From mboxrd@z Thu Jan 1 00:00:00 1970 From: Daniel Vetter Subject: [PATCH 2/2] drm/fb-helper: improve drm_fb_helper_initial_config locking Date: Thu, 20 Mar 2014 14:01:22 +0100 Message-ID: <1395320482-3962-2-git-send-email-daniel.vetter@ffwll.ch> References: <1395320482-3962-1-git-send-email-daniel.vetter@ffwll.ch> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1395320482-3962-1-git-send-email-daniel.vetter@ffwll.ch> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-gfx-bounces@lists.freedesktop.org Sender: "Intel-gfx" To: DRI Development Cc: Daniel Vetter , Intel Graphics Development List-Id: dri-devel@lists.freedesktop.org The locking in drm_fb_helper_initial_config is a bit troublesome for a few reasons: - We can't just wrap the entire function up into modeset locks since the fbdev registration might call down into fbcon code, which then through our ->set_par implementation needs to be able to grab all modeset locks. So we'd have a neat deadlock. - This implies though that all current callers don't hold any modeset locks by necessity, so we have free reign to grab any modeset locks we need to grab. - The private state of the fbdev helper doesn't need any protection through locks, since once we have the fbdev registered it is mostly invariant or protected through the modeset locking in ->set_par and other callbacks. We can fully rely on driver having non-racy setup sequences here. For the initial config computation we actually may not grab locks since drivers which provide their own magic sauce (like i915) might need to grab locks themselves. - We should grab locks though when we probe outputs. Currently there's not much risk, but already now userspace could start poking at sysfs files and so probe concurrently. I expect that in the future driver init will be much more async, and since probing is really time-consuming this is a prime candidate. - We must not hold any crtc->mutex locks while calling probe functions since those might need to lock a crtc for e.g. load detection. i915 is such a driver. Also it's the probing calls which hit upon piles of new locking asserts I've recently added in commit 62ff94a5492175759546f8bc61383189d6b49122 Author: Daniel Vetter Date: Thu Jan 23 22:18:47 2014 +0100 drm/crtc-helper: remove LOCKING from kerneldoc and commit 63951385052f7974155fa38f962f0f4e9847f90a Author: Daniel Vetter Date: Thu Jan 23 15:14:15 2014 +0100 drm/doc: Repleace LOCKING kerneldoc sections in drm_modes.c Hence the right fix is to grab the mode_config mutex, but only that and only right around the probe calls. It seems to be sufficient to shut up all the locking WARNINGs I see on i915 and nouveau in drm_fb_helper_initial_config. Signed-off-by: Daniel Vetter --- drivers/gpu/drm/drm_fb_helper.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/gpu/drm/drm_fb_helper.c b/drivers/gpu/drm/drm_fb_helper.c index 87876198801d..16f271e21b9c 100644 --- a/drivers/gpu/drm/drm_fb_helper.c +++ b/drivers/gpu/drm/drm_fb_helper.c @@ -1536,9 +1536,11 @@ bool drm_fb_helper_initial_config(struct drm_fb_helper *fb_helper, int bpp_sel) drm_fb_helper_parse_command_line(fb_helper); + mutex_lock(&dev->mode_config.mutex); count = drm_fb_helper_probe_connector_modes(fb_helper, dev->mode_config.max_width, dev->mode_config.max_height); + mutex_unlock(&dev->mode_config.mutex); /* * we shouldn't end up with no modes here. */ -- 1.8.4.rc3