The Allwinner backend has a commit bit in order to push the new configuration to the actual hardware. We've always been using that bit.
However, we also should poll for that bit to clear, which we don't. Accessing any register while a commit is pending is forbidden, and will for example show a symptom of reading another, random, register.
If you get this during a read/modify/write cycle, this will result in random register corruption, which are pretty bad.
This can be shown using the following program (while the backend is active): http://code.bulix.org/gdl44p-161437?raw
Fortunately for us, this is not really likely to happen. The window where it can happen is quite thin, and it only happens during a modeset, since it's the only time we commit some new configuration.
Unfortunately for us, QT does a ridiculous amount of modeset, and will just hit that window after a while, creating a distorded (since the register we read/modify/write also has scaling attributes) or with weird colors (since it also has a invertion of red and blue components). And might fix itself later on by reading another random register with proper values for these fields.
Let me know what you think, Maxime
Changes from v1: - Renamed the function drm_atomic_helper_commit_tail_rpm - Dropped the changes in the rcar-du driver - Enhanced the documentation based on Daniel's comments
Maxime Ripard (4): drm/atomic: implement drm_atomic_helper_commit_tail for runtime_pm users drm/sun4i: Use the runtime_pm commit_tail variant drm/sun4i: engine: Add commit_poll function drm/sun4i: make sure we don't have a commit pending
drivers/gpu/drm/drm_atomic_helper.c | 49 +++++++++++++++-------- drivers/gpu/drm/exynos/exynos_drm_fb.c | 27 +------------- drivers/gpu/drm/rockchip/rockchip_drm_fb.c | 21 +---------- drivers/gpu/drm/sun4i/sun4i_backend.c | 14 +++++++- drivers/gpu/drm/sun4i/sun4i_crtc.c | 3 +- drivers/gpu/drm/sun4i/sun4i_framebuffer.c | 6 +++- drivers/gpu/drm/sun4i/sunxi_engine.h | 14 +++++++- include/drm/drm_atomic_helper.h | 1 +- 8 files changed, 74 insertions(+), 61 deletions(-)
base-commit: 5771a8c08880cdca3bfb4a3fc6d309d6bba20877
The current drm_atomic_helper_commit_tail helper works only if the CRTC is accessible, and documents an alternative implementation that is supposed to be used if that happens.
That implementation is then duplicated by some drivers. Instead of documenting it, let's implement an helper that all the relevant users can use directly.
Signed-off-by: Maxime Ripard maxime.ripard@free-electrons.com --- drivers/gpu/drm/drm_atomic_helper.c | 49 +++++++++++++++-------- drivers/gpu/drm/exynos/exynos_drm_fb.c | 27 +------------- drivers/gpu/drm/rockchip/rockchip_drm_fb.c | 21 +---------- include/drm/drm_atomic_helper.h | 1 +- 4 files changed, 37 insertions(+), 61 deletions(-)
diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c index 86d3093c6c9b..d06a65ed3a57 100644 --- a/drivers/gpu/drm/drm_atomic_helper.c +++ b/drivers/gpu/drm/drm_atomic_helper.c @@ -1245,23 +1245,13 @@ EXPORT_SYMBOL(drm_atomic_helper_wait_for_vblanks); * @old_state: atomic state object with old state structures * * This is the default implementation for the - * &drm_mode_config_helper_funcs.atomic_commit_tail hook. + * &drm_mode_config_helper_funcs.atomic_commit_tail hook, for drivers + * that do not support runtime_pm or do not need the CRTC to be + * enabled to perform a commit. Otherwise, see + * drm_atomic_helper_commit_tail_rpm(). * * Note that the default ordering of how the various stages are called is to - * match the legacy modeset helper library closest. One peculiarity of that is - * that it doesn't mesh well with runtime PM at all. - * - * For drivers supporting runtime PM the recommended sequence is instead :: - * - * drm_atomic_helper_commit_modeset_disables(dev, old_state); - * - * drm_atomic_helper_commit_modeset_enables(dev, old_state); - * - * drm_atomic_helper_commit_planes(dev, old_state, - * DRM_PLANE_COMMIT_ACTIVE_ONLY); - * - * for committing the atomic update to hardware. See the kerneldoc entries for - * these three functions for more details. + * match the legacy modeset helper library closest. */ void drm_atomic_helper_commit_tail(struct drm_atomic_state *old_state) { @@ -1281,6 +1271,35 @@ void drm_atomic_helper_commit_tail(struct drm_atomic_state *old_state) } EXPORT_SYMBOL(drm_atomic_helper_commit_tail);
+/** + * drm_atomic_helper_commit_tail_rpm - commit atomic update to hardware + * @old_state: new modeset state to be committed + * + * This is an alternative implementation for the + * &drm_mode_config_helper_funcs.atomic_commit_tail hook, for drivers + * that support runtime_pm or need the CRTC to be enabled to perform a + * commit. Otherwise, one should use the default implementation + * drm_atomic_helper_commit_tail(). + */ +void drm_atomic_helper_commit_tail_rpm(struct drm_atomic_state *old_state) +{ + struct drm_device *dev = old_state->dev; + + drm_atomic_helper_commit_modeset_disables(dev, old_state); + + drm_atomic_helper_commit_modeset_enables(dev, old_state); + + drm_atomic_helper_commit_planes(dev, old_state, + DRM_PLANE_COMMIT_ACTIVE_ONLY); + + drm_atomic_helper_commit_hw_done(old_state); + + drm_atomic_helper_wait_for_vblanks(dev, old_state); + + drm_atomic_helper_cleanup_planes(dev, old_state); +} +EXPORT_SYMBOL(drm_atomic_helper_commit_tail_rpm); + static void commit_tail(struct drm_atomic_state *old_state) { struct drm_device *dev = old_state->dev; diff --git a/drivers/gpu/drm/exynos/exynos_drm_fb.c b/drivers/gpu/drm/exynos/exynos_drm_fb.c index d48fd7c918f8..ed1a648d518c 100644 --- a/drivers/gpu/drm/exynos/exynos_drm_fb.c +++ b/drivers/gpu/drm/exynos/exynos_drm_fb.c @@ -187,33 +187,8 @@ dma_addr_t exynos_drm_fb_dma_addr(struct drm_framebuffer *fb, int index) return exynos_fb->dma_addr[index]; }
-static void exynos_drm_atomic_commit_tail(struct drm_atomic_state *state) -{ - struct drm_device *dev = state->dev; - - drm_atomic_helper_commit_modeset_disables(dev, state); - - drm_atomic_helper_commit_modeset_enables(dev, state); - - /* - * Exynos can't update planes with CRTCs and encoders disabled, - * its updates routines, specially for FIMD, requires the clocks - * to be enabled. So it is necessary to handle the modeset operations - * *before* the commit_planes() step, this way it will always - * have the relevant clocks enabled to perform the update. - */ - drm_atomic_helper_commit_planes(dev, state, - DRM_PLANE_COMMIT_ACTIVE_ONLY); - - drm_atomic_helper_commit_hw_done(state); - - drm_atomic_helper_wait_for_vblanks(dev, state); - - drm_atomic_helper_cleanup_planes(dev, state); -} - static struct drm_mode_config_helper_funcs exynos_drm_mode_config_helpers = { - .atomic_commit_tail = exynos_drm_atomic_commit_tail, + .atomic_commit_tail = drm_atomic_helper_commit_tail_rpm, };
static const struct drm_mode_config_funcs exynos_drm_mode_config_funcs = { diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_fb.c b/drivers/gpu/drm/rockchip/rockchip_drm_fb.c index 81f9548672b0..35ad386c401e 100644 --- a/drivers/gpu/drm/rockchip/rockchip_drm_fb.c +++ b/drivers/gpu/drm/rockchip/rockchip_drm_fb.c @@ -174,27 +174,8 @@ static void rockchip_drm_output_poll_changed(struct drm_device *dev) drm_fb_helper_hotplug_event(fb_helper); }
-static void -rockchip_atomic_commit_tail(struct drm_atomic_state *state) -{ - struct drm_device *dev = state->dev; - - drm_atomic_helper_commit_modeset_disables(dev, state); - - drm_atomic_helper_commit_modeset_enables(dev, state); - - drm_atomic_helper_commit_planes(dev, state, - DRM_PLANE_COMMIT_ACTIVE_ONLY); - - drm_atomic_helper_commit_hw_done(state); - - drm_atomic_helper_wait_for_vblanks(dev, state); - - drm_atomic_helper_cleanup_planes(dev, state); -} - static const struct drm_mode_config_helper_funcs rockchip_mode_config_helpers = { - .atomic_commit_tail = rockchip_atomic_commit_tail, + .atomic_commit_tail = drm_atomic_helper_commit_tail_rpm, };
static const struct drm_mode_config_funcs rockchip_drm_mode_config_funcs = { diff --git a/include/drm/drm_atomic_helper.h b/include/drm/drm_atomic_helper.h index f0a8678ae98e..af4aaff9ec0b 100644 --- a/include/drm/drm_atomic_helper.h +++ b/include/drm/drm_atomic_helper.h @@ -41,6 +41,7 @@ int drm_atomic_helper_check_planes(struct drm_device *dev, int drm_atomic_helper_check(struct drm_device *dev, struct drm_atomic_state *state); void drm_atomic_helper_commit_tail(struct drm_atomic_state *state); +void drm_atomic_helper_commit_tail_rpm(struct drm_atomic_state *state); int drm_atomic_helper_commit(struct drm_device *dev, struct drm_atomic_state *state, bool nonblock);
On Thu, Jul 20, 2017 at 03:01:16PM +0200, Maxime Ripard wrote:
Reviewed-by: Daniel Vetter daniel.vetter@ffwll.ch
Should I throw this into drm-misc, or do you want to merge this through some driver tree? -Daniel
Hi Daniel,
On Thu, Jul 20, 2017 at 08:46:28PM +0200, Daniel Vetter wrote:
Yes, you can put them in drm-misc, it will the be the easiest option.
Thanks! Maxime
On Tue, Jul 25, 2017 at 08:57:34AM +0200, Maxime Ripard wrote:
Merged, thanks. -Daniel
Hi Maxime,
On Thu, Jul 20, 2017 at 03:01:16PM +0200, Maxime Ripard wrote:
Sorry for not noticing the series earlier, I was fooled by the series top commit summary into thinking it was a driver specific change and only paid proper attention when it ended up in drm-next. I have a comment to make though.
Given that this function is supposed to be used by runtime PM aware drivers and that the hook is called from the DRM core, should there not be some pm_runtime_{get,put} calls wrapping the body of this function?
Best regards, Liviu
On Wed, Aug 2, 2017 at 1:20 PM, Liviu Dudau Liviu.Dudau@arm.com wrote:
No, because the drm atomic helpers have no idea which device is backing which part of the drm device. Some drivers might on have one device for the entire driver, some one device for just the display part (which might or might not match drm_device->dev). And many arm drivers have a device for each block separately (and you need to call rpm_get/put on each). And some something in-between (e.g. core device + external encoders).
I don't think there's anything the helpers can to to help support this.
Also, just wrapping functions with rpm_get/put only papers over bugs in your driver - either you enable something, then the rpm_get needs to be done anyway (since the hw will be shut down otherwise), or you disable something, same reasons. The only thing a rpm_get/put does is paper over the bugs where you try to access the hw when it's off. As soon as this function finishes, the hw is shut down again, drops all register values on the floor, so either that access wasn't needed, or your driver has a bug (because it assumes the register values survive when they don't).
So imo all around a bad idea, at least from my experience of doing rpm enabling on i915 hw. -Daniel
On Wed, Aug 02, 2017 at 01:27:23PM +0200, Daniel Vetter wrote:
Hi Daniel,
Hmm, I understand your point about this function not having to care about pm_runtime_{get,put}, but I do not agree with your view that if it wanted to care about it, it wouldn't be able to do the right thing because it doesn't have the right device. After all, this function is about handling the updates that this atomic commit is concerned about, so having the old_state->dev drm_device pointer and its associated device should be enough. Am I missing something?
Understood. Thanks!
Best regards, Liviu
Hi Livu,
On Wednesday 02 Aug 2017 13:46:48 Liviu Dudau wrote:
In embedded system it's quite common for display hardware to be made of multiple IP cores. Depending on the SoC you could have to manage runtime PM at the CRTC level for instance. The DRM core doesn't know about that, and sees a single device only.
I've expressed my doubts previously about the need for a RPM version of drm_atomic_helper_commit_tail(), as the resulting order of CRTC enable/disable and plane update operations can lead to corrupt frames when enabling the CRTC. I had a commit tail implementation in the rcar-du driver that was very similar to drm_atomic_helper_commit_tail_rpm(), and had to move back to the standard order to fix the corrupt frame issue. The result isn't as clean as I would like, as power handling is split between the CRTC enable/disable and the atomic begin/flush in a way that is not straightforward.
It now occurred to me that a simpler implementation could be possible. I'll have to experiment with it first, but the idea is as follows.
1 Call pm_runtime_get_sync() at the beginning of .commit_tail() and pm_runtime_put() at the end.
2. Use the standard CRTC disable, plane update, CRTC enable order in .commit_tail().
3. Call pm_runtime_get() in the CRTC .enable() handler and pm_runtime_put() in the CRTC .disable() handler;
This should guarantee that the device won't be suspended between CRTC disable and CRTC enable during a mode set, without requiring any special collaboration between CRTC enable/disable and atomic begin/flush to handle runtime PM. If drivers implement this, there should be no need for drm_atomic_helper_commit_tail_rpm().
Maxime, Daniel, what do you think about this ?
On Wed, Aug 02, 2017 at 04:27:27PM +0300, Laurent Pinchart wrote:
Well, that is what mali-dp driver currently does, but according to Daniel (and I can see his POV) that is wrong. I'm playing with removing all of that to see if there are any side effects in Mali DP like the ones you mentioned for RCAR.
Best regards, Liviu
Hi Liviu,
On Wednesday 02 Aug 2017 14:32:06 Liviu Dudau wrote:
Is it ? I might not have understood his arguments the same way (or possibly failed to even see them). Are you referring to this comments in this mail thread, or to something else ?
I'm playing with removing all of that to see if there are any side effects in Mali DP like the ones you mentioned for RCAR.
Note that the first frame will usually not be noticed as it often takes a few frames for the display to turn on.
On Wed, Aug 02, 2017 at 04:49:16PM +0300, Laurent Pinchart wrote:
I'm talking about his reply above. My understanding: you can't do pm_runtime_{get,set} in the atomic_commit_tail hook because:
1. you don't know if you have the correct drm_device->dev (or all the relevant devices) to call pm_runtime_get() on. 2. If pm_runtime_get() affects in any way the atomic commit behaviour by being at the top of the commit_tail_rpm() function then you are: a) papering over bugs in one's driver b) going to turn off things at the end of commit_tail_rpm() function when you call pm_runtime_put() so your changes are going to be lost.
Yes, and I have a TV connected to the output that takes ages to sync. But I can usually run some quick rmmmod+insmod tests and the TV would be too slow to turn off the output, so I can see if there are any artifacts.
Best regards, Liviu
Hi Liviu,
On Wednesday 02 Aug 2017 14:57:30 Liviu Dudau wrote:
You can't call pm_runtime_get() in the DRM core for that exact reason, but you can call it in a driver's implementation of .atomic_commit_tail(), which was my proposal. The .atomic_commit_tail() handler would then become something like
{ for_each_ip_affected(ip, ...) pm_runtime_get_sync(ip);
drm_atomic_helper_commit_tail(...);
for_each_ip_affected(ip, ...) pm_runtime_put(ip); }
You won't, because you also call pm_runtime_get() in the CRTC .enable() handler.
One good way to test this is to implement support for CRC calculation if your hardware supports it.
[snip]
On Wed, Aug 2, 2017 at 3:27 PM, Laurent Pinchart laurent.pinchart@ideasonboard.com wrote:
Ok, since you just said on irc that the corrupted frame is some random color, that's indeed not cool. What all other drivers (well at least i915) do is:
1. Enable the screen, but scan out nothing as a pure black. This might mean you need to set up the blend unit with a background color of black. 2. Do an atomic flip to the new screen contents.
This way you get the sequence implement in the new _rpm implementation. And for most normal hw this gives you the simplest implementation, and has the benefit that writing to disabled hw blocks indicates a bug. Wrapping an entire sequence with rpm_get/put like Laurent describes is a really good way to hide bugs (e.g. write new stuff to hw you're about to disable and then drop the hw settings on the floor). Yes it makes implementation a bit easier, but easier != more correct in many cases. Nasty self-checks are good for getting kms implementation rights.
Another benefit of this sequence is that the initial plane enable after a crtc enable is like any other "background black -> planes" atomic switch, and atomic does allow you to disable all planes. So your driver better supports that (or is one of the few which has to reject a config without any planes), since userspace might want to enable the CRTC with no planes, and then you have the exact same bug again.
It sounds like Laurent's 3rd gen rcar-du is special, and so needs special code (or well just a bugfix to enable the composer with all black in the CRTC enable code.
Cheers, Daniel
The backend (planes) commit can only happen when the TCON (CRTC) is enabled, which is not guaranteed with the default commit_tail helper.
Let's use the runtime_pm version that is designed specifically to deal with that case.
Signed-off-by: Maxime Ripard maxime.ripard@free-electrons.com --- drivers/gpu/drm/sun4i/sun4i_framebuffer.c | 6 ++++++ 1 file changed, 6 insertions(+)
diff --git a/drivers/gpu/drm/sun4i/sun4i_framebuffer.c b/drivers/gpu/drm/sun4i/sun4i_framebuffer.c index 9872e0fc03b0..53ab61665390 100644 --- a/drivers/gpu/drm/sun4i/sun4i_framebuffer.c +++ b/drivers/gpu/drm/sun4i/sun4i_framebuffer.c @@ -10,6 +10,7 @@ * the License, or (at your option) any later version. */
+#include <drm/drm_atomic.h> #include <drm/drm_atomic_helper.h> #include <drm/drm_fb_cma_helper.h> #include <drm/drmP.h> @@ -31,6 +32,10 @@ static const struct drm_mode_config_funcs sun4i_de_mode_config_funcs = { .fb_create = drm_fb_cma_create, };
+static struct drm_mode_config_helper_funcs sun4i_de_mode_config_helpers = { + .atomic_commit_tail = drm_atomic_helper_commit_tail_rpm, +}; + struct drm_fbdev_cma *sun4i_framebuffer_init(struct drm_device *drm) { drm_mode_config_reset(drm); @@ -39,6 +44,7 @@ struct drm_fbdev_cma *sun4i_framebuffer_init(struct drm_device *drm) drm->mode_config.max_height = 8192;
drm->mode_config.funcs = &sun4i_de_mode_config_funcs; + drm->mode_config.helper_private = &sun4i_de_mode_config_helpers;
return drm_fbdev_cma_init(drm, 32, drm->mode_config.num_connector); }
On the earlier Allwinner chips, with the first iteration of the display engine, the backend commit bit needs to be polled before making any register access to the backend.
Add an operation for that, that will be called in atomic_begin in order to be sure to have that bit cleared before we do any modifications.
Signed-off-by: Maxime Ripard maxime.ripard@free-electrons.com --- drivers/gpu/drm/sun4i/sun4i_crtc.c | 2 ++ drivers/gpu/drm/sun4i/sunxi_engine.h | 14 ++++++++++++++ 2 files changed, 16 insertions(+)
diff --git a/drivers/gpu/drm/sun4i/sun4i_crtc.c b/drivers/gpu/drm/sun4i/sun4i_crtc.c index f8c70439d1e2..2eba1d8639d8 100644 --- a/drivers/gpu/drm/sun4i/sun4i_crtc.c +++ b/drivers/gpu/drm/sun4i/sun4i_crtc.c @@ -45,6 +45,8 @@ static void sun4i_crtc_atomic_begin(struct drm_crtc *crtc, spin_unlock_irqrestore(&dev->event_lock, flags); crtc->state->event = NULL; } + + WARN_ON(sunxi_engine_commit_poll(engine)); }
static void sun4i_crtc_atomic_flush(struct drm_crtc *crtc, diff --git a/drivers/gpu/drm/sun4i/sunxi_engine.h b/drivers/gpu/drm/sun4i/sunxi_engine.h index 4cb70ae65c79..6618e182613c 100644 --- a/drivers/gpu/drm/sun4i/sunxi_engine.h +++ b/drivers/gpu/drm/sun4i/sunxi_engine.h @@ -17,6 +17,7 @@ struct sunxi_engine;
struct sunxi_engine_ops { void (*commit)(struct sunxi_engine *engine); + int (*commit_poll)(struct sunxi_engine *engine); struct drm_plane **(*layers_init)(struct drm_device *drm, struct sunxi_engine *engine);
@@ -55,6 +56,19 @@ sunxi_engine_commit(struct sunxi_engine *engine) }
/** + * sunxi_engine_commit_poll() - wait for all changes to be committed + * @engine: pointer to the engine + */ +static inline int +sunxi_engine_commit_poll(struct sunxi_engine *engine) +{ + if (engine->ops && engine->ops->commit_poll) + return engine->ops->commit_poll(engine); + + return 0; +} + +/** * sunxi_engine_layers_init() - Create planes (layers) for the engine * @drm: pointer to the drm_device for which planes will be created * @engine: pointer to the engine
In the earlier display engine designs, any register access while a commit is pending is forbidden.
One of the symptoms is that reading a register will return another, random, register value which can lead to register corruptions if we ever do a read/modify/write cycle.
Signed-off-by: Maxime Ripard maxime.ripard@free-electrons.com --- drivers/gpu/drm/sun4i/sun4i_backend.c | 14 ++++++++++++++ drivers/gpu/drm/sun4i/sun4i_crtc.c | 1 + 2 files changed, 15 insertions(+)
diff --git a/drivers/gpu/drm/sun4i/sun4i_backend.c b/drivers/gpu/drm/sun4i/sun4i_backend.c index cf480218daa5..ce1f40f7511e 100644 --- a/drivers/gpu/drm/sun4i/sun4i_backend.c +++ b/drivers/gpu/drm/sun4i/sun4i_backend.c @@ -67,6 +67,19 @@ static void sun4i_backend_commit(struct sunxi_engine *engine) SUN4I_BACKEND_REGBUFFCTL_LOADCTL); }
+static int sun4i_backend_commit_poll(struct sunxi_engine *engine) +{ + u32 val; + + DRM_DEBUG_DRIVER("Polling for the commit to end\n"); + + return regmap_read_poll_timeout(engine->regs, + SUN4I_BACKEND_REGBUFFCTL_REG, + val, + !(val & SUN4I_BACKEND_REGBUFFCTL_LOADCTL), + 100, 50000); +} + void sun4i_backend_layer_enable(struct sun4i_backend *backend, int layer, bool enable) { @@ -330,6 +343,7 @@ static int sun4i_backend_of_get_id(struct device_node *node)
static const struct sunxi_engine_ops sun4i_backend_engine_ops = { .commit = sun4i_backend_commit, + .commit_poll = sun4i_backend_commit_poll, .layers_init = sun4i_layers_init, .apply_color_correction = sun4i_backend_apply_color_correction, .disable_color_correction = sun4i_backend_disable_color_correction, diff --git a/drivers/gpu/drm/sun4i/sun4i_crtc.c b/drivers/gpu/drm/sun4i/sun4i_crtc.c index 2eba1d8639d8..31550493fa1d 100644 --- a/drivers/gpu/drm/sun4i/sun4i_crtc.c +++ b/drivers/gpu/drm/sun4i/sun4i_crtc.c @@ -34,6 +34,7 @@ static void sun4i_crtc_atomic_begin(struct drm_crtc *crtc, struct drm_crtc_state *old_state) { struct sun4i_crtc *scrtc = drm_crtc_to_sun4i_crtc(crtc); + struct sunxi_engine *engine = scrtc->engine; struct drm_device *dev = crtc->dev; unsigned long flags;
dri-devel@lists.freedesktop.org
-
Daniel Vetter -
Laurent Pinchart -
Liviu Dudau -
Maxime Ripard