| Message ID | 1347874683-21191-1-git-send-email-chris@chris-wilson.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, 17 Sep 2012 10:38:03 +0100 Chris Wilson <chris@chris-wilson.co.uk> wrote: > As during the plane cleanup, we wish to disable the hardware and > so may modify state on the associated CRTC, that CRTC must continue to > exist until we are finished. > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Jesse Barnes <jbarnes@virtuousgeek.org> > Cc: stable@vger.kernel.org > --- > drivers/gpu/drm/drm_crtc.c | 8 ++++---- > 1 file changed, 4 insertions(+), 4 deletions(-) > > diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c > index 6fbfc24..af81f77 100644 > --- a/drivers/gpu/drm/drm_crtc.c > +++ b/drivers/gpu/drm/drm_crtc.c > @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device *dev) > fb->funcs->destroy(fb); > } > > - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, head) { > - crtc->funcs->destroy(crtc); > - } > - > list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, > head) { > plane->funcs->destroy(plane); > } > > + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, head) { > + crtc->funcs->destroy(crtc); > + } > + > idr_remove_all(&dev->mode_config.crtc_idr); > idr_destroy(&dev->mode_config.crtc_idr); > } Yeah, looks good. Did we used to not touch the crtc in the plane cleanup? Either way this is safer. Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
On Mon, 17 Sep 2012 07:55:59 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote: > Yeah, looks good. Did we used to not touch the crtc in the plane > cleanup? Either way this is safer. I think the bug only came to light because QA is now routinely running with slub_debug. -Chris
On Mon, Sep 17, 2012 at 04:04:23PM +0100, Chris Wilson wrote: > On Mon, 17 Sep 2012 07:55:59 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote: > > Yeah, looks good. Did we used to not touch the crtc in the plane > > cleanup? Either way this is safer. > > I think the bug only came to light because QA is now routinely running > with slub_debug. And QA now confirmed that this patch fixes things: Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 Tested-by: lu hua <huax.lu@intel.com> -Daniel
Am Mittwoch, den 19.09.2012, 11:34 +0200 schrieb Daniel Vetter: > On Mon, Sep 17, 2012 at 04:04:23PM +0100, Chris Wilson wrote: > > On Mon, 17 Sep 2012 07:55:59 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote: > > > Yeah, looks good. Did we used to not touch the crtc in the plane > > > cleanup? Either way this is safer. > > > > I think the bug only came to light because QA is now routinely running > > with slub_debug. > > And QA now confirmed that this patch fixes things: > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 > Tested-by: lu hua <huax.lu@intel.com> Lu Hua*x*? Thanks, Paul
On Wed, 19 Sep 2012 11:45:34 +0200, Paul Menzel <paulepanter@users.sourceforge.net> wrote: > Am Mittwoch, den 19.09.2012, 11:34 +0200 schrieb Daniel Vetter: > > On Mon, Sep 17, 2012 at 04:04:23PM +0100, Chris Wilson wrote: > > > On Mon, 17 Sep 2012 07:55:59 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote: > > > > Yeah, looks good. Did we used to not touch the crtc in the plane > > > > cleanup? Either way this is safer. > > > > > > I think the bug only came to light because QA is now routinely running > > > with slub_debug. > > > > And QA now confirmed that this patch fixes things: > > > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 > > Tested-by: lu hua <huax.lu@intel.com> > > Lu Hua*x*? The 'x' is just part of Intel's policy for indicating a contractor in email addresses, which can be quite confusing. -Chris
On 09/17/2012 06:38 PM, Chris Wilson wrote: > As during the plane cleanup, we wish to disable the hardware and > so may modify state on the associated CRTC, that CRTC must continue to > exist until we are finished. A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and plane use same framebuffer and the framebuffer is destroyed, crtc is turned off prior to turning off plane. > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Jesse Barnes <jbarnes@virtuousgeek.org> > Cc: stable@vger.kernel.org > --- > drivers/gpu/drm/drm_crtc.c | 8 ++++---- > 1 file changed, 4 insertions(+), 4 deletions(-) > > diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c > index 6fbfc24..af81f77 100644 > --- a/drivers/gpu/drm/drm_crtc.c > +++ b/drivers/gpu/drm/drm_crtc.c > @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device *dev) > fb->funcs->destroy(fb); > } > > - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, head) { > - crtc->funcs->destroy(crtc); > - } > - > list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, > head) { > plane->funcs->destroy(plane); > } > > + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, head) { > + crtc->funcs->destroy(crtc); > + } > + > idr_remove_all(&dev->mode_config.crtc_idr); > idr_destroy(&dev->mode_config.crtc_idr); > }
On Wed, Sep 19, 2012 at 9:52 PM, Joonyoung Shim <jy0922.shim@samsung.com> wrote: > On 09/17/2012 06:38 PM, Chris Wilson wrote: >> >> As during the plane cleanup, we wish to disable the hardware and >> so may modify state on the associated CRTC, that CRTC must continue to >> exist until we are finished. > > > A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and > plane use same framebuffer and the framebuffer is destroyed, crtc is > turned off prior to turning off plane. > I imagine my patch to add refcnt'ing to fb would help in this case.. BR, -R >> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 >> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> >> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> >> Cc: stable@vger.kernel.org >> --- >> drivers/gpu/drm/drm_crtc.c | 8 ++++---- >> 1 file changed, 4 insertions(+), 4 deletions(-) >> >> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c >> index 6fbfc24..af81f77 100644 >> --- a/drivers/gpu/drm/drm_crtc.c >> +++ b/drivers/gpu/drm/drm_crtc.c >> @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device >> *dev) >> fb->funcs->destroy(fb); >> } >> - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, >> head) { >> - crtc->funcs->destroy(crtc); >> - } >> - >> list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, >> head) { >> plane->funcs->destroy(plane); >> } >> + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, >> head) { >> + crtc->funcs->destroy(crtc); >> + } >> + >> idr_remove_all(&dev->mode_config.crtc_idr); >> idr_destroy(&dev->mode_config.crtc_idr); >> } > > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/dri-devel
On 09/20/2012 02:38 PM, Rob Clark wrote: > On Wed, Sep 19, 2012 at 9:52 PM, Joonyoung Shim <jy0922.shim@samsung.com> wrote: >> On 09/17/2012 06:38 PM, Chris Wilson wrote: >>> As during the plane cleanup, we wish to disable the hardware and >>> so may modify state on the associated CRTC, that CRTC must continue to >>> exist until we are finished. >> >> A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and >> plane use same framebuffer and the framebuffer is destroyed, crtc is >> turned off prior to turning off plane. >> > I imagine my patch to add refcnt'ing to fb would help in this case.. > > BR, > -R Even if the patch to add refcnt'ing to fb is applied, same issue will occur in the drm_framebuffer_remove(). It can delay to destroy the fb, but cannot change crtc and plane disable order. >>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 >>> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> >>> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> >>> Cc: stable@vger.kernel.org >>> --- >>> drivers/gpu/drm/drm_crtc.c | 8 ++++---- >>> 1 file changed, 4 insertions(+), 4 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c >>> index 6fbfc24..af81f77 100644 >>> --- a/drivers/gpu/drm/drm_crtc.c >>> +++ b/drivers/gpu/drm/drm_crtc.c >>> @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device >>> *dev) >>> fb->funcs->destroy(fb); >>> } >>> - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, >>> head) { >>> - crtc->funcs->destroy(crtc); >>> - } >>> - >>> list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, >>> head) { >>> plane->funcs->destroy(plane); >>> } >>> + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, >>> head) { >>> + crtc->funcs->destroy(crtc); >>> + } >>> + >>> idr_remove_all(&dev->mode_config.crtc_idr); >>> idr_destroy(&dev->mode_config.crtc_idr); >>> } >> >> _______________________________________________ >> dri-devel mailing list >> dri-devel@lists.freedesktop.org >> http://lists.freedesktop.org/mailman/listinfo/dri-devel
On Thu, 20 Sep 2012 15:10:39 +0900, Joonyoung Shim <jy0922.shim@samsung.com> wrote: > On 09/20/2012 02:38 PM, Rob Clark wrote: > > On Wed, Sep 19, 2012 at 9:52 PM, Joonyoung Shim <jy0922.shim@samsung.com> wrote: > >> On 09/17/2012 06:38 PM, Chris Wilson wrote: > >>> As during the plane cleanup, we wish to disable the hardware and > >>> so may modify state on the associated CRTC, that CRTC must continue to > >>> exist until we are finished. > >> > >> A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and > >> plane use same framebuffer and the framebuffer is destroyed, crtc is > >> turned off prior to turning off plane. > >> > > I imagine my patch to add refcnt'ing to fb would help in this case.. > > > > BR, > > -R > > Even if the patch to add refcnt'ing to fb is applied, same issue will > occur in the drm_framebuffer_remove(). It can delay to destroy the fb, > but cannot change crtc and plane disable order. > > >>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 > >>> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > >>> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> > >>> Cc: stable@vger.kernel.org > >>> --- > >>> drivers/gpu/drm/drm_crtc.c | 8 ++++---- > >>> 1 file changed, 4 insertions(+), 4 deletions(-) > >>> > >>> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c > >>> index 6fbfc24..af81f77 100644 > >>> --- a/drivers/gpu/drm/drm_crtc.c > >>> +++ b/drivers/gpu/drm/drm_crtc.c > >>> @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device > >>> *dev) > >>> fb->funcs->destroy(fb); > >>> } > >>> - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, > >>> head) { > >>> - crtc->funcs->destroy(crtc); > >>> - } > >>> - > >>> list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, > >>> head) { > >>> plane->funcs->destroy(plane); > >>> } > >>> + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, > >>> head) { > >>> + crtc->funcs->destroy(crtc); > >>> + } > >>> + > >>> idr_remove_all(&dev->mode_config.crtc_idr); > >>> idr_destroy(&dev->mode_config.crtc_idr); > >>> } > >> > >> _______________________________________________ > >> dri-devel mailing list > >> dri-devel@lists.freedesktop.org > >> http://lists.freedesktop.org/mailman/listinfo/dri-devel > From: Chris Wilson <chris@chris-wilson.co.uk> Subject: Re: [PATCH] drm: Destroy the planes prior to destroying the associated CRTC To: Rob Clark <rob.clark@linaro.org>, Joonyoung Shim <jy0922.shim@samsung.com> Cc: airlied@redhat.com, stable@vger.kernel.org, dri-devel@lists.freedesktop.org In-Reply-To: <CAF6AEGvOZa_ZhRJN212Rsn-gMMUyWoLT6UFY9iindi8AWx9GvA@mail.gmail.com> References: <1347874683-21191-1-git-send-email-chris@chris-wilson.co.uk> <505A850A.4010205@samsung.com> <CAF6AEGvOZa_ZhRJN212Rsn-gMMUyWoLT6UFY9iindi8AWx9GvA@mail.gmail.com> On Thu, 20 Sep 2012 00:38:04 -0500, Rob Clark <rob.clark@linaro.org> wrote: > On Wed, Sep 19, 2012 at 9:52 PM, Joonyoung Shim <jy0922.shim@samsung.com> wrote: > > On 09/17/2012 06:38 PM, Chris Wilson wrote: > >> > >> As during the plane cleanup, we wish to disable the hardware and > >> so may modify state on the associated CRTC, that CRTC must continue to > >> exist until we are finished. > > > > > > A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and > > plane use same framebuffer and the framebuffer is destroyed, crtc is > > turned off prior to turning off plane. > > > > I imagine my patch to add refcnt'ing to fb would help in this case.. > > BR, > -R > > >> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 > >> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > >> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> > >> Cc: stable@vger.kernel.org > >> --- > >> drivers/gpu/drm/drm_crtc.c | 8 ++++---- > >> 1 file changed, 4 insertions(+), 4 deletions(-) > >> > >> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c > >> index 6fbfc24..af81f77 100644 > >> --- a/drivers/gpu/drm/drm_crtc.c > >> +++ b/drivers/gpu/drm/drm_crtc.c > >> @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device > >> *dev) > >> fb->funcs->destroy(fb); > >> } > >> - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, > >> head) { > >> - crtc->funcs->destroy(crtc); > >> - } > >> - > >> list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, > >> head) { > >> plane->funcs->destroy(plane); > >> } > >> + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, > >> head) { > >> + crtc->funcs->destroy(crtc); > >> + } > >> + > >> idr_remove_all(&dev->mode_config.crtc_idr); > >> idr_destroy(&dev->mode_config.crtc_idr); > >> } > > > > > > _______________________________________________ > > dri-devel mailing list > > dri-devel@lists.freedesktop.org > > http://lists.freedesktop.org/mailman/listinfo/dri-devel From: Chris Wilson <chris@chris-wilson.co.uk> Subject: Re: [PATCH] drm: Destroy the planes prior to destroying the associated CRTC To: Joonyoung Shim <jy0922.shim@samsung.com> Cc: dri-devel@lists.freedesktop.org, airlied@redhat.com, stable@vger.kernel.org In-Reply-To: <505A850A.4010205@samsung.com> References: <1347874683-21191-1-git-send-email-chris@chris-wilson.co.uk> <505A850A.4010205@samsung.com> On Thu, 20 Sep 2012 11:52:58 +0900, Joonyoung Shim <jy0922.shim@samsung.com> wrote: > On 09/17/2012 06:38 PM, Chris Wilson wrote: > > As during the plane cleanup, we wish to disable the hardware and > > so may modify state on the associated CRTC, that CRTC must continue to > > exist until we are finished. > > A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and > plane use same framebuffer and the framebuffer is destroyed, crtc is > turned off prior to turning off plane. This is not an issue in our code as disabling the CRTCs should automatically disable any associated planes, and so the second pass over the planes should be a no-op. The issue during destroy drm_mode_config_cleanup() is that we actually free the CRTC objects and then try to decouple the planes which causes us to reference the just-freed objects in order to see if the hw needs updating. However, reordering the sequence to be CRTCs last would help for consistency. -Chris
On 09/20/2012 04:49 PM, Chris Wilson wrote: > On Thu, 20 Sep 2012 15:10:39 +0900, Joonyoung Shim <jy0922.shim@samsung.com> wrote: >> On 09/20/2012 02:38 PM, Rob Clark wrote: >>> On Wed, Sep 19, 2012 at 9:52 PM, Joonyoung Shim <jy0922.shim@samsung.com> wrote: >>>> On 09/17/2012 06:38 PM, Chris Wilson wrote: >>>>> As during the plane cleanup, we wish to disable the hardware and >>>>> so may modify state on the associated CRTC, that CRTC must continue to >>>>> exist until we are finished. >>>> A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and >>>> plane use same framebuffer and the framebuffer is destroyed, crtc is >>>> turned off prior to turning off plane. >>>> >>> I imagine my patch to add refcnt'ing to fb would help in this case.. >>> >>> BR, >>> -R >> Even if the patch to add refcnt'ing to fb is applied, same issue will >> occur in the drm_framebuffer_remove(). It can delay to destroy the fb, >> but cannot change crtc and plane disable order. >> >>>>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 >>>>> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> >>>>> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> >>>>> Cc: stable@vger.kernel.org >>>>> --- >>>>> drivers/gpu/drm/drm_crtc.c | 8 ++++---- >>>>> 1 file changed, 4 insertions(+), 4 deletions(-) >>>>> >>>>> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c >>>>> index 6fbfc24..af81f77 100644 >>>>> --- a/drivers/gpu/drm/drm_crtc.c >>>>> +++ b/drivers/gpu/drm/drm_crtc.c >>>>> @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device >>>>> *dev) >>>>> fb->funcs->destroy(fb); >>>>> } >>>>> - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, >>>>> head) { >>>>> - crtc->funcs->destroy(crtc); >>>>> - } >>>>> - >>>>> list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, >>>>> head) { >>>>> plane->funcs->destroy(plane); >>>>> } >>>>> + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, >>>>> head) { >>>>> + crtc->funcs->destroy(crtc); >>>>> + } >>>>> + >>>>> idr_remove_all(&dev->mode_config.crtc_idr); >>>>> idr_destroy(&dev->mode_config.crtc_idr); >>>>> } >>>> _______________________________________________ >>>> dri-devel mailing list >>>> dri-devel@lists.freedesktop.org >>>> http://lists.freedesktop.org/mailman/listinfo/dri-devel > From: Chris Wilson <chris@chris-wilson.co.uk> > Subject: Re: [PATCH] drm: Destroy the planes prior to destroying the associated CRTC > To: Rob Clark <rob.clark@linaro.org>, Joonyoung Shim <jy0922.shim@samsung.com> > Cc: airlied@redhat.com, stable@vger.kernel.org, dri-devel@lists.freedesktop.org > In-Reply-To: <CAF6AEGvOZa_ZhRJN212Rsn-gMMUyWoLT6UFY9iindi8AWx9GvA@mail.gmail.com> > References: <1347874683-21191-1-git-send-email-chris@chris-wilson.co.uk> <505A850A.4010205@samsung.com> <CAF6AEGvOZa_ZhRJN212Rsn-gMMUyWoLT6UFY9iindi8AWx9GvA@mail.gmail.com> > > On Thu, 20 Sep 2012 00:38:04 -0500, Rob Clark <rob.clark@linaro.org> wrote: >> On Wed, Sep 19, 2012 at 9:52 PM, Joonyoung Shim <jy0922.shim@samsung.com> wrote: >>> On 09/17/2012 06:38 PM, Chris Wilson wrote: >>>> As during the plane cleanup, we wish to disable the hardware and >>>> so may modify state on the associated CRTC, that CRTC must continue to >>>> exist until we are finished. >>> >>> A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and >>> plane use same framebuffer and the framebuffer is destroyed, crtc is >>> turned off prior to turning off plane. >>> >> I imagine my patch to add refcnt'ing to fb would help in this case.. >> >> BR, >> -R >> >>>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 >>>> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> >>>> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> >>>> Cc: stable@vger.kernel.org >>>> --- >>>> drivers/gpu/drm/drm_crtc.c | 8 ++++---- >>>> 1 file changed, 4 insertions(+), 4 deletions(-) >>>> >>>> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c >>>> index 6fbfc24..af81f77 100644 >>>> --- a/drivers/gpu/drm/drm_crtc.c >>>> +++ b/drivers/gpu/drm/drm_crtc.c >>>> @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device >>>> *dev) >>>> fb->funcs->destroy(fb); >>>> } >>>> - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, >>>> head) { >>>> - crtc->funcs->destroy(crtc); >>>> - } >>>> - >>>> list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, >>>> head) { >>>> plane->funcs->destroy(plane); >>>> } >>>> + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, >>>> head) { >>>> + crtc->funcs->destroy(crtc); >>>> + } >>>> + >>>> idr_remove_all(&dev->mode_config.crtc_idr); >>>> idr_destroy(&dev->mode_config.crtc_idr); >>>> } >>> >>> _______________________________________________ >>> dri-devel mailing list >>> dri-devel@lists.freedesktop.org >>> http://lists.freedesktop.org/mailman/listinfo/dri-devel > From: Chris Wilson <chris@chris-wilson.co.uk> > Subject: Re: [PATCH] drm: Destroy the planes prior to destroying the associated CRTC > To: Joonyoung Shim <jy0922.shim@samsung.com> > Cc: dri-devel@lists.freedesktop.org, airlied@redhat.com, stable@vger.kernel.org > In-Reply-To: <505A850A.4010205@samsung.com> > References: <1347874683-21191-1-git-send-email-chris@chris-wilson.co.uk> <505A850A.4010205@samsung.com> > > On Thu, 20 Sep 2012 11:52:58 +0900, Joonyoung Shim <jy0922.shim@samsung.com> wrote: >> On 09/17/2012 06:38 PM, Chris Wilson wrote: >>> As during the plane cleanup, we wish to disable the hardware and >>> so may modify state on the associated CRTC, that CRTC must continue to >>> exist until we are finished. >> A similar issue can occur in the drm_framebuffer_cleanup(). If crtc and >> plane use same framebuffer and the framebuffer is destroyed, crtc is >> turned off prior to turning off plane. > This is not an issue in our code as disabling the CRTCs should > automatically disable any associated planes, and so the second pass over > the planes should be a no-op. Right, but it is decided by each hw specific driver implementation. If drm core can prevent any problem, it's better. > The issue during destroy drm_mode_config_cleanup() is that we actually > free the CRTC objects and then try to decouple the planes which causes > us to reference the just-freed objects in order to see if the hw needs > updating. > > However, reordering the sequence to be CRTCs last would help for > consistency. Agree.
diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c index 6fbfc24..af81f77 100644 --- a/drivers/gpu/drm/drm_crtc.c +++ b/drivers/gpu/drm/drm_crtc.c @@ -1034,15 +1034,15 @@ void drm_mode_config_cleanup(struct drm_device *dev) fb->funcs->destroy(fb); } - list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, head) { - crtc->funcs->destroy(crtc); - } - list_for_each_entry_safe(plane, plt, &dev->mode_config.plane_list, head) { plane->funcs->destroy(plane); } + list_for_each_entry_safe(crtc, ct, &dev->mode_config.crtc_list, head) { + crtc->funcs->destroy(crtc); + } + idr_remove_all(&dev->mode_config.crtc_idr); idr_destroy(&dev->mode_config.crtc_idr); }
As during the plane cleanup, we wish to disable the hardware and so may modify state on the associated CRTC, that CRTC must continue to exist until we are finished. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54101 Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> Cc: stable@vger.kernel.org --- drivers/gpu/drm/drm_crtc.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)