Message ID | 20240927204616.697467-1-lyude@redhat.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | drm/atomic_helper: Add missing NULL check for drm_plane_helper_funcs.atomic_update | expand |
Am 27.09.24 um 22:46 schrieb Lyude Paul: > Something I discovered while writing rvkms since some versions of the > driver didn't have a filled out atomic_update function - we mention that > this callback is "optional", but we don't actually check whether it's NULL > or not before calling it. As a result, we'll segfault if it's not filled > in. > > rvkms rvkms.0: [drm:drm_atomic_helper_commit_modeset_disables] modeset on [ENCODER:36:Virtual-36] > BUG: kernel NULL pointer dereference, address: 0000000000000000 > PGD 0 P4D 0 > Oops: Oops: 0010 [#1] PREEMPT SMP NOPTI > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS edk2-20240813-1.fc40 08/13/2024 > RIP: 0010:0x0 > > So, let's fix that. > > Signed-off-by: Lyude Paul <lyude@redhat.com> > Fixes: c2fcd274bce5 ("drm: Add atomic/plane helpers") > Cc: dri-devel@lists.freedesktop.org > Cc: <stable@vger.kernel.org> # v3.19+ Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de> > --- > drivers/gpu/drm/drm_atomic_helper.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c > index 43cdf39019a44..b3c507040c6d6 100644 > --- a/drivers/gpu/drm/drm_atomic_helper.c > +++ b/drivers/gpu/drm/drm_atomic_helper.c > @@ -2797,7 +2797,8 @@ void drm_atomic_helper_commit_planes(struct drm_device *dev, > > funcs->atomic_disable(plane, old_state); > } else if (new_plane_state->crtc || disabling) { > - funcs->atomic_update(plane, old_state); > + if (funcs->atomic_update) > + funcs->atomic_update(plane, old_state); > > if (!disabling && funcs->atomic_enable) { > if (drm_atomic_plane_enabling(old_plane_state, new_plane_state)) > @@ -2889,7 +2890,8 @@ drm_atomic_helper_commit_planes_on_crtc(struct drm_crtc_state *old_crtc_state) > if (disabling && plane_funcs->atomic_disable) { > plane_funcs->atomic_disable(plane, old_state); > } else if (new_plane_state->crtc || disabling) { > - plane_funcs->atomic_update(plane, old_state); > + if (plane_funcs->atomic_update) > + plane_funcs->atomic_update(plane, old_state); > > if (!disabling && plane_funcs->atomic_enable) { > if (drm_atomic_plane_enabling(old_plane_state, new_plane_state)) > > base-commit: 22512c3ee0f47faab5def71c4453638923c62522
Hi, On Fri, Sep 27, 2024 at 04:46:16PM GMT, Lyude Paul wrote: > Something I discovered while writing rvkms since some versions of the > driver didn't have a filled out atomic_update function - we mention that > this callback is "optional", but we don't actually check whether it's NULL > or not before calling it. As a result, we'll segfault if it's not filled > in. > > rvkms rvkms.0: [drm:drm_atomic_helper_commit_modeset_disables] modeset on [ENCODER:36:Virtual-36] > BUG: kernel NULL pointer dereference, address: 0000000000000000 > PGD 0 P4D 0 > Oops: Oops: 0010 [#1] PREEMPT SMP NOPTI > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS edk2-20240813-1.fc40 08/13/2024 > RIP: 0010:0x0 > > So, let's fix that. > > Signed-off-by: Lyude Paul <lyude@redhat.com> > Fixes: c2fcd274bce5 ("drm: Add atomic/plane helpers") > Cc: dri-devel@lists.freedesktop.org > Cc: <stable@vger.kernel.org> # v3.19+ So we had kind of a similar argument with drm_connector_init early this year, but I do agree we shouldn't fault if we're missing a callback. I do wonder how we can implement a plane without atomic_update though? Do we have drivers in such a case? If not, a better solution would be to make it mandatory and check it when registering. Maxime
Hi Am 30.09.24 um 09:01 schrieb Maxime Ripard: > Hi, > > On Fri, Sep 27, 2024 at 04:46:16PM GMT, Lyude Paul wrote: >> Something I discovered while writing rvkms since some versions of the >> driver didn't have a filled out atomic_update function - we mention that >> this callback is "optional", but we don't actually check whether it's NULL >> or not before calling it. As a result, we'll segfault if it's not filled >> in. >> >> rvkms rvkms.0: [drm:drm_atomic_helper_commit_modeset_disables] modeset on [ENCODER:36:Virtual-36] >> BUG: kernel NULL pointer dereference, address: 0000000000000000 >> PGD 0 P4D 0 >> Oops: Oops: 0010 [#1] PREEMPT SMP NOPTI >> Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS edk2-20240813-1.fc40 08/13/2024 >> RIP: 0010:0x0 >> >> So, let's fix that. >> >> Signed-off-by: Lyude Paul <lyude@redhat.com> >> Fixes: c2fcd274bce5 ("drm: Add atomic/plane helpers") >> Cc: dri-devel@lists.freedesktop.org >> Cc: <stable@vger.kernel.org> # v3.19+ > So we had kind of a similar argument with drm_connector_init early this > year, but I do agree we shouldn't fault if we're missing a callback. > > I do wonder how we can implement a plane without atomic_update though? > Do we have drivers in such a case? That would likely be an output with an entirely static display. Hard to imaging, I think. > > If not, a better solution would be to make it mandatory and check it > when registering. Although I r-b'ed the patch already, I'd also prefer this solution. > > Maxime
On Mon, 2024-09-30 at 09:06 +0200, Thomas Zimmermann wrote: > Hi > > Am 30.09.24 um 09:01 schrieb Maxime Ripard: > > Hi, > > > > On Fri, Sep 27, 2024 at 04:46:16PM GMT, Lyude Paul wrote: > > > Something I discovered while writing rvkms since some versions of the > > > driver didn't have a filled out atomic_update function - we mention that > > > this callback is "optional", but we don't actually check whether it's NULL > > > or not before calling it. As a result, we'll segfault if it's not filled > > > in. > > > > > > rvkms rvkms.0: [drm:drm_atomic_helper_commit_modeset_disables] modeset on [ENCODER:36:Virtual-36] > > > BUG: kernel NULL pointer dereference, address: 0000000000000000 > > > PGD 0 P4D 0 > > > Oops: Oops: 0010 [#1] PREEMPT SMP NOPTI > > > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS edk2-20240813-1.fc40 08/13/2024 > > > RIP: 0010:0x0 > > > > > > So, let's fix that. > > > > > > Signed-off-by: Lyude Paul <lyude@redhat.com> > > > Fixes: c2fcd274bce5 ("drm: Add atomic/plane helpers") > > > Cc: dri-devel@lists.freedesktop.org > > > Cc: <stable@vger.kernel.org> # v3.19+ > > So we had kind of a similar argument with drm_connector_init early this > > year, but I do agree we shouldn't fault if we're missing a callback. > > > > I do wonder how we can implement a plane without atomic_update though? > > Do we have drivers in such a case? > > That would likely be an output with an entirely static display. Hard to > imaging, I think. > > > > > If not, a better solution would be to make it mandatory and check it > > when registering. > > Although I r-b'ed the patch already, I'd also prefer this solution. Gotcha, FWIW the reason I went with this patch: * atomic_update is actually documented as being optional in the kernel docs, so we'd want to remove that if we make it mandatory * rvkms currently doesn't have an atomic_update. We will likely have one whenever I get a chance to actually add CRC and/or writeback connector supports - but for the time being all we do is register a KMS device with vblank support. I am fine with either solution though > > > > > > Maxime >
Hi, On Mon, Sep 30, 2024 at 03:45:13PM -0400, Lyude Paul wrote: > On Mon, 2024-09-30 at 09:06 +0200, Thomas Zimmermann wrote: > > Hi > > > > Am 30.09.24 um 09:01 schrieb Maxime Ripard: > > > Hi, > > > > > > On Fri, Sep 27, 2024 at 04:46:16PM GMT, Lyude Paul wrote: > > > > Something I discovered while writing rvkms since some versions of the > > > > driver didn't have a filled out atomic_update function - we mention that > > > > this callback is "optional", but we don't actually check whether it's NULL > > > > or not before calling it. As a result, we'll segfault if it's not filled > > > > in. > > > > > > > > rvkms rvkms.0: [drm:drm_atomic_helper_commit_modeset_disables] modeset on [ENCODER:36:Virtual-36] > > > > BUG: kernel NULL pointer dereference, address: 0000000000000000 > > > > PGD 0 P4D 0 > > > > Oops: Oops: 0010 [#1] PREEMPT SMP NOPTI > > > > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS edk2-20240813-1.fc40 08/13/2024 > > > > RIP: 0010:0x0 > > > > > > > > So, let's fix that. > > > > > > > > Signed-off-by: Lyude Paul <lyude@redhat.com> > > > > Fixes: c2fcd274bce5 ("drm: Add atomic/plane helpers") > > > > Cc: dri-devel@lists.freedesktop.org > > > > Cc: <stable@vger.kernel.org> # v3.19+ > > > So we had kind of a similar argument with drm_connector_init early this > > > year, but I do agree we shouldn't fault if we're missing a callback. > > > > > > I do wonder how we can implement a plane without atomic_update though? > > > Do we have drivers in such a case? > > > > That would likely be an output with an entirely static display. Hard to > > imaging, I think. > > > > > > > > If not, a better solution would be to make it mandatory and check it > > > when registering. > > > > Although I r-b'ed the patch already, I'd also prefer this solution. > > Gotcha, FWIW the reason I went with this patch: > * atomic_update is actually documented as being optional in the kernel docs, > so we'd want to remove that if we make it mandatory Sure, that makes total sense :) > * rvkms currently doesn't have an atomic_update. We will likely have one > whenever I get a chance to actually add CRC and/or writeback connector > supports - but for the time being all we do is register a KMS device with > vblank support. WIP drivers can provide an empty implementation. And even if actually didn't need it for $REASONS, I'd argue that an empty implementation (and a comment) makes that explicit instead of making the reader guess why it's not needed. Maxime
diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c index 43cdf39019a44..b3c507040c6d6 100644 --- a/drivers/gpu/drm/drm_atomic_helper.c +++ b/drivers/gpu/drm/drm_atomic_helper.c @@ -2797,7 +2797,8 @@ void drm_atomic_helper_commit_planes(struct drm_device *dev, funcs->atomic_disable(plane, old_state); } else if (new_plane_state->crtc || disabling) { - funcs->atomic_update(plane, old_state); + if (funcs->atomic_update) + funcs->atomic_update(plane, old_state); if (!disabling && funcs->atomic_enable) { if (drm_atomic_plane_enabling(old_plane_state, new_plane_state)) @@ -2889,7 +2890,8 @@ drm_atomic_helper_commit_planes_on_crtc(struct drm_crtc_state *old_crtc_state) if (disabling && plane_funcs->atomic_disable) { plane_funcs->atomic_disable(plane, old_state); } else if (new_plane_state->crtc || disabling) { - plane_funcs->atomic_update(plane, old_state); + if (plane_funcs->atomic_update) + plane_funcs->atomic_update(plane, old_state); if (!disabling && plane_funcs->atomic_enable) { if (drm_atomic_plane_enabling(old_plane_state, new_plane_state))
Something I discovered while writing rvkms since some versions of the driver didn't have a filled out atomic_update function - we mention that this callback is "optional", but we don't actually check whether it's NULL or not before calling it. As a result, we'll segfault if it's not filled in. rvkms rvkms.0: [drm:drm_atomic_helper_commit_modeset_disables] modeset on [ENCODER:36:Virtual-36] BUG: kernel NULL pointer dereference, address: 0000000000000000 PGD 0 P4D 0 Oops: Oops: 0010 [#1] PREEMPT SMP NOPTI Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS edk2-20240813-1.fc40 08/13/2024 RIP: 0010:0x0 So, let's fix that. Signed-off-by: Lyude Paul <lyude@redhat.com> Fixes: c2fcd274bce5 ("drm: Add atomic/plane helpers") Cc: dri-devel@lists.freedesktop.org Cc: <stable@vger.kernel.org> # v3.19+ --- drivers/gpu/drm/drm_atomic_helper.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) base-commit: 22512c3ee0f47faab5def71c4453638923c62522