From mboxrd@z Thu Jan 1 00:00:00 1970 From: Inki Dae Subject: Re: [PATCH 5/9] drm/exynos/crtc: fix framebuffer reference sequence Date: Fri, 12 Sep 2014 19:45:51 +0900 Message-ID: <5412CEDF.7030806@samsung.com> References: <1410268573-2297-1-git-send-email-a.hajda@samsung.com> <1410268573-2297-6-git-send-email-a.hajda@samsung.com> <5412B02A.6040609@samsung.com> <20140912085710.GD4740@phenom.ffwll.local> <5412BC8D.2030007@samsung.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Received: from mailout2.samsung.com ([203.254.224.25]:47402 "EHLO mailout2.samsung.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751629AbaILKpx (ORCPT ); Fri, 12 Sep 2014 06:45:53 -0400 In-reply-to: <5412BC8D.2030007@samsung.com> Sender: linux-samsung-soc-owner@vger.kernel.org List-Id: linux-samsung-soc@vger.kernel.org To: Andrzej Hajda Cc: "moderated list:ARM/S5P EXYNOS AR..." , Seung-Woo Kim , open list , dri-devel@lists.freedesktop.org, Kyungmin Park , Marek Szyprowski On 2014=EB=85=84 09=EC=9B=94 12=EC=9D=BC 18:27, Andrzej Hajda wrote: > On 09/12/2014 10:57 AM, Daniel Vetter wrote: >> On Fri, Sep 12, 2014 at 05:34:50PM +0900, Inki Dae wrote: >>> Hi Andrzej, >>> >>> On 2014=EB=85=84 09=EC=9B=94 09=EC=9D=BC 22:16, Andrzej Hajda wrote= : >>>> Adding reference to framebuffer should be accompanied with removin= g >>>> reference to old framebuffer assigned to the plane. >>>> This patch removes following warning: >>>> >>>> [ 95.038017] WARNING: CPU: 1 PID: 3067 at drivers/gpu/drm/drm_cr= tc.c:5115 drm_mode_config_cleanup+0x258/0x268() >>>> [ 95.048086] Modules linked in: >>>> [ 95.051430] CPU: 1 PID: 3067 Comm: bash Tainted: G W = 3.16.0-11355-g7a6eca5-dirty #3015 >>>> [ 95.060058] [] (unwind_backtrace) from [] (= show_stack+0x10/0x14) >>>> [ 95.067766] [] (show_stack) from [] (dump_s= tack+0x70/0xbc) >>>> [ 95.074953] [] (dump_stack) from [] (warn_s= lowpath_common+0x64/0x88) >>>> [ 95.083005] [] (warn_slowpath_common) from [] (warn_slowpath_null+0x1c/0x24) >>>> [ 95.091780] [] (warn_slowpath_null) from []= (drm_mode_config_cleanup+0x258/0x268) >>>> [ 95.100983] [] (drm_mode_config_cleanup) from [] (exynos_drm_unload+0x38/0x50) >>>> [ 95.109915] [] (exynos_drm_unload) from [] = (drm_dev_unregister+0x24/0x98) >>>> [ 95.118414] [] (drm_dev_unregister) from []= (drm_put_dev+0x28/0x64) >>>> [ 95.126412] [] (drm_put_dev) from [] (take_= down_master+0x24/0x44) >>>> [ 95.134218] [] (take_down_master) from [] (= component_del+0x8c/0xc8) >>>> [ 95.142201] [] (component_del) from [] (exy= nos_dsi_remove+0x18/0x2c) >>>> [ 95.150294] [] (exynos_dsi_remove) from [] = (platform_drv_remove+0x18/0x1c) >>>> [ 95.158872] [] (platform_drv_remove) from [= ] (__device_release_driver+0x70/0xc4) >>>> [ 95.167981] [] (__device_release_driver) from [] (device_release_driver+0x20/0x2c) >>>> [ 95.177268] [] (device_release_driver) from [] (unbind_store+0x5c/0x94) >>>> [ 95.185597] [] (unbind_store) from [] (drv_= attr_store+0x20/0x2c) >>>> [ 95.193323] [] (drv_attr_store) from [] (sy= sfs_kf_write+0x4c/0x50) >>>> [ 95.201224] [] (sysfs_kf_write) from [] (ke= rnfs_fop_write+0xc4/0x184) >>>> [ 95.209393] [] (kernfs_fop_write) from [] (= vfs_write+0xa0/0x1a8) >>>> [ 95.217111] [] (vfs_write) from [] (SyS_wri= te+0x40/0x8c) >>>> [ 95.224146] [] (SyS_write) from [] (ret_fas= t_syscall+0x0/0x48) >>>> >>>> Signed-off-by: Andrzej Hajda >>>> --- >>>> drivers/gpu/drm/exynos/exynos_drm_crtc.c | 6 ++++++ >>>> 1 file changed, 6 insertions(+) >>>> >>>> diff --git a/drivers/gpu/drm/exynos/exynos_drm_crtc.c b/drivers/gp= u/drm/exynos/exynos_drm_crtc.c >>>> index b68e58f..bde19f4 100644 >>>> --- a/drivers/gpu/drm/exynos/exynos_drm_crtc.c >>>> +++ b/drivers/gpu/drm/exynos/exynos_drm_crtc.c >>>> @@ -145,10 +145,16 @@ exynos_drm_crtc_mode_set(struct drm_crtc *cr= tc, struct drm_display_mode *mode, >>>> if (ret) >>>> return ret; >>>> =20 >>>> + /* we need to unreference current fb after replacing it with new= one */ >>>> + old_fb =3D plane->fb; >>>> + >>>> plane->crtc =3D crtc; >>>> plane->fb =3D crtc->primary->fb; >>>> drm_framebuffer_reference(plane->fb); >>>> =20 >>>> + if (old_fb) >>>> + drm_framebuffer_unreference(old_fb); >>> This time would be a good chance that we can consider drm flip queu= e to >>> make sure that whole memory region to old_fb is scanned out complet= ely >>> before dropping a reference of old_fb. the reference of old_fb shou= ld be >>> dropped at irq handler of each crtc devices, fimd and mixer. >> Generally it's not a good idea to drop fb references from irq contex= t, >> since if you actually drop the last reference it'll blow up: fb clea= nup >> needs a bunch of mutexes. >=20 > I agree with that. >=20 >> >> Also the drm core really should be taking care of this for you, you = only >> need to grab references yourself for async flips if you want the buf= fer to >> survive a bit. crtc_mode_set has not need for this. I expect that th= e >> refcounting bug is somewhere else, at least from my experience chasi= ng >> such issues in i915 ;-) >=20 > Hmm, maybe I miss something but I do not see the core grabbing fb ref= erence > on plane->fb update. On the other side drm_framebuffer_remove calls > drm_plane_force_disable which drops plane->fb reference. > I am not yet familiar with this code so maybe there is better solutio= n. >=20 > If not I guess it would be better to move this code to > exynos_plane_mode_set. > At least it is done this way in omap and msm, in fact it seems better= place I cannot see it in msm. In case of msm, drm flip queue is used in mdp4/5_crtc_mode_set function. See update_fb function there. > for such things. What do you think? >=20 > Regards > Andrzej >=20 >> -Daniel >=20 > -- > To unsubscribe from this list: send the line "unsubscribe linux-samsu= ng-soc" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html >=20