| Message ID | 1345403595-9678-3-git-send-email-daniel.vetter@ffwll.ch (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
On Sun, 19 Aug 2012 21:12:19 +0200 Daniel Vetter <daniel.vetter@ffwll.ch> wrote: > Just impendance matching with the the crtc helper stuff. > > ... and somehow the design of this all ended up in this commit here, > too ;-) > > The big plan is that this new set of crtc display_funcs take full > responsibility of modeset operations for the entire display output > pipeline (by calling down into object-specific callbacks and > functions). The platform-specific callbacks simply know best what the > proper order is. > > This has the drawback that we can't do minimal change-overs any more > if a modeset just disables one encoder in a cloned configuration > (because we will only expose a disable/enable action that takes > down/sets up the entire crtc including all encoders). Imo that's the > only sane way to do it though: > - The use-case for this is pretty minimal, even when presenting (at > least sane people) should use a dual-screen output so that you can > see your notes on your panel. Clone mode is imo BS. > - With all the clone mode constrains, shared resources, and special > ordering requirements (which differ even on the same platform > sometimes for different outputs) there's no way we'd get this right > for all cases. Especially since this is a under-used feature. > - And to top it off: On haswell even dp link re-training requires us > to take down the entire display pipe - otherwise the chip dies. > > So the only sane way is to do a full modeset on every crtc where the > output config changes in any way. > > To support global modeset (i.e. set the configuration for all crtcs at > once) we'd then add one more function to allocate global and shared > objects in the best ways (e.g. fdi links, pch plls, ...). The crtc > functions would then simply use the pre-allocated stuff (and shouldn't > be able to fail, ever). We could even do all the object pinning in > there (and maybe try to defragment the global gtt if we fail)! > > Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch> > --- > drivers/gpu/drm/i915/intel_display.c | 37 > ++---------------------------------- 1 file changed, 2 insertions(+), > 35 deletions(-) > > diff --git a/drivers/gpu/drm/i915/intel_display.c > b/drivers/gpu/drm/i915/intel_display.c index adc9868..04bec4b 100644 > --- a/drivers/gpu/drm/i915/intel_display.c > +++ b/drivers/gpu/drm/i915/intel_display.c > @@ -3500,34 +3500,6 @@ static void intel_crtc_disable(struct drm_crtc > *crtc) } > } > > -/* Prepare for a mode set. > - * > - * Note we could be a lot smarter here. We need to figure out which > outputs > - * will be enabled, which disabled (in short, how the config will > changes) > - * and perform the minimum necessary steps to accomplish that, e.g. > updating > - * watermarks, FBC configuration, making sure PLLs are programmed > correctly, > - * panel fitting is in the proper state, etc. > - */ > -static void i9xx_crtc_prepare(struct drm_crtc *crtc) > -{ > - i9xx_crtc_disable(crtc); > -} > - > -static void i9xx_crtc_commit(struct drm_crtc *crtc) > -{ > - i9xx_crtc_enable(crtc); > -} > - > -static void ironlake_crtc_prepare(struct drm_crtc *crtc) > -{ > - ironlake_crtc_disable(crtc); > -} > - > -static void ironlake_crtc_commit(struct drm_crtc *crtc) > -{ > - ironlake_crtc_enable(crtc); > -} > - > void intel_encoder_prepare(struct drm_encoder *encoder) > { > struct drm_encoder_helper_funcs *encoder_funcs = > encoder->helper_private; @@ -6626,13 +6598,8 @@ static void > intel_crtc_init(struct drm_device *dev, int pipe) intel_crtc->active > = true; /* force the pipe off on setup_init_config */ intel_crtc->bpp > = 24; /* default for pre-Ironlake */ > - if (HAS_PCH_SPLIT(dev)) { > - intel_helper_funcs.prepare = ironlake_crtc_prepare; > - intel_helper_funcs.commit = ironlake_crtc_commit; > - } else { > - intel_helper_funcs.prepare = i9xx_crtc_prepare; > - intel_helper_funcs.commit = i9xx_crtc_commit; > - } > + intel_helper_funcs.prepare = dev_priv->display.crtc_disable; > + intel_helper_funcs.commit = dev_priv->display.crtc_enable; > > drm_crtc_helper_add(&intel_crtc->base, &intel_helper_funcs); > } Looks fine. Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c index adc9868..04bec4b 100644 --- a/drivers/gpu/drm/i915/intel_display.c +++ b/drivers/gpu/drm/i915/intel_display.c @@ -3500,34 +3500,6 @@ static void intel_crtc_disable(struct drm_crtc *crtc) } } -/* Prepare for a mode set. - * - * Note we could be a lot smarter here. We need to figure out which outputs - * will be enabled, which disabled (in short, how the config will changes) - * and perform the minimum necessary steps to accomplish that, e.g. updating - * watermarks, FBC configuration, making sure PLLs are programmed correctly, - * panel fitting is in the proper state, etc. - */ -static void i9xx_crtc_prepare(struct drm_crtc *crtc) -{ - i9xx_crtc_disable(crtc); -} - -static void i9xx_crtc_commit(struct drm_crtc *crtc) -{ - i9xx_crtc_enable(crtc); -} - -static void ironlake_crtc_prepare(struct drm_crtc *crtc) -{ - ironlake_crtc_disable(crtc); -} - -static void ironlake_crtc_commit(struct drm_crtc *crtc) -{ - ironlake_crtc_enable(crtc); -} - void intel_encoder_prepare(struct drm_encoder *encoder) { struct drm_encoder_helper_funcs *encoder_funcs = encoder->helper_private; @@ -6626,13 +6598,8 @@ static void intel_crtc_init(struct drm_device *dev, int pipe) intel_crtc->active = true; /* force the pipe off on setup_init_config */ intel_crtc->bpp = 24; /* default for pre-Ironlake */ - if (HAS_PCH_SPLIT(dev)) { - intel_helper_funcs.prepare = ironlake_crtc_prepare; - intel_helper_funcs.commit = ironlake_crtc_commit; - } else { - intel_helper_funcs.prepare = i9xx_crtc_prepare; - intel_helper_funcs.commit = i9xx_crtc_commit; - } + intel_helper_funcs.prepare = dev_priv->display.crtc_disable; + intel_helper_funcs.commit = dev_priv->display.crtc_enable; drm_crtc_helper_add(&intel_crtc->base, &intel_helper_funcs); }
Just impendance matching with the the crtc helper stuff. ... and somehow the design of this all ended up in this commit here, too ;-) The big plan is that this new set of crtc display_funcs take full responsibility of modeset operations for the entire display output pipeline (by calling down into object-specific callbacks and functions). The platform-specific callbacks simply know best what the proper order is. This has the drawback that we can't do minimal change-overs any more if a modeset just disables one encoder in a cloned configuration (because we will only expose a disable/enable action that takes down/sets up the entire crtc including all encoders). Imo that's the only sane way to do it though: - The use-case for this is pretty minimal, even when presenting (at least sane people) should use a dual-screen output so that you can see your notes on your panel. Clone mode is imo BS. - With all the clone mode constrains, shared resources, and special ordering requirements (which differ even on the same platform sometimes for different outputs) there's no way we'd get this right for all cases. Especially since this is a under-used feature. - And to top it off: On haswell even dp link re-training requires us to take down the entire display pipe - otherwise the chip dies. So the only sane way is to do a full modeset on every crtc where the output config changes in any way. To support global modeset (i.e. set the configuration for all crtcs at once) we'd then add one more function to allocate global and shared objects in the best ways (e.g. fdi links, pch plls, ...). The crtc functions would then simply use the pre-allocated stuff (and shouldn't be able to fail, ever). We could even do all the object pinning in there (and maybe try to defragment the global gtt if we fail)! Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch> --- drivers/gpu/drm/i915/intel_display.c | 37 ++---------------------------------- 1 file changed, 2 insertions(+), 35 deletions(-)