really the same). > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > > --- > > drivers/gpu/drm/drm_drv.c | 4 ++-- > > drivers/gpu/drm/drm_mode_object.c | 4 ++++ > > 2 files changed, 6 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c > > index cb6f0245de7c..48c84e3e1931 100644 > > --- a/drivers/gpu/drm/drm_drv.c > > +++ b/drivers/gpu/drm/drm_drv.c > > @@ -997,14 +997,14 @@ int drm_dev_register(struct drm_device *dev, unsigned long flags) > > if (ret) > > goto err_minors; > > > > - dev->registered = true; > > - > > if (dev->driver->load) { > > ret = dev->driver->load(dev, flags); > > if (ret) > > goto err_minors; > > } > > > > + dev->registered = true; > > + > > if (drm_core_check_feature(dev, DRIVER_MODESET)) > > drm_modeset_register_all(dev); > > > > diff --git a/drivers/gpu/drm/drm_mode_object.c b/drivers/gpu/drm/drm_mode_object.c > > index 1c6e51135962..c355ba8e6d5d 100644 > > --- a/drivers/gpu/drm/drm_mode_object.c > > +++ b/drivers/gpu/drm/drm_mode_object.c > > @@ -42,6 +42,8 @@ int __drm_mode_object_add(struct drm_device *dev, struct drm_mode_object *obj, > > { > > int ret; > > > > + WARN_ON(dev->registered && !obj_free_cb); > > These should probably have a comment above them giving some guidance to the > driver developer. > > With some comments, this is: What comment do you expect here? drm_dev_register explains what you should do already, and I expect driver developers to find that one pretty quickly. From there: "This should be done last in the device initialization sequence to make sure userspace can't access an inconsistent state." -Daniel > Reviewed-by: Sean Paul <sean@poorly.run> > > > > + > > mutex_lock(&dev->mode_config.idr_mutex); > > ret = idr_alloc(&dev->mode_config.object_idr, register_obj ? obj : NULL, > > 1, 0, GFP_KERNEL); > > @@ -102,6 +104,8 @@ void drm_mode_object_register(struct drm_device *dev, > > void drm_mode_object_unregister(struct drm_device *dev, > > struct drm_mode_object *object) > > { > > + WARN_ON(dev->registered && !object->free_cb); > > + > > mutex_lock(&dev->mode_config.idr_mutex); > > if (object->id) { > > idr_remove(&dev->mode_config.object_idr, object->id); > > -- > > 2.20.1 > > > > _______________________________________________ > > dri-devel mailing list > > dri-devel@lists.freedesktop.org > > https://lists.freedesktop.org/mailman/listinfo/dri-devel > > -- > Sean Paul, Software Engineer, Google / Chromium OS
On Sat, 2019-06-29 at 17:39 +0200, Daniel Vetter wrote: > On Fri, Jun 28, 2019 at 7:24 PM Sean Paul <sean@poorly.run> wrote: > > On Fri, Jun 14, 2019 at 08:17:23AM +0200, Daniel Vetter wrote: > > > Only dynamic mode objects, i.e. those which are refcounted and > > > have a free > > > callback, can be added while the overall drm_device is visible to > > > userspace. All others must be added before drm_dev_register and > > > removed after drm_dev_unregister. > > > > > > Small issue around drivers still using the load/unload callbacks, > > > we > > > need to make sure we set dev->registered so that load/unload code > > > in > > > these callbacks doesn't trigger false warnings. Only a small > > > adjustement in drm_dev_register was needed. > > > > > > Motivated by some irc discussions about object ids of dynamic > > > objects > > > like blobs become invalid, and me going on a bit an audit spree. > > > > > > > Seems like a very worthwhile change, any idea how many drivers are > > going > > to be sad after this change? > > None I think/hope, really just defense WARN_ON just in case. The main > ones that would be sad are all the ones that have a ->load callback, > but I'm taking care of them. Everyone else does things correctly and > calls drm_dev_register last in their probe function (or around where > they set up fbdev, which is also register the driver at least to the > fbdev world, so really the same). > > > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > > > --- > > > drivers/gpu/drm/drm_drv.c | 4 ++-- > > > drivers/gpu/drm/drm_mode_object.c | 4 ++++ > > > 2 files changed, 6 insertions(+), 2 deletions(-) > > > > > > diff --git a/drivers/gpu/drm/drm_drv.c > > > b/drivers/gpu/drm/drm_drv.c > > > index cb6f0245de7c..48c84e3e1931 100644 > > > --- a/drivers/gpu/drm/drm_drv.c > > > +++ b/drivers/gpu/drm/drm_drv.c > > > @@ -997,14 +997,14 @@ int drm_dev_register(struct drm_device > > > *dev, unsigned long flags) > > > if (ret) > > > goto err_minors; > > > > > > - dev->registered = true; > > > - > > > if (dev->driver->load) { > > > ret = dev->driver->load(dev, flags); > > > if (ret) > > > goto err_minors; > > > } > > > > > > + dev->registered = true; > > > + > > > if (drm_core_check_feature(dev, DRIVER_MODESET)) > > > drm_modeset_register_all(dev); > > > > > > diff --git a/drivers/gpu/drm/drm_mode_object.c > > > b/drivers/gpu/drm/drm_mode_object.c > > > index 1c6e51135962..c355ba8e6d5d 100644 > > > --- a/drivers/gpu/drm/drm_mode_object.c > > > +++ b/drivers/gpu/drm/drm_mode_object.c > > > @@ -42,6 +42,8 @@ int __drm_mode_object_add(struct drm_device > > > *dev, struct drm_mode_object *obj, > > > { > > > int ret; > > > > > > + WARN_ON(dev->registered && !obj_free_cb); Getting warnings on i915 with MST, we add fake MST connectors when a sink with MST support is connected; intel_dp_add_mst_connector()->drm_connector_attach_max_bpc_property() Not sure how to fix that, add a global i915 device property like we do for "audio" and "Broadcast RGB" don't look the right approach here. Any tips? We definitely need a platform with a MST sink on our CI. > > > > These should probably have a comment above them giving some > > guidance to the > > driver developer. > > > > With some comments, this is: > > What comment do you expect here? drm_dev_register explains what you > should do already, and I expect driver developers to find that one > pretty quickly. From there: "This should be done last in the device > initialization sequence to make sure userspace can't access an > inconsistent state." > -Daniel > > > Reviewed-by: Sean Paul <sean@poorly.run> > > > > > > > + > > > mutex_lock(&dev->mode_config.idr_mutex); > > > ret = idr_alloc(&dev->mode_config.object_idr, register_obj > > > ? obj : NULL, > > > 1, 0, GFP_KERNEL); > > > @@ -102,6 +104,8 @@ void drm_mode_object_register(struct > > > drm_device *dev, > > > void drm_mode_object_unregister(struct drm_device *dev, > > > struct drm_mode_object *object) > > > { > > > + WARN_ON(dev->registered && !object->free_cb); > > > + > > > mutex_lock(&dev->mode_config.idr_mutex); > > > if (object->id) { > > > idr_remove(&dev->mode_config.object_idr, object- > > > >id); > > > -- > > > 2.20.1 > > > > > > _______________________________________________ > > > dri-devel mailing list > > > dri-devel@lists.freedesktop.org > > > https://lists.freedesktop.org/mailman/listinfo/dri-devel > > > > -- > > Sean Paul, Software Engineer, Google / Chromium OS > >
On Sat, Aug 17, 2019 at 12:42 AM Souza, Jose <jose.souza@intel.com> wrote: > On Sat, 2019-06-29 at 17:39 +0200, Daniel Vetter wrote: > > On Fri, Jun 28, 2019 at 7:24 PM Sean Paul <sean@poorly.run> wrote: > > > On Fri, Jun 14, 2019 at 08:17:23AM +0200, Daniel Vetter wrote: > > > > Only dynamic mode objects, i.e. those which are refcounted and > > > > have a free > > > > callback, can be added while the overall drm_device is visible to > > > > userspace. All others must be added before drm_dev_register and > > > > removed after drm_dev_unregister. > > > > > > > > Small issue around drivers still using the load/unload callbacks, > > > > we > > > > need to make sure we set dev->registered so that load/unload code > > > > in > > > > these callbacks doesn't trigger false warnings. Only a small > > > > adjustement in drm_dev_register was needed. > > > > > > > > Motivated by some irc discussions about object ids of dynamic > > > > objects > > > > like blobs become invalid, and me going on a bit an audit spree. > > > > > > > > > > Seems like a very worthwhile change, any idea how many drivers are > > > going > > > to be sad after this change? > > > > None I think/hope, really just defense WARN_ON just in case. The main > > ones that would be sad are all the ones that have a ->load callback, > > but I'm taking care of them. Everyone else does things correctly and > > calls drm_dev_register last in their probe function (or around where > > they set up fbdev, which is also register the driver at least to the > > fbdev world, so really the same). > > > > > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > > > > --- > > > > drivers/gpu/drm/drm_drv.c | 4 ++-- > > > > drivers/gpu/drm/drm_mode_object.c | 4 ++++ > > > > 2 files changed, 6 insertions(+), 2 deletions(-) > > > > > > > > diff --git a/drivers/gpu/drm/drm_drv.c > > > > b/drivers/gpu/drm/drm_drv.c > > > > index cb6f0245de7c..48c84e3e1931 100644 > > > > --- a/drivers/gpu/drm/drm_drv.c > > > > +++ b/drivers/gpu/drm/drm_drv.c > > > > @@ -997,14 +997,14 @@ int drm_dev_register(struct drm_device > > > > *dev, unsigned long flags) > > > > if (ret) > > > > goto err_minors; > > > > > > > > - dev->registered = true; > > > > - > > > > if (dev->driver->load) { > > > > ret = dev->driver->load(dev, flags); > > > > if (ret) > > > > goto err_minors; > > > > } > > > > > > > > + dev->registered = true; > > > > + > > > > if (drm_core_check_feature(dev, DRIVER_MODESET)) > > > > drm_modeset_register_all(dev); > > > > > > > > diff --git a/drivers/gpu/drm/drm_mode_object.c > > > > b/drivers/gpu/drm/drm_mode_object.c > > > > index 1c6e51135962..c355ba8e6d5d 100644 > > > > --- a/drivers/gpu/drm/drm_mode_object.c > > > > +++ b/drivers/gpu/drm/drm_mode_object.c > > > > @@ -42,6 +42,8 @@ int __drm_mode_object_add(struct drm_device > > > > *dev, struct drm_mode_object *obj, > > > > { > > > > int ret; > > > > > > > > + WARN_ON(dev->registered && !obj_free_cb); > > Getting warnings on i915 with MST, we add fake MST connectors when a > sink with MST support is connected; > > intel_dp_add_mst_connector()->drm_connector_attach_max_bpc_property() > > Not sure how to fix that, add a global i915 device property like we do > for "audio" and "Broadcast RGB" don't look the right approach here. > Any tips? > > We definitely need a platform with a MST sink on our CI. Uh yeah this is bad. I guess we need to preemptively create all possible bpc properties, so that we can only do an attach. Those really can't be hotplugged. We might also want to revert commit 5ca0ef8a56b8c4812ed78ef9ca53052191dab6e7 Author: Ville Syrjälä <ville.syrjala@linux.intel.com> Date: Tue Mar 26 16:25:54 2019 +0200 drm/i915: Add max_bpc property for DP MST as an interim solution. Adding Ville and Jani. -Daniel > > > > > > These should probably have a comment above them giving some > > > guidance to the > > > driver developer. > > > > > > With some comments, this is: > > > > What comment do you expect here? drm_dev_register explains what you > > should do already, and I expect driver developers to find that one > > pretty quickly. From there: "This should be done last in the device > > initialization sequence to make sure userspace can't access an > > inconsistent state." > > -Daniel > > > > > Reviewed-by: Sean Paul <sean@poorly.run> > > > > > > > > > > + > > > > mutex_lock(&dev->mode_config.idr_mutex); > > > > ret = idr_alloc(&dev->mode_config.object_idr, register_obj > > > > ? obj : NULL, > > > > 1, 0, GFP_KERNEL); > > > > @@ -102,6 +104,8 @@ void drm_mode_object_register(struct > > > > drm_device *dev, > > > > void drm_mode_object_unregister(struct drm_device *dev, > > > > struct drm_mode_object *object) > > > > { > > > > + WARN_ON(dev->registered && !object->free_cb); > > > > + > > > > mutex_lock(&dev->mode_config.idr_mutex); > > > > if (object->id) { > > > > idr_remove(&dev->mode_config.object_idr, object- > > > > >id); > > > > -- > > > > 2.20.1 > > > > > > > > _______________________________________________ > > > > dri-devel mailing list > > > > dri-devel@lists.freedesktop.org > > > > https://lists.freedesktop.org/mailman/listinfo/dri-devel > > > > > > -- > > > Sean Paul, Software Engineer, Google / Chromium OS > > > >
On Tue, 20 Aug 2019, Daniel Vetter <daniel.vetter@ffwll.ch> wrote: > On Sat, Aug 17, 2019 at 12:42 AM Souza, Jose <jose.souza@intel.com> wrote: >> On Sat, 2019-06-29 at 17:39 +0200, Daniel Vetter wrote: >> > On Fri, Jun 28, 2019 at 7:24 PM Sean Paul <sean@poorly.run> wrote: >> > > On Fri, Jun 14, 2019 at 08:17:23AM +0200, Daniel Vetter wrote: >> > > > Only dynamic mode objects, i.e. those which are refcounted and >> > > > have a free >> > > > callback, can be added while the overall drm_device is visible to >> > > > userspace. All others must be added before drm_dev_register and >> > > > removed after drm_dev_unregister. >> > > > >> > > > Small issue around drivers still using the load/unload callbacks, >> > > > we >> > > > need to make sure we set dev->registered so that load/unload code >> > > > in >> > > > these callbacks doesn't trigger false warnings. Only a small >> > > > adjustement in drm_dev_register was needed. >> > > > >> > > > Motivated by some irc discussions about object ids of dynamic >> > > > objects >> > > > like blobs become invalid, and me going on a bit an audit spree. >> > > > >> > > >> > > Seems like a very worthwhile change, any idea how many drivers are >> > > going >> > > to be sad after this change? >> > >> > None I think/hope, really just defense WARN_ON just in case. The main >> > ones that would be sad are all the ones that have a ->load callback, >> > but I'm taking care of them. Everyone else does things correctly and >> > calls drm_dev_register last in their probe function (or around where >> > they set up fbdev, which is also register the driver at least to the >> > fbdev world, so really the same). >> > >> > > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> >> > > > --- >> > > > drivers/gpu/drm/drm_drv.c | 4 ++-- >> > > > drivers/gpu/drm/drm_mode_object.c | 4 ++++ >> > > > 2 files changed, 6 insertions(+), 2 deletions(-) >> > > > >> > > > diff --git a/drivers/gpu/drm/drm_drv.c >> > > > b/drivers/gpu/drm/drm_drv.c >> > > > index cb6f0245de7c..48c84e3e1931 100644 >> > > > --- a/drivers/gpu/drm/drm_drv.c >> > > > +++ b/drivers/gpu/drm/drm_drv.c >> > > > @@ -997,14 +997,14 @@ int drm_dev_register(struct drm_device >> > > > *dev, unsigned long flags) >> > > > if (ret) >> > > > goto err_minors; >> > > > >> > > > - dev->registered = true; >> > > > - >> > > > if (dev->driver->load) { >> > > > ret = dev->driver->load(dev, flags); >> > > > if (ret) >> > > > goto err_minors; >> > > > } >> > > > >> > > > + dev->registered = true; >> > > > + >> > > > if (drm_core_check_feature(dev, DRIVER_MODESET)) >> > > > drm_modeset_register_all(dev); >> > > > >> > > > diff --git a/drivers/gpu/drm/drm_mode_object.c >> > > > b/drivers/gpu/drm/drm_mode_object.c >> > > > index 1c6e51135962..c355ba8e6d5d 100644 >> > > > --- a/drivers/gpu/drm/drm_mode_object.c >> > > > +++ b/drivers/gpu/drm/drm_mode_object.c >> > > > @@ -42,6 +42,8 @@ int __drm_mode_object_add(struct drm_device >> > > > *dev, struct drm_mode_object *obj, >> > > > { >> > > > int ret; >> > > > >> > > > + WARN_ON(dev->registered && !obj_free_cb); >> >> Getting warnings on i915 with MST, we add fake MST connectors when a >> sink with MST support is connected; >> >> intel_dp_add_mst_connector()->drm_connector_attach_max_bpc_property() >> >> Not sure how to fix that, add a global i915 device property like we do >> for "audio" and "Broadcast RGB" don't look the right approach here. >> Any tips? >> >> We definitely need a platform with a MST sink on our CI. > > Uh yeah this is bad. I guess we need to preemptively create all > possible bpc properties, so that we can only do an attach. Those > really can't be hotplugged. Should we plan to make all objects dynamic and refcounted in the long run? It's kind of silly that we've made connectors dynamic yet can't dynamically allocate properties for them. (Except blob properties which are special dynamic snowflakes.) > We might also want to revert > > commit 5ca0ef8a56b8c4812ed78ef9ca53052191dab6e7 > Author: Ville Syrjälä <ville.syrjala@linux.intel.com> > Date: Tue Mar 26 16:25:54 2019 +0200 > > drm/i915: Add max_bpc property for DP MST > > as an interim solution. Adding Ville and Jani. I think we'll want the fixed, pre-created DP MST max BPC property, shoved in dev_priv, as an interim solution before we have everything dynamic. This shouldn't really be more than 50-100 lines, especially given that we'll only need the 6-12 BPC range for MST connectors, and the rest can use the usual methods. Whether we need the revert as an interim solution before the interim solution, I'll leave it up to Ville. How quick do you think you'd get to fixing this? Other than that, ugh. Just ugh. BR, Jani. > -Daniel > >> > > >> > > These should probably have a comment above them giving some >> > > guidance to the >> > > driver developer. >> > > >> > > With some comments, this is: >> > >> > What comment do you expect here? drm_dev_register explains what you >> > should do already, and I expect driver developers to find that one >> > pretty quickly. From there: "This should be done last in the device >> > initialization sequence to make sure userspace can't access an >> > inconsistent state." >> > -Daniel >> > >> > > Reviewed-by: Sean Paul <sean@poorly.run> >> > > >> > > >> > > > + >> > > > mutex_lock(&dev->mode_config.idr_mutex); >> > > > ret = idr_alloc(&dev->mode_config.object_idr, register_obj >> > > > ? obj : NULL, >> > > > 1, 0, GFP_KERNEL); >> > > > @@ -102,6 +104,8 @@ void drm_mode_object_register(struct >> > > > drm_device *dev, >> > > > void drm_mode_object_unregister(struct drm_device *dev, >> > > > struct drm_mode_object *object) >> > > > { >> > > > + WARN_ON(dev->registered && !object->free_cb); >> > > > + >> > > > mutex_lock(&dev->mode_config.idr_mutex); >> > > > if (object->id) { >> > > > idr_remove(&dev->mode_config.object_idr, object- >> > > > >id); >> > > > -- >> > > > 2.20.1 >> > > > >> > > > _______________________________________________ >> > > > dri-devel mailing list >> > > > dri-devel@lists.freedesktop.org >> > > > https://lists.freedesktop.org/mailman/listinfo/dri-devel >> > > >> > > -- >> > > Sean Paul, Software Engineer, Google / Chromium OS >> > >> >
On Tue, Aug 20, 2019 at 11:28 AM Jani Nikula <jani.nikula@intel.com> wrote: > > On Tue, 20 Aug 2019, Daniel Vetter <daniel.vetter@ffwll.ch> wrote: > > On Sat, Aug 17, 2019 at 12:42 AM Souza, Jose <jose.souza@intel.com> wrote: > >> On Sat, 2019-06-29 at 17:39 +0200, Daniel Vetter wrote: > >> > On Fri, Jun 28, 2019 at 7:24 PM Sean Paul <sean@poorly.run> wrote: > >> > > On Fri, Jun 14, 2019 at 08:17:23AM +0200, Daniel Vetter wrote: > >> > > > Only dynamic mode objects, i.e. those which are refcounted and > >> > > > have a free > >> > > > callback, can be added while the overall drm_device is visible to > >> > > > userspace. All others must be added before drm_dev_register and > >> > > > removed after drm_dev_unregister. > >> > > > > >> > > > Small issue around drivers still using the load/unload callbacks, > >> > > > we > >> > > > need to make sure we set dev->registered so that load/unload code > >> > > > in > >> > > > these callbacks doesn't trigger false warnings. Only a small > >> > > > adjustement in drm_dev_register was needed. > >> > > > > >> > > > Motivated by some irc discussions about object ids of dynamic > >> > > > objects > >> > > > like blobs become invalid, and me going on a bit an audit spree. > >> > > > > >> > > > >> > > Seems like a very worthwhile change, any idea how many drivers are > >> > > going > >> > > to be sad after this change? > >> > > >> > None I think/hope, really just defense WARN_ON just in case. The main > >> > ones that would be sad are all the ones that have a ->load callback, > >> > but I'm taking care of them. Everyone else does things correctly and > >> > calls drm_dev_register last in their probe function (or around where > >> > they set up fbdev, which is also register the driver at least to the > >> > fbdev world, so really the same). > >> > > >> > > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > >> > > > --- > >> > > > drivers/gpu/drm/drm_drv.c | 4 ++-- > >> > > > drivers/gpu/drm/drm_mode_object.c | 4 ++++ > >> > > > 2 files changed, 6 insertions(+), 2 deletions(-) > >> > > > > >> > > > diff --git a/drivers/gpu/drm/drm_drv.c > >> > > > b/drivers/gpu/drm/drm_drv.c > >> > > > index cb6f0245de7c..48c84e3e1931 100644 > >> > > > --- a/drivers/gpu/drm/drm_drv.c > >> > > > +++ b/drivers/gpu/drm/drm_drv.c > >> > > > @@ -997,14 +997,14 @@ int drm_dev_register(struct drm_device > >> > > > *dev, unsigned long flags) > >> > > > if (ret) > >> > > > goto err_minors; > >> > > > > >> > > > - dev->registered = true; > >> > > > - > >> > > > if (dev->driver->load) { > >> > > > ret = dev->driver->load(dev, flags); > >> > > > if (ret) > >> > > > goto err_minors; > >> > > > } > >> > > > > >> > > > + dev->registered = true; > >> > > > + > >> > > > if (drm_core_check_feature(dev, DRIVER_MODESET)) > >> > > > drm_modeset_register_all(dev); > >> > > > > >> > > > diff --git a/drivers/gpu/drm/drm_mode_object.c > >> > > > b/drivers/gpu/drm/drm_mode_object.c > >> > > > index 1c6e51135962..c355ba8e6d5d 100644 > >> > > > --- a/drivers/gpu/drm/drm_mode_object.c > >> > > > +++ b/drivers/gpu/drm/drm_mode_object.c > >> > > > @@ -42,6 +42,8 @@ int __drm_mode_object_add(struct drm_device > >> > > > *dev, struct drm_mode_object *obj, > >> > > > { > >> > > > int ret; > >> > > > > >> > > > + WARN_ON(dev->registered && !obj_free_cb); > >> > >> Getting warnings on i915 with MST, we add fake MST connectors when a > >> sink with MST support is connected; > >> > >> intel_dp_add_mst_connector()->drm_connector_attach_max_bpc_property() > >> > >> Not sure how to fix that, add a global i915 device property like we do > >> for "audio" and "Broadcast RGB" don't look the right approach here. > >> Any tips? > >> > >> We definitely need a platform with a MST sink on our CI. > > > > Uh yeah this is bad. I guess we need to preemptively create all > > possible bpc properties, so that we can only do an attach. Those > > really can't be hotplugged. > > Should we plan to make all objects dynamic and refcounted in the long > run? It's kind of silly that we've made connectors dynamic yet can't > dynamically allocate properties for them. (Except blob properties which > are special dynamic snowflakes.) Need to make sure userspace is ok with that. Plus given that it took us years to make drm_connector refcounting not be a pure fireworks show (and Lyude is still busy) I'm not too enthusiastic about this. Also even with the objects which have been dynamic since forever (edid blobs) userspace mixed things up since we reuse ids. It's a mess. > > We might also want to revert > > > > commit 5ca0ef8a56b8c4812ed78ef9ca53052191dab6e7 > > Author: Ville Syrjälä <ville.syrjala@linux.intel.com> > > Date: Tue Mar 26 16:25:54 2019 +0200 > > > > drm/i915: Add max_bpc property for DP MST > > > > as an interim solution. Adding Ville and Jani. > > I think we'll want the fixed, pre-created DP MST max BPC property, > shoved in dev_priv, as an interim solution before we have everything > dynamic. This shouldn't really be more than 50-100 lines, especially > given that we'll only need the 6-12 BPC range for MST connectors, and > the rest can use the usual methods. > > Whether we need the revert as an interim solution before the interim > solution, I'll leave it up to Ville. How quick do you think you'd get to > fixing this? > > Other than that, ugh. Just ugh. Agreed on that :-/ -Daniel > > BR, > Jani. > > > > -Daniel > > > >> > > > >> > > These should probably have a comment above them giving some > >> > > guidance to the > >> > > driver developer. > >> > > > >> > > With some comments, this is: > >> > > >> > What comment do you expect here? drm_dev_register explains what you > >> > should do already, and I expect driver developers to find that one > >> > pretty quickly. From there: "This should be done last in the device > >> > initialization sequence to make sure userspace can't access an > >> > inconsistent state." > >> > -Daniel > >> > > >> > > Reviewed-by: Sean Paul <sean@poorly.run> > >> > > > >> > > > >> > > > + > >> > > > mutex_lock(&dev->mode_config.idr_mutex); > >> > > > ret = idr_alloc(&dev->mode_config.object_idr, register_obj > >> > > > ? obj : NULL, > >> > > > 1, 0, GFP_KERNEL); > >> > > > @@ -102,6 +104,8 @@ void drm_mode_object_register(struct > >> > > > drm_device *dev, > >> > > > void drm_mode_object_unregister(struct drm_device *dev, > >> > > > struct drm_mode_object *object) > >> > > > { > >> > > > + WARN_ON(dev->registered && !object->free_cb); > >> > > > + > >> > > > mutex_lock(&dev->mode_config.idr_mutex); > >> > > > if (object->id) { > >> > > > idr_remove(&dev->mode_config.object_idr, object- > >> > > > >id); > >> > > > -- > >> > > > 2.20.1 > >> > > > > >> > > > _______________________________________________ > >> > > > dri-devel mailing list > >> > > > dri-devel@lists.freedesktop.org > >> > > > https://lists.freedesktop.org/mailman/listinfo/dri-devel > >> > > > >> > > -- > >> > > Sean Paul, Software Engineer, Google / Chromium OS > >> > > >> > > > -- > Jani Nikula, Intel Open Source Graphics Center
diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c index cb6f0245de7c..48c84e3e1931 100644 --- a/drivers/gpu/drm/drm_drv.c +++ b/drivers/gpu/drm/drm_drv.c @@ -997,14 +997,14 @@ int drm_dev_register(struct drm_device *dev, unsigned long flags) if (ret) goto err_minors; - dev->registered = true; - if (dev->driver->load) { ret = dev->driver->load(dev, flags); if (ret) goto err_minors; } + dev->registered = true; + if (drm_core_check_feature(dev, DRIVER_MODESET)) drm_modeset_register_all(dev); diff --git a/drivers/gpu/drm/drm_mode_object.c b/drivers/gpu/drm/drm_mode_object.c index 1c6e51135962..c355ba8e6d5d 100644 --- a/drivers/gpu/drm/drm_mode_object.c +++ b/drivers/gpu/drm/drm_mode_object.c @@ -42,6 +42,8 @@ int __drm_mode_object_add(struct drm_device *dev, struct drm_mode_object *obj, { int ret; + WARN_ON(dev->registered && !obj_free_cb); + mutex_lock(&dev->mode_config.idr_mutex); ret = idr_alloc(&dev->mode_config.object_idr, register_obj ? obj : NULL, 1, 0, GFP_KERNEL); @@ -102,6 +104,8 @@ void drm_mode_object_register(struct drm_device *dev, void drm_mode_object_unregister(struct drm_device *dev, struct drm_mode_object *object) { + WARN_ON(dev->registered && !object->free_cb); + mutex_lock(&dev->mode_config.idr_mutex); if (object->id) { idr_remove(&dev->mode_config.object_idr, object->id);
Only dynamic mode objects, i.e. those which are refcounted and have a free callback, can be added while the overall drm_device is visible to userspace. All others must be added before drm_dev_register and removed after drm_dev_unregister. Small issue around drivers still using the load/unload callbacks, we need to make sure we set dev->registered so that load/unload code in these callbacks doesn't trigger false warnings. Only a small adjustement in drm_dev_register was needed. Motivated by some irc discussions about object ids of dynamic objects like blobs become invalid, and me going on a bit an audit spree. Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> --- drivers/gpu/drm/drm_drv.c | 4 ++-- drivers/gpu/drm/drm_mode_object.c | 4 ++++ 2 files changed, 6 insertions(+), 2 deletions(-)