Merge branch 'msm-next' of git://people.freedesktop.org/~robclark/linux into drm-next

Now that we have the bits needed for mdp5 atomic, here is the followup
pull request I mentioned.  Main highlights are:

1) mdp5 multiple crtc and public plane support (no more hard-coded mixer setup!)
2) mdp5 atomic conversion
3) couple atomic helper fixes for issues found during mdp5 atomic
debug (reviewed by danvet.. but he didn't plane to send an
atomic-fixes pull request so I agreed to tack them on to mine)

* 'msm-next' of git://people.freedesktop.org/~robclark/linux:
  drm/atomic: shutdown *current* encoder
  drm/atomic: check mode_changed *after* atomic_check
  drm/msm/mdp4: fix mixer setup for multi-crtc + planes
  drm/msm/mdp5: dpms(OFF) cleanups
  drm/msm/mdp5: atomic
  drm/msm: atomic fixes
  drm/msm/mdp5: remove global mdp5_ctl_mgr
  drm/msm/mdp5: don't use void * for opaque types
  drm/msm: add multiple CRTC and overlay support
  drm/msm/mdp5: set rate before enabling clk
  drm/msm/mdp5: introduce mdp5_cfg module
  drm/msm/mdp5: make SMP module dynamically configurable
  drm/msm/hdmi: remove useless kref
  drm/msm/mdp5: get the core clock rate from MDP5 config
  drm/msm/mdp5: use irqdomains

24 files changed, 1737 insertions, 739 deletions

diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
index 690360038dc1..a17b8e9c0a81 100644
--- a/drivers/gpu/drm/drm_atomic_helper.c
+++ b/drivers/gpu/drm/drm_atomic_helper.c
@@ -331,7 +331,7 @@ mode_fixup(struct drm_atomic_state *state)
 }
 
 static int
-drm_atomic_helper_check_prepare(struct drm_device *dev,
+drm_atomic_helper_check_modeset(struct drm_device *dev,
 				struct drm_atomic_state *state)
 {
 	int ncrtcs = dev->mode_config.num_crtc;
@@ -428,10 +428,6 @@ int drm_atomic_helper_check(struct drm_device *dev,
 	int ncrtcs = dev->mode_config.num_crtc;
 	int i, ret = 0;
 
-	ret = drm_atomic_helper_check_prepare(dev, state);
-	if (ret)
-		return ret;
-
 	for (i = 0; i < nplanes; i++) {
 		struct drm_plane_helper_funcs *funcs;
 		struct drm_plane *plane = state->planes[i];
@@ -475,6 +471,10 @@ int drm_atomic_helper_check(struct drm_device *dev,
 		}
 	}
 
+	ret = drm_atomic_helper_check_modeset(dev, state);
+	if (ret)
+		return ret;
+
 	return ret;
 }
 EXPORT_SYMBOL(drm_atomic_helper_check);
@@ -499,9 +499,12 @@ disable_outputs(struct drm_device *dev, struct drm_atomic_state *old_state)
 		if (!old_conn_state || !old_conn_state->crtc)
 			continue;
 
-		encoder = connector->state->best_encoder;
+		encoder = old_conn_state->best_encoder;
 
-		if (!encoder)
+		/* We shouldn't get this far if we didn't previously have
+		 * an encoder.. but WARN_ON() rather than explode.
+		 */
+		if (WARN_ON(!encoder))
 			continue;
 
 		funcs = encoder->helper_private;
diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
index 0d96132df059..143d988f8add 100644
--- a/drivers/gpu/drm/msm/Makefile
+++ b/drivers/gpu/drm/msm/Makefile
@@ -25,6 +25,8 @@ msm-y := \
 	mdp/mdp4/mdp4_irq.o \
 	mdp/mdp4/mdp4_kms.o \
 	mdp/mdp4/mdp4_plane.o \
+	mdp/mdp5/mdp5_cfg.o \
+	mdp/mdp5/mdp5_ctl.o \
 	mdp/mdp5/mdp5_crtc.o \
 	mdp/mdp5/mdp5_encoder.o \
 	mdp/mdp5/mdp5_irq.o \
diff --git a/drivers/gpu/drm/msm/hdmi/hdmi.c b/drivers/gpu/drm/msm/hdmi/hdmi.c
index 90077619029d..062c68725376 100644
--- a/drivers/gpu/drm/msm/hdmi/hdmi.c
+++ b/drivers/gpu/drm/msm/hdmi/hdmi.c
@@ -15,6 +15,7 @@
  * this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 
+#include <linux/of_irq.h>
 #include "hdmi.h"
 
 void hdmi_set_mode(struct hdmi *hdmi, bool power_on)
@@ -39,7 +40,7 @@ void hdmi_set_mode(struct hdmi *hdmi, bool power_on)
 			power_on ? "Enable" : "Disable", ctrl);
 }
 
-irqreturn_t hdmi_irq(int irq, void *dev_id)
+static irqreturn_t hdmi_irq(int irq, void *dev_id)
 {
 	struct hdmi *hdmi = dev_id;
 
@@ -54,9 +55,8 @@ irqreturn_t hdmi_irq(int irq, void *dev_id)
 	return IRQ_HANDLED;
 }
 
-void hdmi_destroy(struct kref *kref)
+static void hdmi_destroy(struct hdmi *hdmi)
 {
-	struct hdmi *hdmi = container_of(kref, struct hdmi, refcount);
 	struct hdmi_phy *phy = hdmi->phy;
 
 	if (phy)
@@ -84,8 +84,6 @@ static struct hdmi *hdmi_init(struct platform_device *pdev)
 		goto fail;
 	}
 
-	kref_init(&hdmi->refcount);
-
 	hdmi->pdev = pdev;
 	hdmi->config = config;
 
@@ -182,7 +180,7 @@ static struct hdmi *hdmi_init(struct platform_device *pdev)
 
 fail:
 	if (hdmi)
-		hdmi_destroy(&hdmi->refcount);
+		hdmi_destroy(hdmi);
 
 	return ERR_PTR(ret);
 }
@@ -200,7 +198,6 @@ int hdmi_modeset_init(struct hdmi *hdmi,
 {
 	struct msm_drm_private *priv = dev->dev_private;
 	struct platform_device *pdev = hdmi->pdev;
-	struct hdmi_platform_config *config = pdev->dev.platform_data;
 	int ret;
 
 	hdmi->dev = dev;
@@ -224,22 +221,20 @@ int hdmi_modeset_init(struct hdmi *hdmi,
 		goto fail;
 	}
 
-	if (!config->shared_irq) {
-		hdmi->irq = platform_get_irq(pdev, 0);
-		if (hdmi->irq < 0) {
-			ret = hdmi->irq;
-			dev_err(dev->dev, "failed to get irq: %d\n", ret);
-			goto fail;
-		}
+	hdmi->irq = irq_of_parse_and_map(pdev->dev.of_node, 0);
+	if (hdmi->irq < 0) {
+		ret = hdmi->irq;
+		dev_err(dev->dev, "failed to get irq: %d\n", ret);
+		goto fail;
+	}
 
-		ret = devm_request_threaded_irq(&pdev->dev, hdmi->irq,
-				NULL, hdmi_irq, IRQF_TRIGGER_HIGH | IRQF_ONESHOT,
-				"hdmi_isr", hdmi);
-		if (ret < 0) {
-			dev_err(dev->dev, "failed to request IRQ%u: %d\n",
-					hdmi->irq, ret);
-			goto fail;
-		}
+	ret = devm_request_irq(&pdev->dev, hdmi->irq,
+			hdmi_irq, IRQF_TRIGGER_HIGH | IRQF_ONESHOT,
+			"hdmi_isr", hdmi);
+	if (ret < 0) {
+		dev_err(dev->dev, "failed to request IRQ%u: %d\n",
+				hdmi->irq, ret);
+		goto fail;
 	}
 
 	encoder->bridge = hdmi->bridge;
@@ -271,12 +266,6 @@ fail:
 
 #include <linux/of_gpio.h>
 
-static void set_hdmi(struct drm_device *dev, struct hdmi *hdmi)
-{
-	struct msm_drm_private *priv = dev->dev_private;
-	priv->hdmi = hdmi;
-}
-
 #ifdef CONFIG_OF
 static int get_gpio(struct device *dev, struct device_node *of_node, const char *name)
 {
@@ -297,6 +286,8 @@ static int get_gpio(struct device *dev, struct device_node *of_node, const char
 
 static int hdmi_bind(struct device *dev, struct device *master, void *data)
 {
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct msm_drm_private *priv = drm->dev_private;
 	static struct hdmi_platform_config config = {};
 	struct hdmi *hdmi;
 #ifdef CONFIG_OF
@@ -318,7 +309,6 @@ static int hdmi_bind(struct device *dev, struct device *master, void *data)
 		config.hpd_clk_cnt   = ARRAY_SIZE(hpd_clk_names);
 		config.pwr_clk_names = pwr_clk_names;
 		config.pwr_clk_cnt   = ARRAY_SIZE(pwr_clk_names);
-		config.shared_irq    = true;
 	} else if (of_device_is_compatible(of_node, "qcom,hdmi-tx-8960")) {
 		static const char *hpd_clk_names[] = {"core_clk", "master_iface_clk", "slave_iface_clk"};
 		static const char *hpd_reg_names[] = {"core-vdda", "hdmi-mux"};
@@ -392,14 +382,19 @@ static int hdmi_bind(struct device *dev, struct device *master, void *data)
 	hdmi = hdmi_init(to_platform_device(dev));
 	if (IS_ERR(hdmi))
 		return PTR_ERR(hdmi);
-	set_hdmi(dev_get_drvdata(master), hdmi);
+	priv->hdmi = hdmi;
 	return 0;
 }
 
 static void hdmi_unbind(struct device *dev, struct device *master,
 		void *data)
 {
-	set_hdmi(dev_get_drvdata(master), NULL);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct msm_drm_private *priv = drm->dev_private;
+	if (priv->hdmi) {
+		hdmi_destroy(priv->hdmi);
+		priv->hdmi = NULL;
+	}
 }
 
 static const struct component_ops hdmi_ops = {
diff --git a/drivers/gpu/drm/msm/hdmi/hdmi.h b/drivers/gpu/drm/msm/hdmi/hdmi.h
index b981995410b5..43e654f751b7 100644
--- a/drivers/gpu/drm/msm/hdmi/hdmi.h
+++ b/drivers/gpu/drm/msm/hdmi/hdmi.h
@@ -38,8 +38,6 @@ struct hdmi_audio {
 };
 
 struct hdmi {
-	struct kref refcount;
-
 	struct drm_device *dev;
 	struct platform_device *pdev;
 
@@ -97,13 +95,9 @@ struct hdmi_platform_config {
 	/* gpio's: */
 	int ddc_clk_gpio, ddc_data_gpio, hpd_gpio, mux_en_gpio, mux_sel_gpio;
 	int mux_lpm_gpio;
-
-	/* older devices had their own irq, mdp5+ it is shared w/ mdp: */
-	bool shared_irq;
 };
 
 void hdmi_set_mode(struct hdmi *hdmi, bool power_on);
-void hdmi_destroy(struct kref *kref);
 
 static inline void hdmi_write(struct hdmi *hdmi, u32 reg, u32 data)
 {
@@ -115,17 +109,6 @@ static inline u32 hdmi_read(struct hdmi *hdmi, u32 reg)
 	return msm_readl(hdmi->mmio + reg);
 }
 
-static inline struct hdmi * hdmi_reference(struct hdmi *hdmi)
-{
-	kref_get(&hdmi->refcount);
-	return hdmi;
-}
-
-static inline void hdmi_unreference(struct hdmi *hdmi)
-{
-	kref_put(&hdmi->refcount, hdmi_destroy);
-}
-
 /*
  * The phy appears to be different, for example between 8960 and 8x60,
  * so split the phy related functions out and load the correct one at
diff --git a/drivers/gpu/drm/msm/hdmi/hdmi_bridge.c b/drivers/gpu/drm/msm/hdmi/hdmi_bridge.c
index f6cf745c249e..6902ad6da710 100644
--- a/drivers/gpu/drm/msm/hdmi/hdmi_bridge.c
+++ b/drivers/gpu/drm/msm/hdmi/hdmi_bridge.c
@@ -26,7 +26,6 @@ struct hdmi_bridge {
 static void hdmi_bridge_destroy(struct drm_bridge *bridge)
 {
 	struct hdmi_bridge *hdmi_bridge = to_hdmi_bridge(bridge);
-	hdmi_unreference(hdmi_bridge->hdmi);
 	drm_bridge_cleanup(bridge);
 	kfree(hdmi_bridge);
 }
@@ -218,7 +217,7 @@ struct drm_bridge *hdmi_bridge_init(struct hdmi *hdmi)
 		goto fail;
 	}
 
-	hdmi_bridge->hdmi = hdmi_reference(hdmi);
+	hdmi_bridge->hdmi = hdmi;
 
 	bridge = &hdmi_bridge->base;
 
diff --git a/drivers/gpu/drm/msm/hdmi/hdmi_connector.c b/drivers/gpu/drm/msm/hdmi/hdmi_connector.c
index 0aecb2580072..fbebb0405d76 100644
--- a/drivers/gpu/drm/msm/hdmi/hdmi_connector.c
+++ b/drivers/gpu/drm/msm/hdmi/hdmi_connector.c
@@ -330,8 +330,6 @@ static void hdmi_connector_destroy(struct drm_connector *connector)
 	drm_connector_unregister(connector);
 	drm_connector_cleanup(connector);
 
-	hdmi_unreference(hdmi_connector->hdmi);
-
 	kfree(hdmi_connector);
 }
 
@@ -425,7 +423,7 @@ struct drm_connector *hdmi_connector_init(struct hdmi *hdmi)
 		goto fail;
 	}
 
-	hdmi_connector->hdmi = hdmi_reference(hdmi);
+	hdmi_connector->hdmi = hdmi;
 	INIT_WORK(&hdmi_connector->hpd_work, hotplug_work);
 
 	connector = &hdmi_connector->base;
diff --git a/drivers/gpu/drm/msm/mdp/mdp4/mdp4_crtc.c b/drivers/gpu/drm/msm/mdp/mdp4/mdp4_crtc.c
index fef22e8cabb6..6781aa994613 100644
--- a/drivers/gpu/drm/msm/mdp/mdp4/mdp4_crtc.c
+++ b/drivers/gpu/drm/msm/mdp/mdp4/mdp4_crtc.c
@@ -167,34 +167,54 @@ static bool mdp4_crtc_mode_fixup(struct drm_crtc *crtc,
 	return true;
 }
 
-static void blend_setup(struct drm_crtc *crtc)
+/* statically (for now) map planes to mixer stage (z-order): */
+static const int idxs[] = {
+		[VG1]  = 1,
+		[VG2]  = 2,
+		[RGB1] = 0,
+		[RGB2] = 0,
+		[RGB3] = 0,
+		[VG3]  = 3,
+		[VG4]  = 4,
+
+};
+
+/* setup mixer config, for which we need to consider all crtc's and
+ * the planes attached to them
+ *
+ * TODO may possibly need some extra locking here
+ */
+static void setup_mixer(struct mdp4_kms *mdp4_kms)
 {
-	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);
-	struct mdp4_kms *mdp4_kms = get_kms(crtc);
-	struct drm_plane *plane;
-	int i, ovlp = mdp4_crtc->ovlp;
+	struct drm_mode_config *config = &mdp4_kms->dev->mode_config;
+	struct drm_crtc *crtc;
 	uint32_t mixer_cfg = 0;
 	static const enum mdp_mixer_stage_id stages[] = {
 			STAGE_BASE, STAGE0, STAGE1, STAGE2, STAGE3,
 	};
-	/* statically (for now) map planes to mixer stage (z-order): */
-	static const int idxs[] = {
-			[VG1]  = 1,
-			[VG2]  = 2,
-			[RGB1] = 0,
-			[RGB2] = 0,
-			[RGB3] = 0,
-			[VG3]  = 3,
-			[VG4]  = 4,
 
-	};
-	bool alpha[4]= { false, false, false, false };
+	list_for_each_entry(crtc, &config->crtc_list, head) {
+		struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);
+		struct drm_plane *plane;
 
-	/* Don't rely on value read back from hw, but instead use our
-	 * own shadowed value.  Possibly disable/reenable looses the
-	 * previous value and goes back to power-on default?
-	 */
-	mixer_cfg = mdp4_kms->mixer_cfg;
+		for_each_plane_on_crtc(crtc, plane) {
+			enum mdp4_pipe pipe_id = mdp4_plane_pipe(plane);
+			int idx = idxs[pipe_id];
+			mixer_cfg = mixercfg(mixer_cfg, mdp4_crtc->mixer,
+					pipe_id, stages[idx]);
+		}
+	}
+
+	mdp4_write(mdp4_kms, REG_MDP4_LAYERMIXER_IN_CFG, mixer_cfg);
+}
+
+static void blend_setup(struct drm_crtc *crtc)
+{
+	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);
+	struct mdp4_kms *mdp4_kms = get_kms(crtc);
+	struct drm_plane *plane;
+	int i, ovlp = mdp4_crtc->ovlp;
+	bool alpha[4]= { false, false, false, false };
 
 	mdp4_write(mdp4_kms, REG_MDP4_OVLP_TRANSP_LOW0(ovlp), 0);
 	mdp4_write(mdp4_kms, REG_MDP4_OVLP_TRANSP_LOW1(ovlp), 0);
@@ -209,13 +229,8 @@ static void blend_setup(struct drm_crtc *crtc)
 					to_mdp_format(msm_framebuffer_format(plane->fb));
 			alpha[idx-1] = format->alpha_enable;
 		}
-		mixer_cfg = mixercfg(mixer_cfg, mdp4_crtc->mixer,
-				pipe_id, stages[idx]);
 	}
 
-	/* this shouldn't happen.. and seems to cause underflow: */
-	WARN_ON(!mixer_cfg);
-
 	for (i = 0; i < 4; i++) {
 		uint32_t op;
 
@@ -238,8 +253,7 @@ static void blend_setup(struct drm_crtc *crtc)
 		mdp4_write(mdp4_kms, REG_MDP4_OVLP_STAGE_TRANSP_HIGH1(ovlp, i), 0);
 	}
 
-	mdp4_kms->mixer_cfg = mixer_cfg;
-	mdp4_write(mdp4_kms, REG_MDP4_LAYERMIXER_IN_CFG, mixer_cfg);
+	setup_mixer(mdp4_kms);
 }
 
 static void mdp4_crtc_mode_set_nofb(struct drm_crtc *crtc)
diff --git a/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.h b/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.h
index 770645296f11..cbd77bc626d5 100644
--- a/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.h
+++ b/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.h
@@ -32,13 +32,6 @@ struct mdp4_kms {
 
 	int rev;
 
-	/* Shadow value for MDP4_LAYERMIXER_IN_CFG.. since setup for all
-	 * crtcs/encoders is in one shared register, we need to update it
-	 * via read/modify/write.  But to avoid getting confused by power-
-	 * on-default values after resume, use this shadow value instead:
-	 */
-	uint32_t mixer_cfg;
-
 	/* mapper-id used to request GEM buffer mapped for scanout: */
 	int id;
 
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_cfg.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_cfg.c
new file mode 100644
index 000000000000..b0a44310cf2a
--- /dev/null
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_cfg.c
@@ -0,0 +1,207 @@
+/*
+ * Copyright (c) 2014 The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include "mdp5_kms.h"
+#include "mdp5_cfg.h"
+
+struct mdp5_cfg_handler {
+	int revision;
+	struct mdp5_cfg config;
+};
+
+/* mdp5_cfg must be exposed (used in mdp5.xml.h) */
+const struct mdp5_cfg_hw *mdp5_cfg = NULL;
+
+const struct mdp5_cfg_hw msm8x74_config = {
+	.name = "msm8x74",
+	.smp = {
+		.mmb_count = 22,
+		.mmb_size = 4096,
+	},
+	.ctl = {
+		.count = 5,
+		.base = { 0x00600, 0x00700, 0x00800, 0x00900, 0x00a00 },
+	},
+	.pipe_vig = {
+		.count = 3,
+		.base = { 0x01200, 0x01600, 0x01a00 },
+	},
+	.pipe_rgb = {
+		.count = 3,
+		.base = { 0x01e00, 0x02200, 0x02600 },
+	},
+	.pipe_dma = {
+		.count = 2,
+		.base = { 0x02a00, 0x02e00 },
+	},
+	.lm = {
+		.count = 5,
+		.base = { 0x03200, 0x03600, 0x03a00, 0x03e00, 0x04200 },
+		.nb_stages = 5,
+	},
+	.dspp = {
+		.count = 3,
+		.base = { 0x04600, 0x04a00, 0x04e00 },
+	},
+	.ad = {
+		.count = 2,
+		.base = { 0x13100, 0x13300 }, /* NOTE: no ad in v1.0 */
+	},
+	.intf = {
+		.count = 4,
+		.base = { 0x12500, 0x12700, 0x12900, 0x12b00 },
+	},
+	.max_clk = 200000000,
+};
+
+const struct mdp5_cfg_hw apq8084_config = {
+	.name = "apq8084",
+	.smp = {
+		.mmb_count = 44,
+		.mmb_size = 8192,
+		.reserved_state[0] = GENMASK(7, 0),	/* first 8 MMBs */
+		.reserved[CID_RGB0] = 2,
+		.reserved[CID_RGB1] = 2,
+		.reserved[CID_RGB2] = 2,
+		.reserved[CID_RGB3] = 2,
+	},
+	.ctl = {
+		.count = 5,
+		.base = { 0x00600, 0x00700, 0x00800, 0x00900, 0x00a00 },
+	},
+	.pipe_vig = {
+		.count = 4,
+		.base = { 0x01200, 0x01600, 0x01a00, 0x01e00 },
+	},
+	.pipe_rgb = {
+		.count = 4,
+		.base = { 0x02200, 0x02600, 0x02a00, 0x02e00 },
+	},
+	.pipe_dma = {
+		.count = 2,
+		.base = { 0x03200, 0x03600 },
+	},
+	.lm = {
+		.count = 6,
+		.base = { 0x03a00, 0x03e00, 0x04200, 0x04600, 0x04a00, 0x04e00 },
+		.nb_stages = 5,
+	},
+	.dspp = {
+		.count = 4,
+		.base = { 0x05200, 0x05600, 0x05a00, 0x05e00 },
+
+	},
+	.ad = {
+		.count = 3,
+		.base = { 0x13500, 0x13700, 0x13900 },
+	},
+	.intf = {
+		.count = 5,
+		.base = { 0x12500, 0x12700, 0x12900, 0x12b00, 0x12d00 },
+	},
+	.max_clk = 320000000,
+};
+
+static const struct mdp5_cfg_handler cfg_handlers[] = {
+	{ .revision = 0, .config = { .hw = &msm8x74_config } },
+	{ .revision = 2, .config = { .hw = &msm8x74_config } },
+	{ .revision = 3, .config = { .hw = &apq8084_config } },
+};
+
+
+static struct mdp5_cfg_platform *mdp5_get_config(struct platform_device *dev);
+
+const struct mdp5_cfg_hw *mdp5_cfg_get_hw_config(struct mdp5_cfg_handler *cfg_handler)
+{
+	return cfg_handler->config.hw;
+}
+
+struct mdp5_cfg *mdp5_cfg_get_config(struct mdp5_cfg_handler *cfg_handler)
+{
+	return &cfg_handler->config;
+}
+
+int mdp5_cfg_get_hw_rev(struct mdp5_cfg_handler *cfg_handler)
+{
+	return cfg_handler->revision;
+}
+
+void mdp5_cfg_destroy(struct mdp5_cfg_handler *cfg_handler)
+{
+	kfree(cfg_handler);
+}
+
+struct mdp5_cfg_handler *mdp5_cfg_init(struct mdp5_kms *mdp5_kms,
+		uint32_t major, uint32_t minor)
+{
+	struct drm_device *dev = mdp5_kms->dev;
+	struct platform_device *pdev = dev->platformdev;
+	struct mdp5_cfg_handler *cfg_handler;
+	struct mdp5_cfg_platform *pconfig;
+	int i, ret = 0;
+
+	cfg_handler = kzalloc(sizeof(*cfg_handler), GFP_KERNEL);
+	if (unlikely(!cfg_handler)) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+	if (major != 1) {
+		dev_err(dev->dev, "unexpected MDP major version: v%d.%d\n",
+				major, minor);
+		ret = -ENXIO;
+		goto fail;
+	}
+
+	/* only after mdp5_cfg global pointer's init can we access the hw */
+	for (i = 0; i < ARRAY_SIZE(cfg_handlers); i++) {
+		if (cfg_handlers[i].revision != minor)
+			continue;
+		mdp5_cfg = cfg_handlers[i].config.hw;
+
+		break;
+	}
+	if (unlikely(!mdp5_cfg)) {
+		dev_err(dev->dev, "unexpected MDP minor revision: v%d.%d\n",
+				major, minor);
+		ret = -ENXIO;
+		goto fail;
+	}
+
+	cfg_handler->revision = minor;
+	cfg_handler->config.hw = mdp5_cfg;
+
+	pconfig = mdp5_get_config(pdev);
+	memcpy(&cfg_handler->config.platform, pconfig, sizeof(*pconfig));
+
+	DBG("MDP5: %s hw config selected", mdp5_cfg->name);
+
+	return cfg_handler;
+
+fail:
+	if (cfg_handler)
+		mdp5_cfg_destroy(cfg_handler);
+
+	return NULL;
+}
+
+static struct mdp5_cfg_platform *mdp5_get_config(struct platform_device *dev)
+{
+	static struct mdp5_cfg_platform config = {};
+#ifdef CONFIG_OF
+	/* TODO */
+#endif
+	config.iommu = iommu_domain_alloc(&platform_bus_type);
+
+	return &config;
+}
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_cfg.h b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_cfg.h
new file mode 100644
index 000000000000..dba4d52cceeb
--- /dev/null
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_cfg.h
@@ -0,0 +1,91 @@
+/*
+ * Copyright (c) 2014 The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#ifndef __MDP5_CFG_H__
+#define __MDP5_CFG_H__
+
+#include "msm_drv.h"
+
+/*
+ * mdp5_cfg
+ *
+ * This module configures the dynamic offsets used by mdp5.xml.h
+ * (initialized in mdp5_cfg.c)
+ */
+extern const struct mdp5_cfg_hw *mdp5_cfg;
+
+#define MAX_CTL			8
+#define MAX_BASES		8
+#define MAX_SMP_BLOCKS		44
+#define MAX_CLIENTS		32
+
+typedef DECLARE_BITMAP(mdp5_smp_state_t, MAX_SMP_BLOCKS);
+
+#define MDP5_SUB_BLOCK_DEFINITION \
+	int count; \
+	uint32_t base[MAX_BASES]
+
+struct mdp5_sub_block {
+	MDP5_SUB_BLOCK_DEFINITION;
+};
+
+struct mdp5_lm_block {
+	MDP5_SUB_BLOCK_DEFINITION;
+	uint32_t nb_stages;		/* number of stages per blender */
+};
+
+struct mdp5_smp_block {
+	int mmb_count;			/* number of SMP MMBs */
+	int mmb_size;			/* MMB: size in bytes */
+	mdp5_smp_state_t reserved_state;/* SMP MMBs statically allocated */
+	int reserved[MAX_CLIENTS];	/* # of MMBs allocated per client */
+};
+
+struct mdp5_cfg_hw {
+	char  *name;
+
+	struct mdp5_smp_block smp;
+	struct mdp5_sub_block ctl;
+	struct mdp5_sub_block pipe_vig;
+	struct mdp5_sub_block pipe_rgb;
+	struct mdp5_sub_block pipe_dma;
+	struct mdp5_lm_block  lm;
+	struct mdp5_sub_block dspp;
+	struct mdp5_sub_block ad;
+	struct mdp5_sub_block intf;
+
+	uint32_t max_clk;
+};
+
+/* platform config data (ie. from DT, or pdata) */
+struct mdp5_cfg_platform {
+	struct iommu_domain *iommu;
+};
+
+struct mdp5_cfg {
+	const struct mdp5_cfg_hw *hw;
+	struct mdp5_cfg_platform platform;
+};
+
+struct mdp5_kms;
+struct mdp5_cfg_handler;
+
+const struct mdp5_cfg_hw *mdp5_cfg_get_hw_config(struct mdp5_cfg_handler *cfg_hnd);
+struct mdp5_cfg *mdp5_cfg_get_config(struct mdp5_cfg_handler *cfg_hnd);
+int mdp5_cfg_get_hw_rev(struct mdp5_cfg_handler *cfg_hnd);
+
+struct mdp5_cfg_handler *mdp5_cfg_init(struct mdp5_kms *mdp5_kms,
+		uint32_t major, uint32_t minor);
+void mdp5_cfg_destroy(struct mdp5_cfg_handler *cfg_hnd);
+
+#endif /* __MDP5_CFG_H__ */
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c
index b7b32c47fd71..0598bdea4ff4 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (c) 2014 The Linux Foundation. All rights reserved.
  * Copyright (C) 2013 Red Hat
  * Author: Rob Clark <robdclark@gmail.com>
  *
@@ -17,41 +18,35 @@
 
 #include "mdp5_kms.h"
 
+#include <linux/sort.h>
 #include <drm/drm_mode.h>
 #include "drm_crtc.h"
 #include "drm_crtc_helper.h"
 #include "drm_flip_work.h"
 
+#define SSPP_MAX	(SSPP_RGB3 + 1) /* TODO: Add SSPP_MAX in mdp5.xml.h */
+
 struct mdp5_crtc {
 	struct drm_crtc base;
 	char name[8];
 	int id;
 	bool enabled;
 
-	/* which mixer/encoder we route output to: */
-	int mixer;
+	/* layer mixer used for this CRTC (+ its lock): */
+#define GET_LM_ID(crtc_id)	((crtc_id == 3) ? 5 : crtc_id)
+	int lm;
+	spinlock_t lm_lock;	/* protect REG_MDP5_LM_* registers */
+
+	/* CTL used for this CRTC: */
+	struct mdp5_ctl *ctl;
 
 	/* if there is a pending flip, these will be non-null: */
 	struct drm_pending_vblank_event *event;
-	struct msm_fence_cb pageflip_cb;
 
 #define PENDING_CURSOR 0x1
 #define PENDING_FLIP   0x2
 	atomic_t pending;
 
-	/* the fb that we logically (from PoV of KMS API) hold a ref
-	 * to.  Which we may not yet be scanning out (we may still
-	 * be scanning out previous in case of page_flip while waiting
-	 * for gpu rendering to complete:
-	 */
-	struct drm_framebuffer *fb;
-
-	/* the fb that we currently hold a scanout ref to: */
-	struct drm_framebuffer *scanout_fb;
-
-	/* for unref'ing framebuffers after scanout completes: */
-	struct drm_flip_work unref_fb_work;
-
 	struct mdp_irq vblank;
 	struct mdp_irq err;
 };
@@ -71,66 +66,38 @@ static void request_pending(struct drm_crtc *crtc, uint32_t pending)
 	mdp_irq_register(&get_kms(crtc)->base, &mdp5_crtc->vblank);
 }
 
-static void crtc_flush(struct drm_crtc *crtc)
-{
-	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
-	struct mdp5_kms *mdp5_kms = get_kms(crtc);
-	int id = mdp5_crtc->id;
-	struct drm_plane *plane;
-	uint32_t flush = 0;
-
-	for_each_plane_on_crtc(crtc, plane) {
-		enum mdp5_pipe pipe = mdp5_plane_pipe(plane);
-		flush |= pipe2flush(pipe);
-	}
-
-	flush |= mixer2flush(mdp5_crtc->id);
-	flush |= MDP5_CTL_FLUSH_CTL;
+#define mdp5_lm_get_flush(lm)	mdp_ctl_flush_mask_lm(lm)
 
-	DBG("%s: flush=%08x", mdp5_crtc->name, flush);
-
-	mdp5_write(mdp5_kms, REG_MDP5_CTL_FLUSH(id), flush);
-}
-
-static void update_fb(struct drm_crtc *crtc, struct drm_framebuffer *new_fb)
+static void crtc_flush(struct drm_crtc *crtc, u32 flush_mask)
 {
 	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
-	struct drm_framebuffer *old_fb = mdp5_crtc->fb;
-
-	/* grab reference to incoming scanout fb: */
-	drm_framebuffer_reference(new_fb);
-	mdp5_crtc->base.primary->fb = new_fb;
-	mdp5_crtc->fb = new_fb;
 
-	if (old_fb)
-		drm_flip_work_queue(&mdp5_crtc->unref_fb_work, old_fb);
+	DBG("%s: flush=%08x", mdp5_crtc->name, flush_mask);
+	mdp5_ctl_commit(mdp5_crtc->ctl, flush_mask);
 }
 
-/* unlike update_fb(), take a ref to the new scanout fb *before* updating
- * plane, then call this.  Needed to ensure we don't unref the buffer that
- * is actually still being scanned out.
- *
- * Note that this whole thing goes away with atomic.. since we can defer
- * calling into driver until rendering is done.
+/*
+ * flush updates, to make sure hw is updated to new scanout fb,
+ * so that we can safely queue unref to current fb (ie. next
+ * vblank we know hw is done w/ previous scanout_fb).
  */
-static void update_scanout(struct drm_crtc *crtc, struct drm_framebuffer *fb)
+static void crtc_flush_all(struct drm_crtc *crtc)
 {
 	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
+	struct drm_plane *plane;
+	uint32_t flush_mask = 0;
 
-	/* flush updates, to make sure hw is updated to new scanout fb,
-	 * so that we can safely queue unref to current fb (ie. next
-	 * vblank we know hw is done w/ previous scanout_fb).
-	 */
-	crtc_flush(crtc);
-
-	if (mdp5_crtc->scanout_fb)
-		drm_flip_work_queue(&mdp5_crtc->unref_fb_work,
-				mdp5_crtc->scanout_fb);
+	/* we could have already released CTL in the disable path: */
+	if (!mdp5_crtc->ctl)
+		return;
 
-	mdp5_crtc->scanout_fb = fb;
+	for_each_plane_on_crtc(crtc, plane) {
+		flush_mask |= mdp5_plane_get_flush(plane);
+	}
+	flush_mask |= mdp5_ctl_get_flush(mdp5_crtc->ctl);
+	flush_mask |= mdp5_lm_get_flush(mdp5_crtc->lm);
 
-	/* enable vblank to complete flip: */
-	request_pending(crtc, PENDING_FLIP);
+	crtc_flush(crtc, flush_mask);
 }
 
 /* if file!=NULL, this is preclose potential cancel-flip path */
@@ -151,6 +118,7 @@ static void complete_flip(struct drm_crtc *crtc, struct drm_file *file)
 		 */
 		if (!file || (event->base.file_priv == file)) {
 			mdp5_crtc->event = NULL;
+			DBG("%s: send event: %p", mdp5_crtc->name, event);
 			drm_send_vblank_event(dev, mdp5_crtc->id, event);
 		}
 	}
@@ -160,38 +128,11 @@ static void complete_flip(struct drm_crtc *crtc, struct drm_file *file)
 		mdp5_plane_complete_flip(plane);
 }
 
-static void pageflip_cb(struct msm_fence_cb *cb)
-{
-	struct mdp5_crtc *mdp5_crtc =
-		container_of(cb, struct mdp5_crtc, pageflip_cb);
-	struct drm_crtc *crtc = &mdp5_crtc->base;
-	struct drm_framebuffer *fb = mdp5_crtc->fb;
-
-	if (!fb)
-		return;
-
-	drm_framebuffer_reference(fb);
-	mdp5_plane_set_scanout(crtc->primary, fb);
-	update_scanout(crtc, fb);
-}
-
-static void unref_fb_worker(struct drm_flip_work *work, void *val)
-{
-	struct mdp5_crtc *mdp5_crtc =
-		container_of(work, struct mdp5_crtc, unref_fb_work);
-	struct drm_device *dev = mdp5_crtc->base.dev;
-
-	mutex_lock(&dev->mode_config.mutex);
-	drm_framebuffer_unreference(val);
-	mutex_unlock(&dev->mode_config.mutex);
-}
-
 static void mdp5_crtc_destroy(struct drm_crtc *crtc)
 {
 	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
 
 	drm_crtc_cleanup(crtc);
-	drm_flip_work_cleanup(&mdp5_crtc->unref_fb_work);
 
 	kfree(mdp5_crtc);
 }
@@ -209,6 +150,8 @@ static void mdp5_crtc_dpms(struct drm_crtc *crtc, int mode)
 			mdp5_enable(mdp5_kms);
 			mdp_irq_register(&mdp5_kms->base, &mdp5_crtc->err);
 		} else {
+			/* set STAGE_UNUSED for all layers */
+			mdp5_ctl_blend(mdp5_crtc->ctl, mdp5_crtc->lm, 0x00000000);
 			mdp_irq_unregister(&mdp5_kms->base, &mdp5_crtc->err);
 			mdp5_disable(mdp5_kms);
 		}
@@ -223,54 +166,78 @@ static bool mdp5_crtc_mode_fixup(struct drm_crtc *crtc,
 	return true;
 }
 
+/*
+ * blend_setup() - blend all the planes of a CRTC
+ *
+ * When border is enabled, the border color will ALWAYS be the base layer.
+ * Therefore, the first plane (private RGB pipe) will start at STAGE0.
+ * If disabled, the first plane starts at STAGE_BASE.
+ *
+ * Note:
+ * Border is not enabled here because the private plane is exactly
+ * the CRTC resolution.
+ */
 static void blend_setup(struct drm_crtc *crtc)
 {
 	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
 	struct mdp5_kms *mdp5_kms = get_kms(crtc);
-	int id = mdp5_crtc->id;
+	struct drm_plane *plane;
+	const struct mdp5_cfg_hw *hw_cfg;
+	uint32_t lm = mdp5_crtc->lm, blend_cfg = 0;
+	unsigned long flags;
+#define blender(stage)	((stage) - STAGE_BASE)
 
-	/*
-	 * Hard-coded setup for now until I figure out how the
-	 * layer-mixer works
-	 */
+	hw_cfg = mdp5_cfg_get_hw_config(mdp5_kms->cfg);
 
-	/* LM[id]: */
-	mdp5_write(mdp5_kms, REG_MDP5_LM_BLEND_COLOR_OUT(id),
-			MDP5_LM_BLEND_COLOR_OUT_STAGE0_FG_ALPHA);
-	mdp5_write(mdp5_kms, REG_MDP5_LM_BLEND_OP_MODE(id, 0),
-			MDP5_LM_BLEND_OP_MODE_FG_ALPHA(FG_CONST) |
-			MDP5_LM_BLEND_OP_MODE_BG_ALPHA(FG_PIXEL) |
-			MDP5_LM_BLEND_OP_MODE_BG_INV_ALPHA);
-	mdp5_write(mdp5_kms, REG_MDP5_LM_BLEND_FG_ALPHA(id, 0), 0xff);
-	mdp5_write(mdp5_kms, REG_MDP5_LM_BLEND_BG_ALPHA(id, 0), 0x00);
-
-	/* NOTE: seems that LM[n] and CTL[m], we do not need n==m.. but
-	 * we want to be setting CTL[m].LAYER[n].  Not sure what the
-	 * point of having CTL[m].LAYER[o] (for o!=n).. maybe that is
-	 * used when chaining up mixers for high resolution displays?
-	 */
+	spin_lock_irqsave(&mdp5_crtc->lm_lock, flags);
+
+	/* ctl could be released already when we are shutting down: */
+	if (!mdp5_crtc->ctl)
+		goto out;
 
-	/* CTL[id]: */
-	mdp5_write(mdp5_kms, REG_MDP5_CTL_LAYER_REG(id, 0),
-			MDP5_CTL_LAYER_REG_RGB0(STAGE0) |
-			MDP5_CTL_LAYER_REG_BORDER_COLOR);
-	mdp5_write(mdp5_kms, REG_MDP5_CTL_LAYER_REG(id, 1), 0);
-	mdp5_write(mdp5_kms, REG_MDP5_CTL_LAYER_REG(id, 2), 0);
-	mdp5_write(mdp5_kms, REG_MDP5_CTL_LAYER_REG(id, 3), 0);
-	mdp5_write(mdp5_kms, REG_MDP5_CTL_LAYER_REG(id, 4), 0);
+	for_each_plane_on_crtc(crtc, plane) {
+		enum mdp_mixer_stage_id stage =
+			to_mdp5_plane_state(plane->state)->stage;
+
+		/*
+		 * Note: This cannot happen with current implementation but
+		 * we need to check this condition once z property is added
+		 */
+		BUG_ON(stage > hw_cfg->lm.nb_stages);
+
+		/* LM */
+		mdp5_write(mdp5_kms,
+				REG_MDP5_LM_BLEND_OP_MODE(lm, blender(stage)),
+				MDP5_LM_BLEND_OP_MODE_FG_ALPHA(FG_CONST) |
+				MDP5_LM_BLEND_OP_MODE_BG_ALPHA(BG_CONST));
+		mdp5_write(mdp5_kms, REG_MDP5_LM_BLEND_FG_ALPHA(lm,
+				blender(stage)), 0xff);
+		mdp5_write(mdp5_kms, REG_MDP5_LM_BLEND_BG_ALPHA(lm,
+				blender(stage)), 0x00);
+		/* CTL */
+		blend_cfg |= mdp_ctl_blend_mask(mdp5_plane_pipe(plane), stage);
+		DBG("%s: blending pipe %s on stage=%d", mdp5_crtc->name,
+				pipe2name(mdp5_plane_pipe(plane)), stage);
+	}
+
+	DBG("%s: lm%d: blend config = 0x%08x", mdp5_crtc->name, lm, blend_cfg);
+	mdp5_ctl_blend(mdp5_crtc->ctl, lm, blend_cfg);
+
+out:
+	spin_unlock_irqrestore(&mdp5_crtc->lm_lock, flags);
 }
 
-static int mdp5_crtc_mode_set(struct drm_crtc *crtc,
-		struct drm_display_mode *mode,
-		struct drm_display_mode *adjusted_mode,
-		int x, int y,
-		struct drm_framebuffer *old_fb)
+static void mdp5_crtc_mode_set_nofb(struct drm_crtc *crtc)
 {
 	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
 	struct mdp5_kms *mdp5_kms = get_kms(crtc);
-	int ret;
+	unsigned long flags;
+	struct drm_display_mode *mode;
 
-	mode = adjusted_mode;
+	if (WARN_ON(!crtc->state))
+		return;
+
+	mode = &crtc->state->adjusted_mode;
 
 	DBG("%s: set mode: %d:\"%s\" %d %d %d %d %d %d %d %d %d %d 0x%x 0x%x",
 			mdp5_crtc->name, mode->base.id, mode->name,
@@ -281,28 +248,11 @@ static int mdp5_crtc_mode_set(struct drm_crtc *crtc,
 			mode->vsync_end, mode->vtotal,
 			mode->type, mode->flags);
 
-	/* grab extra ref for update_scanout() */
-	drm_framebuffer_reference(crtc->primary->fb);
-
-	ret = mdp5_plane_mode_set(crtc->primary, crtc, crtc->primary->fb,
-			0, 0, mode->hdisplay, mode->vdisplay,
-			x << 16, y << 16,
-			mode->hdisplay << 16, mode->vdisplay << 16);
-	if (ret) {
-		drm_framebuffer_unreference(crtc->primary->fb);
-		dev_err(crtc->dev->dev, "%s: failed to set mode on plane: %d\n",
-				mdp5_crtc->name, ret);
-		return ret;
-	}
-
-	mdp5_write(mdp5_kms, REG_MDP5_LM_OUT_SIZE(mdp5_crtc->id),
+	spin_lock_irqsave(&mdp5_crtc->lm_lock, flags);
+	mdp5_write(mdp5_kms, REG_MDP5_LM_OUT_SIZE(mdp5_crtc->lm),
 			MDP5_LM_OUT_SIZE_WIDTH(mode->hdisplay) |
 			MDP5_LM_OUT_SIZE_HEIGHT(mode->vdisplay));
-
-	update_fb(crtc, crtc->primary->fb);
-	update_scanout(crtc, crtc->primary->fb);
-
-	return 0;
+	spin_unlock_irqrestore(&mdp5_crtc->lm_lock, flags);
 }
 
 static void mdp5_crtc_prepare(struct drm_crtc *crtc)
@@ -316,65 +266,119 @@ static void mdp5_crtc_prepare(struct drm_crtc *crtc)
 
 static void mdp5_crtc_commit(struct drm_crtc *crtc)
 {
+	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
+	DBG("%s", mdp5_crtc->name);
 	mdp5_crtc_dpms(crtc, DRM_MODE_DPMS_ON);
-	crtc_flush(crtc);
+	crtc_flush_all(crtc);
 	/* drop the ref to mdp clk's that we got in prepare: */
 	mdp5_disable(get_kms(crtc));
 }
 
-static int mdp5_crtc_mode_set_base(struct drm_crtc *crtc, int x, int y,
-		struct drm_framebuffer *old_fb)
+static void mdp5_crtc_load_lut(struct drm_crtc *crtc)
+{
+}
+
+struct plane_state {
+	struct drm_plane *plane;
+	struct mdp5_plane_state *state;
+};
+
+static int pstate_cmp(const void *a, const void *b)
+{
+	struct plane_state *pa = (struct plane_state *)a;
+	struct plane_state *pb = (struct plane_state *)b;
+	return pa->state->zpos - pb->state->zpos;
+}
+
+static int mdp5_crtc_atomic_check(struct drm_crtc *crtc,
+		struct drm_crtc_state *state)
 {
-	struct drm_plane *plane = crtc->primary;
-	struct drm_display_mode *mode = &crtc->mode;
-	int ret;
-
-	/* grab extra ref for update_scanout() */
-	drm_framebuffer_reference(crtc->primary->fb);
-
-	ret = mdp5_plane_mode_set(plane, crtc, crtc->primary->fb,
-			0, 0, mode->hdisplay, mode->vdisplay,
-			x << 16, y << 16,
-			mode->hdisplay << 16, mode->vdisplay << 16);
-	if (ret) {
-		drm_framebuffer_unreference(crtc->primary->fb);
-		return ret;
+	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
+	struct mdp5_kms *mdp5_kms = get_kms(crtc);
+	struct drm_plane *plane;
+	struct drm_device *dev = crtc->dev;
+	struct plane_state pstates[STAGE3 + 1];
+	int cnt = 0, i;
+
+	DBG("%s: check", mdp5_crtc->name);
+
+	if (mdp5_crtc->event) {
+		dev_err(dev->dev, "already pending flip!\n");
+		return -EBUSY;
 	}
 
-	update_fb(crtc, crtc->primary->fb);
-	update_scanout(crtc, crtc->primary->fb);
+	/* request a free CTL, if none is already allocated for this CRTC */
+	if (state->enable && !mdp5_crtc->ctl) {
+		mdp5_crtc->ctl = mdp5_ctlm_request(mdp5_kms->ctlm, crtc);
+		if (WARN_ON(!mdp5_crtc->ctl))
+			return -EINVAL;
+	}
+
+	/* verify that there are not too many planes attached to crtc
+	 * and that we don't have conflicting mixer stages:
+	 */
+	for_each_pending_plane_on_crtc(state->state, crtc, plane) {
+		struct drm_plane_state *pstate;
+
+		if (cnt >= ARRAY_SIZE(pstates)) {
+			dev_err(dev->dev, "too many planes!\n");
+			return -EINVAL;
+		}
+
+		pstate = state->state->plane_states[drm_plane_index(plane)];
+
+		/* plane might not have changed, in which case take
+		 * current state:
+		 */
+		if (!pstate)
+			pstate = plane->state;
+
+		pstates[cnt].plane = plane;
+		pstates[cnt].state = to_mdp5_plane_state(pstate);
+
+		cnt++;
+	}
+
+	sort(pstates, cnt, sizeof(pstates[0]), pstate_cmp, NULL);
+
+	for (i = 0; i < cnt; i++) {
+		pstates[i].state->stage = STAGE_BASE + i;
+		DBG("%s: assign pipe %s on stage=%d", mdp5_crtc->name,
+				pipe2name(mdp5_plane_pipe(pstates[i].plane)),
+				pstates[i].state->stage);
+	}
 
 	return 0;
 }
 
-static void mdp5_crtc_load_lut(struct drm_crtc *crtc)
+static void mdp5_crtc_atomic_begin(struct drm_crtc *crtc)
 {
+	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
+	DBG("%s: begin", mdp5_crtc->name);
 }
 
-static int mdp5_crtc_page_flip(struct drm_crtc *crtc,
-		struct drm_framebuffer *new_fb,
-		struct drm_pending_vblank_event *event,
-		uint32_t page_flip_flags)
+static void mdp5_crtc_atomic_flush(struct drm_crtc *crtc)
 {
 	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
 	struct drm_device *dev = crtc->dev;
-	struct drm_gem_object *obj;
 	unsigned long flags;
 
-	if (mdp5_crtc->event) {
-		dev_err(dev->dev, "already pending flip!\n");
-		return -EBUSY;
-	}
+	DBG("%s: flush", mdp5_crtc->name);
 
-	obj = msm_framebuffer_bo(new_fb, 0);
+	WARN_ON(mdp5_crtc->event);
 
 	spin_lock_irqsave(&dev->event_lock, flags);
-	mdp5_crtc->event = event;
+	mdp5_crtc->event = crtc->state->event;
 	spin_unlock_irqrestore(&dev->event_lock, flags);
 
-	update_fb(crtc, new_fb);
+	blend_setup(crtc);
+	crtc_flush_all(crtc);
+	request_pending(crtc, PENDING_FLIP);
 
-	return msm_gem_queue_inactive_cb(obj, &mdp5_crtc->pageflip_cb);
+	if (mdp5_crtc->ctl && !crtc->state->enable) {
+		mdp5_ctl_release(mdp5_crtc->ctl);
+		mdp5_crtc->ctl = NULL;
+	}
 }
 
 static int mdp5_crtc_set_property(struct drm_crtc *crtc,
@@ -385,27 +389,33 @@ static int mdp5_crtc_set_property(struct drm_crtc *crtc,
 }
 
 static const struct drm_crtc_funcs mdp5_crtc_funcs = {
-	.set_config = drm_crtc_helper_set_config,
+	.set_config = drm_atomic_helper_set_config,
 	.destroy = mdp5_crtc_destroy,
-	.page_flip = mdp5_crtc_page_flip,
+	.page_flip = drm_atomic_helper_page_flip,
 	.set_property = mdp5_crtc_set_property,
+	.reset = drm_atomic_helper_crtc_reset,
+	.atomic_duplicate_state = drm_atomic_helper_crtc_duplicate_state,
+	.atomic_destroy_state = drm_atomic_helper_crtc_destroy_state,
 };
 
 static const struct drm_crtc_helper_funcs mdp5_crtc_helper_funcs = {
 	.dpms = mdp5_crtc_dpms,
 	.mode_fixup = mdp5_crtc_mode_fixup,
-	.mode_set = mdp5_crtc_mode_set,
+	.mode_set_nofb = mdp5_crtc_mode_set_nofb,
+	.mode_set = drm_helper_crtc_mode_set,
+	.mode_set_base = drm_helper_crtc_mode_set_base,
 	.prepare = mdp5_crtc_prepare,
 	.commit = mdp5_crtc_commit,
-	.mode_set_base = mdp5_crtc_mode_set_base,
 	.load_lut = mdp5_crtc_load_lut,
+	.atomic_check = mdp5_crtc_atomic_check,
+	.atomic_begin = mdp5_crtc_atomic_begin,
+	.atomic_flush = mdp5_crtc_atomic_flush,
 };
 
 static void mdp5_crtc_vblank_irq(struct mdp_irq *irq, uint32_t irqstatus)
 {
 	struct mdp5_crtc *mdp5_crtc = container_of(irq, struct mdp5_crtc, vblank);
 	struct drm_crtc *crtc = &mdp5_crtc->base;
-	struct msm_drm_private *priv = crtc->dev->dev_private;
 	unsigned pending;
 
 	mdp_irq_unregister(&get_kms(crtc)->base, &mdp5_crtc->vblank);
@@ -414,16 +424,14 @@ static void mdp5_crtc_vblank_irq(struct mdp_irq *irq, uint32_t irqstatus)
 
 	if (pending & PENDING_FLIP) {
 		complete_flip(crtc, NULL);
-		drm_flip_work_commit(&mdp5_crtc->unref_fb_work, priv->wq);
 	}
 }
 
 static void mdp5_crtc_err_irq(struct mdp_irq *irq, uint32_t irqstatus)
 {
 	struct mdp5_crtc *mdp5_crtc = container_of(irq, struct mdp5_crtc, err);
-	struct drm_crtc *crtc = &mdp5_crtc->base;
+
 	DBG("%s: error: %08x", mdp5_crtc->name, irqstatus);
-	crtc_flush(crtc);
 }
 
 uint32_t mdp5_crtc_vblank(struct drm_crtc *crtc)
@@ -444,10 +452,9 @@ void mdp5_crtc_set_intf(struct drm_crtc *crtc, int intf,
 {
 	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
 	struct mdp5_kms *mdp5_kms = get_kms(crtc);
-	static const enum mdp5_intfnum intfnum[] = {
-			INTF0, INTF1, INTF2, INTF3,
-	};
+	uint32_t flush_mask = 0;
 	uint32_t intf_sel;
+	unsigned long flags;
 
 	/* now that we know what irq's we want: */
 	mdp5_crtc->err.irqmask = intf2err(intf);
@@ -457,6 +464,7 @@ void mdp5_crtc_set_intf(struct drm_crtc *crtc, int intf,
 	if (!mdp5_kms)
 		return;
 
+	spin_lock_irqsave(&mdp5_kms->resource_lock, flags);
 	intf_sel = mdp5_read(mdp5_kms, REG_MDP5_DISP_INTF_SEL);
 
 	switch (intf) {
@@ -481,39 +489,25 @@ void mdp5_crtc_set_intf(struct drm_crtc *crtc, int intf,
 		break;
 	}
 
-	blend_setup(crtc);
+	mdp5_write(mdp5_kms, REG_MDP5_DISP_INTF_SEL, intf_sel);
+	spin_unlock_irqrestore(&mdp5_kms->resource_lock, flags);
 
 	DBG("%s: intf_sel=%08x", mdp5_crtc->name, intf_sel);
+	mdp5_ctl_set_intf(mdp5_crtc->ctl, intf);
+	flush_mask |= mdp5_ctl_get_flush(mdp5_crtc->ctl);
+	flush_mask |= mdp5_lm_get_flush(mdp5_crtc->lm);
 
-	mdp5_write(mdp5_kms, REG_MDP5_DISP_INTF_SEL, intf_sel);
-	mdp5_write(mdp5_kms, REG_MDP5_CTL_OP(mdp5_crtc->id),
-			MDP5_CTL_OP_MODE(MODE_NONE) |
-			MDP5_CTL_OP_INTF_NUM(intfnum[intf]));
-
-	crtc_flush(crtc);
+	crtc_flush(crtc, flush_mask);
 }
 
-static void set_attach(struct drm_crtc *crtc, enum mdp5_pipe pipe_id,
-		struct drm_plane *plane)
+int mdp5_crtc_get_lm(struct drm_crtc *crtc)
 {
 	struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
 
-	blend_setup(crtc);
-	if (mdp5_crtc->enabled && (plane != crtc->primary))
-		crtc_flush(crtc);
-}
+	if (WARN_ON(!crtc))
+		return -EINVAL;
 
-void mdp5_crtc_attach(struct drm_crtc *crtc, struct drm_plane *plane)
-{
-	set_attach(crtc, mdp5_plane_pipe(plane), plane);
-}
-
-void mdp5_crtc_detach(struct drm_crtc *crtc, struct drm_plane *plane)
-{
-	/* don't actually detatch our primary plane: */
-	if (crtc->primary == plane)
-		return;
-	set_attach(crtc, mdp5_plane_pipe(plane), NULL);
+	return mdp5_crtc->lm;
 }
 
 /* initialize crtc */
@@ -530,6 +524,9 @@ struct drm_crtc *mdp5_crtc_init(struct drm_device *dev,
 	crtc = &mdp5_crtc->base;
 
 	mdp5_crtc->id = id;
+	mdp5_crtc->lm = GET_LM_ID(id);
+
+	spin_lock_init(&mdp5_crtc->lm_lock);
 
 	mdp5_crtc->vblank.irq = mdp5_crtc_vblank_irq;
 	mdp5_crtc->err.irq = mdp5_crtc_err_irq;
@@ -537,11 +534,6 @@ struct drm_crtc *mdp5_crtc_init(struct drm_device *dev,
 	snprintf(mdp5_crtc->name, sizeof(mdp5_crtc->name), "%s:%d",
 			pipe2name(mdp5_plane_pipe(plane)), id);
 
-	drm_flip_work_init(&mdp5_crtc->unref_fb_work,
-			"unref fb", unref_fb_worker);
-
-	INIT_FENCE_CB(&mdp5_crtc->pageflip_cb, pageflip_cb);
-
 	drm_crtc_init_with_planes(dev, crtc, plane, NULL, &mdp5_crtc_funcs);
 	drm_crtc_helper_add(crtc, &mdp5_crtc_helper_funcs);
 	plane->crtc = crtc;
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_ctl.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_ctl.c
new file mode 100644
index 000000000000..dea4505ac963
--- /dev/null
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_ctl.c
@@ -0,0 +1,322 @@
+/*
+ * Copyright (c) 2014 The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include "mdp5_kms.h"
+#include "mdp5_ctl.h"
+
+/*
+ * CTL - MDP Control Pool Manager
+ *
+ * Controls are shared between all CRTCs.
+ *
+ * They are intended to be used for data path configuration.
+ * The top level register programming describes the complete data path for
+ * a specific data path ID - REG_MDP5_CTL_*(<id>, ...)
+ *
+ * Hardware capabilities determine the number of concurrent data paths
+ *
+ * In certain use cases (high-resolution dual pipe), one single CTL can be
+ * shared across multiple CRTCs.
+ *
+ * Because the number of CTLs can be less than the number of CRTCs,
+ * CTLs are dynamically allocated from a pool of CTLs, only once a CRTC is
+ * requested by the client (in mdp5_crtc_mode_set()).
+ */
+
+struct mdp5_ctl {
+	struct mdp5_ctl_manager *ctlm;
+
+	u32 id;
+
+	/* whether this CTL has been allocated or not: */
+	bool busy;
+
+	/* memory output connection (@see mdp5_ctl_mode): */
+	u32 mode;
+
+	/* REG_MDP5_CTL_*(<id>) registers access info + lock: */
+	spinlock_t hw_lock;
+	u32 reg_offset;
+
+	/* flush mask used to commit CTL registers */
+	u32 flush_mask;
+
+	bool cursor_on;
+
+	struct drm_crtc *crtc;
+};
+
+struct mdp5_ctl_manager {
+	struct drm_device *dev;
+
+	/* number of CTL / Layer Mixers in this hw config: */
+	u32 nlm;
+	u32 nctl;
+
+	/* pool of CTLs + lock to protect resource allocation (ctls[i].busy) */
+	spinlock_t pool_lock;
+	struct mdp5_ctl ctls[MAX_CTL];
+};
+
+static inline
+struct mdp5_kms *get_kms(struct mdp5_ctl_manager *ctl_mgr)
+{
+	struct msm_drm_private *priv = ctl_mgr->dev->dev_private;
+
+	return to_mdp5_kms(to_mdp_kms(priv->kms));
+}
+
+static inline
+void ctl_write(struct mdp5_ctl *ctl, u32 reg, u32 data)
+{
+	struct mdp5_kms *mdp5_kms = get_kms(ctl->ctlm);
+
+	(void)ctl->reg_offset; /* TODO use this instead of mdp5_write */
+	mdp5_write(mdp5_kms, reg, data);
+}
+
+static inline
+u32 ctl_read(struct mdp5_ctl *ctl, u32 reg)
+{
+	struct mdp5_kms *mdp5_kms = get_kms(ctl->ctlm);
+
+	(void)ctl->reg_offset; /* TODO use this instead of mdp5_write */
+	return mdp5_read(mdp5_kms, reg);
+}
+
+
+int mdp5_ctl_set_intf(struct mdp5_ctl *ctl, enum mdp5_intf intf)
+{
+	unsigned long flags;
+	static const enum mdp5_intfnum intfnum[] = {
+			INTF0, INTF1, INTF2, INTF3,
+	};
+
+	spin_lock_irqsave(&ctl->hw_lock, flags);
+	ctl_write(ctl, REG_MDP5_CTL_OP(ctl->id),
+			MDP5_CTL_OP_MODE(ctl->mode) |
+			MDP5_CTL_OP_INTF_NUM(intfnum[intf]));
+	spin_unlock_irqrestore(&ctl->hw_lock, flags);
+
+	return 0;
+}
+
+int mdp5_ctl_set_cursor(struct mdp5_ctl *ctl, bool enable)
+{
+	struct mdp5_ctl_manager *ctl_mgr = ctl->ctlm;
+	unsigned long flags;
+	u32 blend_cfg;
+	int lm;
+
+	lm = mdp5_crtc_get_lm(ctl->crtc);
+	if (unlikely(WARN_ON(lm < 0))) {
+		dev_err(ctl_mgr->dev->dev, "CTL %d cannot find LM: %d",
+				ctl->id, lm);
+		return -EINVAL;
+	}
+
+	spin_lock_irqsave(&ctl->hw_lock, flags);
+
+	blend_cfg = ctl_read(ctl, REG_MDP5_CTL_LAYER_REG(ctl->id, lm));
+
+	if (enable)
+		blend_cfg |=  MDP5_CTL_LAYER_REG_CURSOR_OUT;
+	else
+		blend_cfg &= ~MDP5_CTL_LAYER_REG_CURSOR_OUT;
+
+	ctl_write(ctl, REG_MDP5_CTL_LAYER_REG(ctl->id, lm), blend_cfg);
+
+	spin_unlock_irqrestore(&ctl->hw_lock, flags);
+
+	ctl->cursor_on = enable;
+
+	return 0;
+}
+
+
+int mdp5_ctl_blend(struct mdp5_ctl *ctl, u32 lm, u32 blend_cfg)
+{
+	unsigned long flags;
+
+	if (ctl->cursor_on)
+		blend_cfg |=  MDP5_CTL_LAYER_REG_CURSOR_OUT;
+	else
+		blend_cfg &= ~MDP5_CTL_LAYER_REG_CURSOR_OUT;
+
+	spin_lock_irqsave(&ctl->hw_lock, flags);
+	ctl_write(ctl, REG_MDP5_CTL_LAYER_REG(ctl->id, lm), blend_cfg);
+	spin_unlock_irqrestore(&ctl->hw_lock, flags);
+
+	return 0;
+}
+
+int mdp5_ctl_commit(struct mdp5_ctl *ctl, u32 flush_mask)
+{
+	struct mdp5_ctl_manager *ctl_mgr = ctl->ctlm;
+	unsigned long flags;
+
+	if (flush_mask & MDP5_CTL_FLUSH_CURSOR_DUMMY) {
+		int lm = mdp5_crtc_get_lm(ctl->crtc);
+
+		if (unlikely(WARN_ON(lm < 0))) {
+			dev_err(ctl_mgr->dev->dev, "CTL %d cannot find LM: %d",
+					ctl->id, lm);
+			return -EINVAL;
+		}
+
+		/* for current targets, cursor bit is the same as LM bit */
+		flush_mask |= mdp_ctl_flush_mask_lm(lm);
+	}
+
+	spin_lock_irqsave(&ctl->hw_lock, flags);
+	ctl_write(ctl, REG_MDP5_CTL_FLUSH(ctl->id), flush_mask);
+	spin_unlock_irqrestore(&ctl->hw_lock, flags);
+
+	return 0;
+}
+
+u32 mdp5_ctl_get_flush(struct mdp5_ctl *ctl)
+{
+	return ctl->flush_mask;
+}
+
+void mdp5_ctl_release(struct mdp5_ctl *ctl)
+{
+	struct mdp5_ctl_manager *ctl_mgr = ctl->ctlm;
+	unsigned long flags;
+
+	if (unlikely(WARN_ON(ctl->id >= MAX_CTL) || !ctl->busy)) {
+		dev_err(ctl_mgr->dev->dev, "CTL %d in bad state (%d)",
+				ctl->id, ctl->busy);
+		return;
+	}
+
+	spin_lock_irqsave(&ctl_mgr->pool_lock, flags);
+	ctl->busy = false;
+	spin_unlock_irqrestore(&ctl_mgr->pool_lock, flags);
+
+	DBG("CTL %d released", ctl->id);
+}
+
+/*
+ * mdp5_ctl_request() - CTL dynamic allocation
+ *
+ * Note: Current implementation considers that we can only have one CRTC per CTL
+ *
+ * @return first free CTL
+ */
+struct mdp5_ctl *mdp5_ctlm_request(struct mdp5_ctl_manager *ctl_mgr,
+		struct drm_crtc *crtc)
+{
+	struct mdp5_ctl *ctl = NULL;
+	unsigned long flags;
+	int c;
+
+	spin_lock_irqsave(&ctl_mgr->pool_lock, flags);
+
+	for (c = 0; c < ctl_mgr->nctl; c++)
+		if (!ctl_mgr->ctls[c].busy)
+			break;
+
+	if (unlikely(c >= ctl_mgr->nctl)) {
+		dev_err(ctl_mgr->dev->dev, "No more CTL available!");
+		goto unlock;
+	}
+
+	ctl = &ctl_mgr->ctls[c];
+
+	ctl->crtc = crtc;
+	ctl->busy = true;
+	DBG("CTL %d allocated", ctl->id);
+
+unlock:
+	spin_unlock_irqrestore(&ctl_mgr->pool_lock, flags);
+	return ctl;
+}
+
+void mdp5_ctlm_hw_reset(struct mdp5_ctl_manager *ctl_mgr)
+{
+	unsigned long flags;
+	int c;
+
+	for (c = 0; c < ctl_mgr->nctl; c++) {
+		struct mdp5_ctl *ctl = &ctl_mgr->ctls[c];
+
+		spin_lock_irqsave(&ctl->hw_lock, flags);
+		ctl_write(ctl, REG_MDP5_CTL_OP(ctl->id), 0);
+		spin_unlock_irqrestore(&ctl->hw_lock, flags);
+	}
+}
+
+void mdp5_ctlm_destroy(struct mdp5_ctl_manager *ctl_mgr)
+{
+	kfree(ctl_mgr);
+}
+
+struct mdp5_ctl_manager *mdp5_ctlm_init(struct drm_device *dev,
+		void __iomem *mmio_base, const struct mdp5_cfg_hw *hw_cfg)
+{
+	struct mdp5_ctl_manager *ctl_mgr;
+	const struct mdp5_sub_block *ctl_cfg = &hw_cfg->ctl;
+	unsigned long flags;
+	int c, ret;
+
+	ctl_mgr = kzalloc(sizeof(*ctl_mgr), GFP_KERNEL);
+	if (!ctl_mgr) {
+		dev_err(dev->dev, "failed to allocate CTL manager\n");
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+	if (unlikely(WARN_ON(ctl_cfg->count > MAX_CTL))) {
+		dev_err(dev->dev, "Increase static pool size to at least %d\n",
+				ctl_cfg->count);
+		ret = -ENOSPC;
+		goto fail;
+	}
+
+	/* initialize the CTL manager: */
+	ctl_mgr->dev = dev;
+	ctl_mgr->nlm = hw_cfg->lm.count;
+	ctl_mgr->nctl = ctl_cfg->count;
+	spin_lock_init(&ctl_mgr->pool_lock);
+
+	/* initialize each CTL of the pool: */
+	spin_lock_irqsave(&ctl_mgr->pool_lock, flags);
+	for (c = 0; c < ctl_mgr->nctl; c++) {
+		struct mdp5_ctl *ctl = &ctl_mgr->ctls[c];
+
+		if (WARN_ON(!ctl_cfg->base[c])) {
+			dev_err(dev->dev, "CTL_%d: base is null!\n", c);
+			ret = -EINVAL;
+			goto fail;
+		}
+		ctl->ctlm = ctl_mgr;
+		ctl->id = c;
+		ctl->mode = MODE_NONE;
+		ctl->reg_offset = ctl_cfg->base[c];
+		ctl->flush_mask = MDP5_CTL_FLUSH_CTL;
+		ctl->busy = false;
+		spin_lock_init(&ctl->hw_lock);
+	}
+	spin_unlock_irqrestore(&ctl_mgr->pool_lock, flags);
+	DBG("Pool of %d CTLs created.", ctl_mgr->nctl);
+
+	return ctl_mgr;
+
+fail:
+	if (ctl_mgr)
+		mdp5_ctlm_destroy(ctl_mgr);
+
+	return ERR_PTR(ret);
+}
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_ctl.h b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_ctl.h
new file mode 100644
index 000000000000..1018519b6af2
--- /dev/null
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_ctl.h
@@ -0,0 +1,122 @@
+/*
+ * Copyright (c) 2014 The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#ifndef __MDP5_CTL_H__
+#define __MDP5_CTL_H__
+
+#include "msm_drv.h"
+
+/*
+ * CTL Manager prototypes:
+ * mdp5_ctlm_init() returns a ctlm (CTL Manager) handler,
+ * which is then used to call the other mdp5_ctlm_*(ctlm, ...) functions.
+ */
+struct mdp5_ctl_manager;
+struct mdp5_ctl_manager *mdp5_ctlm_init(struct drm_device *dev,
+		void __iomem *mmio_base, const struct mdp5_cfg_hw *hw_cfg);
+void mdp5_ctlm_hw_reset(struct mdp5_ctl_manager *ctlm);
+void mdp5_ctlm_destroy(struct mdp5_ctl_manager *ctlm);
+
+/*
+ * CTL prototypes:
+ * mdp5_ctl_request(ctlm, ...) returns a ctl (CTL resource) handler,
+ * which is then used to call the other mdp5_ctl_*(ctl, ...) functions.
+ */
+struct mdp5_ctl *mdp5_ctlm_request(struct mdp5_ctl_manager *ctlm, struct drm_crtc *crtc);
+
+int mdp5_ctl_set_intf(struct mdp5_ctl *ctl, enum mdp5_intf intf);
+
+int mdp5_ctl_set_cursor(struct mdp5_ctl *ctl, bool enable);
+
+/* @blend_cfg: see LM blender config definition below */
+int mdp5_ctl_blend(struct mdp5_ctl *ctl, u32 lm, u32 blend_cfg);
+
+/* @flush_mask: see CTL flush masks definitions below */
+int mdp5_ctl_commit(struct mdp5_ctl *ctl, u32 flush_mask);
+u32 mdp5_ctl_get_flush(struct mdp5_ctl *ctl);
+
+void mdp5_ctl_release(struct mdp5_ctl *ctl);
+
+/*
+ * blend_cfg (LM blender config):
+ *
+ * The function below allows the caller of mdp5_ctl_blend() to specify how pipes
+ * are being blended according to their stage (z-order), through @blend_cfg arg.
+ */
+static inline u32 mdp_ctl_blend_mask(enum mdp5_pipe pipe,
+		enum mdp_mixer_stage_id stage)
+{
+	switch (pipe) {
+	case SSPP_VIG0: return MDP5_CTL_LAYER_REG_VIG0(stage);
+	case SSPP_VIG1: return MDP5_CTL_LAYER_REG_VIG1(stage);
+	case SSPP_VIG2: return MDP5_CTL_LAYER_REG_VIG2(stage);
+	case SSPP_RGB0: return MDP5_CTL_LAYER_REG_RGB0(stage);
+	case SSPP_RGB1: return MDP5_CTL_LAYER_REG_RGB1(stage);
+	case SSPP_RGB2: return MDP5_CTL_LAYER_REG_RGB2(stage);
+	case SSPP_DMA0: return MDP5_CTL_LAYER_REG_DMA0(stage);
+	case SSPP_DMA1: return MDP5_CTL_LAYER_REG_DMA1(stage);
+	case SSPP_VIG3: return MDP5_CTL_LAYER_REG_VIG3(stage);
+	case SSPP_RGB3: return MDP5_CTL_LAYER_REG_RGB3(stage);
+	default:	return 0;
+	}
+}
+
+/*
+ * flush_mask (CTL flush masks):
+ *
+ * The following functions allow each DRM entity to get and store
+ * their own flush mask.
+ * Once stored, these masks will then be accessed through each DRM's
+ * interface and used by the caller of mdp5_ctl_commit() to specify
+ * which block(s) need to be flushed through @flush_mask parameter.
+ */
+
+#define MDP5_CTL_FLUSH_CURSOR_DUMMY	0x80000000
+
+static inline u32 mdp_ctl_flush_mask_cursor(int cursor_id)
+{
+	/* TODO: use id once multiple cursor support is present */
+	(void)cursor_id;
+
+	return MDP5_CTL_FLUSH_CURSOR_DUMMY;
+}
+
+static inline u32 mdp_ctl_flush_mask_lm(int lm)
+{
+	switch (lm) {
+	case 0:  return MDP5_CTL_FLUSH_LM0;
+	case 1:  return MDP5_CTL_FLUSH_LM1;
+	case 2:  return MDP5_CTL_FLUSH_LM2;
+	case 5:  return MDP5_CTL_FLUSH_LM5;
+	default: return 0;
+	}
+}
+
+static inline u32 mdp_ctl_flush_mask_pipe(enum mdp5_pipe pipe)
+{
+	switch (pipe) {
+	case SSPP_VIG0: return MDP5_CTL_FLUSH_VIG0;
+	case SSPP_VIG1: return MDP5_CTL_FLUSH_VIG1;
+	case SSPP_VIG2: return MDP5_CTL_FLUSH_VIG2;
+	case SSPP_RGB0: return MDP5_CTL_FLUSH_RGB0;
+	case SSPP_RGB1: return MDP5_CTL_FLUSH_RGB1;
+	case SSPP_RGB2: return MDP5_CTL_FLUSH_RGB2;
+	case SSPP_DMA0: return MDP5_CTL_FLUSH_DMA0;
+	case SSPP_DMA1: return MDP5_CTL_FLUSH_DMA1;
+	case SSPP_VIG3: return MDP5_CTL_FLUSH_VIG3;
+	case SSPP_RGB3: return MDP5_CTL_FLUSH_RGB3;
+	default:        return 0;
+	}
+}
+
+#endif /* __MDP5_CTL_H__ */
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_encoder.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_encoder.c
index edec7bfaa952..0254bfdeb92f 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_encoder.c
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_encoder.c
@@ -24,6 +24,7 @@ struct mdp5_encoder {
 	struct drm_encoder base;
 	int intf;
 	enum mdp5_intf intf_id;
+	spinlock_t intf_lock;	/* protect REG_MDP5_INTF_* registers */
 	bool enabled;
 	uint32_t bsc;
 };
@@ -115,6 +116,7 @@ static void mdp5_encoder_dpms(struct drm_encoder *encoder, int mode)
 	struct mdp5_kms *mdp5_kms = get_kms(encoder);
 	int intf = mdp5_encoder->intf;
 	bool enabled = (mode == DRM_MODE_DPMS_ON);
+	unsigned long flags;
 
 	DBG("mode=%d", mode);
 
@@ -123,9 +125,24 @@ static void mdp5_encoder_dpms(struct drm_encoder *encoder, int mode)
 
 	if (enabled) {
 		bs_set(mdp5_encoder, 1);
+		spin_lock_irqsave(&mdp5_encoder->intf_lock, flags);
 		mdp5_write(mdp5_kms, REG_MDP5_INTF_TIMING_ENGINE_EN(intf), 1);
+		spin_unlock_irqrestore(&mdp5_encoder->intf_lock, flags);
 	} else {
+		spin_lock_irqsave(&mdp5_encoder->intf_lock, flags);
 		mdp5_write(mdp5_kms, REG_MDP5_INTF_TIMING_ENGINE_EN(intf), 0);
+		spin_unlock_irqrestore(&mdp5_encoder->intf_lock, flags);
+
+		/*
+		 * Wait for a vsync so we know the ENABLE=0 latched before
+		 * the (connector) source of the vsync's gets disabled,
+		 * otherwise we end up in a funny state if we re-enable
+		 * before the disable latches, which results that some of
+		 * the settings changes for the new modeset (like new
+		 * scanout buffer) don't latch properly..
+		 */
+		mdp_irq_wait(&mdp5_kms->base, intf2vblank(intf));
+
 		bs_set(mdp5_encoder, 0);
 	}
 
@@ -150,6 +167,7 @@ static void mdp5_encoder_mode_set(struct drm_encoder *encoder,
 	uint32_t display_v_start, display_v_end;
 	uint32_t hsync_start_x, hsync_end_x;
 	uint32_t format;
+	unsigned long flags;
 
 	mode = adjusted_mode;
 
@@ -180,6 +198,8 @@ static void mdp5_encoder_mode_set(struct drm_encoder *encoder,
 	display_v_start = (mode->vtotal - mode->vsync_start) * mode->htotal + dtv_hsync_skew;
 	display_v_end = vsync_period - ((mode->vsync_start - mode->vdisplay) * mode->htotal) + dtv_hsync_skew - 1;
 
+	spin_lock_irqsave(&mdp5_encoder->intf_lock, flags);
+
 	mdp5_write(mdp5_kms, REG_MDP5_INTF_HSYNC_CTL(intf),
 			MDP5_INTF_HSYNC_CTL_PULSEW(mode->hsync_end - mode->hsync_start) |
 			MDP5_INTF_HSYNC_CTL_PERIOD(mode->htotal));
@@ -201,6 +221,8 @@ static void mdp5_encoder_mode_set(struct drm_encoder *encoder,
 	mdp5_write(mdp5_kms, REG_MDP5_INTF_ACTIVE_VEND_F0(intf), 0);
 	mdp5_write(mdp5_kms, REG_MDP5_INTF_PANEL_FORMAT(intf), format);
 	mdp5_write(mdp5_kms, REG_MDP5_INTF_FRAME_LINE_COUNT_EN(intf), 0x3);  /* frame+line? */
+
+	spin_unlock_irqrestore(&mdp5_encoder->intf_lock, flags);
 }
 
 static void mdp5_encoder_prepare(struct drm_encoder *encoder)
@@ -242,6 +264,8 @@ struct drm_encoder *mdp5_encoder_init(struct drm_device *dev, int intf,
 	mdp5_encoder->intf_id = intf_id;
 	encoder = &mdp5_encoder->base;
 
+	spin_lock_init(&mdp5_encoder->intf_lock);
+
 	drm_encoder_init(dev, encoder, &mdp5_encoder_funcs,
 			 DRM_MODE_ENCODER_TMDS);
 	drm_encoder_helper_add(encoder, &mdp5_encoder_helper_funcs);
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_irq.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_irq.c
index 812c59bbaf7f..70ac81edd40f 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_irq.c
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_irq.c
@@ -15,6 +15,8 @@
  * this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 
+#include <linux/irqdomain.h>
+#include <linux/irq.h>
 
 #include "msm_drv.h"
 #include "mdp5_kms.h"
@@ -82,18 +84,23 @@ irqreturn_t mdp5_irq(struct msm_kms *kms)
 {
 	struct mdp_kms *mdp_kms = to_mdp_kms(kms);
 	struct mdp5_kms *mdp5_kms = to_mdp5_kms(mdp_kms);
-	struct msm_drm_private *priv = mdp5_kms->dev->dev_private;
 	uint32_t intr;
 
 	intr = mdp5_read(mdp5_kms, REG_MDP5_HW_INTR_STATUS);
 
 	VERB("intr=%08x", intr);
 
-	if (intr & MDP5_HW_INTR_STATUS_INTR_MDP)
+	if (intr & MDP5_HW_INTR_STATUS_INTR_MDP) {
 		mdp5_irq_mdp(mdp_kms);
+		intr &= ~MDP5_HW_INTR_STATUS_INTR_MDP;
+	}
 
-	if (intr & MDP5_HW_INTR_STATUS_INTR_HDMI)
-		hdmi_irq(0, priv->hdmi);
+	while (intr) {
+		irq_hw_number_t hwirq = fls(intr) - 1;
+		generic_handle_irq(irq_find_mapping(
+				mdp5_kms->irqcontroller.domain, hwirq));
+		intr &= ~(1 << hwirq);
+	}
 
 	return IRQ_HANDLED;
 }
@@ -110,3 +117,82 @@ void mdp5_disable_vblank(struct msm_kms *kms, struct drm_crtc *crtc)
 	mdp_update_vblank_mask(to_mdp_kms(kms),
 			mdp5_crtc_vblank(crtc), false);
 }
+
+/*
+ * interrupt-controller implementation, so sub-blocks (hdmi/eDP/dsi/etc)
+ * can register to get their irq's delivered
+ */
+
+#define VALID_IRQS  (MDP5_HW_INTR_STATUS_INTR_DSI0 | \
+		MDP5_HW_INTR_STATUS_INTR_DSI1 | \
+		MDP5_HW_INTR_STATUS_INTR_HDMI | \
+		MDP5_HW_INTR_STATUS_INTR_EDP)
+
+static void mdp5_hw_mask_irq(struct irq_data *irqd)
+{
+	struct mdp5_kms *mdp5_kms = irq_data_get_irq_chip_data(irqd);
+	smp_mb__before_atomic();
+	clear_bit(irqd->hwirq, &mdp5_kms->irqcontroller.enabled_mask);
+	smp_mb__after_atomic();
+}
+
+static void mdp5_hw_unmask_irq(struct irq_data *irqd)
+{
+	struct mdp5_kms *mdp5_kms = irq_data_get_irq_chip_data(irqd);
+	smp_mb__before_atomic();
+	set_bit(irqd->hwirq, &mdp5_kms->irqcontroller.enabled_mask);
+	smp_mb__after_atomic();
+}
+
+static struct irq_chip mdp5_hw_irq_chip = {
+	.name		= "mdp5",
+	.irq_mask	= mdp5_hw_mask_irq,
+	.irq_unmask	= mdp5_hw_unmask_irq,
+};
+
+static int mdp5_hw_irqdomain_map(struct irq_domain *d,
+		unsigned int irq, irq_hw_number_t hwirq)
+{
+	struct mdp5_kms *mdp5_kms = d->host_data;
+
+	if (!(VALID_IRQS & (1 << hwirq)))
+		return -EPERM;
+
+	irq_set_chip_and_handler(irq, &mdp5_hw_irq_chip, handle_level_irq);
+	irq_set_chip_data(irq, mdp5_kms);
+	set_irq_flags(irq, IRQF_VALID);
+
+	return 0;
+}
+
+static struct irq_domain_ops mdp5_hw_irqdomain_ops = {
+	.map = mdp5_hw_irqdomain_map,
+	.xlate = irq_domain_xlate_onecell,
+};
+
+
+int mdp5_irq_domain_init(struct mdp5_kms *mdp5_kms)
+{
+	struct device *dev = mdp5_kms->dev->dev;
+	struct irq_domain *d;
+
+	d = irq_domain_add_linear(dev->of_node, 32,
+			&mdp5_hw_irqdomain_ops, mdp5_kms);
+	if (!d) {
+		dev_err(dev, "mdp5 irq domain add failed\n");
+		return -ENXIO;
+	}
+
+	mdp5_kms->irqcontroller.enabled_mask = 0;
+	mdp5_kms->irqcontroller.domain = d;
+
+	return 0;
+}
+
+void mdp5_irq_domain_fini(struct mdp5_kms *mdp5_kms)
+{
+	if (mdp5_kms->irqcontroller.domain) {
+		irq_domain_remove(mdp5_kms->irqcontroller.domain);
+		mdp5_kms->irqcontroller.domain = NULL;
+	}
+}
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.c
index ce0308124a72..a11f1b80c488 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.c
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (c) 2014, The Linux Foundation. All rights reserved.
  * Copyright (C) 2013 Red Hat
  * Author: Rob Clark <robdclark@gmail.com>
  *
@@ -24,145 +25,11 @@ static const char *iommu_ports[] = {
 		"mdp_0",
 };
 
-static struct mdp5_platform_config *mdp5_get_config(struct platform_device *dev);
-
-const struct mdp5_config *mdp5_cfg;
-
-static const struct mdp5_config msm8x74_config = {
-	.name = "msm8x74",
-	.ctl = {
-		.count = 5,
-		.base = { 0x00600, 0x00700, 0x00800, 0x00900, 0x00a00 },
-	},
-	.pipe_vig = {
-		.count = 3,
-		.base = { 0x01200, 0x01600, 0x01a00 },
-	},
-	.pipe_rgb = {
-		.count = 3,
-		.base = { 0x01e00, 0x02200, 0x02600 },
-	},
-	.pipe_dma = {
-		.count = 2,
-		.base = { 0x02a00, 0x02e00 },
-	},
-	.lm = {
-		.count = 5,
-		.base = { 0x03200, 0x03600, 0x03a00, 0x03e00, 0x04200 },
-	},
-	.dspp = {
-		.count = 3,
-		.base = { 0x04600, 0x04a00, 0x04e00 },
-	},
-	.ad = {
-		.count = 2,
-		.base = { 0x13100, 0x13300 }, /* NOTE: no ad in v1.0 */
-	},
-	.intf = {
-		.count = 4,
-		.base = { 0x12500, 0x12700, 0x12900, 0x12b00 },
-	},
-};
-
-static const struct mdp5_config apq8084_config = {
-	.name = "apq8084",
-	.ctl = {
-		.count = 5,
-		.base = { 0x00600, 0x00700, 0x00800, 0x00900, 0x00a00 },
-	},
-	.pipe_vig = {
-		.count = 4,
-		.base = { 0x01200, 0x01600, 0x01a00, 0x01e00 },
-	},
-	.pipe_rgb = {
-		.count = 4,
-		.base = { 0x02200, 0x02600, 0x02a00, 0x02e00 },
-	},
-	.pipe_dma = {
-		.count = 2,
-		.base = { 0x03200, 0x03600 },
-	},
-	.lm = {
-		.count = 6,
-		.base = { 0x03a00, 0x03e00, 0x04200, 0x04600, 0x04a00, 0x04e00 },
-	},
-	.dspp = {
-		.count = 4,
-		.base = { 0x05200, 0x05600, 0x05a00, 0x05e00 },
-
-	},
-	.ad = {
-		.count = 3,
-		.base = { 0x13500, 0x13700, 0x13900 },
-	},
-	.intf = {
-		.count = 5,
-		.base = { 0x12500, 0x12700, 0x12900, 0x12b00, 0x12d00 },
-	},
-};
-
-struct mdp5_config_entry {
-	int revision;
-	const struct mdp5_config *config;
-};
-
-static const struct mdp5_config_entry mdp5_configs[] = {
-	{ .revision = 0, .config = &msm8x74_config },
-	{ .revision = 2, .config = &msm8x74_config },
-	{ .revision = 3, .config = &apq8084_config },
-};
-
-static int mdp5_select_hw_cfg(struct msm_kms *kms)
-{
-	struct mdp5_kms *mdp5_kms = to_mdp5_kms(to_mdp_kms(kms));
-	struct drm_device *dev = mdp5_kms->dev;
-	uint32_t version, major, minor;
-	int i, ret = 0;
-
-	mdp5_enable(mdp5_kms);
-	version = mdp5_read(mdp5_kms, REG_MDP5_MDP_VERSION);
-	mdp5_disable(mdp5_kms);
-
-	major = FIELD(version, MDP5_MDP_VERSION_MAJOR);
-	minor = FIELD(version, MDP5_MDP_VERSION_MINOR);
-
-	DBG("found MDP5 version v%d.%d", major, minor);
-
-	if (major != 1) {
-		dev_err(dev->dev, "unexpected MDP major version: v%d.%d\n",
-				major, minor);
-		ret = -ENXIO;
-		goto out;
-	}
-
-	mdp5_kms->rev = minor;
-
-	/* only after mdp5_cfg global pointer's init can we access the hw */
-	for (i = 0; i < ARRAY_SIZE(mdp5_configs); i++) {
-		if (mdp5_configs[i].revision != minor)
-			continue;
-		mdp5_kms->hw_cfg = mdp5_cfg = mdp5_configs[i].config;
-		break;
-	}
-	if (unlikely(!mdp5_kms->hw_cfg)) {
-		dev_err(dev->dev, "unexpected MDP minor revision: v%d.%d\n",
-				major, minor);
-		ret = -ENXIO;
-		goto out;
-	}
-
-	DBG("MDP5: %s config selected", mdp5_kms->hw_cfg->name);
-
-	return 0;
-out:
-	return ret;
-}
-
 static int mdp5_hw_init(struct msm_kms *kms)
 {
 	struct mdp5_kms *mdp5_kms = to_mdp5_kms(to_mdp_kms(kms));
 	struct drm_device *dev = mdp5_kms->dev;
-	int i;
+	unsigned long flags;
 
 	pm_runtime_get_sync(dev->dev);
 
@@ -190,10 +57,11 @@ static int mdp5_hw_init(struct msm_kms *kms)
 	 * care.
 	 */
 
+	spin_lock_irqsave(&mdp5_kms->resource_lock, flags);
 	mdp5_write(mdp5_kms, REG_MDP5_DISP_INTF_SEL, 0);
+	spin_unlock_irqrestore(&mdp5_kms->resource_lock, flags);
 
-	for (i = 0; i < mdp5_kms->hw_cfg->ctl.count; i++)
-		mdp5_write(mdp5_kms, REG_MDP5_CTL_OP(i), 0);
+	mdp5_ctlm_hw_reset(mdp5_kms->ctlm);
 
 	pm_runtime_put_sync(dev->dev);
 
@@ -221,10 +89,20 @@ static void mdp5_destroy(struct msm_kms *kms)
 	struct mdp5_kms *mdp5_kms = to_mdp5_kms(to_mdp_kms(kms));
 	struct msm_mmu *mmu = mdp5_kms->mmu;
 
+	mdp5_irq_domain_fini(mdp5_kms);
+
 	if (mmu) {
 		mmu->funcs->detach(mmu, iommu_ports, ARRAY_SIZE(iommu_ports));
 		mmu->funcs->destroy(mmu);
 	}
+
+	if (mdp5_kms->ctlm)
+		mdp5_ctlm_destroy(mdp5_kms->ctlm);
+	if (mdp5_kms->smp)
+		mdp5_smp_destroy(mdp5_kms->smp);
+	if (mdp5_kms->cfg)
+		mdp5_cfg_destroy(mdp5_kms->cfg);
+
 	kfree(mdp5_kms);
 }
 
@@ -274,17 +152,31 @@ static int modeset_init(struct mdp5_kms *mdp5_kms)
 	static const enum mdp5_pipe crtcs[] = {
 			SSPP_RGB0, SSPP_RGB1, SSPP_RGB2, SSPP_RGB3,
 	};
+	static const enum mdp5_pipe pub_planes[] = {
+			SSPP_VIG0, SSPP_VIG1, SSPP_VIG2, SSPP_VIG3,
+	};
 	struct drm_device *dev = mdp5_kms->dev;
 	struct msm_drm_private *priv = dev->dev_private;
 	struct drm_encoder *encoder;
+	const struct mdp5_cfg_hw *hw_cfg;
 	int i, ret;
 
-	/* construct CRTCs: */
-	for (i = 0; i < mdp5_kms->hw_cfg->pipe_rgb.count; i++) {
+	hw_cfg = mdp5_cfg_get_hw_config(mdp5_kms->cfg);
+
+	/* register our interrupt-controller for hdmi/eDP/dsi/etc
+	 * to use for irqs routed through mdp:
+	 */
+	ret = mdp5_irq_domain_init(mdp5_kms);
+	if (ret)
+		goto fail;
+
+	/* construct CRTCs and their private planes: */
+	for (i = 0; i < hw_cfg->pipe_rgb.count; i++) {
 		struct drm_plane *plane;
 		struct drm_crtc *crtc;
 
-		plane = mdp5_plane_init(dev, crtcs[i], true);
+		plane = mdp5_plane_init(dev, crtcs[i], true,
+				hw_cfg->pipe_rgb.base[i]);
 		if (IS_ERR(plane)) {
 			ret = PTR_ERR(plane);
 			dev_err(dev->dev, "failed to construct plane for %s (%d)\n",
@@ -302,6 +194,20 @@ static int modeset_init(struct mdp5_kms *mdp5_kms)
 		priv->crtcs[priv->num_crtcs++] = crtc;
 	}
 
+	/* Construct public planes: */
+	for (i = 0; i < hw_cfg->pipe_vig.count; i++) {
+		struct drm_plane *plane;
+
+		plane = mdp5_plane_init(dev, pub_planes[i], false,
+				hw_cfg->pipe_vig.base[i]);
+		if (IS_ERR(plane)) {
+			ret = PTR_ERR(plane);
+			dev_err(dev->dev, "failed to construct %s plane: %d\n",
+					pipe2name(pub_planes[i]), ret);
+			goto fail;
+		}
+	}
+
 	/* Construct encoder for HDMI: */
 	encoder = mdp5_encoder_init(dev, 3, INTF_HDMI);
 	if (IS_ERR(encoder)) {
@@ -338,6 +244,21 @@ fail:
 	return ret;
 }
 
+static void read_hw_revision(struct mdp5_kms *mdp5_kms,
+		uint32_t *major, uint32_t *minor)
+{
+	uint32_t version;
+
+	mdp5_enable(mdp5_kms);
+	version = mdp5_read(mdp5_kms, REG_MDP5_MDP_VERSION);
+	mdp5_disable(mdp5_kms);
+
+	*major = FIELD(version, MDP5_MDP_VERSION_MAJOR);
+	*minor = FIELD(version, MDP5_MDP_VERSION_MINOR);
+
+	DBG("MDP5 version v%d.%d", *major, *minor);
+}
+
 static int get_clk(struct platform_device *pdev, struct clk **clkp,
 		const char *name)
 {
@@ -354,10 +275,11 @@ static int get_clk(struct platform_device *pdev, struct clk **clkp,
 struct msm_kms *mdp5_kms_init(struct drm_device *dev)
 {
 	struct platform_device *pdev = dev->platformdev;
-	struct mdp5_platform_config *config = mdp5_get_config(pdev);
+	struct mdp5_cfg *config;
 	struct mdp5_kms *mdp5_kms;
 	struct msm_kms *kms = NULL;
 	struct msm_mmu *mmu;
+	uint32_t major, minor;
 	int i, ret;
 
 	mdp5_kms = kzalloc(sizeof(*mdp5_kms), GFP_KERNEL);
@@ -367,12 +289,13 @@ struct msm_kms *mdp5_kms_init(struct drm_device *dev)
 		goto fail;
 	}
 
+	spin_lock_init(&mdp5_kms->resource_lock);
+
 	mdp_kms_init(&mdp5_kms->base, &kms_funcs);
 
 	kms = &mdp5_kms->base.base;
 
 	mdp5_kms->dev = dev;
-	mdp5_kms->smp_blk_cnt = config->smp_blk_cnt;
 
 	mdp5_kms->mmio = msm_ioremap(pdev, "mdp_phys", "MDP5");
 	if (IS_ERR(mdp5_kms->mmio)) {
@@ -417,24 +340,52 @@ struct msm_kms *mdp5_kms_init(struct drm_device *dev)
 	if (ret)
 		goto fail;
 
-	ret = clk_set_rate(mdp5_kms->src_clk, config->max_clk);
+	/* we need to set a default rate before enabling.  Set a safe
+	 * rate first, then figure out hw revision, and then set a
+	 * more optimal rate:
+	 */
+	clk_set_rate(mdp5_kms->src_clk, 200000000);
 
-	ret = mdp5_select_hw_cfg(kms);
-	if (ret)
+	read_hw_revision(mdp5_kms, &major, &minor);
+
+	mdp5_kms->cfg = mdp5_cfg_init(mdp5_kms, major, minor);
+	if (IS_ERR(mdp5_kms->cfg)) {
+		ret = PTR_ERR(mdp5_kms->cfg);
+		mdp5_kms->cfg = NULL;
+		goto fail;
+	}
+
+	config = mdp5_cfg_get_config(mdp5_kms->cfg);
+
+	/* TODO: compute core clock rate at runtime */
+	clk_set_rate(mdp5_kms->src_clk, config->hw->max_clk);
+
+	mdp5_kms->smp = mdp5_smp_init(mdp5_kms->dev, &config->hw->smp);
+	if (IS_ERR(mdp5_kms->smp)) {
+		ret = PTR_ERR(mdp5_kms->smp);
+		mdp5_kms->smp = NULL;
 		goto fail;
+	}
+
+	mdp5_kms->ctlm = mdp5_ctlm_init(dev, mdp5_kms->mmio, config->hw);
+	if (IS_ERR(mdp5_kms->ctlm)) {
+		ret = PTR_ERR(mdp5_kms->ctlm);
+		mdp5_kms->ctlm = NULL;
+		goto fail;
+	}
 
 	/* make sure things are off before attaching iommu (bootloader could
 	 * have left things on, in which case we'll start getting faults if
 	 * we don't disable):
 	 */
 	mdp5_enable(mdp5_kms);
-	for (i = 0; i < mdp5_kms->hw_cfg->intf.count; i++)
+	for (i = 0; i < config->hw->intf.count; i++)
 		mdp5_write(mdp5_kms, REG_MDP5_INTF_TIMING_ENGINE_EN(i), 0);
 	mdp5_disable(mdp5_kms);
 	mdelay(16);
 
-	if (config->iommu) {
-		mmu = msm_iommu_new(&pdev->dev, config->iommu);
+	if (config->platform.iommu) {
+		mmu = msm_iommu_new(&pdev->dev, config->platform.iommu);
 		if (IS_ERR(mmu)) {
 			ret = PTR_ERR(mmu);
 			dev_err(dev->dev, "failed to init iommu: %d\n", ret);
@@ -475,18 +426,3 @@ fail:
 		mdp5_destroy(kms);
 	return ERR_PTR(ret);
 }
-
-static struct mdp5_platform_config *mdp5_get_config(struct platform_device *dev)
-{
-	static struct mdp5_platform_config config = {};
-#ifdef CONFIG_OF
-	/* TODO */
-#endif
-	config.iommu = iommu_domain_alloc(&platform_bus_type);
-	/* TODO hard-coded in downstream mdss, but should it be? */
-	config.max_clk = 200000000;
-	/* TODO get from DT: */
-	config.smp_blk_cnt = 22;
-
-	return &config;
-}
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.h b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.h
index c91101d5ac0f..dd69c77c0d64 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.h
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.h
@@ -21,25 +21,9 @@
 #include "msm_drv.h"
 #include "msm_kms.h"
 #include "mdp/mdp_kms.h"
-/* dynamic offsets used by mdp5.xml.h (initialized in mdp5_kms.c) */
-#define MDP5_MAX_BASES		8
-struct mdp5_sub_block {
-	int	count;
-	uint32_t base[MDP5_MAX_BASES];
-};
-struct mdp5_config {
-	char  *name;
-	struct mdp5_sub_block ctl;
-	struct mdp5_sub_block pipe_vig;
-	struct mdp5_sub_block pipe_rgb;
-	struct mdp5_sub_block pipe_dma;
-	struct mdp5_sub_block lm;
-	struct mdp5_sub_block dspp;
-	struct mdp5_sub_block ad;
-	struct mdp5_sub_block intf;
-};
-extern const struct mdp5_config *mdp5_cfg;
+#include "mdp5_cfg.h"	/* must be included before mdp5.xml.h */
 #include "mdp5.xml.h"
+#include "mdp5_ctl.h"
 #include "mdp5_smp.h"
 
 struct mdp5_kms {
@@ -47,17 +31,14 @@ struct mdp5_kms {
 
 	struct drm_device *dev;
 
-	int rev;
-	const struct mdp5_config *hw_cfg;
+	struct mdp5_cfg_handler *cfg;
 
 	/* mapper-id used to request GEM buffer mapped for scanout: */
 	int id;
 	struct msm_mmu *mmu;
 
-	/* for tracking smp allocation amongst pipes: */
-	mdp5_smp_state_t smp_state;
-	struct mdp5_client_smp_state smp_client_state[CID_MAX];
-	int smp_blk_cnt;
+	struct mdp5_smp *smp;
+	struct mdp5_ctl_manager *ctlm;
 
 	/* io/register spaces: */
 	void __iomem *mmio, *vbif;
@@ -71,16 +52,47 @@ struct mdp5_kms {
 	struct clk *lut_clk;
 	struct clk *vsync_clk;
 
+	/*
+	 * lock to protect access to global resources: ie., following register:
+	 *	- REG_MDP5_DISP_INTF_SEL
+	 */
+	spinlock_t resource_lock;
+
 	struct mdp_irq error_handler;
+
+	struct {
+		volatile unsigned long enabled_mask;
+		struct irq_domain *domain;
+	} irqcontroller;
 };
 #define to_mdp5_kms(x) container_of(x, struct mdp5_kms, base)
 
-/* platform config data (ie. from DT, or pdata) */
-struct mdp5_platform_config {
-	struct iommu_domain *iommu;
-	uint32_t max_clk;
-	int smp_blk_cnt;
+struct mdp5_plane_state {
+	struct drm_plane_state base;
+
+	/* "virtual" zpos.. we calculate actual mixer-stage at runtime
+	 * by sorting the attached planes by zpos and then assigning
+	 * mixer stage lowest to highest.  Private planes get default
+	 * zpos of zero, and public planes a unique value that is
+	 * greater than zero.  This way, things work out if a naive
+	 * userspace assigns planes to a crtc without setting zpos.
+	 */
+	int zpos;
+
+	/* the actual mixer stage, calculated in crtc->atomic_check()
+	 * NOTE: this should move to mdp5_crtc_state, when that exists
+	 */
+	enum mdp_mixer_stage_id stage;
+
+	/* some additional transactional status to help us know in the
+	 * apply path whether we need to update SMP allocation, and
+	 * whether current update is still pending:
+	 */
+	bool mode_changed : 1;
+	bool pending : 1;
 };
+#define to_mdp5_plane_state(x) \
+		container_of(x, struct mdp5_plane_state, base)
 
 static inline void mdp5_write(struct mdp5_kms *mdp5_kms, u32 reg, u32 data)
 {
@@ -105,23 +117,6 @@ static inline const char *pipe2name(enum mdp5_pipe pipe)
 	return names[pipe];
 }
 
-static inline uint32_t pipe2flush(enum mdp5_pipe pipe)
-{
-	switch (pipe) {
-	case SSPP_VIG0: return MDP5_CTL_FLUSH_VIG0;
-	case SSPP_VIG1: return MDP5_CTL_FLUSH_VIG1;
-	case SSPP_VIG2: return MDP5_CTL_FLUSH_VIG2;
-	case SSPP_RGB0: return MDP5_CTL_FLUSH_RGB0;
-	case SSPP_RGB1: return MDP5_CTL_FLUSH_RGB1;
-	case SSPP_RGB2: return MDP5_CTL_FLUSH_RGB2;
-	case SSPP_DMA0: return MDP5_CTL_FLUSH_DMA0;
-	case SSPP_DMA1: return MDP5_CTL_FLUSH_DMA1;
-	case SSPP_VIG3: return MDP5_CTL_FLUSH_VIG3;
-	case SSPP_RGB3: return MDP5_CTL_FLUSH_RGB3;
-	default:        return 0;
-	}
-}
-
 static inline int pipe2nclients(enum mdp5_pipe pipe)
 {
 	switch (pipe) {
@@ -135,34 +130,6 @@ static inline int pipe2nclients(enum mdp5_pipe pipe)
 	}
 }
 
-static inline enum mdp5_client_id pipe2client(enum mdp5_pipe pipe, int plane)
-{
-	WARN_ON(plane >= pipe2nclients(pipe));
-	switch (pipe) {
-	case SSPP_VIG0: return CID_VIG0_Y + plane;
-	case SSPP_VIG1: return CID_VIG1_Y + plane;
-	case SSPP_VIG2: return CID_VIG2_Y + plane;
-	case SSPP_RGB0: return CID_RGB0;
-	case SSPP_RGB1: return CID_RGB1;
-	case SSPP_RGB2: return CID_RGB2;
-	case SSPP_DMA0: return CID_DMA0_Y + plane;
-	case SSPP_DMA1: return CID_DMA1_Y + plane;
-	case SSPP_VIG3: return CID_VIG3_Y + plane;
-	case SSPP_RGB3: return CID_RGB3;
-	default:        return CID_UNUSED;
-	}
-}
-
-static inline uint32_t mixer2flush(int lm)
-{
-	switch (lm) {
-	case 0:  return MDP5_CTL_FLUSH_LM0;
-	case 1:  return MDP5_CTL_FLUSH_LM1;
-	case 2:  return MDP5_CTL_FLUSH_LM2;
-	default: return 0;
-	}
-}
-
 static inline uint32_t intf2err(int intf)
 {
 	switch (intf) {
@@ -195,6 +162,8 @@ void mdp5_irq_uninstall(struct msm_kms *kms);
 irqreturn_t mdp5_irq(struct msm_kms *kms);
 int mdp5_enable_vblank(struct msm_kms *kms, struct drm_crtc *crtc);
 void mdp5_disable_vblank(struct msm_kms *kms, struct drm_crtc *crtc);
+int mdp5_irq_domain_init(struct mdp5_kms *mdp5_kms);
+void mdp5_irq_domain_fini(struct mdp5_kms *mdp5_kms);
 
 static inline
 uint32_t mdp5_get_formats(enum mdp5_pipe pipe, uint32_t *pixel_formats,
@@ -208,26 +177,18 @@ uint32_t mdp5_get_formats(enum mdp5_pipe pipe, uint32_t *pixel_formats,
 
 void mdp5_plane_install_properties(struct drm_plane *plane,
 		struct drm_mode_object *obj);
-void mdp5_plane_set_scanout(struct drm_plane *plane,
-		struct drm_framebuffer *fb);
-int mdp5_plane_mode_set(struct drm_plane *plane,
-		struct drm_crtc *crtc, struct drm_framebuffer *fb,
-		int crtc_x, int crtc_y,
-		unsigned int crtc_w, unsigned int crtc_h,
-		uint32_t src_x, uint32_t src_y,
-		uint32_t src_w, uint32_t src_h);
+uint32_t mdp5_plane_get_flush(struct drm_plane *plane);
 void mdp5_plane_complete_flip(struct drm_plane *plane);
 enum mdp5_pipe mdp5_plane_pipe(struct drm_plane *plane);
 struct drm_plane *mdp5_plane_init(struct drm_device *dev,
-		enum mdp5_pipe pipe, bool private_plane);
+		enum mdp5_pipe pipe, bool private_plane, uint32_t reg_offset);
 
 uint32_t mdp5_crtc_vblank(struct drm_crtc *crtc);
 
+int mdp5_crtc_get_lm(struct drm_crtc *crtc);
 void mdp5_crtc_cancel_pending_flip(struct drm_crtc *crtc, struct drm_file *file);
 void mdp5_crtc_set_intf(struct drm_crtc *crtc, int intf,
 		enum mdp5_intf intf_id);
-void mdp5_crtc_attach(struct drm_crtc *crtc, struct drm_plane *plane);
-void mdp5_crtc_detach(struct drm_crtc *crtc, struct drm_plane *plane);
 struct drm_crtc *mdp5_crtc_init(struct drm_device *dev,
 		struct drm_plane *plane, int id);
 
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_plane.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_plane.c
index f3daec4412ad..533df7caa310 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_plane.c
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_plane.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (c) 2014 The Linux Foundation. All rights reserved.
  * Copyright (C) 2013 Red Hat
  * Author: Rob Clark <robdclark@gmail.com>
  *
@@ -17,6 +18,7 @@
 
 #include "mdp5_kms.h"
 
+#define MAX_PLANE	4
 
 struct mdp5_plane {
 	struct drm_plane base;
@@ -24,6 +26,11 @@ struct mdp5_plane {
 
 	enum mdp5_pipe pipe;
 
+	spinlock_t pipe_lock;	/* protect REG_MDP5_PIPE_* registers */
+	uint32_t reg_offset;
+
+	uint32_t flush_mask;	/* used to commit pipe registers */
+
 	uint32_t nformats;
 	uint32_t formats[32];
 
@@ -31,31 +38,24 @@ struct mdp5_plane {
 };
 #define to_mdp5_plane(x) container_of(x, struct mdp5_plane, base)
 
+static int mdp5_plane_mode_set(struct drm_plane *plane,
+		struct drm_crtc *crtc, struct drm_framebuffer *fb,
+		int crtc_x, int crtc_y,
+		unsigned int crtc_w, unsigned int crtc_h,
+		uint32_t src_x, uint32_t src_y,
+		uint32_t src_w, uint32_t src_h);
+static void set_scanout_locked(struct drm_plane *plane,
+		struct drm_framebuffer *fb);
+
 static struct mdp5_kms *get_kms(struct drm_plane *plane)
 {
 	struct msm_drm_private *priv = plane->dev->dev_private;
 	return to_mdp5_kms(to_mdp_kms(priv->kms));
 }
 
-static int mdp5_plane_update(struct drm_plane *plane,
-		struct drm_crtc *crtc, struct drm_framebuffer *fb,
-		int crtc_x, int crtc_y,
-		unsigned int crtc_w, unsigned int crtc_h,
-		uint32_t src_x, uint32_t src_y,
-		uint32_t src_w, uint32_t src_h)
+static bool plane_enabled(struct drm_plane_state *state)
 {
-	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
-
-	mdp5_plane->enabled = true;
-
-	if (plane->fb)
-		drm_framebuffer_unreference(plane->fb);
-
-	drm_framebuffer_reference(fb);
-
-	return mdp5_plane_mode_set(plane, crtc, fb,
-			crtc_x, crtc_y, crtc_w, crtc_h,
-			src_x, src_y, src_w, src_h);
+	return state->fb && state->crtc;
 }
 
 static int mdp5_plane_disable(struct drm_plane *plane)
@@ -63,21 +63,13 @@ static int mdp5_plane_disable(struct drm_plane *plane)
 	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
 	struct mdp5_kms *mdp5_kms = get_kms(plane);
 	enum mdp5_pipe pipe = mdp5_plane->pipe;
-	int i;
 
 	DBG("%s: disable", mdp5_plane->name);
 
-	/* update our SMP request to zero (release all our blks): */
-	for (i = 0; i < pipe2nclients(pipe); i++)
-		mdp5_smp_request(mdp5_kms, pipe2client(pipe, i), 0);
-
-	/* TODO detaching now will cause us not to get the last
-	 * vblank and mdp5_smp_commit().. so other planes will
-	 * still see smp blocks previously allocated to us as
-	 * in-use..
-	 */
-	if (plane->crtc)
-		mdp5_crtc_detach(plane->crtc, plane);
+	if (mdp5_kms) {
+		/* Release the memory we requested earlier from the SMP: */
+		mdp5_smp_release(mdp5_kms->smp, pipe);
+	}
 
 	return 0;
 }
@@ -85,11 +77,8 @@ static int mdp5_plane_disable(struct drm_plane *plane)
 static void mdp5_plane_destroy(struct drm_plane *plane)
 {
 	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
-	struct msm_drm_private *priv = plane->dev->dev_private;
-
-	if (priv->kms)
-		mdp5_plane_disable(plane);
 
+	drm_plane_helper_disable(plane);
 	drm_plane_cleanup(plane);
 
 	kfree(mdp5_plane);
@@ -109,109 +98,185 @@ int mdp5_plane_set_property(struct drm_plane *plane,
 	return -EINVAL;
 }
 
+static void mdp5_plane_reset(struct drm_plane *plane)
+{
+	struct mdp5_plane_state *mdp5_state;
+
+	if (plane->state && plane->state->fb)
+		drm_framebuffer_unreference(plane->state->fb);
+
+	kfree(to_mdp5_plane_state(plane->state));
+	mdp5_state = kzalloc(sizeof(*mdp5_state), GFP_KERNEL);
+
+	if (plane->type == DRM_PLANE_TYPE_PRIMARY) {
+		mdp5_state->zpos = 0;
+	} else {
+		mdp5_state->zpos = 1 + drm_plane_index(plane);
+	}
+
+	plane->state = &mdp5_state->base;
+}
+
+static struct drm_plane_state *
+mdp5_plane_duplicate_state(struct drm_plane *plane)
+{
+	struct mdp5_plane_state *mdp5_state;
+
+	if (WARN_ON(!plane->state))
+		return NULL;
+
+	mdp5_state = kmemdup(to_mdp5_plane_state(plane->state),
+			sizeof(*mdp5_state), GFP_KERNEL);
+
+	if (mdp5_state && mdp5_state->base.fb)
+		drm_framebuffer_reference(mdp5_state->base.fb);
+
+	mdp5_state->mode_changed = false;
+	mdp5_state->pending = false;
+
+	return &mdp5_state->base;
+}
+
+static void mdp5_plane_destroy_state(struct drm_plane *plane,
+		struct drm_plane_state *state)
+{
+	if (state->fb)
+		drm_framebuffer_unreference(state->fb);
+
+	kfree(to_mdp5_plane_state(state));
+}
+
 static const struct drm_plane_funcs mdp5_plane_funcs = {
-		.update_plane = mdp5_plane_update,
-		.disable_plane = mdp5_plane_disable,
+		.update_plane = drm_atomic_helper_update_plane,
+		.disable_plane = drm_atomic_helper_disable_plane,
 		.destroy = mdp5_plane_destroy,
 		.set_property = mdp5_plane_set_property,
+		.reset = mdp5_plane_reset,
+		.atomic_duplicate_state = mdp5_plane_duplicate_state,
+		.atomic_destroy_state = mdp5_plane_destroy_state,
 };
 
-void mdp5_plane_set_scanout(struct drm_plane *plane,
+static int mdp5_plane_prepare_fb(struct drm_plane *plane,
 		struct drm_framebuffer *fb)
 {
 	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
 	struct mdp5_kms *mdp5_kms = get_kms(plane);
-	enum mdp5_pipe pipe = mdp5_plane->pipe;
-	uint32_t nplanes = drm_format_num_planes(fb->pixel_format);
-	uint32_t iova[4];
-	int i;
-
-	for (i = 0; i < nplanes; i++) {
-		struct drm_gem_object *bo = msm_framebuffer_bo(fb, i);
-		msm_gem_get_iova(bo, mdp5_kms->id, &iova[i]);
-	}
-	for (; i < 4; i++)
-		iova[i] = 0;
 
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC_STRIDE_A(pipe),
-			MDP5_PIPE_SRC_STRIDE_A_P0(fb->pitches[0]) |
-			MDP5_PIPE_SRC_STRIDE_A_P1(fb->pitches[1]));
-
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC_STRIDE_B(pipe),
-			MDP5_PIPE_SRC_STRIDE_B_P2(fb->pitches[2]) |
-			MDP5_PIPE_SRC_STRIDE_B_P3(fb->pitches[3]));
-
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC0_ADDR(pipe), iova[0]);
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC1_ADDR(pipe), iova[1]);
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC2_ADDR(pipe), iova[2]);
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC3_ADDR(pipe), iova[3]);
-
-	plane->fb = fb;
+	DBG("%s: prepare: FB[%u]", mdp5_plane->name, fb->base.id);
+	return msm_framebuffer_prepare(fb, mdp5_kms->id);
 }
 
-/* NOTE: looks like if horizontal decimation is used (if we supported that)
- * then the width used to calculate SMP block requirements is the post-
- * decimated width.  Ie. SMP buffering sits downstream of decimation (which
- * presumably happens during the dma from scanout buffer).
- */
-static int request_smp_blocks(struct drm_plane *plane, uint32_t format,
-		uint32_t nplanes, uint32_t width)
+static void mdp5_plane_cleanup_fb(struct drm_plane *plane,
+		struct drm_framebuffer *fb)
 {
-	struct drm_device *dev = plane->dev;
 	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
 	struct mdp5_kms *mdp5_kms = get_kms(plane);
-	enum mdp5_pipe pipe = mdp5_plane->pipe;
-	int i, hsub, nlines, nblks, ret;
 
-	hsub = drm_format_horz_chroma_subsampling(format);
+	DBG("%s: cleanup: FB[%u]", mdp5_plane->name, fb->base.id);
+	msm_framebuffer_cleanup(fb, mdp5_kms->id);
+}
 
-	/* different if BWC (compressed framebuffer?) enabled: */
-	nlines = 2;
+static int mdp5_plane_atomic_check(struct drm_plane *plane,
+		struct drm_plane_state *state)
+{
+	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
+	struct drm_plane_state *old_state = plane->state;
 
-	for (i = 0, nblks = 0; i < nplanes; i++) {
-		int n, fetch_stride, cpp;
+	DBG("%s: check (%d -> %d)", mdp5_plane->name,
+			plane_enabled(old_state), plane_enabled(state));
 
-		cpp = drm_format_plane_cpp(format, i);
-		fetch_stride = width * cpp / (i ? hsub : 1);
+	if (plane_enabled(state) && plane_enabled(old_state)) {
+		/* we cannot change SMP block configuration during scanout: */
+		bool full_modeset = false;
+		if (state->fb->pixel_format != old_state->fb->pixel_format) {
+			DBG("%s: pixel_format change!", mdp5_plane->name);
+			full_modeset = true;
+		}
+		if (state->src_w != old_state->src_w) {
+			DBG("%s: src_w change!", mdp5_plane->name);
+			full_modeset = true;
+		}
+		if (to_mdp5_plane_state(old_state)->pending) {
+			DBG("%s: still pending!", mdp5_plane->name);
+			full_modeset = true;
+		}
+		if (full_modeset) {
+			struct drm_crtc_state *crtc_state =
+					drm_atomic_get_crtc_state(state->state, state->crtc);
+			crtc_state->mode_changed = true;
+			to_mdp5_plane_state(state)->mode_changed = true;
+		}
+	} else {
+		to_mdp5_plane_state(state)->mode_changed = true;
+	}
 
-		n = DIV_ROUND_UP(fetch_stride * nlines, SMP_BLK_SIZE);
+	return 0;
+}
 
-		/* for hw rev v1.00 */
-		if (mdp5_kms->rev == 0)
-			n = roundup_pow_of_two(n);
+static void mdp5_plane_atomic_update(struct drm_plane *plane)
+{
+	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
+	struct drm_plane_state *state = plane->state;
 
-		DBG("%s[%d]: request %d SMP blocks", mdp5_plane->name, i, n);
-		ret = mdp5_smp_request(mdp5_kms, pipe2client(pipe, i), n);
-		if (ret) {
-			dev_err(dev->dev, "Could not allocate %d SMP blocks: %d\n",
-					n, ret);
-			return ret;
-		}
+	DBG("%s: update", mdp5_plane->name);
 
-		nblks += n;
+	if (!plane_enabled(state)) {
+		to_mdp5_plane_state(state)->pending = true;
+		mdp5_plane_disable(plane);
+	} else if (to_mdp5_plane_state(state)->mode_changed) {
+		int ret;
+		to_mdp5_plane_state(state)->pending = true;
+		ret = mdp5_plane_mode_set(plane,
+				state->crtc, state->fb,
+				state->crtc_x, state->crtc_y,
+				state->crtc_w, state->crtc_h,
+				state->src_x,  state->src_y,
+				state->src_w, state->src_h);
+		/* atomic_check should have ensured that this doesn't fail */
+		WARN_ON(ret < 0);
+	} else {
+		unsigned long flags;
+		spin_lock_irqsave(&mdp5_plane->pipe_lock, flags);
+		set_scanout_locked(plane, state->fb);
+		spin_unlock_irqrestore(&mdp5_plane->pipe_lock, flags);
 	}
-
-	/* in success case, return total # of blocks allocated: */
-	return nblks;
 }
 
-static void set_fifo_thresholds(struct drm_plane *plane, int nblks)
+static const struct drm_plane_helper_funcs mdp5_plane_helper_funcs = {
+		.prepare_fb = mdp5_plane_prepare_fb,
+		.cleanup_fb = mdp5_plane_cleanup_fb,
+		.atomic_check = mdp5_plane_atomic_check,
+		.atomic_update = mdp5_plane_atomic_update,
+};
+
+static void set_scanout_locked(struct drm_plane *plane,
+		struct drm_framebuffer *fb)
 {
 	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
 	struct mdp5_kms *mdp5_kms = get_kms(plane);
 	enum mdp5_pipe pipe = mdp5_plane->pipe;
-	uint32_t val;
 
-	/* 1/4 of SMP pool that is being fetched */
-	val = (nblks * SMP_ENTRIES_PER_BLK) / 4;
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC_STRIDE_A(pipe),
+			MDP5_PIPE_SRC_STRIDE_A_P0(fb->pitches[0]) |
+			MDP5_PIPE_SRC_STRIDE_A_P1(fb->pitches[1]));
 
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_REQPRIO_FIFO_WM_0(pipe), val * 1);
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_REQPRIO_FIFO_WM_1(pipe), val * 2);
-	mdp5_write(mdp5_kms, REG_MDP5_PIPE_REQPRIO_FIFO_WM_2(pipe), val * 3);
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC_STRIDE_B(pipe),
+			MDP5_PIPE_SRC_STRIDE_B_P2(fb->pitches[2]) |
+			MDP5_PIPE_SRC_STRIDE_B_P3(fb->pitches[3]));
+
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC0_ADDR(pipe),
+			msm_framebuffer_iova(fb, mdp5_kms->id, 0));
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC1_ADDR(pipe),
+			msm_framebuffer_iova(fb, mdp5_kms->id, 1));
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC2_ADDR(pipe),
+			msm_framebuffer_iova(fb, mdp5_kms->id, 2));
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC3_ADDR(pipe),
+			msm_framebuffer_iova(fb, mdp5_kms->id, 4));
 
+	plane->fb = fb;
 }
 
-int mdp5_plane_mode_set(struct drm_plane *plane,
+static int mdp5_plane_mode_set(struct drm_plane *plane,
 		struct drm_crtc *crtc, struct drm_framebuffer *fb,
 		int crtc_x, int crtc_y,
 		unsigned int crtc_w, unsigned int crtc_h,
@@ -225,7 +290,8 @@ int mdp5_plane_mode_set(struct drm_plane *plane,
 	uint32_t nplanes, config = 0;
 	uint32_t phasex_step = 0, phasey_step = 0;
 	uint32_t hdecm = 0, vdecm = 0;
-	int i, nblks;
+	unsigned long flags;
+	int ret;
 
 	nplanes = drm_format_num_planes(fb->pixel_format);
 
@@ -243,12 +309,11 @@ int mdp5_plane_mode_set(struct drm_plane *plane,
 			fb->base.id, src_x, src_y, src_w, src_h,
 			crtc->base.id, crtc_x, crtc_y, crtc_w, crtc_h);
 
-	/*
-	 * Calculate and request required # of smp blocks:
-	 */
-	nblks = request_smp_blocks(plane, fb->pixel_format, nplanes, src_w);
-	if (nblks < 0)
-		return nblks;
+	/* Request some memory from the SMP: */
+	ret = mdp5_smp_request(mdp5_kms->smp,
+			mdp5_plane->pipe, fb->pixel_format, src_w);
+	if (ret)
+		return ret;
 
 	/*
 	 * Currently we update the hw for allocations/requests immediately,
@@ -256,8 +321,7 @@ int mdp5_plane_mode_set(struct drm_plane *plane,
 	 * would move into atomic->check_plane_state(), while updating the
 	 * hw would remain here:
 	 */
-	for (i = 0; i < pipe2nclients(pipe); i++)
-		mdp5_smp_configure(mdp5_kms, pipe2client(pipe, i));
+	mdp5_smp_configure(mdp5_kms->smp, pipe);
 
 	if (src_w != crtc_w) {
 		config |= MDP5_PIPE_SCALE_CONFIG_SCALEX_EN;
@@ -269,6 +333,8 @@ int mdp5_plane_mode_set(struct drm_plane *plane,
 		/* TODO calc phasey_step, vdecm */
 	}
 
+	spin_lock_irqsave(&mdp5_plane->pipe_lock, flags);
+
 	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC_IMG_SIZE(pipe),
 			MDP5_PIPE_SRC_IMG_SIZE_WIDTH(src_w) |
 			MDP5_PIPE_SRC_IMG_SIZE_HEIGHT(src_h));
@@ -289,8 +355,6 @@ int mdp5_plane_mode_set(struct drm_plane *plane,
 			MDP5_PIPE_OUT_XY_X(crtc_x) |
 			MDP5_PIPE_OUT_XY_Y(crtc_y));
 
-	mdp5_plane_set_scanout(plane, fb);
-
 	format = to_mdp_format(msm_framebuffer_format(fb));
 
 	mdp5_write(mdp5_kms, REG_MDP5_PIPE_SRC_FORMAT(pipe),
@@ -330,22 +394,24 @@ int mdp5_plane_mode_set(struct drm_plane *plane,
 			MDP5_PIPE_SCALE_CONFIG_SCALEX_MAX_FILTER(SCALE_FILTER_NEAREST) |
 			MDP5_PIPE_SCALE_CONFIG_SCALEY_MAX_FILTER(SCALE_FILTER_NEAREST));
 
-	set_fifo_thresholds(plane, nblks);
+	set_scanout_locked(plane, fb);
 
-	/* TODO detach from old crtc (if we had more than one) */
-	mdp5_crtc_attach(crtc, plane);
+	spin_unlock_irqrestore(&mdp5_plane->pipe_lock, flags);
 
-	return 0;
+	return ret;
 }
 
 void mdp5_plane_complete_flip(struct drm_plane *plane)
 {
 	struct mdp5_kms *mdp5_kms = get_kms(plane);
-	enum mdp5_pipe pipe = to_mdp5_plane(plane)->pipe;
-	int i;
+	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
+	enum mdp5_pipe pipe = mdp5_plane->pipe;
+
+	DBG("%s: complete flip", mdp5_plane->name);
 
-	for (i = 0; i < pipe2nclients(pipe); i++)
-		mdp5_smp_commit(mdp5_kms, pipe2client(pipe, i));
+	mdp5_smp_commit(mdp5_kms->smp, pipe);
+
+	to_mdp5_plane_state(plane->state)->pending = false;
 }
 
 enum mdp5_pipe mdp5_plane_pipe(struct drm_plane *plane)
@@ -354,9 +420,16 @@ enum mdp5_pipe mdp5_plane_pipe(struct drm_plane *plane)
 	return mdp5_plane->pipe;
 }
 
+uint32_t mdp5_plane_get_flush(struct drm_plane *plane)
+{
+	struct mdp5_plane *mdp5_plane = to_mdp5_plane(plane);
+
+	return mdp5_plane->flush_mask;
+}
+
 /* initialize plane */
 struct drm_plane *mdp5_plane_init(struct drm_device *dev,
-		enum mdp5_pipe pipe, bool private_plane)
+		enum mdp5_pipe pipe, bool private_plane, uint32_t reg_offset)
 {
 	struct drm_plane *plane = NULL;
 	struct mdp5_plane *mdp5_plane;
@@ -377,10 +450,18 @@ struct drm_plane *mdp5_plane_init(struct drm_device *dev,
 	mdp5_plane->nformats = mdp5_get_formats(pipe, mdp5_plane->formats,
 			ARRAY_SIZE(mdp5_plane->formats));
 
+	mdp5_plane->flush_mask = mdp_ctl_flush_mask_pipe(pipe);
+	mdp5_plane->reg_offset = reg_offset;
+	spin_lock_init(&mdp5_plane->pipe_lock);
+
 	type = private_plane ? DRM_PLANE_TYPE_PRIMARY : DRM_PLANE_TYPE_OVERLAY;
-	drm_universal_plane_init(dev, plane, 0xff, &mdp5_plane_funcs,
+	ret = drm_universal_plane_init(dev, plane, 0xff, &mdp5_plane_funcs,
 				 mdp5_plane->formats, mdp5_plane->nformats,
 				 type);
+	if (ret)
+		goto fail;
+
+	drm_plane_helper_add(plane, &mdp5_plane_helper_funcs);
 
 	mdp5_plane_install_properties(plane, &plane->base);
 
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_smp.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_smp.c
index 2d0236b963a6..bf551885e019 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_smp.c
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_smp.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (c) 2014, The Linux Foundation. All rights reserved.
  * Copyright (C) 2013 Red Hat
  * Author: Rob Clark <robdclark@gmail.com>
  *
@@ -29,8 +30,11 @@
  * Based on the size of the attached scanout buffer, a certain # of
  * blocks must be allocated to that client out of the shared pool.
  *
- * For each block, it can be either free, or pending/in-use by a
- * client.  The updates happen in three steps:
+ * In some hw, some blocks are statically allocated for certain pipes
+ * and CANNOT be re-allocated (eg: MMB0 and MMB1 both tied to RGB0).
+ *
+ * For each block that can be dynamically allocated, it can be either
+ * free, or pending/in-use by a client. The updates happen in three steps:
  *
  *  1) mdp5_smp_request():
  *     When plane scanout is setup, calculate required number of
@@ -61,21 +65,68 @@
  * inuse and pending state of all clients..
  */
 
-static DEFINE_SPINLOCK(smp_lock);
+struct mdp5_smp {
+	struct drm_device *dev;
+
+	int blk_cnt;
+	int blk_size;
+
+	spinlock_t state_lock;
+	mdp5_smp_state_t state; /* to track smp allocation amongst pipes: */
+
+	struct mdp5_client_smp_state client_state[CID_MAX];
+};
 
+static inline
+struct mdp5_kms *get_kms(struct mdp5_smp *smp)
+{
+	struct msm_drm_private *priv = smp->dev->dev_private;
+
+	return to_mdp5_kms(to_mdp_kms(priv->kms));
+}
+
+static inline enum mdp5_client_id pipe2client(enum mdp5_pipe pipe, int plane)
+{
+	WARN_ON(plane >= pipe2nclients(pipe));
+	switch (pipe) {
+	case SSPP_VIG0: return CID_VIG0_Y + plane;
+	case SSPP_VIG1: return CID_VIG1_Y + plane;
+	case SSPP_VIG2: return CID_VIG2_Y + plane;
+	case SSPP_RGB0: return CID_RGB0;
+	case SSPP_RGB1: return CID_RGB1;
+	case SSPP_RGB2: return CID_RGB2;
+	case SSPP_DMA0: return CID_DMA0_Y + plane;
+	case SSPP_DMA1: return CID_DMA1_Y + plane;
+	case SSPP_VIG3: return CID_VIG3_Y + plane;
+	case SSPP_RGB3: return CID_RGB3;
+	default:        return CID_UNUSED;
+	}
+}
 
 /* step #1: update # of blocks pending for the client: */
-int mdp5_smp_request(struct mdp5_kms *mdp5_kms,
+static int smp_request_block(struct mdp5_smp *smp,
 		enum mdp5_client_id cid, int nblks)
 {
-	struct mdp5_client_smp_state *ps = &mdp5_kms->smp_client_state[cid];
-	int i, ret, avail, cur_nblks, cnt = mdp5_kms->smp_blk_cnt;
+	struct mdp5_kms *mdp5_kms = get_kms(smp);
+	const struct mdp5_cfg_hw *hw_cfg;
+	struct mdp5_client_smp_state *ps = &smp->client_state[cid];
+	int i, ret, avail, cur_nblks, cnt = smp->blk_cnt;
+	int reserved;
 	unsigned long flags;
 
-	spin_lock_irqsave(&smp_lock, flags);
+	hw_cfg = mdp5_cfg_get_hw_config(mdp5_kms->cfg);
+	reserved = hw_cfg->smp.reserved[cid];
+
+	spin_lock_irqsave(&smp->state_lock, flags);
 
-	avail = cnt - bitmap_weight(mdp5_kms->smp_state, cnt);
+	nblks -= reserved;
+	if (reserved)
+		DBG("%d MMBs allocated (%d reserved)", nblks, reserved);
+
+	avail = cnt - bitmap_weight(smp->state, cnt);
 	if (nblks > avail) {
+		dev_err(mdp5_kms->dev->dev, "out of blks (req=%d > avail=%d)\n",
+				nblks, avail);
 		ret = -ENOSPC;
 		goto fail;
 	}
@@ -84,9 +135,9 @@ int mdp5_smp_request(struct mdp5_kms *mdp5_kms,
 	if (nblks > cur_nblks) {
 		/* grow the existing pending reservation: */
 		for (i = cur_nblks; i < nblks; i++) {
-			int blk = find_first_zero_bit(mdp5_kms->smp_state, cnt);
+			int blk = find_first_zero_bit(smp->state, cnt);
 			set_bit(blk, ps->pending);
-			set_bit(blk, mdp5_kms->smp_state);
+			set_bit(blk, smp->state);
 		}
 	} else {
 		/* shrink the existing pending reservation: */
@@ -98,15 +149,88 @@ int mdp5_smp_request(struct mdp5_kms *mdp5_kms,
 	}
 
 fail:
-	spin_unlock_irqrestore(&smp_lock, flags);
+	spin_unlock_irqrestore(&smp->state_lock, flags);
+	return 0;
+}
+
+static void set_fifo_thresholds(struct mdp5_smp *smp,
+		enum mdp5_pipe pipe, int nblks)
+{
+	struct mdp5_kms *mdp5_kms = get_kms(smp);
+	u32 smp_entries_per_blk = smp->blk_size / (128 / BITS_PER_BYTE);
+	u32 val;
+
+	/* 1/4 of SMP pool that is being fetched */
+	val = (nblks * smp_entries_per_blk) / 4;
+
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_REQPRIO_FIFO_WM_0(pipe), val * 1);
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_REQPRIO_FIFO_WM_1(pipe), val * 2);
+	mdp5_write(mdp5_kms, REG_MDP5_PIPE_REQPRIO_FIFO_WM_2(pipe), val * 3);
+}
+
+/*
+ * NOTE: looks like if horizontal decimation is used (if we supported that)
+ * then the width used to calculate SMP block requirements is the post-
+ * decimated width.  Ie. SMP buffering sits downstream of decimation (which
+ * presumably happens during the dma from scanout buffer).
+ */
+int mdp5_smp_request(struct mdp5_smp *smp, enum mdp5_pipe pipe, u32 fmt, u32 width)
+{
+	struct mdp5_kms *mdp5_kms = get_kms(smp);
+	struct drm_device *dev = mdp5_kms->dev;
+	int rev = mdp5_cfg_get_hw_rev(mdp5_kms->cfg);
+	int i, hsub, nplanes, nlines, nblks, ret;
+
+	nplanes = drm_format_num_planes(fmt);
+	hsub = drm_format_horz_chroma_subsampling(fmt);
+
+	/* different if BWC (compressed framebuffer?) enabled: */
+	nlines = 2;
+
+	for (i = 0, nblks = 0; i < nplanes; i++) {
+		int n, fetch_stride, cpp;
+
+		cpp = drm_format_plane_cpp(fmt, i);
+		fetch_stride = width * cpp / (i ? hsub : 1);
+
+		n = DIV_ROUND_UP(fetch_stride * nlines, smp->blk_size);
+
+		/* for hw rev v1.00 */
+		if (rev == 0)
+			n = roundup_pow_of_two(n);
+
+		DBG("%s[%d]: request %d SMP blocks", pipe2name(pipe), i, n);
+		ret = smp_request_block(smp, pipe2client(pipe, i), n);
+		if (ret) {
+			dev_err(dev->dev, "Cannot allocate %d SMP blocks: %d\n",
+					n, ret);
+			return ret;
+		}
+
+		nblks += n;
+	}
+
+	set_fifo_thresholds(smp, pipe, nblks);
+
 	return 0;
 }
 
-static void update_smp_state(struct mdp5_kms *mdp5_kms,
+/* Release SMP blocks for all clients of the pipe */
+void mdp5_smp_release(struct mdp5_smp *smp, enum mdp5_pipe pipe)
+{
+	int i, nblks;
+
+	for (i = 0, nblks = 0; i < pipe2nclients(pipe); i++)
+		smp_request_block(smp, pipe2client(pipe, i), 0);
+	set_fifo_thresholds(smp, pipe, 0);
+}
+
+static void update_smp_state(struct mdp5_smp *smp,
 		enum mdp5_client_id cid, mdp5_smp_state_t *assigned)
 {
-	int cnt = mdp5_kms->smp_blk_cnt;
-	uint32_t blk, val;
+	struct mdp5_kms *mdp5_kms = get_kms(smp);
+	int cnt = smp->blk_cnt;
+	u32 blk, val;
 
 	for_each_set_bit(blk, *assigned, cnt) {
 		int idx = blk / 3;
@@ -135,39 +259,80 @@ static void update_smp_state(struct mdp5_kms *mdp5_kms,
 }
 
 /* step #2: configure hw for union(pending, inuse): */
-void mdp5_smp_configure(struct mdp5_kms *mdp5_kms, enum mdp5_client_id cid)
+void mdp5_smp_configure(struct mdp5_smp *smp, enum mdp5_pipe pipe)
 {
-	struct mdp5_client_smp_state *ps = &mdp5_kms->smp_client_state[cid];
-	int cnt = mdp5_kms->smp_blk_cnt;
+	int cnt = smp->blk_cnt;
 	mdp5_smp_state_t assigned;
+	int i;
 
-	bitmap_or(assigned, ps->inuse, ps->pending, cnt);
-	update_smp_state(mdp5_kms, cid, &assigned);
+	for (i = 0; i < pipe2nclients(pipe); i++) {
+		enum mdp5_client_id cid = pipe2client(pipe, i);
+		struct mdp5_client_smp_state *ps = &smp->client_state[cid];
+
+		bitmap_or(assigned, ps->inuse, ps->pending, cnt);
+		update_smp_state(smp, cid, &assigned);
+	}
 }
 
 /* step #3: after vblank, copy pending -> inuse: */
-void mdp5_smp_commit(struct mdp5_kms *mdp5_kms, enum mdp5_client_id cid)
+void mdp5_smp_commit(struct mdp5_smp *smp, enum mdp5_pipe pipe)
 {
-	struct mdp5_client_smp_state *ps = &mdp5_kms->smp_client_state[cid];
-	int cnt = mdp5_kms->smp_blk_cnt;
+	int cnt = smp->blk_cnt;
 	mdp5_smp_state_t released;
+	int i;
+
+	for (i = 0; i < pipe2nclients(pipe); i++) {
+		enum mdp5_client_id cid = pipe2client(pipe, i);
+		struct mdp5_client_smp_state *ps = &smp->client_state[cid];
+
+		/*
+		 * Figure out if there are any blocks we where previously
+		 * using, which can be released and made available to other
+		 * clients:
+		 */
+		if (bitmap_andnot(released, ps->inuse, ps->pending, cnt)) {
+			unsigned long flags;
+
+			spin_lock_irqsave(&smp->state_lock, flags);
+			/* clear released blocks: */
+			bitmap_andnot(smp->state, smp->state, released, cnt);
+			spin_unlock_irqrestore(&smp->state_lock, flags);
 
-	/*
-	 * Figure out if there are any blocks we where previously
-	 * using, which can be released and made available to other
-	 * clients:
-	 */
-	if (bitmap_andnot(released, ps->inuse, ps->pending, cnt)) {
-		unsigned long flags;
-
-		spin_lock_irqsave(&smp_lock, flags);
-		/* clear released blocks: */
-		bitmap_andnot(mdp5_kms->smp_state, mdp5_kms->smp_state,
-				released, cnt);
-		spin_unlock_irqrestore(&smp_lock, flags);
-
-		update_smp_state(mdp5_kms, CID_UNUSED, &released);
+			update_smp_state(smp, CID_UNUSED, &released);
+		}
+
+		bitmap_copy(ps->inuse, ps->pending, cnt);
 	}
+}
+
+void mdp5_smp_destroy(struct mdp5_smp *smp)
+{
+	kfree(smp);
+}
+
+struct mdp5_smp *mdp5_smp_init(struct drm_device *dev, const struct mdp5_smp_block *cfg)
+{
+	struct mdp5_smp *smp = NULL;
+	int ret;
+
+	smp = kzalloc(sizeof(*smp), GFP_KERNEL);
+	if (unlikely(!smp)) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+	smp->dev = dev;
+	smp->blk_cnt = cfg->mmb_count;
+	smp->blk_size = cfg->mmb_size;
+
+	/* statically tied MMBs cannot be re-allocated: */
+	bitmap_copy(smp->state, cfg->reserved_state, smp->blk_cnt);
+	spin_lock_init(&smp->state_lock);
+
+	return smp;
+fail:
+	if (smp)
+		mdp5_smp_destroy(smp);
 
-	bitmap_copy(ps->inuse, ps->pending, cnt);
+	return ERR_PTR(ret);
 }
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_smp.h b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_smp.h
index 0ab739e1a1dd..e47179f63585 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_smp.h
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_smp.h
@@ -1,4 +1,5 @@
 /*
+ * Copyright (c) 2014, The Linux Foundation. All rights reserved.
  * Copyright (C) 2013 Red Hat
  * Author: Rob Clark <robdclark@gmail.com>
  *
@@ -20,22 +21,26 @@
 
 #include "msm_drv.h"
 
-#define MAX_SMP_BLOCKS  22
-#define SMP_BLK_SIZE    4096
-#define SMP_ENTRIES_PER_BLK (SMP_BLK_SIZE / 16)
-
-typedef DECLARE_BITMAP(mdp5_smp_state_t, MAX_SMP_BLOCKS);
-
 struct mdp5_client_smp_state {
 	mdp5_smp_state_t inuse;
 	mdp5_smp_state_t pending;
 };
 
 struct mdp5_kms;
+struct mdp5_smp;
+
+/*
+ * SMP module prototypes:
+ * mdp5_smp_init() returns a SMP @handler,
+ * which is then used to call the other mdp5_smp_*(handler, ...) functions.
+ */
 
-int mdp5_smp_request(struct mdp5_kms *mdp5_kms, enum mdp5_client_id cid, int nblks);
-void mdp5_smp_configure(struct mdp5_kms *mdp5_kms, enum mdp5_client_id cid);
-void mdp5_smp_commit(struct mdp5_kms *mdp5_kms, enum mdp5_client_id cid);
+struct mdp5_smp *mdp5_smp_init(struct drm_device *dev, const struct mdp5_smp_block *cfg);
+void  mdp5_smp_destroy(struct mdp5_smp *smp);
 
+int  mdp5_smp_request(struct mdp5_smp *smp, enum mdp5_pipe pipe, u32 fmt, u32 width);
+void mdp5_smp_configure(struct mdp5_smp *smp, enum mdp5_pipe pipe);
+void mdp5_smp_commit(struct mdp5_smp *smp, enum mdp5_pipe pipe);
+void mdp5_smp_release(struct mdp5_smp *smp, enum mdp5_pipe pipe);
 
 #endif /* __MDP5_SMP_H__ */
diff --git a/drivers/gpu/drm/msm/msm_atomic.c b/drivers/gpu/drm/msm/msm_atomic.c
index 8cf3361daba3..f0de412e13dc 100644
--- a/drivers/gpu/drm/msm/msm_atomic.c
+++ b/drivers/gpu/drm/msm/msm_atomic.c
@@ -117,7 +117,7 @@ int msm_atomic_commit(struct drm_device *dev,
 		if (!plane)
 			continue;
 
-		if (plane->state->fb != new_state->fb)
+		if ((plane->state->fb != new_state->fb) && new_state->fb)
 			add_fb(c, new_state->fb);
 	}
 
diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
index 4b52d752bb6e..136303818436 100644
--- a/drivers/gpu/drm/msm/msm_drv.h
+++ b/drivers/gpu/drm/msm/msm_drv.h
@@ -215,7 +215,6 @@ struct drm_fb_helper *msm_fbdev_init(struct drm_device *dev);
 struct hdmi;
 int hdmi_modeset_init(struct hdmi *hdmi, struct drm_device *dev,
 		struct drm_encoder *encoder);
-irqreturn_t hdmi_irq(int irq, void *dev_id);
 void __init hdmi_register(void);
 void __exit hdmi_unregister(void);
 
diff --git a/drivers/gpu/drm/msm/msm_fb.c b/drivers/gpu/drm/msm/msm_fb.c
index f4e42d506ff7..84dec161d836 100644
--- a/drivers/gpu/drm/msm/msm_fb.c
+++ b/drivers/gpu/drm/msm/msm_fb.c
@@ -120,6 +120,8 @@ void msm_framebuffer_cleanup(struct drm_framebuffer *fb, int id)
 uint32_t msm_framebuffer_iova(struct drm_framebuffer *fb, int id, int plane)
 {
 	struct msm_framebuffer *msm_fb = to_msm_framebuffer(fb);
+	if (!msm_fb->planes[plane])
+		return 0;
 	return msm_gem_iova(msm_fb->planes[plane], id);
 }
 
diff --git a/drivers/gpu/drm/msm/msm_kms.h b/drivers/gpu/drm/msm/msm_kms.h
index 15a0fec99c70..7fb4876388e7 100644
--- a/drivers/gpu/drm/msm/msm_kms.h
+++ b/drivers/gpu/drm/msm/msm_kms.h
@@ -68,6 +68,24 @@ struct msm_kms *mdp5_kms_init(struct drm_device *dev);
 /* TODO move these helper iterator macro somewhere common: */
 #define for_each_plane_on_crtc(_crtc, _plane) \
 	list_for_each_entry((_plane), &(_crtc)->dev->mode_config.plane_list, head) \
-		if ((_plane)->crtc == (_crtc))
+		if ((_plane)->state->crtc == (_crtc))
+
+static inline bool
+__plane_will_be_attached_to_crtc(struct drm_atomic_state *state,
+		struct drm_plane *plane, struct drm_crtc *crtc)
+{
+	int idx = drm_plane_index(plane);
+
+	/* if plane is modified in incoming state, use the new state: */
+	if (state->plane_states[idx])
+		return state->plane_states[idx]->crtc == crtc;
+
+	/* otherwise, current state: */
+	return plane->state->crtc == crtc;
+}
+
+#define for_each_pending_plane_on_crtc(_state, _crtc, _plane) \
+	list_for_each_entry((_plane), &(_crtc)->dev->mode_config.plane_list, head) \
+		if (__plane_will_be_attached_to_crtc((_state), (_plane), (_crtc)))
 
 #endif /* __MSM_KMS_H__ */