[PATCH] drm/core: Do not preserve framebuffer on rmfb, v3.

[Maarten Lankhorst]

It turns out that preserving framebuffers after the rmfb call breaks
vmwgfx userspace. This was originally introduced because it was thought
nobody relied on the behavior, but unfortunately it seems there are
exceptions.

drm_framebuffer_remove may fail with -EINTR now, so a straight revert
is impossible. There is no way to remove the framebuffer from the lists
and active planes without introducing a race because of the different
locking requirements. Instead call drm_framebuffer_remove from a
workqueue, which is unaffected by signals.

Changes since v1:
- Add comment.
Changes since v2:
- Add fastpath for refcount = 1. (danvet)

Cc: stable at vger.kernel.org #v4.4+
Fixes: 13803132818c ("drm/core: Preserve the framebuffer after removing it.")
Testcase: kms_flip.flip-vs-rmfb-interruptible
References: https://lists.freedesktop.org/archives/dri-devel/2016-March/102876.html
Cc: Thomas Hellstrom <thellstrom at vmware.com>
Cc: David Herrmann <dh.herrmann at gmail.com>
---
 drivers/gpu/drm/drm_crtc.c | 29 ++++++++++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c
index 55ffde5a3a4a..743bece1f579 100644
--- a/drivers/gpu/drm/drm_crtc.c
+++ b/drivers/gpu/drm/drm_crtc.c
@@ -3434,6 +3434,18 @@ int drm_mode_addfb2(struct drm_device *dev,
 	return 0;
 }
 
+struct drm_mode_rmfb_work {
+	struct work_struct work;
+	struct drm_framebuffer *fb;
+};
+
+static void drm_mode_rmfb_work_fn(struct work_struct *w)
+{
+	struct drm_mode_rmfb_work *arg = container_of(w, typeof(*arg), work);
+
+	drm_framebuffer_remove(arg->fb);
+}
+
 /**
  * drm_mode_rmfb - remove an FB from the configuration
  * @dev: drm device for the ioctl
@@ -3474,7 +3486,22 @@ int drm_mode_rmfb(struct drm_device *dev,
 	mutex_unlock(&dev->mode_config.fb_lock);
 	mutex_unlock(&file_priv->fbs_lock);
 
-	drm_framebuffer_unreference(fb);
+	/*
+	 * drm_framebuffer_remove may fail with -EINTR on pending signals,
+	 * so run this in a separate stack as there's no way to correctly
+	 * handle this after the fb is already removed from the lookup table.
+	 */
+	if (atomic_read(&fb->refcount.refcount) > 1) {
+		struct drm_mode_rmfb_work arg;
+
+		INIT_WORK_ONSTACK(&arg.work, drm_mode_rmfb_work_fn);
+		arg.fb = fb;
+
+		schedule_work(&arg.work);
+		flush_work(&arg.work);
+		destroy_work_on_stack(&arg.work);
+	} else
+		drm_framebuffer_unreference(fb);
 
 	return 0;
 
-- 
2.1.0