When initially turning a crtc on, drm_reset_vblank_timestamp will set the vblank timestamp to 0 for any driver that doesn't provide a ->get_vblank_timestamp() hook.
Unfortunately, the FLIP_COMPLETE event depends on that timestamp, and the only way to regenerate a valid one is to have vblank interrupts enabled and have a valid in-ISR call to drm_crtc_handle_vblank.
Wrap the call to komeda_crtc_do_flush in ->atomic_enable() with a drm_crtc_vblank_{get,put} pair so we can have a vblank ISR prior to the FLIP_COMPLETE ISR (or more likely, they'll get handled in the same ISR, which is equally valid).
Cc: Daniel Vetter daniel@ffwll.ch Cc: Liviu Dudau Liviu.Dudau@arm.com Signed-off-by: Mihail Atanassov mihail.atanassov@arm.com --- drivers/gpu/drm/arm/display/komeda/komeda_crtc.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c index f4400788ab94..87420a767bc4 100644 --- a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c +++ b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c @@ -258,7 +258,9 @@ komeda_crtc_atomic_enable(struct drm_crtc *crtc, { komeda_crtc_prepare(to_kcrtc(crtc)); drm_crtc_vblank_on(crtc); + WARN_ON(drm_crtc_vblank_get(crtc)); komeda_crtc_do_flush(crtc, old); + drm_crtc_vblank_put(crtc); }
static void
On Thursday, 19 September 2019 14:30:02 BST Mihail Atanassov wrote:
When initially turning a crtc on, drm_reset_vblank_timestamp will set the vblank timestamp to 0 for any driver that doesn't provide a ->get_vblank_timestamp() hook.
Unfortunately, the FLIP_COMPLETE event depends on that timestamp, and the only way to regenerate a valid one is to have vblank interrupts enabled and have a valid in-ISR call to drm_crtc_handle_vblank.
Wrap the call to komeda_crtc_do_flush in ->atomic_enable() with a drm_crtc_vblank_{get,put} pair so we can have a vblank ISR prior to the FLIP_COMPLETE ISR (or more likely, they'll get handled in the same ISR, which is equally valid).
Cc: Daniel Vetter daniel@ffwll.ch Cc: Liviu Dudau Liviu.Dudau@arm.com Signed-off-by: Mihail Atanassov mihail.atanassov@arm.com
drivers/gpu/drm/arm/display/komeda/komeda_crtc.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c index f4400788ab94..87420a767bc4 100644 --- a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c +++ b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c @@ -258,7 +258,9 @@ komeda_crtc_atomic_enable(struct drm_crtc *crtc, { komeda_crtc_prepare(to_kcrtc(crtc)); drm_crtc_vblank_on(crtc);
- WARN_ON(drm_crtc_vblank_get(crtc)); komeda_crtc_do_flush(crtc, old);
- drm_crtc_vblank_put(crtc);
}
static void
Please ignore, this doesn't work (page flips after 5 seconds don't get an updated timestamp because of the delayed vsync disable). I'll push a v2 soon where vsync on/off tracks crtc enable/disable.
When initially turning a crtc on, drm_reset_vblank_timestamp will set the vblank timestamp to 0 for any driver that doesn't provide a ->get_vblank_timestamp() hook.
Unfortunately, the FLIP_COMPLETE event depends on that timestamp, and the only way to regenerate a valid one is to have vblank interrupts enabled and have a valid in-ISR call to drm_crtc_handle_vblank.
Additionally, if the user doesn't request vblanks but _does_ request FLIP_COMPLETE events, we still don't have a good timestamp: it'll be the same stamp as the last vblank one.
Work around the issue by always enabling vblanks when the CRTC is on. Reducing the amount of time that PL0 has to be unmasked would be nice to fix at a later time.
Changes since v1 [https://patchwork.freedesktop.org/patch/331727/]: - moved drm_crtc_vblank_put call to the ->atomic_disable() hook
Cc: Daniel Vetter daniel@ffwll.ch Cc: Liviu Dudau Liviu.Dudau@arm.com Signed-off-by: Mihail Atanassov mihail.atanassov@arm.com --- drivers/gpu/drm/arm/display/komeda/komeda_crtc.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c index 34bc73ca18bc..d06679afb228 100644 --- a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c +++ b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c @@ -489,6 +489,7 @@ komeda_crtc_atomic_enable(struct drm_crtc *crtc, pm_runtime_get_sync(crtc->dev->dev); komeda_crtc_prepare(to_kcrtc(crtc)); drm_crtc_vblank_on(crtc); + WARN_ON(drm_crtc_vblank_get(crtc)); komeda_crtc_do_flush(crtc, old); }
@@ -581,6 +582,7 @@ komeda_crtc_atomic_disable(struct drm_crtc *crtc, komeda_crtc_flush_and_wait_for_flip_done(kcrtc, disable_done); }
+ drm_crtc_vblank_put(crtc); drm_crtc_vblank_off(crtc); komeda_crtc_unprepare(kcrtc); pm_runtime_put(crtc->dev->dev);
On Mon, Sep 23, 2019 at 10:10:26AM +0000, Mihail Atanassov wrote:
When initially turning a crtc on, drm_reset_vblank_timestamp will set the vblank timestamp to 0 for any driver that doesn't provide a ->get_vblank_timestamp() hook.
Unfortunately, the FLIP_COMPLETE event depends on that timestamp, and the only way to regenerate a valid one is to have vblank interrupts enabled and have a valid in-ISR call to drm_crtc_handle_vblank.
Additionally, if the user doesn't request vblanks but _does_ request FLIP_COMPLETE events, we still don't have a good timestamp: it'll be the same stamp as the last vblank one.
Work around the issue by always enabling vblanks when the CRTC is on. Reducing the amount of time that PL0 has to be unmasked would be nice to fix at a later time.
Changes since v1 [https://patchwork.freedesktop.org/patch/331727/]:
- moved drm_crtc_vblank_put call to the ->atomic_disable() hook
Cc: Daniel Vetter daniel@ffwll.ch Cc: Liviu Dudau Liviu.Dudau@arm.com Signed-off-by: Mihail Atanassov mihail.atanassov@arm.com
drivers/gpu/drm/arm/display/komeda/komeda_crtc.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c index 34bc73ca18bc..d06679afb228 100644 --- a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c +++ b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c @@ -489,6 +489,7 @@ komeda_crtc_atomic_enable(struct drm_crtc *crtc, pm_runtime_get_sync(crtc->dev->dev); komeda_crtc_prepare(to_kcrtc(crtc)); drm_crtc_vblank_on(crtc);
- WARN_ON(drm_crtc_vblank_get(crtc)); komeda_crtc_do_flush(crtc, old);
}
@@ -581,6 +582,7 @@ komeda_crtc_atomic_disable(struct drm_crtc *crtc, komeda_crtc_flush_and_wait_for_flip_done(kcrtc, disable_done); }
- drm_crtc_vblank_put(crtc); drm_crtc_vblank_off(crtc); komeda_crtc_unprepare(kcrtc); pm_runtime_put(crtc->dev->dev);
Looks good to me.
Reviewed-by: James Qian Wang (Arm Technology China) james.qian.wang@arm.com
When initially turning a crtc on, drm_reset_vblank_timestamp will set the vblank timestamp to 0 for any driver that doesn't provide a ->get_vblank_timestamp() hook.
Unfortunately, the FLIP_COMPLETE event depends on that timestamp, and the only way to regenerate a valid one is to have vblank interrupts enabled and have a valid in-ISR call to drm_crtc_handle_vblank.
Additionally, if the user doesn't request vblanks but _does_ request FLIP_COMPLETE events, we still don't have a good timestamp: it'll be the same stamp as the last vblank one.
Work around the issue by always enabling vblanks when the CRTC is on. Reducing the amount of time that PL0 has to be unmasked would be nice to fix at a later time.
Changes since v1 [https://patchwork.freedesktop.org/patch/331727/]: - moved drm_crtc_vblank_put call to the ->atomic_disable() hook
Cc: Daniel Vetter daniel@ffwll.ch Cc: Liviu Dudau Liviu.Dudau@arm.com Signed-off-by: Mihail Atanassov mihail.atanassov@arm.com Reviewed-by: James Qian Wang (Arm Technology China) james.qian.wang@arm.com --- drivers/gpu/drm/arm/display/komeda/komeda_crtc.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c index 9ca5dbfd0723..75263d8cd0bd 100644 --- a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c +++ b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c @@ -249,6 +249,7 @@ komeda_crtc_atomic_enable(struct drm_crtc *crtc, { komeda_crtc_prepare(to_kcrtc(crtc)); drm_crtc_vblank_on(crtc); + WARN_ON(drm_crtc_vblank_get(crtc)); komeda_crtc_do_flush(crtc, old); }
@@ -341,6 +342,7 @@ komeda_crtc_atomic_disable(struct drm_crtc *crtc, komeda_crtc_flush_and_wait_for_flip_done(kcrtc, disable_done); }
+ drm_crtc_vblank_put(crtc); drm_crtc_vblank_off(crtc); komeda_crtc_unprepare(kcrtc); }
On Tue, Oct 01, 2019 at 02:21:40PM +0000, Mihail Atanassov wrote:
When initially turning a crtc on, drm_reset_vblank_timestamp will set the vblank timestamp to 0 for any driver that doesn't provide a ->get_vblank_timestamp() hook.
Unfortunately, the FLIP_COMPLETE event depends on that timestamp, and the only way to regenerate a valid one is to have vblank interrupts enabled and have a valid in-ISR call to drm_crtc_handle_vblank.
Additionally, if the user doesn't request vblanks but _does_ request FLIP_COMPLETE events, we still don't have a good timestamp: it'll be the same stamp as the last vblank one.
Work around the issue by always enabling vblanks when the CRTC is on. Reducing the amount of time that PL0 has to be unmasked would be nice to fix at a later time.
Changes since v1 [https://patchwork.freedesktop.org/patch/331727/]:
- moved drm_crtc_vblank_put call to the ->atomic_disable() hook
Cc: Daniel Vetter daniel@ffwll.ch Cc: Liviu Dudau Liviu.Dudau@arm.com Signed-off-by: Mihail Atanassov mihail.atanassov@arm.com Reviewed-by: James Qian Wang (Arm Technology China) james.qian.wang@arm.com
Pushed to drm-misc-next f59769c52cd7d158df53487ec2936f5592073340
Thanks, Ayan
drivers/gpu/drm/arm/display/komeda/komeda_crtc.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c index 9ca5dbfd0723..75263d8cd0bd 100644 --- a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c +++ b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c @@ -249,6 +249,7 @@ komeda_crtc_atomic_enable(struct drm_crtc *crtc, { komeda_crtc_prepare(to_kcrtc(crtc)); drm_crtc_vblank_on(crtc);
- WARN_ON(drm_crtc_vblank_get(crtc)); komeda_crtc_do_flush(crtc, old);
}
@@ -341,6 +342,7 @@ komeda_crtc_atomic_disable(struct drm_crtc *crtc, komeda_crtc_flush_and_wait_for_flip_done(kcrtc, disable_done); }
- drm_crtc_vblank_put(crtc); drm_crtc_vblank_off(crtc); komeda_crtc_unprepare(kcrtc);
}
2.23.0
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
dri-devel@lists.freedesktop.org
-
Ayan Halder -
james qian wang (Arm Technology China) -
Mihail Atanassov