dri-devel.lists.freedesktop.org archive mirror
 help / color / mirror / Atom feed
From: Daniel Vetter <daniel.vetter-/w4YWyX8dFk@public.gmane.org>
To: DRI Development
	<dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>
Cc: Nouveau Dev
	<nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>,
	Intel Graphics Development
	<intel-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>,
	Radeon Dev
	<xorg-driver-ati-go0+a7rfsptAfugRpC6u6w@public.gmane.org>,
	Daniel Vetter <daniel.vetter-/w4YWyX8dFk@public.gmane.org>
Subject: [PATCH 32/37] drm: optimize drm_framebuffer_remove
Date: Wed, 12 Dec 2012 14:07:12 +0100	[thread overview]
Message-ID: <1355317637-16742-33-git-send-email-daniel.vetter@ffwll.ch> (raw)
In-Reply-To: <1355317637-16742-1-git-send-email-daniel.vetter-/w4YWyX8dFk@public.gmane.org>

Now that all framebuffer usage is properly refcounted, we are no
longer required to hold the modeset locks while dropping the last
reference. Hence implemented a fastpath which avoids the potential
stalls associated with grabbing mode_config.lock for the case where
there's no other reference around.

Explain in a big comment why it is safe. Also update kerneldocs with
the new locking rules around drm_framebuffer_remove.

Signed-off-by: Daniel Vetter <daniel.vetter-/w4YWyX8dFk@public.gmane.org>
---
 drivers/gpu/drm/drm_crtc.c |   72 +++++++++++++++++++++++++++-----------------
 1 file changed, 45 insertions(+), 27 deletions(-)

diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c
index 6562eba..6dd441c 100644
--- a/drivers/gpu/drm/drm_crtc.c
+++ b/drivers/gpu/drm/drm_crtc.c
@@ -516,7 +516,11 @@ EXPORT_SYMBOL(drm_framebuffer_cleanup);
  *
  * Scans all the CRTCs and planes in @dev's mode_config.  If they're
  * using @fb, removes it, setting it to NULL. Then drops the reference to the
- * passed-in framebuffer.
+ * passed-in framebuffer. Might take the modeset locks.
+ *
+ * Note that this function optimizes the cleanup away if the caller holds the
+ * last reference to the framebuffer. It is also guaranteed to not take the
+ * modeset locks in this case.
  */
 void drm_framebuffer_remove(struct drm_framebuffer *fb)
 {
@@ -526,33 +530,51 @@ void drm_framebuffer_remove(struct drm_framebuffer *fb)
 	struct drm_mode_set set;
 	int ret;
 
-	WARN_ON(!drm_modeset_is_locked(dev));
 	WARN_ON(!list_empty(&fb->filp_head));
 
-	/* remove from any CRTC */
-	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
-		if (crtc->fb == fb) {
-			/* should turn off the crtc */
-			memset(&set, 0, sizeof(struct drm_mode_set));
-			set.crtc = crtc;
-			set.fb = NULL;
-			ret = drm_mode_set_config_internal(&set);
-			if (ret)
-				DRM_ERROR("failed to reset crtc %p when fb was deleted\n", crtc);
+	/*
+	 * drm ABI mandates that we remove any deleted framebuffers from active
+	 * useage. But since most sane clients only remove framebuffers they no
+	 * longer need, try to optimize this away.
+	 *
+	 * Since we're holding a reference ourselves, observing a refcount of 1
+	 * means that we're the last holder and can skip it. Also, the refcount
+	 * can never increase from 1 again, so we don't need any barriers or
+	 * locks.
+	 *
+	 * Note that userspace could try to race with use and instate a new
+	 * usage _after_ we've cleared all current ones. End result will be an
+	 * in-use fb with fb-id == 0. Userspace is allowed to shoot its own foot
+	 * in this manner.
+	 */
+	if (atomic_read(&fb->refcount.refcount) > 1) {
+		drm_modeset_lock_all(dev);
+		/* remove from any CRTC */
+		list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
+			if (crtc->fb == fb) {
+				/* should turn off the crtc */
+				memset(&set, 0, sizeof(struct drm_mode_set));
+				set.crtc = crtc;
+				set.fb = NULL;
+				ret = drm_mode_set_config_internal(&set);
+				if (ret)
+					DRM_ERROR("failed to reset crtc %p when fb was deleted\n", crtc);
+			}
 		}
-	}
 
-	list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
-		if (plane->fb == fb) {
-			/* should turn off the crtc */
-			ret = plane->funcs->disable_plane(plane);
-			if (ret)
-				DRM_ERROR("failed to disable plane with busy fb\n");
-			/* disconnect the plane from the fb and crtc: */
-			__drm_framebuffer_unreference(plane->fb);
-			plane->fb = NULL;
-			plane->crtc = NULL;
+		list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
+			if (plane->fb == fb) {
+				/* should turn off the crtc */
+				ret = plane->funcs->disable_plane(plane);
+				if (ret)
+					DRM_ERROR("failed to disable plane with busy fb\n");
+				/* disconnect the plane from the fb and crtc: */
+				__drm_framebuffer_unreference(plane->fb);
+				plane->fb = NULL;
+				plane->crtc = NULL;
+			}
 		}
+		drm_modeset_unlock_all(dev);
 	}
 
 	drm_framebuffer_unreference(fb);
@@ -2537,9 +2559,7 @@ int drm_mode_rmfb(struct drm_device *dev,
 	mutex_unlock(&dev->mode_config.fb_lock);
 	mutex_unlock(&file_priv->fbs_lock);
 
-	drm_modeset_lock_all(dev);
 	drm_framebuffer_remove(fb);
-	drm_modeset_unlock_all(dev);
 
 	return 0;
 
@@ -2687,9 +2707,7 @@ void drm_fb_release(struct drm_file *priv)
 		list_del_init(&fb->filp_head);
 
 		/* This will also drop the fpriv->fbs reference. */
-		drm_modeset_lock_all(dev);
 		drm_framebuffer_remove(fb);
-		drm_modeset_unlock_all(dev);
 	}
 	mutex_unlock(&priv->fbs_lock);
 }
-- 
1.7.10.4

  parent reply	other threads:[~2012-12-12 13:07 UTC|newest]

Thread overview: 59+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-12-12 13:06 [PATCH 00/37] [RFC] revamped modeset locking Daniel Vetter
2012-12-12 13:06 ` [PATCH 01/37] drm: review locking rules in drm_crtc.c Daniel Vetter
2012-12-12 13:06 ` [PATCH 05/37] drm/i915: use drm_modeset_lock_all Daniel Vetter
2012-12-12 13:06 ` [PATCH 08/37] drm/shmobile: " Daniel Vetter
2012-12-12 13:06 ` [PATCH 10/37] drm: add per-crtc locks Daniel Vetter
2012-12-13 11:38   ` [Intel-gfx] " Ville Syrjälä
2012-12-13 11:54     ` Daniel Vetter
2012-12-13 14:25       ` Ville Syrjälä
2012-12-12 13:06 ` [PATCH 11/37] drm/radeon: add W|RREG32_IDX for MM_INDEX|DATA based mmio accesss Daniel Vetter
2012-12-12 13:07 ` [PATCH 25/37] drm: don't take modeset locks in getfb ioctl Daniel Vetter
2012-12-12 13:07 ` [PATCH 26/37] drm: fb refcounting for dirtyfb_ioctl Daniel Vetter
2012-12-12 13:07 ` [PATCH 31/37] drm/vmwgfx: add proper framebuffer refcounting Daniel Vetter
2012-12-12 13:07 ` [PATCH 33/37] drm/nouveau: try to protect nbo->pin_refcount Daniel Vetter
     [not found] ` <1355317637-16742-1-git-send-email-daniel.vetter-/w4YWyX8dFk@public.gmane.org>
2012-12-12 13:06   ` [PATCH 02/37] drm/doc: integrate drm_crtc.c kerneldoc Daniel Vetter
2012-12-12 13:06   ` [PATCH 03/37] drm: add drm_modeset_lock|unlock_all Daniel Vetter
2012-12-12 13:06   ` [PATCH 04/37] drm/i915: rework locking for intel_dpio|sbi_read|write Daniel Vetter
2012-12-12 20:54     ` [Intel-gfx] " Jesse Barnes
2012-12-12 22:00       ` Daniel Vetter
2012-12-12 22:05         ` Jesse Barnes
2012-12-12 13:06   ` [PATCH 06/37] drm/gma500: use drm_modeset_lock_all Daniel Vetter
2012-12-12 13:06   ` [PATCH 07/37] drm/ast: " Daniel Vetter
2012-12-12 13:06   ` [PATCH 09/37] drm/vmgfx: " Daniel Vetter
2012-12-12 13:06   ` [PATCH 12/37] drm/radeon: make indirect register access concurrency-safe Daniel Vetter
2012-12-12 13:06   ` [PATCH 13/37] drm/nouveau: protect evo_wait/evo_kick sections with a channel mutex Daniel Vetter
2012-12-12 13:06   ` [PATCH 14/37] drm: only take the crtc lock for ->cursor_set Daniel Vetter
2012-12-12 13:06   ` [PATCH 15/37] drm: only take the crtc lock for ->cursor_move Daniel Vetter
2012-12-13 11:03     ` [PATCH] " Daniel Vetter
2012-12-12 13:06   ` [PATCH 16/37] drm/<drivers>: reorder framebuffer init sequence Daniel Vetter
2012-12-13 11:05     ` [PATCH 1/2] " Daniel Vetter
     [not found]       ` <1355396719-25286-1-git-send-email-daniel.vetter-/w4YWyX8dFk@public.gmane.org>
2012-12-13 11:05         ` [PATCH 2/2] drm/exynos: " Daniel Vetter
2012-12-13 12:26           ` Inki Dae
2012-12-13 12:43             ` Daniel Vetter
2012-12-13 15:16               ` Inki Dae
2012-12-13 16:38                 ` Daniel Vetter
2012-12-14  4:57                   ` Inki Dae
2012-12-14  8:40                     ` Daniel Vetter
2012-12-12 13:06   ` [PATCH 17/37] drm: revamp locking around fb creation/destruction Daniel Vetter
2012-12-12 13:06   ` [PATCH 18/37] drm: create drm_framebuffer_lookup Daniel Vetter
2012-12-12 13:06   ` [PATCH 19/37] drm/gma500: move fbcon restore to lastclose Daniel Vetter
2012-12-12 13:07   ` [PATCH 20/37] drm: revamp framebuffer cleanup interfaces Daniel Vetter
2012-12-12 13:07   ` [PATCH 21/37] drm: reference framebuffers which are on the idr Daniel Vetter
2012-12-12 13:07   ` [PATCH 22/37] drm: nest modeset locks within fpriv->fbs_lock Daniel Vetter
2012-12-12 13:07   ` [PATCH 23/37] drm/i915: fixup overlay stolen memory leak Daniel Vetter
2012-12-12 13:07   ` [PATCH 24/37] drm: push modeset_lock_all into ->fb_create driver callbacks Daniel Vetter
2012-12-12 13:07   ` [PATCH 27/37] drm: refcounting for sprite framebuffers Daniel Vetter
2012-12-12 13:07   ` [PATCH 28/37] drm: encapsulate crtc->set_config calls Daniel Vetter
2012-12-12 13:07   ` [PATCH 29/37] drm: refcounting for crtc framebuffers Daniel Vetter
2012-12-12 13:07   ` [PATCH 30/37] drm/i915: dump refcount into framebuffer debugfs file Daniel Vetter
2012-12-12 13:07   ` Daniel Vetter [this message]
2012-12-12 13:07   ` [PATCH 34/37] drm/ttm: fix fence locking in ttm_buffer_object_transfer Daniel Vetter
2012-12-12 14:48     ` Jerome Glisse
     [not found]       ` <CAH3drwbL34+omj2S3ArOPgoYRQ3uzPEiRPFLn7YkP3cOMnPiZg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2012-12-12 15:42         ` Daniel Vetter
2012-12-13 11:06     ` [PATCH] allow shmob+imx drm drivers to be compiled Daniel Vetter
2012-12-13 11:18     ` Daniel Vetter
2012-12-13 11:26     ` [PATCH] drm/ttm: fix fence locking in ttm_buffer_object_transfer Daniel Vetter
2012-12-12 13:07   ` [PATCH 35/37] drm/radeon: fix fence locking in the pageflip callback Daniel Vetter
2012-12-12 13:07   ` [PATCH 36/37] drm: only grab the crtc lock for pageflips Daniel Vetter
2012-12-12 13:07   ` [PATCH 37/37] drm: don't hold crtc mutexes for connector ->detect callbacks Daniel Vetter
2012-12-12 14:18 ` [PATCH 00/37] [RFC] revamped modeset locking Daniel Vetter

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1355317637-16742-33-git-send-email-daniel.vetter@ffwll.ch \
    --to=daniel.vetter-/w4ywyx8dfk@public.gmane.org \
    --cc=dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org \
    --cc=intel-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org \
    --cc=nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org \
    --cc=xorg-driver-ati-go0+a7rfsptAfugRpC6u6w@public.gmane.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).