Hi Am 17.02.23 um 20:47 schrieb Daniel Vetter:
Some vague evidences suggests this can go wrong. Try to prevent it by holding the right mutex and clearing ->deferred_setup to make sure we later on don't accidentally try to re-register the fbdev when the driver thought it had it all cleaned up already. v2: I realized that this is fundamentally butchered, and CI complained about lockdep splats. So limit the critical section again and just add a few notes what the proper fix is. References: https://intel-gfx-ci.01.org/tree/linux-next/next-20201215/fi-byt-j1900/igt@i915_pm_rpm@xxxxxxxxxxxxxxxxxx Signed-off-by: Daniel Vetter <daniel.vetter@xxxxxxxxx> Cc: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx> Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Cc: Maarten Lankhorst <maarten.lankhorst@xxxxxxxxxxxxxxx> Cc: Maxime Ripard <mripard@xxxxxxxxxx> Cc: Thomas Zimmermann <tzimmermann@xxxxxxx> Cc: David Airlie <airlied@xxxxxxxx> Cc: Daniel Vetter <daniel@xxxxxxxx> --- drivers/gpu/drm/drm_fb_helper.c | 6 ++++++ drivers/gpu/drm/drm_fbdev_generic.c | 5 +++++ 2 files changed, 11 insertions(+) diff --git a/drivers/gpu/drm/drm_fb_helper.c b/drivers/gpu/drm/drm_fb_helper.c index 3e17261a12b6..2415a2c7ca44 100644 --- a/drivers/gpu/drm/drm_fb_helper.c +++ b/drivers/gpu/drm/drm_fb_helper.c @@ -545,6 +545,9 @@ EXPORT_SYMBOL(drm_fb_helper_alloc_info); * A wrapper around unregister_framebuffer, to release the fb_info * framebuffer device. This must be called before releasing all resources for * @fb_helper by calling drm_fb_helper_fini(). + * + * Note that this is fundamentally racy on hotunload because it doen't handle + * open fbdev file descriptors at all. Use drm_fbdev_generic_setup() instead. */ void drm_fb_helper_unregister_info(struct drm_fb_helper *fb_helper) { @@ -558,6 +561,9 @@ EXPORT_SYMBOL(drm_fb_helper_unregister_info); * @fb_helper: driver-allocated fbdev helper, can be NULL * * This cleans up all remaining resources associated with @fb_helper. + * + * Note that this is fundamentally racy on hotunload because it doen't handle + * open fbdev file descriptors at all. Use drm_fbdev_generic_setup() instead. */ void drm_fb_helper_fini(struct drm_fb_helper *fb_helper) { diff --git a/drivers/gpu/drm/drm_fbdev_generic.c b/drivers/gpu/drm/drm_fbdev_generic.c index 365f80717fa1..1618109592ce 100644 --- a/drivers/gpu/drm/drm_fbdev_generic.c +++ b/drivers/gpu/drm/drm_fbdev_generic.c @@ -347,7 +347,12 @@ static void drm_fbdev_client_unregister(struct drm_client_dev *client) { struct drm_fb_helper *fb_helper = drm_fb_helper_from_client(client);+ mutex_lock(&fb_helper->lock);+ fb_helper->deferred_setup = false; + mutex_unlock(&fb_helper->lock);
The unregister code runs as part of the client cleanup. And the client also goes through a number of helpers that handle display hotplug and/or restore, which are affected by ->deferred_setup. But it's all mutually exclusive. AFAICT nothing runs parallel to the unregister code. See the use of dev->clientlist_mutex in drm_client.c. The patch is not necessary IMHO.
Best regards Thomas
+ if (fb_helper->info) { + /* drm_fbdev_fb_destroy() takes care of cleanup */ drm_fb_helper_unregister_info(fb_helper); } else { drm_client_release(&fb_helper->client);
-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Ivo Totev
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature