<div dir="ltr"><div dir="ltr">I think there are no direct issues with initialization in the state how it is now. I suspect it's because "drm_kms_helper_poll_enable()" starts the first worker thread with a delay, which gives enough time to initialize required resources. I changed the initialization part to keep it consistent with the finish part, which is the one causing troubles.<div> </div><div>I am not sure where I could move "drm_kms_helper_poll_enable/disable()", since it is defined in "drm/drm_probe_helper.c", which is only included in "nouveau_display.c" and "nouveau_connector.c". Both creating a new function in "nouveau_display.c", and including "probe_helper.h" and using poll_enable in a different file like "nouveau_fbcon.c" seem like too big changes for such small fix. I don't know.</div><div> </div><div>Can this new proposed order break something in the finish part as well? Maybe it would be just better to change the order of "nouveau_drm_finish" and keep the current order of "noueau_drm_init"?</div></div></div> <div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Thu, May 5, 2022 at 9:57 PM Lyude Paul <<a href="mailto:lyude@redhat.com">lyude@redhat.com</a>> wrote: </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hmm, I think we might just need to move the drm_kms_helper_poll_enable() call to the end here instead of all of nouveau_display_init(). I realized this because in nouveau_display_init() it seems that we actually rely on nouveau_display_init() to setup hotplug interrupts - which we do actually need this early on in the driver probe process. That being said though, drm_kms_helper_poll_enable() shouldn't be required for MST short HPD IRQs from working so moving that instead should work. On Wed, 2022-05-04 at 19:18 +0200, Mark Menzynski wrote: > Resources needed for output poll workers are destroyed in > nouveau_fbcon_fini() before output poll workers are cleared in > nouveau_display_fini(). This means there is a time between fbcon_fini > and display_fini, where if output poll happens, it crashes. > > BUG: KASAN: use-after-free in > __drm_fb_helper_initial_config_and_unlock.cold+0x1f3/0x291 > [drm_kms_helper] > > Cc: Ben Skeggs <<a href="mailto:bskeggs@redhat.com" target="_blank">bskeggs@redhat.com</a>> > Cc: Karol Herbst <<a href="mailto:kherbst@redhat.com" target="_blank">kherbst@redhat.com</a>> > Cc: Lyude Paul <<a href="mailto:lyude@redhat.com" target="_blank">lyude@redhat.com</a>> > Cc: David Airlie <<a href="mailto:airlied@linux.ie" target="_blank">airlied@linux.ie</a>> > Cc: Daniel Vetter <<a href="mailto:daniel@ffwll.ch" target="_blank">daniel@ffwll.ch</a>> > Cc: Sumit Semwal <<a href="mailto:sumit.semwal@linaro.org" target="_blank">sumit.semwal@linaro.org</a>> > Cc: "Christian König" <<a href="mailto:christian.koenig@amd.com" target="_blank">christian.koenig@amd.com</a>> > Cc: <a href="mailto:dri-devel@lists.freedesktop.org" target="_blank">dri-devel@lists.freedesktop.org</a> > Cc: <a href="mailto:nouveau@lists.freedesktop.org" target="_blank">nouveau@lists.freedesktop.org</a> > Cc: <a href="mailto:linux-kernel@vger.kernel.org" target="_blank">linux-kernel@vger.kernel.org</a> > Cc: <a href="mailto:linux-media@vger.kernel.org" target="_blank">linux-media@vger.kernel.org</a> > Cc: <a href="mailto:linaro-mm-sig@lists.linaro.org" target="_blank">linaro-mm-sig@lists.linaro.org</a> > Signed-off-by: Mark Menzynski <<a href="mailto:mmenzyns@redhat.com" target="_blank">mmenzyns@redhat.com</a>> > --- > drivers/gpu/drm/nouveau/nouveau_drm.c | 17 ++++++++--------- > 1 file changed, 8 insertions(+), 9 deletions(-) > > diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c > b/drivers/gpu/drm/nouveau/nouveau_drm.c > index 561309d447e0..773efdd20d2f 100644 > --- a/drivers/gpu/drm/nouveau/nouveau_drm.c > +++ b/drivers/gpu/drm/nouveau/nouveau_drm.c > @@ -588,12 +588,6 @@ nouveau_drm_device_init(struct drm_device *dev) > if (ret) > goto fail_dispctor; > > - if (dev->mode_config.num_crtc) { > - ret = nouveau_display_init(dev, false, false); > - if (ret) > - goto fail_dispinit; > - } > - > nouveau_debugfs_init(drm); > nouveau_hwmon_init(dev); > nouveau_svm_init(drm); > @@ -601,6 +595,12 @@ nouveau_drm_device_init(struct drm_device *dev) > nouveau_fbcon_init(dev); > nouveau_led_init(dev); > > + if (dev->mode_config.num_crtc) { > + ret = nouveau_display_init(dev, false, false); > + if (ret) > + goto fail_dispinit; > + } > + > if (nouveau_pmops_runtime()) { > pm_runtime_use_autosuspend(dev->dev); > pm_runtime_set_autosuspend_delay(dev->dev, 5000); > @@ -641,15 +641,14 @@ nouveau_drm_device_fini(struct drm_device *dev) > pm_runtime_forbid(dev->dev); > } > > + if (dev->mode_config.num_crtc) > + nouveau_display_fini(dev, false, false); > nouveau_led_fini(dev); > nouveau_fbcon_fini(dev); > nouveau_dmem_fini(drm); > nouveau_svm_fini(drm); > nouveau_hwmon_fini(dev); > nouveau_debugfs_fini(drm); > - > - if (dev->mode_config.num_crtc) > - nouveau_display_fini(dev, false, false); > nouveau_display_destroy(dev); > > nouveau_accel_fini(drm); -- Cheers, Lyude Paul (she/her) Software Engineer at Red Hat </blockquote></div>