Merge "msm: vidc: Set proper verbosity in EBD function logs"
diff --git a/arch/arm/boot/dts/apq8074-v2-cdp.dts b/arch/arm/boot/dts/apq8074-v2.0-1-cdp.dts
similarity index 92%
rename from arch/arm/boot/dts/apq8074-v2-cdp.dts
rename to arch/arm/boot/dts/apq8074-v2.0-1-cdp.dts
index 1dc0912..0489b55 100644
--- a/arch/arm/boot/dts/apq8074-v2-cdp.dts
+++ b/arch/arm/boot/dts/apq8074-v2.0-1-cdp.dts
@@ -12,11 +12,11 @@
 
 /dts-v1/;
 
-/include/ "apq8074-v2.dtsi"
+/include/ "apq8074-v2.0-1.dtsi"
 /include/ "msm8974-cdp.dtsi"
 
 / {
-	model = "Qualcomm APQ 8074v2 CDP";
+	model = "Qualcomm APQ 8074v2.0-1 CDP";
 	compatible = "qcom,apq8074-cdp", "qcom,apq8074", "qcom,cdp";
 	qcom,msm-id = <184 1 0x20000>;
 };
diff --git a/arch/arm/boot/dts/apq8074-v2-dragonboard.dts b/arch/arm/boot/dts/apq8074-v2.0-1-dragonboard.dts
similarity index 89%
rename from arch/arm/boot/dts/apq8074-v2-dragonboard.dts
rename to arch/arm/boot/dts/apq8074-v2.0-1-dragonboard.dts
index 5a6f5f3..128d8bd 100644
--- a/arch/arm/boot/dts/apq8074-v2-dragonboard.dts
+++ b/arch/arm/boot/dts/apq8074-v2.0-1-dragonboard.dts
@@ -12,11 +12,11 @@
 
 /dts-v1/;
 
-/include/ "apq8074-v2.dtsi"
+/include/ "apq8074-v2.0-1.dtsi"
 /include/ "apq8074-dragonboard.dtsi"
 
 / {
-	model = "Qualcomm APQ 8074v2 DRAGONBOARD";
+	model = "Qualcomm APQ 8074v2.0-1 DRAGONBOARD";
 	compatible = "qcom,apq8074-dragonboard", "qcom,apq8074", "qcom,dragonboard";
 	qcom,msm-id = <184 10 0x20000>;
 };
diff --git a/arch/arm/boot/dts/apq8074-v2-liquid.dts b/arch/arm/boot/dts/apq8074-v2.0-1-liquid.dts
similarity index 89%
rename from arch/arm/boot/dts/apq8074-v2-liquid.dts
rename to arch/arm/boot/dts/apq8074-v2.0-1-liquid.dts
index a0ecb50..63c32f3 100644
--- a/arch/arm/boot/dts/apq8074-v2-liquid.dts
+++ b/arch/arm/boot/dts/apq8074-v2.0-1-liquid.dts
@@ -12,11 +12,11 @@
 
 /dts-v1/;
 
-/include/ "apq8074-v2.dtsi"
+/include/ "apq8074-v2.0-1.dtsi"
 /include/ "msm8974-liquid.dtsi"
 
 / {
-	model = "Qualcomm APQ 8074v2 LIQUID";
+	model = "Qualcomm APQ 8074v2.0-1 LIQUID";
 	compatible = "qcom,apq8074-liquid", "qcom,apq8074", "qcom,liquid";
 	qcom,msm-id = <184 9 0x20000>;
 };
diff --git a/arch/arm/boot/dts/apq8074-v2.dtsi b/arch/arm/boot/dts/apq8074-v2.0-1.dtsi
similarity index 97%
rename from arch/arm/boot/dts/apq8074-v2.dtsi
rename to arch/arm/boot/dts/apq8074-v2.0-1.dtsi
index c700a5c..8314fab 100644
--- a/arch/arm/boot/dts/apq8074-v2.dtsi
+++ b/arch/arm/boot/dts/apq8074-v2.0-1.dtsi
@@ -16,7 +16,7 @@
  * msm8974.dtsi file.
  */
 
-/include/ "msm8974-v2.dtsi"
+/include/ "msm8974-v2.0-1.dtsi"
 
 &soc {
 	qcom,qseecom@a700000 {
diff --git a/arch/arm/boot/dts/apq8084.dtsi b/arch/arm/boot/dts/apq8084.dtsi
index b027f7d..943f2a3 100644
--- a/arch/arm/boot/dts/apq8084.dtsi
+++ b/arch/arm/boot/dts/apq8084.dtsi
@@ -344,6 +344,12 @@
 		qcom,pet-time = <10000>;
 		qcom,ipi-ping;
 	};
+
+	qcom,msm-rng@f9bff000{
+		compatible = "qcom,msm-rng";
+		reg = <0xf9bff000 0x200>;
+		qcom,msm-rng-iface-clk;
+	};
 };
 
 &gdsc_venus {
diff --git a/arch/arm/boot/dts/msm8974-v1.dtsi b/arch/arm/boot/dts/msm8974-v1.dtsi
index 7b801da..86a61cd 100644
--- a/arch/arm/boot/dts/msm8974-v1.dtsi
+++ b/arch/arm/boot/dts/msm8974-v1.dtsi
@@ -148,3 +148,20 @@
 &usb_otg {
 	qcom,hsusb-otg-pnoc-errata-fix;
 };
+
+&gdsc_venus {
+	qcom,skip-logic-collapse;
+	qcom,retain-periph;
+	qcom,retain-mem;
+};
+
+&gdsc_mdss {
+	qcom,skip-logic-collapse;
+	qcom,retain-periph;
+	qcom,retain-mem;
+};
+
+&gdsc_oxili_gx {
+	qcom,retain-mem;
+	qcom,retain-periph;
+};
diff --git a/arch/arm/boot/dts/msm8974-v2-cdp.dts b/arch/arm/boot/dts/msm8974-v2.0-1-cdp.dts
similarity index 90%
rename from arch/arm/boot/dts/msm8974-v2-cdp.dts
rename to arch/arm/boot/dts/msm8974-v2.0-1-cdp.dts
index f4014aa..875b3fc 100644
--- a/arch/arm/boot/dts/msm8974-v2-cdp.dts
+++ b/arch/arm/boot/dts/msm8974-v2.0-1-cdp.dts
@@ -12,11 +12,11 @@
 
 /dts-v1/;
 
-/include/ "msm8974-v2.dtsi"
+/include/ "msm8974-v2.0-1.dtsi"
 /include/ "msm8974-cdp.dtsi"
 
 / {
-	model = "Qualcomm MSM 8974v2 CDP";
+	model = "Qualcomm MSM 8974v2.0/1 CDP";
 	compatible = "qcom,msm8974-cdp", "qcom,msm8974", "qcom,cdp";
 	qcom,msm-id = <126 1 0x20000>,
 		      <185 1 0x20000>,
diff --git a/arch/arm/boot/dts/msm8974-v2-fluid.dts b/arch/arm/boot/dts/msm8974-v2.0-1-fluid.dts
similarity index 90%
rename from arch/arm/boot/dts/msm8974-v2-fluid.dts
rename to arch/arm/boot/dts/msm8974-v2.0-1-fluid.dts
index 9c9e3c0..236593d 100644
--- a/arch/arm/boot/dts/msm8974-v2-fluid.dts
+++ b/arch/arm/boot/dts/msm8974-v2.0-1-fluid.dts
@@ -12,11 +12,11 @@
 
 /dts-v1/;
 
-/include/ "msm8974-v2.dtsi"
+/include/ "msm8974-v2.0-1.dtsi"
 /include/ "msm8974-fluid.dtsi"
 
 / {
-	model = "Qualcomm MSM 8974v2 FLUID";
+	model = "Qualcomm MSM 8974v2.0/1 FLUID";
 	compatible = "qcom,msm8974-fluid", "qcom,msm8974", "qcom,fluid";
 	qcom,msm-id = <126 3 0x20000>,
 		      <185 3 0x20000>,
diff --git a/arch/arm/boot/dts/msm8974-v2-liquid.dts b/arch/arm/boot/dts/msm8974-v2.0-1-liquid.dts
similarity index 90%
rename from arch/arm/boot/dts/msm8974-v2-liquid.dts
rename to arch/arm/boot/dts/msm8974-v2.0-1-liquid.dts
index ddae6fe..23292f6 100644
--- a/arch/arm/boot/dts/msm8974-v2-liquid.dts
+++ b/arch/arm/boot/dts/msm8974-v2.0-1-liquid.dts
@@ -12,11 +12,11 @@
 
 /dts-v1/;
 
-/include/ "msm8974-v2.dtsi"
+/include/ "msm8974-v2.0-1.dtsi"
 /include/ "msm8974-liquid.dtsi"
 
 / {
-	model = "Qualcomm MSM 8974v2 LIQUID";
+	model = "Qualcomm MSM 8974v2.0/1 LIQUID";
 	compatible = "qcom,msm8974-liquid", "qcom,msm8974", "qcom,liquid";
 	qcom,msm-id = <126 9 0x20000>,
 		      <185 9 0x20000>,
diff --git a/arch/arm/boot/dts/msm8974-v2-mtp.dts b/arch/arm/boot/dts/msm8974-v2.0-1-mtp.dts
similarity index 91%
rename from arch/arm/boot/dts/msm8974-v2-mtp.dts
rename to arch/arm/boot/dts/msm8974-v2.0-1-mtp.dts
index 021b626..de9e6a3 100644
--- a/arch/arm/boot/dts/msm8974-v2-mtp.dts
+++ b/arch/arm/boot/dts/msm8974-v2.0-1-mtp.dts
@@ -12,11 +12,11 @@
 
 /dts-v1/;
 
-/include/ "msm8974-v2.dtsi"
+/include/ "msm8974-v2.0-1.dtsi"
 /include/ "msm8974-mtp.dtsi"
 
 / {
-	model = "Qualcomm MSM 8974v2 MTP";
+	model = "Qualcomm MSM 8974v2.0/1 MTP";
 	compatible = "qcom,msm8974-mtp", "qcom,msm8974", "qcom,mtp";
 	qcom,msm-id = <126 8 0x20000>,
 		      <185 8 0x20000>,
diff --git a/arch/arm/boot/dts/msm8974-v2.0-1.dtsi b/arch/arm/boot/dts/msm8974-v2.0-1.dtsi
new file mode 100644
index 0000000..1fad868
--- /dev/null
+++ b/arch/arm/boot/dts/msm8974-v2.0-1.dtsi
@@ -0,0 +1,36 @@
+/* Copyright (c) 2013, 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.
+ */
+
+/*
+ * As a general rule, only version-specific property overrides should be placed
+ * inside this file. However, device definitions should be placed inside the
+ * msm8974.dtsi file.
+ */
+
+/include/ "msm8974-v2.dtsi"
+
+&gdsc_venus {
+	qcom,skip-logic-collapse;
+	qcom,retain-periph;
+	qcom,retain-mem;
+};
+
+&gdsc_mdss {
+	qcom,skip-logic-collapse;
+	qcom,retain-periph;
+	qcom,retain-mem;
+};
+
+&gdsc_oxili_gx {
+	qcom,retain-mem;
+	qcom,retain-periph;
+};
diff --git a/arch/arm/boot/dts/msm8974-v2.2.dtsi b/arch/arm/boot/dts/msm8974-v2.2.dtsi
index 09455b1..0ca021b 100644
--- a/arch/arm/boot/dts/msm8974-v2.2.dtsi
+++ b/arch/arm/boot/dts/msm8974-v2.2.dtsi
@@ -103,3 +103,8 @@
 		};
 	};
 };
+
+&gdsc_mdss {
+	qcom,retain-periph;
+	qcom,retain-mem;
+};
diff --git a/arch/arm/boot/dts/msm8974.dtsi b/arch/arm/boot/dts/msm8974.dtsi
index 59e8dac..4360fe0 100644
--- a/arch/arm/boot/dts/msm8974.dtsi
+++ b/arch/arm/boot/dts/msm8974.dtsi
@@ -1651,17 +1651,11 @@
 
 &gdsc_venus {
 	qcom,clock-names = "core_clk";
-	qcom,skip-logic-collapse;
-	qcom,retain-periph;
-	qcom,retain-mem;
 	status = "ok";
 };
 
 &gdsc_mdss {
 	qcom,clock-names = "core_clk", "lut_clk";
-	qcom,skip-logic-collapse;
-	qcom,retain-periph;
-	qcom,retain-mem;
 	status = "ok";
 };
 
@@ -1678,8 +1672,6 @@
 
 &gdsc_oxili_gx {
 	qcom,clock-names = "core_clk";
-	qcom,retain-mem;
-	qcom,retain-periph;
 	status = "ok";
 };
 
diff --git a/arch/arm/mach-msm/Makefile.boot b/arch/arm/mach-msm/Makefile.boot
index 3505afe..72472f9 100644
--- a/arch/arm/mach-msm/Makefile.boot
+++ b/arch/arm/mach-msm/Makefile.boot
@@ -53,13 +53,13 @@
         dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v1-mtp.dtb
         dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v1-rumi.dtb
         dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v1-sim.dtb
-        dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v2-cdp.dtb
-        dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v2-fluid.dtb
-        dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v2-liquid.dtb
-        dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v2-mtp.dtb
-        dtb-$(CONFIG_ARCH_MSM8974)	+= apq8074-v2-cdp.dtb
-        dtb-$(CONFIG_ARCH_MSM8974)	+= apq8074-v2-liquid.dtb
-        dtb-$(CONFIG_ARCH_MSM8974)	+= apq8074-v2-dragonboard.dtb
+        dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v2.0-1-cdp.dtb
+        dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v2.0-1-fluid.dtb
+        dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v2.0-1-liquid.dtb
+        dtb-$(CONFIG_ARCH_MSM8974)	+= msm8974-v2.0-1-mtp.dtb
+        dtb-$(CONFIG_ARCH_MSM8974)	+= apq8074-v2.0-1-cdp.dtb
+        dtb-$(CONFIG_ARCH_MSM8974)	+= apq8074-v2.0-1-liquid.dtb
+        dtb-$(CONFIG_ARCH_MSM8974)	+= apq8074-v2.0-1-dragonboard.dtb
 
 # APQ8084
    zreladdr-$(CONFIG_ARCH_APQ8084)	:= 0x00008000
diff --git a/drivers/gpu/msm/Makefile b/drivers/gpu/msm/Makefile
index 118e033..aac183b 100644
--- a/drivers/gpu/msm/Makefile
+++ b/drivers/gpu/msm/Makefile
@@ -22,9 +22,11 @@
 msm_adreno-y += \
 	adreno_ringbuffer.o \
 	adreno_drawctxt.o \
+	adreno_dispatch.o \
 	adreno_postmortem.o \
 	adreno_snapshot.o \
 	adreno_coresight.o \
+	adreno_trace.o \
 	adreno_a2xx.o \
 	adreno_a2xx_trace.o \
 	adreno_a2xx_snapshot.o \
diff --git a/drivers/gpu/msm/adreno.c b/drivers/gpu/msm/adreno.c
index b964620..184dd982 100644
--- a/drivers/gpu/msm/adreno.c
+++ b/drivers/gpu/msm/adreno.c
@@ -32,6 +32,7 @@
 
 #include "adreno.h"
 #include "adreno_pm4types.h"
+#include "adreno_trace.h"
 
 #include "a2xx_reg.h"
 #include "a3xx_reg.h"
@@ -213,7 +214,7 @@
 		512, 0, 2, SZ_128K, 0x3FF037, 0x3FF016 },
 };
 
-static unsigned int adreno_isidle(struct kgsl_device *device);
+static bool adreno_isidle(struct kgsl_device *device);
 
 /**
  * adreno_perfcounter_init: Reserve kernel performance counters
@@ -598,23 +599,9 @@
 
 static irqreturn_t adreno_irq_handler(struct kgsl_device *device)
 {
-	irqreturn_t result;
 	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
 
-	result = adreno_dev->gpudev->irq_handler(adreno_dev);
-
-	device->pwrctrl.irq_last = 1;
-	if (device->requested_state == KGSL_STATE_NONE) {
-		kgsl_pwrctrl_request_state(device, KGSL_STATE_NAP);
-		queue_work(device->work_queue, &device->idle_check_ws);
-	}
-
-	/* Reset the time-out in our idle timer */
-	mod_timer_pending(&device->idle_timer,
-		jiffies + device->pwrctrl.interval_timeout);
-	mod_timer_pending(&device->hang_timer,
-		(jiffies + msecs_to_jiffies(KGSL_TIMEOUT_PART)));
-	return result;
+	return adreno_dev->gpudev->irq_handler(adreno_dev);
 }
 
 static void adreno_cleanup_pt(struct kgsl_device *device,
@@ -921,7 +908,7 @@
 		adreno_dev->dev.cff_dump_enable);
 }
 
-static void adreno_iommu_setstate(struct kgsl_device *device,
+static int adreno_iommu_setstate(struct kgsl_device *device,
 					unsigned int context_id,
 					uint32_t flags)
 {
@@ -934,22 +921,24 @@
 	struct kgsl_context *context;
 	struct adreno_context *adreno_ctx = NULL;
 	struct adreno_ringbuffer *rb = &adreno_dev->ringbuffer;
+	unsigned int result;
 
 	if (adreno_use_default_setstate(adreno_dev)) {
 		kgsl_mmu_device_setstate(&device->mmu, flags);
-		return;
+		return 0;
 	}
 	num_iommu_units = kgsl_mmu_get_num_iommu_units(&device->mmu);
 
 	context = kgsl_context_get(device, context_id);
 	if (context == NULL)
-		return;
+		return -EINVAL;
 
 	adreno_ctx = ADRENO_CONTEXT(context);
 
-	if (kgsl_mmu_enable_clk(&device->mmu,
-				KGSL_IOMMU_CONTEXT_USER))
-		return;
+	result = kgsl_mmu_enable_clk(&device->mmu, KGSL_IOMMU_CONTEXT_USER);
+
+	if (result)
+		goto done;
 
 	pt_val = kgsl_mmu_get_pt_base_addr(&device->mmu,
 				device->mmu.hwpagetable);
@@ -983,14 +972,24 @@
 	 * This returns the per context timestamp but we need to
 	 * use the global timestamp for iommu clock disablement
 	 */
-	adreno_ringbuffer_issuecmds(device, adreno_ctx, KGSL_CMD_FLAGS_PMODE,
-			&link[0], sizedwords);
+	result = adreno_ringbuffer_issuecmds(device, adreno_ctx,
+			KGSL_CMD_FLAGS_PMODE, &link[0], sizedwords);
 
-	kgsl_mmu_disable_clk_on_ts(&device->mmu, rb->global_ts, true);
+	/*
+	 * On error disable the IOMMU clock right away otherwise turn it off
+	 * after the command has been retired
+	 */
+	if (result)
+		kgsl_mmu_disable_clk_on_ts(&device->mmu, 0, false);
+	else
+		kgsl_mmu_disable_clk_on_ts(&device->mmu, rb->global_ts, true);
+
+done:
 	kgsl_context_put(context);
+	return result;
 }
 
-static void adreno_gpummu_setstate(struct kgsl_device *device,
+static int adreno_gpummu_setstate(struct kgsl_device *device,
 					unsigned int context_id,
 					uint32_t flags)
 {
@@ -1001,6 +1000,7 @@
 	unsigned int mh_mmu_invalidate = 0x00000003; /*invalidate all and tc */
 	struct kgsl_context *context;
 	struct adreno_context *adreno_ctx = NULL;
+	int ret = 0;
 
 	/*
 	 * Fix target freeze issue by adding TLB flush for each submit
@@ -1017,7 +1017,8 @@
 	if (!adreno_use_default_setstate(adreno_dev)) {
 		context = kgsl_context_get(device, context_id);
 		if (context == NULL)
-			return;
+			return -EINVAL;
+
 		adreno_ctx = ADRENO_CONTEXT(context);
 
 		if (flags & KGSL_MMUFLAGS_PTUPDATE) {
@@ -1092,7 +1093,7 @@
 			sizedwords += 2;
 		}
 
-		adreno_ringbuffer_issuecmds(device, adreno_ctx,
+		ret = adreno_ringbuffer_issuecmds(device, adreno_ctx,
 					KGSL_CMD_FLAGS_PMODE,
 					&link[0], sizedwords);
 
@@ -1100,9 +1101,11 @@
 	} else {
 		kgsl_mmu_device_setstate(&device->mmu, flags);
 	}
+
+	return ret;
 }
 
-static void adreno_setstate(struct kgsl_device *device,
+static int adreno_setstate(struct kgsl_device *device,
 			unsigned int context_id,
 			uint32_t flags)
 {
@@ -1111,6 +1114,8 @@
 		return adreno_gpummu_setstate(device, context_id, flags);
 	else if (KGSL_MMU_TYPE_IOMMU == kgsl_mmu_get_mmutype())
 		return adreno_iommu_setstate(device, context_id, flags);
+
+	return 0;
 }
 
 static unsigned int
@@ -1586,6 +1591,10 @@
 	if (status)
 		goto error_close_rb;
 
+	status = adreno_dispatcher_init(adreno_dev);
+	if (status)
+		goto error_close_device;
+
 	adreno_debugfs_init(device);
 	adreno_profile_init(device);
 
@@ -1601,6 +1610,8 @@
 
 	return 0;
 
+error_close_device:
+	kgsl_device_platform_remove(device);
 error_close_rb:
 	adreno_ringbuffer_close(&adreno_dev->ringbuffer);
 error:
@@ -1623,6 +1634,7 @@
 	kgsl_pwrscale_detach_policy(device);
 	kgsl_pwrscale_close(device);
 
+	adreno_dispatcher_close(adreno_dev);
 	adreno_ringbuffer_close(&adreno_dev->ringbuffer);
 	kgsl_device_platform_remove(device);
 
@@ -1634,8 +1646,7 @@
 	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
 	int i;
 
-	if (KGSL_STATE_DUMP_AND_FT != device->state)
-		kgsl_pwrctrl_set_state(device, KGSL_STATE_INIT);
+	kgsl_pwrctrl_set_state(device, KGSL_STATE_INIT);
 
 	/* Power up the device */
 	kgsl_pwrctrl_enable(device);
@@ -1705,8 +1716,7 @@
 
 	kgsl_cffdump_open(device);
 
-	if (KGSL_STATE_DUMP_AND_FT != device->state)
-		kgsl_pwrctrl_set_state(device, KGSL_STATE_INIT);
+	kgsl_pwrctrl_set_state(device, KGSL_STATE_INIT);
 
 	regulator_left_on = (regulator_is_enabled(device->pwrctrl.gpu_reg) ||
 				(device->pwrctrl.gpu_cx &&
@@ -1757,11 +1767,11 @@
 	if (status)
 		goto error_irq_off;
 
-	mod_timer(&device->hang_timer,
-		(jiffies + msecs_to_jiffies(KGSL_TIMEOUT_PART)));
-
 	adreno_perfcounter_start(adreno_dev);
 
+	/* Start the dispatcher */
+	adreno_dispatcher_start(adreno_dev);
+
 	device->reset_counter++;
 
 	return 0;
@@ -1791,6 +1801,7 @@
 
 	adreno_dev->drawctxt_active = NULL;
 
+	adreno_dispatcher_stop(adreno_dev);
 	adreno_ringbuffer_stop(&adreno_dev->ringbuffer);
 
 	kgsl_mmu_stop(&device->mmu);
@@ -1798,7 +1809,6 @@
 	device->ftbl->irqctrl(device, 0);
 	kgsl_pwrctrl_irq(device, KGSL_PWRFLAGS_OFF);
 	del_timer_sync(&device->idle_timer);
-	del_timer_sync(&device->hang_timer);
 
 	adreno_ocmem_gmem_free(adreno_dev);
 
@@ -1810,917 +1820,41 @@
 	return 0;
 }
 
-/*
- * Set the reset status of all contexts to
- * INNOCENT_CONTEXT_RESET_EXT except for the bad context
- * since thats the guilty party, if fault tolerance failed then
- * mark all as guilty
- */
-
-static int _mark_context_status(int id, void *ptr, void *data)
-{
-	unsigned int ft_status = *((unsigned int *) data);
-	struct kgsl_context *context = ptr;
-	struct adreno_context *adreno_context = ADRENO_CONTEXT(context);
-
-	if (ft_status) {
-		context->reset_status =
-				KGSL_CTX_STAT_GUILTY_CONTEXT_RESET_EXT;
-		adreno_context->flags |= CTXT_FLAGS_GPU_HANG;
-	} else if (KGSL_CTX_STAT_GUILTY_CONTEXT_RESET_EXT !=
-		context->reset_status) {
-		if (adreno_context->flags & (CTXT_FLAGS_GPU_HANG |
-			CTXT_FLAGS_GPU_HANG_FT))
-			context->reset_status =
-			KGSL_CTX_STAT_GUILTY_CONTEXT_RESET_EXT;
-		else
-			context->reset_status =
-			KGSL_CTX_STAT_INNOCENT_CONTEXT_RESET_EXT;
-	}
-
-	return 0;
-}
-
-static void adreno_mark_context_status(struct kgsl_device *device,
-					int ft_status)
-{
-	/* Mark the status for all the contexts in the device */
-
-	read_lock(&device->context_lock);
-	idr_for_each(&device->context_idr, _mark_context_status, &ft_status);
-	read_unlock(&device->context_lock);
-}
-
-/*
- * For hung contexts set the current memstore value to the most recent issued
- * timestamp - this resets the status and lets the system continue on
- */
-
-static int _set_max_ts(int id, void *ptr, void *data)
-{
-	struct kgsl_device *device = data;
-	struct kgsl_context *context = ptr;
-	struct adreno_context *drawctxt = ADRENO_CONTEXT(context);
-
-	if (drawctxt && drawctxt->flags & CTXT_FLAGS_GPU_HANG) {
-		kgsl_sharedmem_writel(device, &device->memstore,
-			KGSL_MEMSTORE_OFFSET(context->id,
-			soptimestamp), drawctxt->timestamp);
-		kgsl_sharedmem_writel(device, &device->memstore,
-			KGSL_MEMSTORE_OFFSET(context->id,
-			eoptimestamp), drawctxt->timestamp);
-	}
-
-	return 0;
-}
-
-static void adreno_set_max_ts_for_bad_ctxs(struct kgsl_device *device)
-{
-	read_lock(&device->context_lock);
-	idr_for_each(&device->context_idr, _set_max_ts, device);
-	read_unlock(&device->context_lock);
-}
-
-static void adreno_destroy_ft_data(struct adreno_ft_data *ft_data)
-{
-	vfree(ft_data->rb_buffer);
-	vfree(ft_data->bad_rb_buffer);
-	vfree(ft_data->good_rb_buffer);
-}
-
-static int _find_start_of_cmd_seq(struct adreno_ringbuffer *rb,
-					unsigned int *ptr,
-					bool inc)
-{
-	int status = -EINVAL;
-	unsigned int val1;
-	unsigned int size = rb->buffer_desc.size;
-	unsigned int start_ptr = *ptr;
-
-	while ((start_ptr / sizeof(unsigned int)) != rb->wptr) {
-		if (inc)
-			start_ptr = adreno_ringbuffer_inc_wrapped(start_ptr,
-									size);
-		else
-			start_ptr = adreno_ringbuffer_dec_wrapped(start_ptr,
-									size);
-		kgsl_sharedmem_readl(&rb->buffer_desc, &val1, start_ptr);
-		/* Ensure above read is finished before next read */
-		rmb();
-		if (KGSL_CMD_IDENTIFIER == val1) {
-			if ((start_ptr / sizeof(unsigned int)) != rb->wptr)
-				start_ptr = adreno_ringbuffer_dec_wrapped(
-							start_ptr, size);
-				*ptr = start_ptr;
-				status = 0;
-				break;
-		}
-	}
-	return status;
-}
-
-static int _find_cmd_seq_after_eop_ts(struct adreno_ringbuffer *rb,
-					unsigned int *rb_rptr,
-					unsigned int global_eop,
-					bool inc)
-{
-	int status = -EINVAL;
-	unsigned int temp_rb_rptr = *rb_rptr;
-	unsigned int size = rb->buffer_desc.size;
-	unsigned int val[3];
-	int i = 0;
-	bool check = false;
-
-	if (inc && temp_rb_rptr / sizeof(unsigned int) != rb->wptr)
-		return status;
-
-	do {
-		/*
-		 * when decrementing we need to decrement first and
-		 * then read make sure we cover all the data
-		 */
-		if (!inc)
-			temp_rb_rptr = adreno_ringbuffer_dec_wrapped(
-					temp_rb_rptr, size);
-		kgsl_sharedmem_readl(&rb->buffer_desc, &val[i],
-					temp_rb_rptr);
-		/* Ensure above read is finished before next read */
-		rmb();
-
-		if (check && ((inc && val[i] == global_eop) ||
-			(!inc && (val[i] ==
-			cp_type3_packet(CP_MEM_WRITE, 2) ||
-			val[i] == CACHE_FLUSH_TS)))) {
-			/* decrement i, i.e i = (i - 1 + 3) % 3 if
-			 * we are going forward, else increment i */
-			i = (i + 2) % 3;
-			if (val[i] == rb->device->memstore.gpuaddr +
-				KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
-						eoptimestamp)) {
-				int j = ((i + 2) % 3);
-				if ((inc && (val[j] == CACHE_FLUSH_TS ||
-						val[j] == cp_type3_packet(
-							CP_MEM_WRITE, 2))) ||
-					(!inc && val[j] == global_eop)) {
-						/* Found the global eop */
-						status = 0;
-						break;
-				}
-			}
-			/* if no match found then increment i again
-			 * since we decremented before matching */
-			i = (i + 1) % 3;
-		}
-		if (inc)
-			temp_rb_rptr = adreno_ringbuffer_inc_wrapped(
-						temp_rb_rptr, size);
-
-		i = (i + 1) % 3;
-		if (2 == i)
-			check = true;
-	} while (temp_rb_rptr / sizeof(unsigned int) != rb->wptr);
-	/* temp_rb_rptr points to the command stream after global eop,
-	 * move backward till the start of command sequence */
-	if (!status) {
-		status = _find_start_of_cmd_seq(rb, &temp_rb_rptr, false);
-		if (!status) {
-			*rb_rptr = temp_rb_rptr;
-			KGSL_FT_INFO(rb->device,
-			"Offset of cmd sequence after eop timestamp: 0x%x\n",
-			temp_rb_rptr / sizeof(unsigned int));
-		}
-	}
-	if (status)
-		KGSL_FT_ERR(rb->device,
-		"Failed to find the command sequence after eop timestamp %x\n",
-		global_eop);
-	return status;
-}
-
-static int _find_hanging_ib_sequence(struct adreno_ringbuffer *rb,
-				unsigned int *rb_rptr,
-				unsigned int ib1)
-{
-	int status = -EINVAL;
-	unsigned int temp_rb_rptr = *rb_rptr;
-	unsigned int size = rb->buffer_desc.size;
-	unsigned int val[2];
-	int i = 0;
-	bool check = false;
-	bool ctx_switch = false;
-
-	while (temp_rb_rptr / sizeof(unsigned int) != rb->wptr) {
-		kgsl_sharedmem_readl(&rb->buffer_desc, &val[i], temp_rb_rptr);
-		/* Ensure above read is finished before next read */
-		rmb();
-
-		if (check && val[i] == ib1) {
-			/* decrement i, i.e i = (i - 1 + 2) % 2 */
-			i = (i + 1) % 2;
-			if (adreno_cmd_is_ib(val[i])) {
-				/* go till start of command sequence */
-				status = _find_start_of_cmd_seq(rb,
-						&temp_rb_rptr, false);
-
-				KGSL_FT_INFO(rb->device,
-				"Found the hanging IB at offset 0x%x\n",
-				temp_rb_rptr / sizeof(unsigned int));
-				break;
-			}
-			/* if no match the increment i since we decremented
-			 * before checking */
-			i = (i + 1) % 2;
-		}
-		/* Make sure you do not encounter a context switch twice, we can
-		 * encounter it once for the bad context as the start of search
-		 * can point to the context switch */
-		if (val[i] == KGSL_CONTEXT_TO_MEM_IDENTIFIER) {
-			if (ctx_switch) {
-				KGSL_FT_ERR(rb->device,
-				"Context switch encountered before bad "
-				"IB found\n");
-				break;
-			}
-			ctx_switch = true;
-		}
-		i = (i + 1) % 2;
-		if (1 == i)
-			check = true;
-		temp_rb_rptr = adreno_ringbuffer_inc_wrapped(temp_rb_rptr,
-								size);
-	}
-	if  (!status)
-		*rb_rptr = temp_rb_rptr;
-	return status;
-}
-
-static void adreno_setup_ft_data(struct kgsl_device *device,
-					struct adreno_ft_data *ft_data)
-{
-	int ret = 0;
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	struct adreno_ringbuffer *rb = &adreno_dev->ringbuffer;
-	struct kgsl_context *context;
-	struct adreno_context *adreno_context;
-	unsigned int rb_rptr = rb->wptr * sizeof(unsigned int);
-
-	memset(ft_data, 0, sizeof(*ft_data));
-	ft_data->start_of_replay_cmds = 0xFFFFFFFF;
-	ft_data->replay_for_snapshot = 0xFFFFFFFF;
-
-	adreno_readreg(adreno_dev, ADRENO_REG_CP_IB1_BASE, &ft_data->ib1);
-
-	kgsl_sharedmem_readl(&device->memstore, &ft_data->context_id,
-			KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
-			current_context));
-
-	kgsl_sharedmem_readl(&device->memstore,
-			&ft_data->global_eop,
-			KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
-			eoptimestamp));
-
-	/* Ensure context id and global eop ts read complete */
-	rmb();
-
-	ft_data->rb_buffer = vmalloc(rb->buffer_desc.size);
-	if (!ft_data->rb_buffer) {
-		KGSL_MEM_ERR(device, "vmalloc(%d) failed\n",
-				rb->buffer_desc.size);
-		return;
-	}
-
-	ft_data->bad_rb_buffer = vmalloc(rb->buffer_desc.size);
-	if (!ft_data->bad_rb_buffer) {
-		KGSL_MEM_ERR(device, "vmalloc(%d) failed\n",
-				rb->buffer_desc.size);
-		return;
-	}
-
-	ft_data->good_rb_buffer = vmalloc(rb->buffer_desc.size);
-	if (!ft_data->good_rb_buffer) {
-		KGSL_MEM_ERR(device, "vmalloc(%d) failed\n",
-				rb->buffer_desc.size);
-		return;
-	}
-	ft_data->status = 0;
-
-	/* find the start of bad command sequence in rb */
-	context = kgsl_context_get(device, ft_data->context_id);
-
-	ft_data->ft_policy = adreno_dev->ft_policy;
-
-	if (!ft_data->ft_policy)
-		ft_data->ft_policy = KGSL_FT_DEFAULT_POLICY;
-
-	/* Look for the command stream that is right after the global eop */
-	ret = _find_cmd_seq_after_eop_ts(rb, &rb_rptr,
-					ft_data->global_eop + 1, false);
-	if (ret) {
-		ft_data->ft_policy |= KGSL_FT_TEMP_DISABLE;
-		goto done;
-	} else {
-		ft_data->start_of_replay_cmds = rb_rptr;
-		ft_data->ft_policy &= ~KGSL_FT_TEMP_DISABLE;
-	}
-
-	if (context) {
-		adreno_context = ADRENO_CONTEXT(context);
-		if (adreno_context->flags & CTXT_FLAGS_PREAMBLE) {
-			if (ft_data->ib1) {
-				ret = _find_hanging_ib_sequence(rb,
-						&rb_rptr, ft_data->ib1);
-				if (ret) {
-					KGSL_FT_ERR(device,
-					"Start not found for replay IB seq\n");
-					goto done;
-				}
-				ft_data->start_of_replay_cmds = rb_rptr;
-				ft_data->replay_for_snapshot = rb_rptr;
-			}
-		}
-	}
-
-done:
-	kgsl_context_put(context);
-}
-
-static int
-_adreno_check_long_ib(struct kgsl_device *device)
-{
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	unsigned int curr_global_ts = 0;
-
-	/* check if the global ts is still the same */
-	kgsl_sharedmem_readl(&device->memstore,
-			&curr_global_ts,
-			KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
-			eoptimestamp));
-	/* Ensure above read is finished before long ib check */
-	rmb();
-
-	/* Mark long ib as handled */
-	adreno_dev->long_ib = 0;
-
-	if (curr_global_ts == adreno_dev->long_ib_ts) {
-		KGSL_FT_ERR(device,
-			"IB ran too long, invalidate ctxt\n");
-		return 1;
-	} else {
-		/* Do nothing GPU has gone ahead */
-		KGSL_FT_INFO(device, "false long ib detection return\n");
-		return 0;
-	}
-}
-
 /**
- * adreno_soft_reset() -  Do a soft reset of the GPU hardware
- * @device: KGSL device to soft reset
+ * adreno_reset() - Helper function to reset the GPU
+ * @device: Pointer to the KGSL device structure for the GPU
  *
- * "soft reset" the GPU hardware - this is a fast path GPU reset
- * The GPU hardware is reset but we never pull power so we can skip
- * a lot of the standard adreno_stop/adreno_start sequence
+ * Try to reset the GPU to recover from a fault.  First, try to do a low latency
+ * soft reset.  If the soft reset fails for some reason, then bring out the big
+ * guns and toggle the footswitch.
  */
-int adreno_soft_reset(struct kgsl_device *device)
+int adreno_reset(struct kgsl_device *device)
 {
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
 	int ret;
 
-	/* If the jump table index is 0 soft reset is not supported */
-	if ((!adreno_dev->pm4_jt_idx) || (!adreno_dev->gpudev->soft_reset)) {
-		dev_WARN_ONCE(device->dev, 1, "Soft reset not supported");
-		return -EINVAL;
-	}
+	/* Try soft reset first */
+	if (adreno_soft_reset(device) == 0)
+		return 0;
 
-	if (adreno_dev->drawctxt_active)
-		kgsl_context_put(&adreno_dev->drawctxt_active->base);
-
-	adreno_dev->drawctxt_active = NULL;
-
-	/* Stop the ringbuffer */
-	adreno_ringbuffer_stop(&adreno_dev->ringbuffer);
-
-	/* Delete the idle timer */
-	del_timer_sync(&device->idle_timer);
-
-	/* Make sure we are totally awake */
-	kgsl_pwrctrl_enable(device);
-
-	/* Reset the GPU */
-	adreno_dev->gpudev->soft_reset(adreno_dev);
-
-	/* Reinitialize the GPU */
-	adreno_dev->gpudev->start(adreno_dev);
-
-	/* Enable IRQ */
-	kgsl_pwrctrl_irq(device, KGSL_PWRFLAGS_ON);
-	device->ftbl->irqctrl(device, 1);
-
-	/*
-	 * Restart the ringbuffer - we can go down the warm start path because
-	 * power was never yanked
-	 */
-	ret = adreno_ringbuffer_warm_start(&adreno_dev->ringbuffer);
+	/* If it failed, then pull the power */
+	ret = adreno_stop(device);
 	if (ret)
 		return ret;
 
-	device->reset_counter++;
+	ret = adreno_start(device);
 
-	return 0;
-}
-
-static int
-_adreno_ft_restart_device(struct kgsl_device *device,
-			   struct kgsl_context *context)
-{
-	/* If device soft reset fails try hard reset */
-	if (adreno_soft_reset(device))
-		KGSL_DEV_ERR_ONCE(device, "Device soft reset failed\n");
-	else
-		/* Soft reset is successful */
-		goto reset_done;
-
-	/* restart device */
-	if (adreno_stop(device)) {
-		KGSL_FT_ERR(device, "Device stop failed\n");
-		return 1;
-	}
-
-	if (adreno_init(device)) {
-		KGSL_FT_ERR(device, "Device init failed\n");
-		return 1;
-	}
-
-	if (adreno_start(device)) {
-		KGSL_FT_ERR(device, "Device start failed\n");
-		return 1;
-	}
-
-reset_done:
-	if (context)
-		kgsl_mmu_setstate(&device->mmu, context->pagetable,
-				KGSL_MEMSTORE_GLOBAL);
-
-	/* If iommu is used then we need to make sure that the iommu clocks
-	 * are on since there could be commands in pipeline that touch iommu */
-	if (KGSL_MMU_TYPE_IOMMU == kgsl_mmu_get_mmutype()) {
-		if (kgsl_mmu_enable_clk(&device->mmu,
-				KGSL_IOMMU_CONTEXT_USER))
-			return 1;
-	}
-
-	return 0;
-}
-
-static inline void
-_adreno_debug_ft_info(struct kgsl_device *device,
-			struct adreno_ft_data *ft_data)
-{
-
-	/*
-	 * Dumping rb is a very useful tool to debug FT.
-	 * It will tell us if we are extracting the rb correctly
-	 * NOP'ing the right IB, skipping the EOF correctly etc.
-	 */
-	if (device->ft_log >= 7)  {
-
-		/* Print fault tolerance data here */
-		KGSL_FT_INFO(device, "Temp RB buffer size 0x%X\n",
-			ft_data->rb_size);
-		adreno_dump_rb(device, ft_data->rb_buffer,
-			ft_data->rb_size<<2, 0, ft_data->rb_size);
-
-		KGSL_FT_INFO(device, "Bad RB buffer size 0x%X\n",
-			ft_data->bad_rb_size);
-		adreno_dump_rb(device, ft_data->bad_rb_buffer,
-			ft_data->bad_rb_size<<2, 0, ft_data->bad_rb_size);
-
-		KGSL_FT_INFO(device, "Good RB buffer size 0x%X\n",
-			ft_data->good_rb_size);
-		adreno_dump_rb(device, ft_data->good_rb_buffer,
-			ft_data->good_rb_size<<2, 0, ft_data->good_rb_size);
-
-	}
-}
-
-static int
-_adreno_ft_resubmit_rb(struct kgsl_device *device,
-			struct adreno_ringbuffer *rb,
-			struct kgsl_context *context,
-			struct adreno_ft_data *ft_data,
-			unsigned int *buff, unsigned int size)
-{
-	unsigned int ret = 0;
-	unsigned int retry_num = 0;
-
-	_adreno_debug_ft_info(device, ft_data);
-
-	do {
-		ret = _adreno_ft_restart_device(device, context);
-		if (ret == 0)
-			break;
+	if (ret == 0) {
 		/*
-		 * If device restart fails sleep for 20ms before
-		 * attempting restart. This allows GPU HW to settle
-		 * and improve the chances of next restart to be
-		 * successful.
+		 * If active_cnt is non-zero then the system was active before
+		 * going into a reset - put it back in that state
 		 */
-		msleep(20);
-		KGSL_FT_ERR(device, "Retry device restart %d\n", retry_num);
-		retry_num++;
-	} while (retry_num < 4);
 
-	if (ret) {
-		KGSL_FT_ERR(device, "Device restart failed\n");
-		BUG_ON(1);
-		goto done;
-	}
-
-	if (size) {
-
-		/* submit commands and wait for them to pass */
-		adreno_ringbuffer_restore(rb, buff, size);
-
-		ret = adreno_idle(device);
-	}
-
-done:
-	return ret;
-}
-
-
-static int
-_adreno_ft(struct kgsl_device *device,
-			struct adreno_ft_data *ft_data)
-{
-	int ret = 0, i;
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	struct adreno_ringbuffer *rb = &adreno_dev->ringbuffer;
-	struct kgsl_context *context;
-	struct adreno_context *adreno_context = NULL;
-	struct adreno_context *last_active_ctx = adreno_dev->drawctxt_active;
-	unsigned int long_ib = 0;
-	static int no_context_ft;
-	struct kgsl_mmu *mmu = &device->mmu;
-
-	context = kgsl_context_get(device, ft_data->context_id);
-
-	if (context == NULL) {
-		KGSL_FT_ERR(device, "Last context unknown id:%d\n",
-			ft_data->context_id);
-		if (no_context_ft) {
-			/*
-			 * If 2 consecutive no context ft occurred then
-			 * just reset GPU
-			 */
-			no_context_ft = 0;
-			goto play_good_cmds;
-		}
-	} else {
-		no_context_ft = 0;
-		adreno_context = ADRENO_CONTEXT(context);
-		adreno_context->flags |= CTXT_FLAGS_GPU_HANG;
-		/*
-		 * set the invalid ts flag to 0 for this context since we have
-		 * detected a hang for it
-		 */
-		context->wait_on_invalid_ts = false;
-
-		if (!(adreno_context->flags & CTXT_FLAGS_PER_CONTEXT_TS)) {
-			ft_data->status = 1;
-			KGSL_FT_ERR(device, "Fault tolerance not supported\n");
-			goto play_good_cmds;
-		}
-
-		/*
-		 *  This flag will be set by userspace for contexts
-		 *  that do not want to be fault tolerant (ex: OPENCL)
-		 */
-		if (adreno_context->flags & CTXT_FLAGS_NO_FAULT_TOLERANCE) {
-			ft_data->status = 1;
-			KGSL_FT_ERR(device,
-			"No FT set for this context play good cmds\n");
-			goto play_good_cmds;
-		}
-
-	}
-
-	/* Check if we detected a long running IB, if false return */
-	if ((adreno_context) && (adreno_dev->long_ib)) {
-		long_ib = _adreno_check_long_ib(device);
-		if (!long_ib) {
-			adreno_context->flags &= ~CTXT_FLAGS_GPU_HANG;
-			return 0;
-		}
-	}
-
-	/*
-	 * Extract valid contents from rb which can still be executed after
-	 * hang
-	 */
-	adreno_ringbuffer_extract(rb, ft_data);
-
-	/* If long IB detected do not attempt replay of bad cmds */
-	if (long_ib) {
-		ft_data->status = 1;
-		_adreno_debug_ft_info(device, ft_data);
-		goto play_good_cmds;
-	}
-
-	if ((ft_data->ft_policy & KGSL_FT_DISABLE) ||
-		(ft_data->ft_policy & KGSL_FT_TEMP_DISABLE)) {
-		KGSL_FT_ERR(device, "NO FT policy play only good cmds\n");
-		ft_data->status = 1;
-		goto play_good_cmds;
-	}
-
-	/* Do not try to replay if hang is due to a pagefault */
-	if (context && test_bit(KGSL_CONTEXT_PAGEFAULT, &context->priv)) {
-		/* Resume MMU */
-		mmu->mmu_ops->mmu_pagefault_resume(mmu);
-		if ((ft_data->context_id == context->id) &&
-			(ft_data->global_eop == context->pagefault_ts)) {
-			ft_data->ft_policy &= ~KGSL_FT_REPLAY;
-			KGSL_FT_ERR(device, "MMU fault skipping replay\n");
-		}
-		clear_bit(KGSL_CONTEXT_PAGEFAULT, &context->priv);
-	}
-
-	if (ft_data->ft_policy & KGSL_FT_REPLAY) {
-		ret = _adreno_ft_resubmit_rb(device, rb, context, ft_data,
-				ft_data->bad_rb_buffer, ft_data->bad_rb_size);
-
-		if (ret) {
-			KGSL_FT_ERR(device, "Replay status: 1\n");
-			ft_data->status = 1;
-		} else
-			goto play_good_cmds;
-	}
-
-	if (ft_data->ft_policy & KGSL_FT_SKIPIB) {
-		for (i = 0; i < ft_data->bad_rb_size; i++) {
-			if ((ft_data->bad_rb_buffer[i] ==
-					CP_HDR_INDIRECT_BUFFER_PFD) &&
-				(ft_data->bad_rb_buffer[i+1] == ft_data->ib1)) {
-
-				ft_data->bad_rb_buffer[i] = cp_nop_packet(2);
-				ft_data->bad_rb_buffer[i+1] =
-							KGSL_NOP_IB_IDENTIFIER;
-				ft_data->bad_rb_buffer[i+2] =
-							KGSL_NOP_IB_IDENTIFIER;
-				break;
-			}
-		}
-
-		if ((i == (ft_data->bad_rb_size)) || (!ft_data->ib1)) {
-			KGSL_FT_ERR(device, "Bad IB to NOP not found\n");
-			ft_data->status = 1;
-			goto play_good_cmds;
-		}
-
-		ret = _adreno_ft_resubmit_rb(device, rb, context, ft_data,
-				ft_data->bad_rb_buffer, ft_data->bad_rb_size);
-
-		if (ret) {
-			KGSL_FT_ERR(device, "NOP faulty IB status: 1\n");
-			ft_data->status = 1;
-		} else {
-			ft_data->status = 0;
-			goto play_good_cmds;
-		}
-	}
-
-	if (ft_data->ft_policy & KGSL_FT_SKIPFRAME) {
-		for (i = 0; i < ft_data->bad_rb_size; i++) {
-			if (ft_data->bad_rb_buffer[i] ==
-					KGSL_END_OF_FRAME_IDENTIFIER) {
-				ft_data->bad_rb_buffer[0] = cp_nop_packet(i);
-				break;
-			}
-		}
-
-		/* EOF not found in RB, discard till EOF in
-		   next IB submission */
-		if (adreno_context && (i == ft_data->bad_rb_size)) {
-			adreno_context->flags |= CTXT_FLAGS_SKIP_EOF;
-			KGSL_FT_INFO(device,
-			"EOF not found in RB, skip next issueib till EOF\n");
-			ft_data->bad_rb_buffer[0] = cp_nop_packet(i);
-		}
-
-		ret = _adreno_ft_resubmit_rb(device, rb, context, ft_data,
-				ft_data->bad_rb_buffer, ft_data->bad_rb_size);
-
-		if (ret) {
-			KGSL_FT_ERR(device, "Skip EOF status: 1\n");
-			ft_data->status = 1;
-		} else {
-			ft_data->status = 0;
-			goto play_good_cmds;
-		}
-	}
-
-play_good_cmds:
-
-	if (ft_data->status)
-		KGSL_FT_ERR(device, "Bad context commands failed\n");
-	else {
-		KGSL_FT_INFO(device, "Bad context commands success\n");
-
-		if (adreno_context) {
-			adreno_context->flags = (adreno_context->flags &
-				~CTXT_FLAGS_GPU_HANG) | CTXT_FLAGS_GPU_HANG_FT;
-		}
-
-		if (last_active_ctx)
-			_kgsl_context_get(&last_active_ctx->base);
-
-		adreno_dev->drawctxt_active = last_active_ctx;
-	}
-
-	ret = _adreno_ft_resubmit_rb(device, rb, context, ft_data,
-			ft_data->good_rb_buffer, ft_data->good_rb_size);
-
-	if (ret) {
-		/*
-		 * If we fail here we can try to invalidate another
-		 * context and try fault tolerance again, although
-		 * we will only try ft with no context once to avoid
-		 * going into continuous loop of trying ft with no context
-		 */
-		if (!context)
-			no_context_ft = 1;
-		ret = -EAGAIN;
-		KGSL_FT_ERR(device, "Playing good commands unsuccessful\n");
-		goto done;
-	} else
-		KGSL_FT_INFO(device, "Playing good commands successful\n");
-
-	/* ringbuffer now has data from the last valid context id,
-	 * so restore the active_ctx to the last valid context */
-	if (ft_data->last_valid_ctx_id) {
-		struct kgsl_context *last_ctx = kgsl_context_get(device,
-			ft_data->last_valid_ctx_id);
-
-		adreno_dev->drawctxt_active = ADRENO_CONTEXT(last_ctx);
-	}
-
-done:
-	/* Turn off iommu clocks */
-	if (KGSL_MMU_TYPE_IOMMU == kgsl_mmu_get_mmutype())
-		kgsl_mmu_disable_clk_on_ts(&device->mmu, 0, false);
-
-	kgsl_context_put(context);
-	return ret;
-}
-
-static int
-adreno_ft(struct kgsl_device *device,
-			struct adreno_ft_data *ft_data)
-{
-	int ret = 0;
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	struct adreno_ringbuffer *rb = &adreno_dev->ringbuffer;
-
-	/*
-	 * If GPU FT is turned off do not run FT.
-	 * If GPU stall detection is suspected to be false,
-	 * we can use this option to confirm stall detection.
-	 */
-	if (ft_data->ft_policy & KGSL_FT_OFF) {
-		KGSL_FT_ERR(device, "GPU FT turned off\n");
-		return 0;
-	}
-
-	KGSL_FT_INFO(device,
-	"Start Parameters: IB1: 0x%X, "
-	"Bad context_id: %u, global_eop: 0x%x\n",
-	ft_data->ib1, ft_data->context_id, ft_data->global_eop);
-
-	KGSL_FT_INFO(device, "Last issued global timestamp: %x\n",
-			rb->global_ts);
-
-	/* We may need to replay commands multiple times based on whether
-	 * multiple contexts hang the GPU */
-	while (true) {
-
-		ret = _adreno_ft(device, ft_data);
-
-		if (-EAGAIN == ret) {
-			/* setup new fault tolerance parameters and retry, this
-			 * means more than 1 contexts are causing hang */
-			adreno_destroy_ft_data(ft_data);
-			adreno_setup_ft_data(device, ft_data);
-			KGSL_FT_INFO(device,
-			"Retry. Parameters: "
-			"IB1: 0x%X, Bad context_id: %u, global_eop: 0x%x\n",
-			ft_data->ib1, ft_data->context_id,
-			ft_data->global_eop);
-		} else {
-			break;
-		}
-	}
-
-	if (ret)
-		goto done;
-
-	/* Restore correct states after fault tolerance */
-	if (adreno_dev->drawctxt_active)
-		device->mmu.hwpagetable =
-			adreno_dev->drawctxt_active->base.pagetable;
-	else
-		device->mmu.hwpagetable = device->mmu.defaultpagetable;
-	kgsl_sharedmem_writel(device, &device->memstore,
-			KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
-			eoptimestamp), rb->global_ts);
-
-	/* switch to NULL ctxt */
-	if (adreno_dev->drawctxt_active != NULL)
-		adreno_drawctxt_switch(adreno_dev, NULL, 0);
-
-done:
-	adreno_set_max_ts_for_bad_ctxs(device);
-	adreno_mark_context_status(device, ret);
-	KGSL_FT_ERR(device, "policy 0x%X status 0x%x\n",
-			ft_data->ft_policy, ret);
-	return ret;
-}
-
-int
-adreno_dump_and_exec_ft(struct kgsl_device *device)
-{
-	int result = -ETIMEDOUT;
-	struct adreno_ft_data ft_data;
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	struct kgsl_pwrctrl *pwr = &device->pwrctrl;
-	unsigned int curr_pwrlevel;
-
-	if (device->state == KGSL_STATE_HUNG)
-		goto done;
-	if (device->state == KGSL_STATE_DUMP_AND_FT) {
-		mutex_unlock(&device->mutex);
-		wait_for_completion(&device->ft_gate);
-		mutex_lock(&device->mutex);
-		if (device->state != KGSL_STATE_HUNG)
-			result = 0;
-	} else {
-		/*
-		 * While fault tolerance is happening we do not want the
-		 * idle_timer to fire and attempt to change any device state
-		 */
-		del_timer_sync(&device->idle_timer);
-
-		kgsl_pwrctrl_set_state(device, KGSL_STATE_DUMP_AND_FT);
-		INIT_COMPLETION(device->ft_gate);
-		/* Detected a hang */
-
-		kgsl_cffdump_hang(device);
-		/* Run fault tolerance at max power level */
-		curr_pwrlevel = pwr->active_pwrlevel;
-		kgsl_pwrctrl_pwrlevel_change(device, pwr->max_pwrlevel);
-
-		/* Get the fault tolerance data as soon as hang is detected */
-		adreno_setup_ft_data(device, &ft_data);
-
-		/*
-		 * If long ib is detected, do not attempt postmortem or
-		 * snapshot, if GPU is still executing commands
-		 * we will get errors
-		 */
-		if (!adreno_dev->long_ib) {
-			/*
-			 * Trigger an automatic dump of the state to
-			 * the console
-			 */
-			kgsl_postmortem_dump(device, 0);
-
-			/*
-			* Make a GPU snapshot.  For now, do it after the
-			* PM dump so we can at least be sure the PM dump
-			* will work as it always has
-			*/
-			kgsl_device_snapshot(device, 1);
-		}
-
-		result = adreno_ft(device, &ft_data);
-		adreno_destroy_ft_data(&ft_data);
-
-		/* restore power level */
-		kgsl_pwrctrl_pwrlevel_change(device, curr_pwrlevel);
-
-		if (result) {
-			kgsl_pwrctrl_set_state(device, KGSL_STATE_HUNG);
-		} else {
+		if (atomic_read(&device->active_cnt))
 			kgsl_pwrctrl_set_state(device, KGSL_STATE_ACTIVE);
-			mod_timer(&device->hang_timer,
-				(jiffies +
-				msecs_to_jiffies(KGSL_TIMEOUT_PART)));
-		}
-		complete_all(&device->ft_gate);
 	}
-done:
-	return result;
+
+	return ret;
 }
-EXPORT_SYMBOL(adreno_dump_and_exec_ft);
 
 /**
  * _ft_sysfs_store() -  Common routine to write to FT sysfs files
@@ -3119,140 +2253,166 @@
 	return status;
 }
 
-static int adreno_ringbuffer_drain(struct kgsl_device *device,
-	unsigned int *regs)
+/**
+ * adreno_hw_isidle() - Check if the GPU core is idle
+ * @device: Pointer to the KGSL device structure for the GPU
+ *
+ * Return true if the RBBM status register for the GPU type indicates that the
+ * hardware is idle
+ */
+static bool adreno_hw_isidle(struct kgsl_device *device)
+{
+	unsigned int reg_rbbm_status;
+	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
+
+	/* Don't consider ourselves idle if there is an IRQ pending */
+	if (adreno_dev->gpudev->irq_pending(adreno_dev))
+		return false;
+
+	adreno_readreg(adreno_dev, ADRENO_REG_RBBM_STATUS,
+		&reg_rbbm_status);
+
+	if (adreno_is_a2xx(adreno_dev)) {
+		if (reg_rbbm_status == 0x110)
+			return true;
+	} else if (adreno_is_a3xx(adreno_dev)) {
+		if (!(reg_rbbm_status & 0x80000000))
+			return true;
+	}
+
+	return false;
+}
+
+/**
+ * adreno_soft_reset() -  Do a soft reset of the GPU hardware
+ * @device: KGSL device to soft reset
+ *
+ * "soft reset" the GPU hardware - this is a fast path GPU reset
+ * The GPU hardware is reset but we never pull power so we can skip
+ * a lot of the standard adreno_stop/adreno_start sequence
+ */
+int adreno_soft_reset(struct kgsl_device *device)
 {
 	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	struct adreno_ringbuffer *rb = &adreno_dev->ringbuffer;
-	unsigned long wait = jiffies;
-	unsigned long timeout = jiffies + msecs_to_jiffies(ADRENO_IDLE_TIMEOUT);
-	unsigned int rptr;
+	int ret;
 
-	do {
-		/*
-		 * Wait is "jiffies" first time in the loop to start
-		 * GPU stall detection immediately.
-		 */
-		if (time_after(jiffies, wait)) {
-			/* Check to see if the core is hung */
-			if (adreno_ft_detect(device, regs))
-				return -ETIMEDOUT;
+	/* If the jump table index is 0 soft reset is not supported */
+	if ((!adreno_dev->pm4_jt_idx) || (!adreno_dev->gpudev->soft_reset)) {
+		dev_WARN_ONCE(device->dev, 1, "Soft reset not supported");
+		return -EINVAL;
+	}
 
-			wait = jiffies + msecs_to_jiffies(KGSL_TIMEOUT_PART);
-		}
-		rptr = adreno_get_rptr(rb);
-		if (time_after(jiffies, timeout)) {
-			KGSL_DRV_ERR(device, "rptr: %x, wptr: %x\n",
-				rptr, rb->wptr);
-			return -ETIMEDOUT;
-		}
-	} while (rptr != rb->wptr);
+	if (adreno_dev->drawctxt_active)
+		kgsl_context_put(&adreno_dev->drawctxt_active->base);
+
+	adreno_dev->drawctxt_active = NULL;
+
+	/* Stop the ringbuffer */
+	adreno_ringbuffer_stop(&adreno_dev->ringbuffer);
+
+	/* Delete the idle timer */
+	del_timer_sync(&device->idle_timer);
+
+	/* Make sure we are totally awake */
+	kgsl_pwrctrl_enable(device);
+
+	/* Reset the GPU */
+	adreno_dev->gpudev->soft_reset(adreno_dev);
+
+	/* Reinitialize the GPU */
+	adreno_dev->gpudev->start(adreno_dev);
+
+	/* Enable IRQ */
+	kgsl_pwrctrl_irq(device, KGSL_PWRFLAGS_ON);
+	device->ftbl->irqctrl(device, 1);
+
+	/*
+	 * Restart the ringbuffer - we can go down the warm start path because
+	 * power was never yanked
+	 */
+	ret = adreno_ringbuffer_warm_start(&adreno_dev->ringbuffer);
+	if (ret)
+		return ret;
+
+	device->reset_counter++;
 
 	return 0;
 }
 
-/* Caller must hold the device mutex. */
+/*
+ * adreno_isidle() - return true if the GPU hardware is idle
+ * @device: Pointer to the KGSL device structure for the GPU
+ *
+ * Return true if the GPU hardware is idle and there are no commands pending in
+ * the ringbuffer
+ */
+static bool adreno_isidle(struct kgsl_device *device)
+{
+	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
+	unsigned int rptr;
+
+	if (!kgsl_pwrctrl_isenabled(device))
+		return true;
+
+	rptr = adreno_get_rptr(&adreno_dev->ringbuffer);
+
+	if (rptr == adreno_dev->ringbuffer.wptr)
+		return adreno_hw_isidle(device);
+
+	return false;
+}
+
+/**
+ * adreno_idle() - wait for the GPU hardware to go idle
+ * @device: Pointer to the KGSL device structure for the GPU
+ *
+ * Wait up to ADRENO_IDLE_TIMEOUT milliseconds for the GPU hardware to go quiet.
+ */
+
 int adreno_idle(struct kgsl_device *device)
 {
-	unsigned long wait_time;
-	unsigned long wait_time_part;
-	unsigned int prev_reg_val[FT_DETECT_REGS_COUNT];
 	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
+	unsigned long wait = jiffies + msecs_to_jiffies(ADRENO_IDLE_TIMEOUT);
 
-	memset(prev_reg_val, 0, sizeof(prev_reg_val));
+	/*
+	 * Make sure the device mutex is held so the dispatcher can't send any
+	 * more commands to the hardware
+	 */
 
-	kgsl_cffdump_regpoll(device,
-		adreno_getreg(adreno_dev, ADRENO_REG_RBBM_STATUS) << 2,
-		0x00000000, 0x80000000);
+	BUG_ON(!mutex_is_locked(&device->mutex));
 
-retry:
-	/* First, wait for the ringbuffer to drain */
-	if (adreno_ringbuffer_drain(device, prev_reg_val))
-		goto err;
+	if (adreno_is_a3xx(adreno_dev))
+		kgsl_cffdump_regpoll(device,
+			adreno_getreg(adreno_dev, ADRENO_REG_RBBM_STATUS) << 2,
+			0x00000000, 0x80000000);
+	else
+		kgsl_cffdump_regpoll(device,
+			adreno_getreg(adreno_dev, ADRENO_REG_RBBM_STATUS) << 2,
+			0x110, 0x110);
 
-	/* now, wait for the GPU to finish its operations */
-	wait_time = jiffies + msecs_to_jiffies(ADRENO_IDLE_TIMEOUT);
-	wait_time_part = jiffies + msecs_to_jiffies(KGSL_TIMEOUT_PART);
-
-	while (time_before(jiffies, wait_time)) {
+	while (time_before(jiffies, wait)) {
 		if (adreno_isidle(device))
 			return 0;
-
-		/* Dont wait for timeout, detect hang faster.  */
-		if (time_after(jiffies, wait_time_part)) {
-			wait_time_part = jiffies +
-				msecs_to_jiffies(KGSL_TIMEOUT_PART);
-			if ((adreno_ft_detect(device, prev_reg_val)))
-				goto err;
-		}
-
 	}
 
-err:
-	KGSL_DRV_ERR(device, "spun too long waiting for RB to idle\n");
-	if (KGSL_STATE_DUMP_AND_FT != device->state &&
-		!adreno_dump_and_exec_ft(device)) {
-		wait_time = jiffies + ADRENO_IDLE_TIMEOUT;
-		goto retry;
-	}
+	kgsl_postmortem_dump(device, 0);
+
 	return -ETIMEDOUT;
 }
 
 /**
- * is_adreno_rbbm_status_idle - Check if GPU core is idle by probing
- * rbbm_status register
- * @device - Pointer to the GPU device whose idle status is to be
- * checked
- * @returns - Returns whether the core is idle (based on rbbm_status)
- * false if the core is active, true if the core is idle
+ * adreno_drain() - Drain the dispatch queue
+ * @device: Pointer to the KGSL device structure for the GPU
+ *
+ * Tell the dispatcher to pause - this has the effect of draining the inflight
+ * command batches
  */
-static bool is_adreno_rbbm_status_idle(struct kgsl_device *device)
+static int adreno_drain(struct kgsl_device *device)
 {
-	unsigned int reg_rbbm_status;
-	bool status = false;
 	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
 
-	/* Is the core idle? */
-	adreno_readreg(adreno_dev, ADRENO_REG_RBBM_STATUS,
-				&reg_rbbm_status);
-
-	if (adreno_is_a2xx(adreno_dev)) {
-		if (reg_rbbm_status == 0x110)
-			status = true;
-	} else {
-		if (!(reg_rbbm_status & 0x80000000))
-			status = true;
-	}
-	return status;
-}
-
-static unsigned int adreno_isidle(struct kgsl_device *device)
-{
-	int status = false;
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	struct adreno_ringbuffer *rb = &adreno_dev->ringbuffer;
-
-	/* If the device isn't active, don't force it on. */
-	if (kgsl_pwrctrl_isenabled(device)) {
-		/* Is the ring buffer is empty? */
-		unsigned int rptr = adreno_get_rptr(rb);
-		if (rptr == rb->wptr) {
-			/*
-			 * Are there interrupts pending? If so then pretend we
-			 * are not idle - this avoids the possiblity that we go
-			 * to a lower power state without handling interrupts
-			 * first.
-			 */
-
-			if (!adreno_dev->gpudev->irq_pending(adreno_dev)) {
-				/* Is the core idle? */
-				status = is_adreno_rbbm_status_idle(device);
-			}
-		}
-	} else {
-		status = true;
-	}
-	return status;
+	adreno_dispatcher_pause(adreno_dev);
+	return 0;
 }
 
 /* Caller must hold the device mutex. */
@@ -3423,342 +2583,6 @@
 	__raw_writel(value, reg);
 }
 
-static unsigned int _get_context_id(struct kgsl_context *k_ctxt)
-{
-	unsigned int context_id = KGSL_MEMSTORE_GLOBAL;
-
-	if (k_ctxt != NULL) {
-		struct adreno_context *a_ctxt = ADRENO_CONTEXT(k_ctxt);
-		if (kgsl_context_detached(k_ctxt))
-			context_id = KGSL_CONTEXT_INVALID;
-		else if (a_ctxt->flags & CTXT_FLAGS_PER_CONTEXT_TS)
-			context_id = k_ctxt->id;
-	}
-
-	return context_id;
-}
-
-static unsigned int adreno_check_hw_ts(struct kgsl_device *device,
-		struct kgsl_context *context, unsigned int timestamp)
-{
-	int status = 0;
-	unsigned int ref_ts, enableflag;
-	unsigned int context_id = _get_context_id(context);
-
-	/*
-	 * If the context ID is invalid, we are in a race with
-	 * the context being destroyed by userspace so bail.
-	 */
-	if (context_id == KGSL_CONTEXT_INVALID) {
-		KGSL_DRV_WARN(device, "context was detached");
-		return -EINVAL;
-	}
-
-	status = kgsl_check_timestamp(device, context, timestamp);
-	if (status)
-		return status;
-
-	kgsl_sharedmem_readl(&device->memstore, &enableflag,
-			KGSL_MEMSTORE_OFFSET(context_id, ts_cmp_enable));
-	/*
-	 * Barrier is needed here to make sure the read from memstore
-	 * has posted
-	 */
-
-	mb();
-
-	if (enableflag) {
-		kgsl_sharedmem_readl(&device->memstore, &ref_ts,
-				KGSL_MEMSTORE_OFFSET(context_id,
-					ref_wait_ts));
-
-		/* Make sure the memstore read has posted */
-		mb();
-		if (timestamp_cmp(ref_ts, timestamp) >= 0) {
-			kgsl_sharedmem_writel(device, &device->memstore,
-					KGSL_MEMSTORE_OFFSET(context_id,
-						ref_wait_ts), timestamp);
-			/* Make sure the memstore write is posted */
-			wmb();
-		}
-	} else {
-		kgsl_sharedmem_writel(device, &device->memstore,
-				KGSL_MEMSTORE_OFFSET(context_id,
-					ref_wait_ts), timestamp);
-		enableflag = 1;
-		kgsl_sharedmem_writel(device, &device->memstore,
-				KGSL_MEMSTORE_OFFSET(context_id,
-					ts_cmp_enable), enableflag);
-
-		/* Make sure the memstore write gets posted */
-		wmb();
-
-		/*
-		 * submit a dummy packet so that even if all
-		 * commands upto timestamp get executed we will still
-		 * get an interrupt
-		 */
-
-		if (context && device->state != KGSL_STATE_SLUMBER) {
-			adreno_ringbuffer_issuecmds(device,
-					ADRENO_CONTEXT(context),
-					KGSL_CMD_FLAGS_GET_INT, NULL, 0);
-		}
-	}
-
-	return 0;
-}
-
-/* Return 1 if the event timestmp has already passed, 0 if it was marked */
-static int adreno_next_event(struct kgsl_device *device,
-		struct kgsl_event *event)
-{
-	return adreno_check_hw_ts(device, event->context, event->timestamp);
-}
-
-static int adreno_check_interrupt_timestamp(struct kgsl_device *device,
-		struct kgsl_context *context, unsigned int timestamp)
-{
-	int status;
-
-	mutex_lock(&device->mutex);
-	status = adreno_check_hw_ts(device, context, timestamp);
-	mutex_unlock(&device->mutex);
-
-	return status;
-}
-
-/*
- wait_event_interruptible_timeout checks for the exit condition before
- placing a process in wait q. For conditional interrupts we expect the
- process to already be in its wait q when its exit condition checking
- function is called.
-*/
-#define kgsl_wait_event_interruptible_timeout(wq, condition, timeout, io)\
-({									\
-	long __ret = timeout;						\
-	if (io)						\
-		__wait_io_event_interruptible_timeout(wq, condition, __ret);\
-	else						\
-		__wait_event_interruptible_timeout(wq, condition, __ret);\
-	__ret;								\
-})
-
-
-
-unsigned int adreno_ft_detect(struct kgsl_device *device,
-						unsigned int *prev_reg_val)
-{
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	struct adreno_ringbuffer *rb = &adreno_dev->ringbuffer;
-	unsigned int curr_reg_val[FT_DETECT_REGS_COUNT];
-	unsigned int fast_hang_detected = 1;
-	unsigned int long_ib_detected = 1;
-	unsigned int i;
-	static unsigned long next_hang_detect_time;
-	static unsigned int prev_global_ts;
-	unsigned int curr_global_ts = 0;
-	unsigned int curr_context_id = 0;
-	static struct adreno_context *curr_context;
-	static struct kgsl_context *context;
-	static char pid_name[TASK_COMM_LEN] = "unknown";
-
-	if (!adreno_dev->fast_hang_detect)
-		fast_hang_detected = 0;
-
-	if (!adreno_dev->long_ib_detect)
-		long_ib_detected = 0;
-
-	if (!(adreno_dev->ringbuffer.flags & KGSL_FLAGS_STARTED))
-		return 0;
-
-	if (is_adreno_rbbm_status_idle(device) &&
-		(kgsl_readtimestamp(device, NULL, KGSL_TIMESTAMP_RETIRED)
-		 == rb->global_ts)) {
-
-		/*
-		 * On A2XX if the RPTR != WPTR and the device is idle, then
-		 * the last write to WPTR probably failed to latch so write it
-		 * again
-		 */
-
-		if (adreno_is_a2xx(adreno_dev)) {
-			unsigned int rptr;
-			adreno_readreg(adreno_dev, ADRENO_REG_CP_RB_RPTR,
-						&rptr);
-			if (rptr != adreno_dev->ringbuffer.wptr)
-				adreno_writereg(adreno_dev,
-					ADRENO_REG_CP_RB_WPTR,
-					adreno_dev->ringbuffer.wptr);
-		}
-
-		return 0;
-	}
-
-	/*
-	 * Time interval between hang detection should be KGSL_TIMEOUT_PART
-	 * or more, if next hang detection is requested < KGSL_TIMEOUT_PART
-	 * from the last time do nothing.
-	 */
-	if ((next_hang_detect_time) &&
-		(time_before(jiffies, next_hang_detect_time)))
-			return 0;
-	else
-		next_hang_detect_time = (jiffies +
-			msecs_to_jiffies(KGSL_TIMEOUT_PART-1));
-
-	/* Read the current Hang detect reg values here */
-	for (i = 0; i < FT_DETECT_REGS_COUNT; i++) {
-		if (ft_detect_regs[i] == 0)
-			continue;
-		kgsl_regread(device, ft_detect_regs[i],
-			&curr_reg_val[i]);
-	}
-
-	/* Read the current global timestamp here */
-	kgsl_sharedmem_readl(&device->memstore,
-			&curr_global_ts,
-			KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
-			eoptimestamp));
-	/* Make sure the memstore read has posted */
-	mb();
-
-	if (curr_global_ts == prev_global_ts) {
-
-		/* If we don't already have a good context, get it. */
-		if (kgsl_context_detached(context)) {
-			kgsl_context_put(context);
-			context = NULL;
-			curr_context = NULL;
-			strlcpy(pid_name, "unknown", sizeof(pid_name));
-
-			kgsl_sharedmem_readl(&device->memstore,
-				&curr_context_id,
-				KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
-				current_context));
-			/* Make sure the memstore read has posted */
-			mb();
-
-			context = kgsl_context_get(device, curr_context_id);
-			if (context != NULL) {
-				struct task_struct *task;
-				curr_context = ADRENO_CONTEXT(context);
-				curr_context->ib_gpu_time_used = 0;
-				task = find_task_by_vpid(context->pid);
-				if (task)
-					get_task_comm(pid_name, task);
-			} else {
-				KGSL_DRV_ERR(device,
-					"Fault tolerance no context found\n");
-			}
-		}
-		for (i = 0; i < FT_DETECT_REGS_COUNT; i++) {
-			if (curr_reg_val[i] != prev_reg_val[i]) {
-				fast_hang_detected = 0;
-
-				/* Check for long IB here */
-				if ((i >=
-					LONG_IB_DETECT_REG_INDEX_START)
-					&&
-					(i <=
-					LONG_IB_DETECT_REG_INDEX_END))
-					long_ib_detected = 0;
-			}
-		}
-
-		if (fast_hang_detected) {
-			KGSL_FT_ERR(device,
-				"Proc %s, ctxt_id %d ts %d triggered fault tolerance"
-				" on global ts %d\n",
-				pid_name, context ? context->id : 0,
-				(kgsl_readtimestamp(device, context,
-				KGSL_TIMESTAMP_RETIRED) + 1),
-				curr_global_ts + 1);
-			return 1;
-		}
-
-		if (curr_context != NULL) {
-
-			curr_context->ib_gpu_time_used += KGSL_TIMEOUT_PART;
-			KGSL_FT_INFO(device,
-			"Proc %s used GPU Time %d ms on timestamp 0x%X\n",
-			pid_name, curr_context->ib_gpu_time_used,
-			curr_global_ts+1);
-
-			if ((long_ib_detected) &&
-				(!(curr_context->flags &
-				 CTXT_FLAGS_NO_FAULT_TOLERANCE))) {
-				curr_context->ib_gpu_time_used +=
-					KGSL_TIMEOUT_PART;
-				if (curr_context->ib_gpu_time_used >
-					KGSL_TIMEOUT_LONG_IB_DETECTION) {
-					if (adreno_dev->long_ib_ts !=
-						curr_global_ts) {
-						KGSL_FT_ERR(device,
-						"Proc %s, ctxt_id %d ts %d"
-						"used GPU for %d ms long ib "
-						"detected on global ts %d\n",
-						pid_name, context->id,
-						(kgsl_readtimestamp(device,
-						context,
-						KGSL_TIMESTAMP_RETIRED)+1),
-						curr_context->ib_gpu_time_used,
-						curr_global_ts+1);
-						adreno_dev->long_ib = 1;
-						adreno_dev->long_ib_ts =
-								curr_global_ts;
-						curr_context->ib_gpu_time_used =
-								0;
-						return 1;
-					}
-				}
-			}
-		}
-	} else {
-		/* GPU is moving forward */
-		prev_global_ts = curr_global_ts;
-		kgsl_context_put(context);
-		context = NULL;
-		curr_context = NULL;
-		strlcpy(pid_name, "unknown", sizeof(pid_name));
-		adreno_dev->long_ib = 0;
-		adreno_dev->long_ib_ts = 0;
-	}
-
-
-	/* If hangs are not detected copy the current reg values
-	 * to previous values and return no hang */
-	for (i = 0; i < FT_DETECT_REGS_COUNT; i++)
-			prev_reg_val[i] = curr_reg_val[i];
-	return 0;
-}
-
-static int _check_pending_timestamp(struct kgsl_device *device,
-		struct kgsl_context *context, unsigned int timestamp)
-{
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	unsigned int context_id = _get_context_id(context);
-	unsigned int ts_issued;
-
-	if (context_id == KGSL_CONTEXT_INVALID)
-		return -EINVAL;
-
-	ts_issued = adreno_context_timestamp(context, &adreno_dev->ringbuffer);
-
-	if (timestamp_cmp(timestamp, ts_issued) <= 0)
-		return 0;
-
-	if (context && !context->wait_on_invalid_ts) {
-		KGSL_DRV_ERR(device, "Cannot wait for invalid ts <%d:0x%x>, last issued ts <%d:0x%x>\n",
-			context_id, timestamp, context_id, ts_issued);
-
-			/* Only print this message once */
-			context->wait_on_invalid_ts = true;
-	}
-
-	return -EINVAL;
-}
-
 /**
  * adreno_waittimestamp - sleep while waiting for the specified timestamp
  * @device - pointer to a KGSL device structure
@@ -3766,147 +2590,35 @@
  * @timestamp - GPU timestamp to wait for
  * @msecs - amount of time to wait (in milliseconds)
  *
- * Wait 'msecs' milliseconds for the specified timestamp to expire. Wake up
- * every KGSL_TIMEOUT_PART milliseconds to check for a device hang and process
- * one if it happened.  Otherwise, spend most of our time in an interruptible
- * wait for the timestamp interrupt to be processed.  This function must be
- * called with the mutex already held.
+ * Wait up to 'msecs' milliseconds for the specified timestamp to expire.
  */
 static int adreno_waittimestamp(struct kgsl_device *device,
-				struct kgsl_context *context,
-				unsigned int timestamp,
-				unsigned int msecs)
+		struct kgsl_context *context,
+		unsigned int timestamp,
+		unsigned int msecs)
 {
-	static unsigned int io_cnt;
-	struct adreno_context *adreno_ctx = context ? ADRENO_CONTEXT(context) :
-		NULL;
-	struct kgsl_pwrctrl *pwr = &device->pwrctrl;
-	unsigned int context_id = _get_context_id(context);
-	unsigned int time_elapsed = 0;
-	unsigned int wait;
-	int ts_compare = 1;
-	int io, ret = -ETIMEDOUT;
+	int ret;
+	struct adreno_context *drawctxt;
 
-	if (context_id == KGSL_CONTEXT_INVALID) {
-		KGSL_DRV_WARN(device, "context was detached");
+	if (context == NULL) {
+		/* If they are doing then complain once */
+		dev_WARN_ONCE(device->dev, 1,
+			"IOCTL_KGSL_DEVICE_WAITTIMESTAMP is deprecated\n");
 		return -EINVAL;
 	}
 
-	/*
-	 * Check to see if the requested timestamp is "newer" then the last
-	 * timestamp issued. If it is complain once and return error.  Only
-	 * print the message once per context so that badly behaving
-	 * applications don't spam the logs
-	 */
+	/* Return -EINVAL if the context has been detached */
+	if (kgsl_context_detached(context))
+		return -EINVAL;
 
-	if (adreno_ctx && !(adreno_ctx->flags & CTXT_FLAGS_USER_GENERATED_TS)) {
-		if (_check_pending_timestamp(device, context, timestamp))
-			return -EINVAL;
+	ret = adreno_drawctxt_wait(ADRENO_DEVICE(device), context,
+		timestamp, msecs_to_jiffies(msecs));
 
-		/* Reset the invalid timestamp flag on a valid wait */
-		context->wait_on_invalid_ts = false;
-	}
+	/* If the context got invalidated then return a specific error */
+	drawctxt = ADRENO_CONTEXT(context);
 
-	/*
-	 * On the first time through the loop only wait 100ms.
-	 * this gives enough time for the engine to start moving and oddly
-	 * provides better hang detection results than just going the full
-	 * KGSL_TIMEOUT_PART right off the bat. The exception to this rule
-	 * is if msecs happens to be < 100ms then just use 20ms or the msecs,
-	 * whichever is larger because anything less than 20 is unreliable
-	 */
-
-	if (msecs == 0 || msecs >= 100)
-		wait = 100;
-	else
-		wait = (msecs > 20) ? msecs : 20;
-
-	do {
-		long status;
-
-		/*
-		 * if the timestamp happens while we're not
-		 * waiting, there's a chance that an interrupt
-		 * will not be generated and thus the timestamp
-		 * work needs to be queued.
-		 */
-
-		if (kgsl_check_timestamp(device, context, timestamp)) {
-			queue_work(device->work_queue, &device->ts_expired_ws);
-			ret = 0;
-			break;
-		}
-
-		/*
-		 * For proper power accounting sometimes we need to call
-		 * io_wait_interruptible_timeout and sometimes we need to call
-		 * plain old wait_interruptible_timeout. We call the regular
-		 * timeout N times out of 100, where N is a number specified by
-		 * the current power level
-		 */
-
-		io_cnt = (io_cnt + 1) % 100;
-		io = (io_cnt < pwr->pwrlevels[pwr->active_pwrlevel].io_fraction)
-			? 0 : 1;
-
-		mutex_unlock(&device->mutex);
-
-		/* Wait for a timestamp event */
-		status = kgsl_wait_event_interruptible_timeout(
-			device->wait_queue,
-			adreno_check_interrupt_timestamp(device, context,
-				timestamp), msecs_to_jiffies(wait), io);
-
-		mutex_lock(&device->mutex);
-
-		/*
-		 * If status is non zero then either the condition was satisfied
-		 * or there was an error.  In either event, this is the end of
-		 * the line for us
-		 */
-
-		if (status != 0) {
-			ret = (status > 0) ? 0 : (int) status;
-			break;
-		}
-		time_elapsed += wait;
-
-		/* If user specified timestamps are being used, wait at least
-		 * KGSL_SYNCOBJ_SERVER_TIMEOUT msecs for the user driver to
-		 * issue a IB for a timestamp before checking to see if the
-		 * current timestamp we are waiting for is valid or not
-		 */
-
-		if (ts_compare && (adreno_ctx &&
-			(adreno_ctx->flags & CTXT_FLAGS_USER_GENERATED_TS))) {
-			if (time_elapsed > KGSL_SYNCOBJ_SERVER_TIMEOUT) {
-				ret = _check_pending_timestamp(device, context,
-					timestamp);
-				if (ret)
-					break;
-
-				/* Don't do this check again */
-				ts_compare = 0;
-
-				/*
-				 * Reset the invalid timestamp flag on a valid
-				 * wait
-				 */
-				context->wait_on_invalid_ts = false;
-			}
-		}
-
-		/*
-		 * We want to wait the floor of KGSL_TIMEOUT_PART
-		 * and (msecs - time_elapsed).
-		 */
-
-		if (KGSL_TIMEOUT_PART < (msecs - time_elapsed))
-			wait = KGSL_TIMEOUT_PART;
-		else
-			wait = (msecs - time_elapsed);
-
-	} while (!msecs || time_elapsed < msecs);
+	if (drawctxt->state == ADRENO_CONTEXT_STATE_INVALID)
+		ret = -EDEADLK;
 
 	return ret;
 }
@@ -3915,13 +2627,13 @@
 		struct kgsl_context *context, enum kgsl_timestamp_type type)
 {
 	unsigned int timestamp = 0;
-	unsigned int context_id = _get_context_id(context);
+	unsigned int id = context ? context->id : KGSL_MEMSTORE_GLOBAL;
 
 	/*
-	 * If the context ID is invalid, we are in a race with
+	 * If the context is detached we are in a race with
 	 * the context being destroyed by userspace so bail.
 	 */
-	if (context_id == KGSL_CONTEXT_INVALID) {
+	if (context && kgsl_context_detached(context)) {
 		KGSL_DRV_WARN(device, "context was detached");
 		return timestamp;
 	}
@@ -3935,11 +2647,11 @@
 	}
 	case KGSL_TIMESTAMP_CONSUMED:
 		kgsl_sharedmem_readl(&device->memstore, &timestamp,
-			KGSL_MEMSTORE_OFFSET(context_id, soptimestamp));
+			KGSL_MEMSTORE_OFFSET(id, soptimestamp));
 		break;
 	case KGSL_TIMESTAMP_RETIRED:
 		kgsl_sharedmem_readl(&device->memstore, &timestamp,
-			KGSL_MEMSTORE_OFFSET(context_id, eoptimestamp));
+			KGSL_MEMSTORE_OFFSET(id, eoptimestamp));
 		break;
 	}
 
@@ -4099,6 +2811,7 @@
 	.gpuid = adreno_gpuid,
 	.snapshot = adreno_snapshot,
 	.irq_handler = adreno_irq_handler,
+	.drain = adreno_drain,
 	/* Optional functions */
 	.setstate = adreno_setstate,
 	.drawctxt_create = adreno_drawctxt_create,
@@ -4106,7 +2819,6 @@
 	.drawctxt_destroy = adreno_drawctxt_destroy,
 	.setproperty = adreno_setproperty,
 	.postmortem_dump = adreno_dump,
-	.next_event = adreno_next_event,
 };
 
 static struct platform_driver adreno_platform_driver = {
diff --git a/drivers/gpu/msm/adreno.h b/drivers/gpu/msm/adreno.h
index 72f15e7..32e43b2 100644
--- a/drivers/gpu/msm/adreno.h
+++ b/drivers/gpu/msm/adreno.h
@@ -96,6 +96,46 @@
 	TRACE_BUS_CTL,
 };
 
+/*
+ * Maximum size of the dispatcher ringbuffer - the actual inflight size will be
+ * smaller then this but this size will allow for a larger range of inflight
+ * sizes that can be chosen at runtime
+ */
+
+#define ADRENO_DISPATCH_CMDQUEUE_SIZE 128
+
+/**
+ * struct adreno_dispatcher - container for the adreno GPU dispatcher
+ * @mutex: Mutex to protect the structure
+ * @state: Current state of the dispatcher (active or paused)
+ * @timer: Timer to monitor the progress of the command batches
+ * @inflight: Number of command batch operations pending in the ringbuffer
+ * @fault: True if a HW fault was detected
+ * @pending: Priority list of contexts waiting to submit command batches
+ * @plist_lock: Spin lock to protect the pending queue
+ * @cmdqueue: Queue of command batches currently flight
+ * @head: pointer to the head of of the cmdqueue.  This is the oldest pending
+ * operation
+ * @tail: pointer to the tail of the cmdqueue.  This is the most recently
+ * submitted operation
+ * @work: work_struct to put the dispatcher in a work queue
+ * @kobj: kobject for the dispatcher directory in the device sysfs node
+ */
+struct adreno_dispatcher {
+	struct mutex mutex;
+	unsigned int state;
+	struct timer_list timer;
+	unsigned int inflight;
+	int fault;
+	struct plist_head pending;
+	spinlock_t plist_lock;
+	struct kgsl_cmdbatch *cmdqueue[ADRENO_DISPATCH_CMDQUEUE_SIZE];
+	unsigned int head;
+	unsigned int tail;
+	struct work_struct work;
+	struct kobject kobj;
+};
+
 struct adreno_gpudev;
 
 struct adreno_device {
@@ -136,6 +176,7 @@
 	unsigned int ocmem_base;
 	unsigned int gpu_cycles;
 	struct adreno_profile profile;
+	struct adreno_dispatcher dispatcher;
 };
 
 #define PERFCOUNTER_FLAG_NONE 0x0
@@ -266,9 +307,9 @@
 
 	/* GPU specific function hooks */
 	int (*ctxt_create)(struct adreno_device *, struct adreno_context *);
-	void (*ctxt_save)(struct adreno_device *, struct adreno_context *);
-	void (*ctxt_restore)(struct adreno_device *, struct adreno_context *);
-	void (*ctxt_draw_workaround)(struct adreno_device *,
+	int (*ctxt_save)(struct adreno_device *, struct adreno_context *);
+	int (*ctxt_restore)(struct adreno_device *, struct adreno_context *);
+	int (*ctxt_draw_workaround)(struct adreno_device *,
 					struct adreno_context *);
 	irqreturn_t (*irq_handler)(struct adreno_device *);
 	void (*irq_control)(struct adreno_device *, int);
@@ -291,46 +332,6 @@
 	void (*postmortem_dump)(struct adreno_device *adreno_dev);
 };
 
-/*
- * struct adreno_ft_data - Structure that contains all information to
- * perform gpu fault tolerance
- * @ib1 - IB1 that the GPU was executing when hang happened
- * @context_id - Context which caused the hang
- * @global_eop - eoptimestamp at time of hang
- * @rb_buffer - Buffer that holds the commands from good contexts
- * @rb_size - Number of valid dwords in rb_buffer
- * @bad_rb_buffer - Buffer that holds commands from the hanging context
- * bad_rb_size - Number of valid dwords in bad_rb_buffer
- * @good_rb_buffer - Buffer that holds commands from good contexts
- * good_rb_size - Number of valid dwords in good_rb_buffer
- * @last_valid_ctx_id - The last context from which commands were placed in
- * ringbuffer before the GPU hung
- * @step - Current fault tolerance step being executed
- * @err_code - Fault tolerance error code
- * @fault - Indicates whether the hang was caused due to a pagefault
- * @start_of_replay_cmds - Offset in ringbuffer from where commands can be
- * replayed during fault tolerance
- * @replay_for_snapshot - Offset in ringbuffer where IB's can be saved for
- * replaying with snapshot
- */
-struct adreno_ft_data {
-	unsigned int ib1;
-	unsigned int context_id;
-	unsigned int global_eop;
-	unsigned int *rb_buffer;
-	unsigned int rb_size;
-	unsigned int *bad_rb_buffer;
-	unsigned int bad_rb_size;
-	unsigned int *good_rb_buffer;
-	unsigned int good_rb_size;
-	unsigned int last_valid_ctx_id;
-	unsigned int status;
-	unsigned int ft_policy;
-	unsigned int err_code;
-	unsigned int start_of_replay_cmds;
-	unsigned int replay_for_snapshot;
-};
-
 #define FT_DETECT_REGS_COUNT 12
 
 struct log_field {
@@ -410,13 +411,21 @@
 void *adreno_snapshot(struct kgsl_device *device, void *snapshot, int *remain,
 		int hang);
 
-int adreno_dump_and_exec_ft(struct kgsl_device *device);
+void adreno_dispatcher_start(struct adreno_device *adreno_dev);
+int adreno_dispatcher_init(struct adreno_device *adreno_dev);
+void adreno_dispatcher_close(struct adreno_device *adreno_dev);
+int adreno_dispatcher_idle(struct adreno_device *adreno_dev,
+		unsigned int timeout);
+void adreno_dispatcher_irq_fault(struct kgsl_device *device);
+void adreno_dispatcher_stop(struct adreno_device *adreno_dev);
 
-void adreno_dump_rb(struct kgsl_device *device, const void *buf,
-			 size_t len, int start, int size);
+int adreno_dispatcher_queue_cmd(struct adreno_device *adreno_dev,
+		struct adreno_context *drawctxt, struct kgsl_cmdbatch *cmdbatch,
+		uint32_t *timestamp);
 
-unsigned int adreno_ft_detect(struct kgsl_device *device,
-						unsigned int *prev_reg_val);
+void adreno_dispatcher_schedule(struct kgsl_device *device);
+void adreno_dispatcher_pause(struct adreno_device *adreno_dev);
+int adreno_reset(struct kgsl_device *device);
 
 int adreno_ft_init_sysfs(struct kgsl_device *device);
 void adreno_ft_uninit_sysfs(struct kgsl_device *device);
@@ -533,9 +542,7 @@
 {
 	if (k_ctxt) {
 		struct adreno_context *a_ctxt = ADRENO_CONTEXT(k_ctxt);
-
-		if (a_ctxt->flags & CTXT_FLAGS_PER_CONTEXT_TS)
-			return a_ctxt->timestamp;
+		return a_ctxt->timestamp;
 	}
 	return rb->global_ts;
 }
diff --git a/drivers/gpu/msm/adreno_a2xx.c b/drivers/gpu/msm/adreno_a2xx.c
index 3d72c5c..cce4f91 100644
--- a/drivers/gpu/msm/adreno_a2xx.c
+++ b/drivers/gpu/msm/adreno_a2xx.c
@@ -1451,7 +1451,7 @@
 	return ret;
 }
 
-static void a2xx_drawctxt_draw_workaround(struct adreno_device *adreno_dev,
+static int a2xx_drawctxt_draw_workaround(struct adreno_device *adreno_dev,
 					struct adreno_context *context)
 {
 	struct kgsl_device *device = &adreno_dev->dev;
@@ -1468,7 +1468,7 @@
 				ADRENO_NUM_CTX_SWITCH_ALLOWED_BEFORE_DRAW)
 			adreno_dev->gpudev->ctx_switches_since_last_draw = 0;
 		else
-			return;
+			return 0;
 		/*
 		 * Issue an empty draw call to avoid possible hangs due to
 		 * repeated idles without intervening draw calls.
@@ -1499,41 +1499,46 @@
 					| adreno_dev->pix_shader_start;
 	}
 
-	adreno_ringbuffer_issuecmds(device, context, KGSL_CMD_FLAGS_PMODE,
-			&cmd[0], cmds - cmd);
+	return adreno_ringbuffer_issuecmds(device, context,
+			KGSL_CMD_FLAGS_PMODE, &cmd[0], cmds - cmd);
 }
 
-static void a2xx_drawctxt_save(struct adreno_device *adreno_dev,
+static int a2xx_drawctxt_save(struct adreno_device *adreno_dev,
 			struct adreno_context *context)
 {
 	struct kgsl_device *device = &adreno_dev->dev;
+	int ret;
 
 	if (context == NULL || (context->flags & CTXT_FLAGS_BEING_DESTROYED))
-		return;
+		return 0;
 
-	if (context->flags & CTXT_FLAGS_GPU_HANG)
-		KGSL_CTXT_WARN(device,
-			"Current active context has caused gpu hang\n");
+	if (context->state == ADRENO_CONTEXT_STATE_INVALID)
+		return 0;
 
 	if (!(context->flags & CTXT_FLAGS_PREAMBLE)) {
 		kgsl_cffdump_syncmem(context->base.device, &context->gpustate,
 			context->reg_save[1],
 			context->reg_save[2] << 2, true);
 		/* save registers and constants. */
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE,
 			context->reg_save, 3);
 
+		if (ret)
+			return ret;
+
 		if (context->flags & CTXT_FLAGS_SHADER_SAVE) {
 			kgsl_cffdump_syncmem(context->base.device,
 				&context->gpustate,
 				context->shader_save[1],
 				context->shader_save[2] << 2, true);
 			/* save shader partitioning and instructions. */
-			adreno_ringbuffer_issuecmds(device, context,
+			ret = adreno_ringbuffer_issuecmds(device, context,
 				KGSL_CMD_FLAGS_PMODE,
 				context->shader_save, 3);
 
+			if (ret)
+				return ret;
 			kgsl_cffdump_syncmem(context->base.device,
 				&context->gpustate,
 				context->shader_fixup[1],
@@ -1542,10 +1547,13 @@
 			 * fixup shader partitioning parameter for
 			 *  SET_SHADER_BASES.
 			 */
-			adreno_ringbuffer_issuecmds(device, context,
+			ret = adreno_ringbuffer_issuecmds(device, context,
 				KGSL_CMD_FLAGS_NONE,
 				context->shader_fixup, 3);
 
+			if (ret)
+				return ret;
+
 			context->flags |= CTXT_FLAGS_SHADER_RESTORE;
 		}
 	}
@@ -1558,32 +1566,41 @@
 		/* save gmem.
 		 * (note: changes shader. shader must already be saved.)
 		 */
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_PMODE,
 			context->context_gmem_shadow.gmem_save, 3);
 
+		if (ret)
+			return ret;
+
 		kgsl_cffdump_syncmem(context->base.device, &context->gpustate,
 			context->chicken_restore[1],
 			context->chicken_restore[2] << 2, true);
 
 		/* Restore TP0_CHICKEN */
 		if (!(context->flags & CTXT_FLAGS_PREAMBLE)) {
-			adreno_ringbuffer_issuecmds(device, context,
+			ret = adreno_ringbuffer_issuecmds(device, context,
 				KGSL_CMD_FLAGS_NONE,
 				context->chicken_restore, 3);
+
+			if (ret)
+				return ret;
 		}
 		adreno_dev->gpudev->ctx_switches_since_last_draw = 0;
 
 		context->flags |= CTXT_FLAGS_GMEM_RESTORE;
 	} else if (adreno_is_a2xx(adreno_dev))
-		a2xx_drawctxt_draw_workaround(adreno_dev, context);
+		return a2xx_drawctxt_draw_workaround(adreno_dev, context);
+
+	return 0;
 }
 
-static void a2xx_drawctxt_restore(struct adreno_device *adreno_dev,
+static int a2xx_drawctxt_restore(struct adreno_device *adreno_dev,
 			struct adreno_context *context)
 {
 	struct kgsl_device *device = &adreno_dev->dev;
 	unsigned int cmds[5];
+	int ret = 0;
 
 	if (context == NULL) {
 		/* No context - set the default pagetable and thats it */
@@ -1598,7 +1615,7 @@
 			: KGSL_CONTEXT_INVALID;
 		kgsl_mmu_setstate(&device->mmu, device->mmu.defaultpagetable,
 				  id);
-		return;
+		return 0;
 	}
 
 	cmds[0] = cp_nop_packet(1);
@@ -1607,8 +1624,11 @@
 	cmds[3] = device->memstore.gpuaddr +
 		KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL, current_context);
 	cmds[4] = context->base.id;
-	adreno_ringbuffer_issuecmds(device, context, KGSL_CMD_FLAGS_NONE,
+	ret = adreno_ringbuffer_issuecmds(device, context, KGSL_CMD_FLAGS_NONE,
 					cmds, 5);
+	if (ret)
+		return ret;
+
 	kgsl_mmu_setstate(&device->mmu, context->base.pagetable,
 			context->base.id);
 
@@ -1621,9 +1641,11 @@
 			context->context_gmem_shadow.gmem_restore[2] << 2,
 			true);
 
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_PMODE,
 			context->context_gmem_shadow.gmem_restore, 3);
+		if (ret)
+			return ret;
 
 		if (!(context->flags & CTXT_FLAGS_PREAMBLE)) {
 			kgsl_cffdump_syncmem(context->base.device,
@@ -1632,9 +1654,11 @@
 				context->chicken_restore[2] << 2, true);
 
 			/* Restore TP0_CHICKEN */
-			adreno_ringbuffer_issuecmds(device, context,
+			ret = adreno_ringbuffer_issuecmds(device, context,
 				KGSL_CMD_FLAGS_NONE,
 				context->chicken_restore, 3);
+			if (ret)
+				return ret;
 		}
 
 		context->flags &= ~CTXT_FLAGS_GMEM_RESTORE;
@@ -1646,8 +1670,10 @@
 			context->reg_restore[2] << 2, true);
 
 		/* restore registers and constants. */
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE, context->reg_restore, 3);
+		if (ret)
+			return ret;
 
 		/* restore shader instructions & partitioning. */
 		if (context->flags & CTXT_FLAGS_SHADER_RESTORE) {
@@ -1656,18 +1682,22 @@
 				context->shader_restore[1],
 				context->shader_restore[2] << 2, true);
 
-			adreno_ringbuffer_issuecmds(device, context,
+			ret = adreno_ringbuffer_issuecmds(device, context,
 				KGSL_CMD_FLAGS_NONE,
 				context->shader_restore, 3);
+			if (ret)
+				return ret;
 		}
 	}
 
 	if (adreno_is_a20x(adreno_dev)) {
 		cmds[0] = cp_type3_packet(CP_SET_BIN_BASE_OFFSET, 1);
 		cmds[1] = context->bin_base_offset;
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE, cmds, 2);
 	}
+
+	return ret;
 }
 
 /*
@@ -1734,13 +1764,14 @@
 
 	if (!status) {
 		if (master_status & MASTER_INT_SIGNAL__CP_INT_STAT) {
-			/* This indicates that we could not read CP_INT_STAT.
-			 * As a precaution just wake up processes so
-			 * they can check their timestamps. Since, we
-			 * did not ack any interrupts this interrupt will
-			 * be generated again */
+			/*
+			 * This indicates that we could not read CP_INT_STAT.
+			 * As a precaution schedule the dispatcher to check
+			 * things out. Since we did not ack any interrupts this
+			 * interrupt will be generated again
+			 */
 			KGSL_DRV_WARN(device, "Unable to read CP_INT_STATUS\n");
-			wake_up_interruptible_all(&device->wait_queue);
+			adreno_dispatcher_schedule(device);
 		} else
 			KGSL_DRV_WARN(device, "Spurious interrput detected\n");
 		return;
@@ -1766,7 +1797,7 @@
 
 	if (status & (CP_INT_CNTL__IB1_INT_MASK | CP_INT_CNTL__RB_INT_MASK)) {
 		queue_work(device->work_queue, &device->ts_expired_ws);
-		wake_up_interruptible_all(&device->wait_queue);
+		adreno_dispatcher_schedule(device);
 	}
 }
 
diff --git a/drivers/gpu/msm/adreno_a3xx.c b/drivers/gpu/msm/adreno_a3xx.c
index d96965c..8b75c4e 100644
--- a/drivers/gpu/msm/adreno_a3xx.c
+++ b/drivers/gpu/msm/adreno_a3xx.c
@@ -2382,32 +2382,38 @@
 	return ret;
 }
 
-static void a3xx_drawctxt_save(struct adreno_device *adreno_dev,
+static int a3xx_drawctxt_save(struct adreno_device *adreno_dev,
 			   struct adreno_context *context)
 {
 	struct kgsl_device *device = &adreno_dev->dev;
+	int ret;
 
 	if (context == NULL || (context->flags & CTXT_FLAGS_BEING_DESTROYED))
-		return;
+		return 0;
 
-	if (context->flags & CTXT_FLAGS_GPU_HANG)
-		KGSL_CTXT_WARN(device,
-			       "Current active context has caused gpu hang\n");
+	if (context->state == ADRENO_CONTEXT_STATE_INVALID)
+		return 0;
 
 	if (!(context->flags & CTXT_FLAGS_PREAMBLE)) {
 		/* Fixup self modifying IBs for save operations */
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE, context->save_fixup, 3);
+		if (ret)
+			return ret;
 
 		/* save registers and constants. */
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE,
 			context->regconstant_save, 3);
+		if (ret)
+			return ret;
 
 		if (context->flags & CTXT_FLAGS_SHADER_SAVE) {
 			/* Save shader instructions */
-			adreno_ringbuffer_issuecmds(device, context,
+			ret = adreno_ringbuffer_issuecmds(device, context,
 				KGSL_CMD_FLAGS_PMODE, context->shader_save, 3);
+			if (ret)
+				return ret;
 
 			context->flags |= CTXT_FLAGS_SHADER_RESTORE;
 		}
@@ -2425,19 +2431,25 @@
 			context->context_gmem_shadow.gmem_save[1],
 			context->context_gmem_shadow.gmem_save[2] << 2, true);
 
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 					KGSL_CMD_FLAGS_PMODE,
 					    context->context_gmem_shadow.
 					    gmem_save, 3);
+		if (ret)
+			return ret;
+
 		context->flags |= CTXT_FLAGS_GMEM_RESTORE;
 	}
+
+	return 0;
 }
 
-static void a3xx_drawctxt_restore(struct adreno_device *adreno_dev,
+static int a3xx_drawctxt_restore(struct adreno_device *adreno_dev,
 			      struct adreno_context *context)
 {
 	struct kgsl_device *device = &adreno_dev->dev;
 	unsigned int cmds[5];
+	int ret = 0;
 
 	if (context == NULL) {
 		/* No context - set the default pagetable and thats it */
@@ -2452,7 +2464,7 @@
 			: KGSL_CONTEXT_INVALID;
 		kgsl_mmu_setstate(&device->mmu, device->mmu.defaultpagetable,
 				  id);
-		return;
+		return 0;
 	}
 
 	cmds[0] = cp_nop_packet(1);
@@ -2461,8 +2473,11 @@
 	cmds[3] = device->memstore.gpuaddr +
 		KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL, current_context);
 	cmds[4] = context->base.id;
-	adreno_ringbuffer_issuecmds(device, context, KGSL_CMD_FLAGS_NONE,
+	ret = adreno_ringbuffer_issuecmds(device, context, KGSL_CMD_FLAGS_NONE,
 					cmds, 5);
+	if (ret)
+		return ret;
+
 	kgsl_mmu_setstate(&device->mmu, context->base.pagetable,
 			context->base.id);
 
@@ -2478,36 +2493,47 @@
 			context->context_gmem_shadow.gmem_restore[2] << 2,
 			true);
 
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 					KGSL_CMD_FLAGS_PMODE,
 					    context->context_gmem_shadow.
 					    gmem_restore, 3);
+		if (ret)
+			return ret;
 		context->flags &= ~CTXT_FLAGS_GMEM_RESTORE;
 	}
 
 	if (!(context->flags & CTXT_FLAGS_PREAMBLE)) {
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE, context->reg_restore, 3);
+		if (ret)
+			return ret;
 
 		/* Fixup self modifying IBs for restore operations */
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE,
 			context->restore_fixup, 3);
+		if (ret)
+			return ret;
 
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE,
 			context->constant_restore, 3);
+		if (ret)
+			return ret;
 
 		if (context->flags & CTXT_FLAGS_SHADER_RESTORE)
-			adreno_ringbuffer_issuecmds(device, context,
+			ret = adreno_ringbuffer_issuecmds(device, context,
 				KGSL_CMD_FLAGS_NONE,
 				context->shader_restore, 3);
-
+			if (ret)
+				return ret;
 		/* Restore HLSQ_CONTROL_0 register */
-		adreno_ringbuffer_issuecmds(device, context,
+		ret = adreno_ringbuffer_issuecmds(device, context,
 			KGSL_CMD_FLAGS_NONE,
 			context->hlsqcontrol_restore, 3);
 	}
+
+	return ret;
 }
 
 static int a3xx_rb_init(struct adreno_device *adreno_dev,
@@ -2621,11 +2647,8 @@
 {
 	struct kgsl_device *device = &adreno_dev->dev;
 
-	/* Wake up everybody waiting for the interrupt */
-	wake_up_interruptible_all(&device->wait_queue);
-
-	/* Schedule work to free mem and issue ibs */
 	queue_work(device->work_queue, &device->ts_expired_ws);
+	adreno_dispatcher_schedule(device);
 }
 
 /**
diff --git a/drivers/gpu/msm/adreno_dispatch.c b/drivers/gpu/msm/adreno_dispatch.c
new file mode 100644
index 0000000..e429934
--- /dev/null
+++ b/drivers/gpu/msm/adreno_dispatch.c
@@ -0,0 +1,1038 @@
+/* Copyright (c) 2013, 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 <linux/wait.h>
+#include <linux/delay.h>
+#include <linux/sched.h>
+#include <linux/jiffies.h>
+
+#include "kgsl.h"
+#include "adreno.h"
+#include "adreno_ringbuffer.h"
+#include "adreno_trace.h"
+
+#define ADRENO_DISPATCHER_ACTIVE 0
+#define ADRENO_DISPATCHER_PAUSE 1
+
+#define CMDQUEUE_NEXT(_i, _s) (((_i) + 1) % (_s))
+
+/* Number of commands that can be queued in a context before it sleeps */
+static unsigned int _context_cmdqueue_size = 50;
+
+/* Number of milliseconds to wait for the context queue to clear */
+static unsigned int _context_queue_wait = 10000;
+
+/* Number of command batches sent at a time from a single context */
+static unsigned int _context_cmdbatch_burst = 5;
+
+/* Number of command batches inflight in the ringbuffer at any time */
+static unsigned int _dispatcher_inflight = 15;
+
+/* Command batch timeout (in milliseconds) */
+static unsigned int _cmdbatch_timeout = 2000;
+
+/**
+ * adreno_dispatcher_get_cmdbatch() - Get a new command from a context queue
+ * @drawctxt: Pointer to the adreno draw context
+ *
+ * Dequeue a new command batch from the context list
+ */
+static inline struct kgsl_cmdbatch *adreno_dispatcher_get_cmdbatch(
+		struct adreno_context *drawctxt)
+{
+	struct kgsl_cmdbatch *cmdbatch = NULL;
+
+	mutex_lock(&drawctxt->mutex);
+	if (drawctxt->cmdqueue_head != drawctxt->cmdqueue_tail) {
+		cmdbatch = drawctxt->cmdqueue[drawctxt->cmdqueue_head];
+		drawctxt->cmdqueue_head =
+			CMDQUEUE_NEXT(drawctxt->cmdqueue_head,
+			ADRENO_CONTEXT_CMDQUEUE_SIZE);
+		drawctxt->queued--;
+	}
+
+	mutex_unlock(&drawctxt->mutex);
+
+	return cmdbatch;
+}
+
+/**
+ * adreno_dispatcher_requeue_cmdbatch() - Put a command back on the context
+ * queue
+ * @drawctxt: Pointer to the adreno draw context
+ * @cmdbatch: Pointer to the KGSL cmdbatch to requeue
+ *
+ * Failure to submit a command to the ringbuffer isn't the fault of the command
+ * being submitted so if a failure happens, push it back on the head of the the
+ * context queue to be reconsidered again
+ */
+static inline void adreno_dispatcher_requeue_cmdbatch(
+		struct adreno_context *drawctxt, struct kgsl_cmdbatch *cmdbatch)
+{
+	unsigned int prev;
+	mutex_lock(&drawctxt->mutex);
+
+	if (kgsl_context_detached(&drawctxt->base) ||
+		drawctxt->state == ADRENO_CONTEXT_STATE_INVALID) {
+		mutex_unlock(&drawctxt->mutex);
+		return;
+	}
+
+	prev = drawctxt->cmdqueue_head - 1;
+
+	if (prev < 0)
+		prev = ADRENO_CONTEXT_CMDQUEUE_SIZE - 1;
+
+	/*
+	 * The maximum queue size always needs to be one less then the size of
+	 * the ringbuffer queue so there is "room" to put the cmdbatch back in
+	 */
+
+	BUG_ON(prev == drawctxt->cmdqueue_tail);
+
+	drawctxt->cmdqueue[prev] = cmdbatch;
+	drawctxt->queued++;
+
+	/* Reset the command queue head to reflect the newly requeued change */
+	drawctxt->cmdqueue_head = prev;
+	mutex_unlock(&drawctxt->mutex);
+}
+
+/**
+ * dispatcher_queue_context() - Queue a context in the dispatcher pending list
+ * @dispatcher: Pointer to the adreno dispatcher struct
+ * @drawctxt: Pointer to the adreno draw context
+ *
+ * Add a context to the dispatcher pending list.
+ */
+static void  dispatcher_queue_context(struct adreno_device *adreno_dev,
+		struct adreno_context *drawctxt)
+{
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+
+	spin_lock(&dispatcher->plist_lock);
+
+	if (plist_node_empty(&drawctxt->pending)) {
+		/* Get a reference to the context while it sits on the list */
+		_kgsl_context_get(&drawctxt->base);
+		trace_dispatch_queue_context(drawctxt);
+		plist_add(&drawctxt->pending, &dispatcher->pending);
+	}
+
+	spin_unlock(&dispatcher->plist_lock);
+}
+
+/**
+ * sendcmd() - Send a command batch to the GPU hardware
+ * @dispatcher: Pointer to the adreno dispatcher struct
+ * @cmdbatch: Pointer to the KGSL cmdbatch being sent
+ *
+ * Send a KGSL command batch to the GPU hardware
+ */
+static int sendcmd(struct adreno_device *adreno_dev,
+	struct kgsl_cmdbatch *cmdbatch)
+{
+	struct kgsl_device *device = &adreno_dev->dev;
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+	int ret;
+
+	dispatcher->inflight++;
+
+	mutex_lock(&device->mutex);
+
+	if (dispatcher->inflight == 1) {
+		/* Time to make the donuts.  Turn on the GPU */
+		ret = kgsl_active_count_get(device);
+		if (ret) {
+			dispatcher->inflight--;
+			mutex_unlock(&device->mutex);
+			return ret;
+		}
+	}
+
+	ret = adreno_ringbuffer_submitcmd(adreno_dev, cmdbatch);
+
+	/* Turn the GPU back off on failure.  Sad face. */
+	if (ret && dispatcher->inflight == 1)
+		kgsl_active_count_put(device);
+
+	mutex_unlock(&device->mutex);
+
+	if (ret) {
+		dispatcher->inflight--;
+		KGSL_DRV_ERR(device,
+			"Unable to submit command to the ringbuffer\n");
+		return ret;
+	}
+
+	trace_adreno_cmdbatch_submitted(cmdbatch, dispatcher->inflight);
+
+	dispatcher->cmdqueue[dispatcher->tail] = cmdbatch;
+	dispatcher->tail = (dispatcher->tail + 1) %
+		ADRENO_DISPATCH_CMDQUEUE_SIZE;
+
+	/*
+	 * If this is the first command in the pipe then the GPU will
+	 * immediately start executing it so we can start the expiry timeout on
+	 * the command batch here.  Subsequent command batches will have their
+	 * timer started when the previous command batch is retired
+	 */
+	if (dispatcher->inflight == 1) {
+		cmdbatch->expires = jiffies +
+			msecs_to_jiffies(_cmdbatch_timeout);
+		mod_timer(&dispatcher->timer, cmdbatch->expires);
+	}
+
+	return 0;
+}
+
+/**
+ * dispatcher_context_sendcmds() - Send commands from a context to the GPU
+ * @adreno_dev: Pointer to the adreno device struct
+ * @drawctxt: Pointer to the adreno context to dispatch commands from
+ *
+ * Dequeue and send a burst of commands from the specified context to the GPU
+ */
+static int dispatcher_context_sendcmds(struct adreno_device *adreno_dev,
+		struct adreno_context *drawctxt)
+{
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+	int count = 0;
+
+	/*
+	 * Each context can send a specific number of command batches per cycle
+	 */
+	for ( ; count < _context_cmdbatch_burst &&
+		dispatcher->inflight < _dispatcher_inflight; count++) {
+		int ret;
+		struct kgsl_cmdbatch *cmdbatch =
+			adreno_dispatcher_get_cmdbatch(drawctxt);
+
+		if (cmdbatch == NULL)
+			break;
+
+		ret = sendcmd(adreno_dev, cmdbatch);
+
+		/*
+		 * There are various reasons why we can't submit a command (no
+		 * memory for the commands, full ringbuffer, etc) but none of
+		 * these are actually the current command's fault.  Requeue it
+		 * back on the context and let it come back around again if
+		 * conditions improve
+		 */
+		if (ret) {
+			adreno_dispatcher_requeue_cmdbatch(drawctxt, cmdbatch);
+			break;
+		}
+	}
+
+	/*
+	 * If the context successfully submitted commands, then
+	 * unconditionally put it back on the queue to be considered the
+	 * next time around. This might seem a little wasteful but it is
+	 * reasonable to think that a busy context will stay busy.
+	 */
+
+	if (count) {
+		dispatcher_queue_context(adreno_dev, drawctxt);
+
+		/*
+		 * If we submitted something there will be room in the
+		 * context queue so ping the context wait queue on the
+		 * chance that the context is snoozing
+		 */
+
+		wake_up_interruptible_all(&drawctxt->wq);
+	}
+
+	return count;
+}
+
+/**
+ * _adreno_dispatcher_issuecmds() - Issue commmands from pending contexts
+ * @adreno_dev: Pointer to the adreno device struct
+ *
+ * Issue as many commands as possible (up to inflight) from the pending contexts
+ * This function assumes the dispatcher mutex has been locked.
+ */
+static int _adreno_dispatcher_issuecmds(struct adreno_device *adreno_dev)
+{
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+
+	/* Don't do anything if the dispatcher is paused */
+	if (dispatcher->state != ADRENO_DISPATCHER_ACTIVE)
+		return 0;
+
+	while (dispatcher->inflight < _dispatcher_inflight) {
+		struct adreno_context *drawctxt = NULL;
+
+		spin_lock(&dispatcher->plist_lock);
+
+		if (!plist_head_empty(&dispatcher->pending)) {
+			drawctxt = plist_first_entry(&dispatcher->pending,
+				struct adreno_context, pending);
+
+			plist_del(&drawctxt->pending, &dispatcher->pending);
+		}
+
+		spin_unlock(&dispatcher->plist_lock);
+
+		if (drawctxt == NULL)
+			break;
+
+		if (kgsl_context_detached(&drawctxt->base) ||
+			drawctxt->state == ADRENO_CONTEXT_STATE_INVALID) {
+			kgsl_context_put(&drawctxt->base);
+			continue;
+		}
+
+		dispatcher_context_sendcmds(adreno_dev, drawctxt);
+		kgsl_context_put(&drawctxt->base);
+	}
+
+	return 0;
+}
+
+/**
+ * adreno_dispatcher_issuecmds() - Issue commmands from pending contexts
+ * @adreno_dev: Pointer to the adreno device struct
+ *
+ * Lock the dispatcher and call _adreno_dispatcher_issueibcmds
+ */
+int adreno_dispatcher_issuecmds(struct adreno_device *adreno_dev)
+{
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+	int ret;
+
+	mutex_lock(&dispatcher->mutex);
+	ret = _adreno_dispatcher_issuecmds(adreno_dev);
+	mutex_unlock(&dispatcher->mutex);
+
+	return ret;
+}
+
+static int _check_context_queue(struct adreno_context *drawctxt)
+{
+	int ret;
+
+	mutex_lock(&drawctxt->mutex);
+
+	/*
+	 * Wake up if there is room in the context or if the whole thing got
+	 * invalidated while we were asleep
+	 */
+
+	if (drawctxt->state == ADRENO_CONTEXT_STATE_INVALID)
+		ret = 1;
+	else
+		ret = drawctxt->queued < _context_cmdqueue_size ? 1 : 0;
+
+	mutex_unlock(&drawctxt->mutex);
+
+	return ret;
+}
+
+/**
+ * adreno_dispatcher_replay() - Replay commands from the dispatcher queue
+ * @adreno_dev: Pointer to the adreno device struct
+ *
+ * Replay the commands from the dispatcher inflight queue.  This is called after
+ * a power down/up to recover from a fault
+ */
+int adreno_dispatcher_replay(struct adreno_device *adreno_dev)
+{
+	struct kgsl_device *device = &adreno_dev->dev;
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+	struct kgsl_cmdbatch **replay;
+	int i, ptr, count = 0;
+
+	BUG_ON(!mutex_is_locked(&dispatcher->mutex));
+
+	replay = kzalloc(sizeof(*replay) * dispatcher->inflight, GFP_KERNEL);
+
+	/*
+	 * If we can't allocate enough memory for the replay commands then we
+	 * are in a bad way.  Invalidate everything, reset the GPU and see ya
+	 * later alligator
+	 */
+
+	if (replay == NULL) {
+
+		ptr = dispatcher->head;
+
+		while (ptr != dispatcher->tail) {
+			struct kgsl_context *context =
+				dispatcher->cmdqueue[ptr]->context;
+
+			adreno_drawctxt_invalidate(device, context);
+			ptr = CMDQUEUE_NEXT(ptr, ADRENO_DISPATCH_CMDQUEUE_SIZE);
+		}
+
+		/* Reset the dispatcher queue */
+		dispatcher->inflight = 0;
+		dispatcher->head = dispatcher->tail = 0;
+
+		/* Reset the hardware */
+		mutex_lock(&device->mutex);
+
+		/*
+		 * If adreno_reset fails then the GPU is not alive and there
+		 * isn't anything we can do to recover at this point
+		 */
+
+		BUG_ON(adreno_reset(device));
+		mutex_unlock(&device->mutex);
+
+		return 0;
+	}
+
+	ptr = dispatcher->head;
+
+	while (ptr != dispatcher->tail) {
+		struct kgsl_cmdbatch *cmdbatch = dispatcher->cmdqueue[ptr];
+		struct adreno_context *drawctxt =
+			ADRENO_CONTEXT(cmdbatch->context);
+
+		if (cmdbatch->invalid)
+			adreno_drawctxt_invalidate(device, cmdbatch->context);
+
+		if (!kgsl_context_detached(cmdbatch->context) &&
+			drawctxt->state == ADRENO_CONTEXT_STATE_ACTIVE) {
+			/*
+			 * The context for the command batch is still valid -
+			 * add it to the replay list
+			 */
+			replay[count++] = dispatcher->cmdqueue[ptr];
+		} else {
+			/*
+			 * Skip over invaliated or detached contexts - cancel
+			 * any pending events for the timestamp and destroy the
+			 * command batch
+			 */
+			mutex_lock(&device->mutex);
+			kgsl_cancel_events_timestamp(device, cmdbatch->context,
+				cmdbatch->timestamp);
+			mutex_unlock(&device->mutex);
+
+			kgsl_cmdbatch_destroy(cmdbatch);
+		}
+
+		ptr = CMDQUEUE_NEXT(ptr, ADRENO_DISPATCH_CMDQUEUE_SIZE);
+	}
+
+	/* Reset the dispatcher queue */
+	dispatcher->inflight = 0;
+	dispatcher->head = dispatcher->tail = 0;
+
+	mutex_lock(&device->mutex);
+	BUG_ON(adreno_reset(device));
+	mutex_unlock(&device->mutex);
+
+	/* Replay the pending command buffers */
+	for (i = 0; i < count; i++) {
+		int ret = sendcmd(adreno_dev, replay[i]);
+
+		/*
+		 * I'm afraid that if we get an error during replay we
+		 * are not going to space today
+		 */
+
+		BUG_ON(ret);
+	}
+
+	/*
+	 * active_count will be set when we come into this function because
+	 * there were inflight commands.  By virtue of setting ->inflight back
+	 * to 0 sendcmd() will increase the active count again on the first
+	 * submission.  This active_count_put is needed to put the universe back
+	 * in balance and as a bonus it ensures that the hardware stays up for
+	 * the entire reset process
+	 */
+	mutex_lock(&device->mutex);
+	kgsl_active_count_put(device);
+	mutex_unlock(&device->mutex);
+
+	kfree(replay);
+	return 0;
+}
+
+/**
+ * adreno_dispatcher_queue_cmd() - Queue a new command in the context
+ * @adreno_dev: Pointer to the adreno device struct
+ * @drawctxt: Pointer to the adreno draw context
+ * @cmdbatch: Pointer to the command batch being submitted
+ * @timestamp: Pointer to the requested timestamp
+ *
+ * Queue a command in the context - if there isn't any room in the queue, then
+ * block until there is
+ */
+int adreno_dispatcher_queue_cmd(struct adreno_device *adreno_dev,
+		struct adreno_context *drawctxt, struct kgsl_cmdbatch *cmdbatch,
+		uint32_t *timestamp)
+{
+	int ret;
+
+	mutex_lock(&drawctxt->mutex);
+
+	if (drawctxt->flags & CTXT_FLAGS_BEING_DESTROYED) {
+		mutex_unlock(&drawctxt->mutex);
+		return -EINVAL;
+	}
+
+	/* Wait for room in the context queue */
+
+	while (drawctxt->queued >= _context_cmdqueue_size) {
+		trace_adreno_drawctxt_sleep(drawctxt);
+		mutex_unlock(&drawctxt->mutex);
+
+		ret = wait_event_interruptible_timeout(drawctxt->wq,
+			_check_context_queue(drawctxt),
+			msecs_to_jiffies(_context_queue_wait));
+
+		mutex_lock(&drawctxt->mutex);
+		trace_adreno_drawctxt_wake(drawctxt);
+
+		if (ret <= 0) {
+			mutex_unlock(&drawctxt->mutex);
+			return (ret == 0) ? -ETIMEDOUT : (int) ret;
+		}
+
+		/*
+		 * Account for the possiblity that the context got invalidated
+		 * while we were sleeping
+		 */
+
+		if (drawctxt->state == ADRENO_CONTEXT_STATE_INVALID) {
+			mutex_unlock(&drawctxt->mutex);
+			return -EDEADLK;
+		}
+	}
+
+	/*
+	 * If the UMD specified a timestamp then use that under the condition
+	 * that it is greater then the last queued timestamp in the context.
+	 */
+
+	if (drawctxt->flags & CTXT_FLAGS_USER_GENERATED_TS) {
+		if (timestamp_cmp(drawctxt->timestamp, *timestamp) >= 0) {
+			mutex_unlock(&drawctxt->mutex);
+			return -ERANGE;
+		}
+
+		drawctxt->timestamp = *timestamp;
+	} else
+		drawctxt->timestamp++;
+
+	cmdbatch->timestamp = drawctxt->timestamp;
+	*timestamp = drawctxt->timestamp;
+
+	/* Put the command into the queue */
+	drawctxt->cmdqueue[drawctxt->cmdqueue_tail] = cmdbatch;
+	drawctxt->cmdqueue_tail = (drawctxt->cmdqueue_tail + 1) %
+		ADRENO_CONTEXT_CMDQUEUE_SIZE;
+
+	drawctxt->queued++;
+	trace_adreno_cmdbatch_queued(cmdbatch, drawctxt->queued);
+
+
+	mutex_unlock(&drawctxt->mutex);
+
+	/* Add the context to the dispatcher pending list */
+	dispatcher_queue_context(adreno_dev, drawctxt);
+
+	/*
+	 * Only issue commands if inflight is less than burst -this prevents us
+	 * from sitting around waiting for the mutex on a busy system - the work
+	 * loop will schedule it for us. Inflight is mutex protected but the
+	 * worse that can happen is that it will go to 0 after we check and if
+	 * it goes to 0 it is because the work loop decremented it and the work
+	 * queue will try to schedule new commands anyway.
+	 */
+
+	if (adreno_dev->dispatcher.inflight < _context_cmdbatch_burst)
+		adreno_dispatcher_issuecmds(adreno_dev);
+
+	return 0;
+}
+
+/**
+ * dispatcher_do_fault() - Handle a GPU fault and reset the GPU
+ * @device: Pointer to the KGSL device
+ * @cmdbatch: Pointer to the command batch believed to be responsible for the
+ * fault
+ * @invalidate: Non zero if the current command should be invalidated
+ *
+ * Trigger a fault in the dispatcher and start the replay process
+ */
+static void dispatcher_do_fault(struct kgsl_device *device,
+		struct kgsl_cmdbatch *cmdbatch, int invalidate)
+{
+	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+	unsigned int reg;
+
+	/* Stop the timers */
+	del_timer_sync(&dispatcher->timer);
+
+	mutex_lock(&device->mutex);
+
+	/*
+	 * There is an interesting race condition here - when a command batch
+	 * expires and we invaliate before we recover we run the risk of having
+	 * the UMD clean up the context and free memory that the GPU is still
+	 * using.  Not that it is dangerous because we are a few microseconds
+	 * away from resetting, but it still ends up in pagefaults and log
+	 * messages and so on and so forth. To avoid this we mark the command
+	 * batch itself as invalid and then reset - the context will get
+	 * invalidated in the replay.
+	 */
+
+	if (invalidate)
+		cmdbatch->invalid = 1;
+
+	/*
+	 * Stop the CP in its tracks - this ensures that we don't get activity
+	 * while we are trying to dump the state of the system
+	 */
+
+
+	adreno_readreg(adreno_dev, ADRENO_REG_CP_ME_CNTL, &reg);
+	reg |= (1 << 27) | (1 << 28);
+	adreno_writereg(adreno_dev, ADRENO_REG_CP_ME_CNTL, reg);
+
+	kgsl_postmortem_dump(device, 0);
+	kgsl_device_snapshot(device, 1);
+	mutex_unlock(&device->mutex);
+
+	/* If we can't replay then bravely run away and die */
+	if (adreno_dispatcher_replay(adreno_dev))
+		BUG();
+}
+
+static inline int cmdbatch_consumed(struct kgsl_cmdbatch *cmdbatch,
+		unsigned int consumed, unsigned int retired)
+{
+	return ((timestamp_cmp(cmdbatch->timestamp, consumed) >= 0) &&
+		(timestamp_cmp(retired, cmdbatch->timestamp) < 0));
+}
+
+/**
+ * adreno_dispatcher_work() - Master work handler for the dispatcher
+ * @work: Pointer to the work struct for the current work queue
+ *
+ * Process expired commands and send new ones.
+ */
+static void adreno_dispatcher_work(struct work_struct *work)
+{
+	struct adreno_dispatcher *dispatcher =
+		container_of(work, struct adreno_dispatcher, work);
+	struct adreno_device *adreno_dev =
+		container_of(dispatcher, struct adreno_device, dispatcher);
+	struct kgsl_device *device = &adreno_dev->dev;
+	int inv, count = 0;
+
+	mutex_lock(&dispatcher->mutex);
+
+	while (dispatcher->head != dispatcher->tail) {
+		uint32_t consumed, retired = 0;
+		struct kgsl_cmdbatch *cmdbatch =
+			dispatcher->cmdqueue[dispatcher->head];
+		struct adreno_context *drawctxt;
+		BUG_ON(cmdbatch == NULL);
+
+		drawctxt = ADRENO_CONTEXT(cmdbatch->context);
+
+		/*
+		 * First try to expire the timestamp. This happens if the
+		 * context is valid and the timestamp expired normally or if the
+		 * context was destroyed before the command batch was finished
+		 * in the GPU.  Either way retire the command batch advance the
+		 * pointers and continue processing the queue
+		 */
+
+		if (!kgsl_context_detached(cmdbatch->context))
+			retired = kgsl_readtimestamp(device, cmdbatch->context,
+				KGSL_TIMESTAMP_RETIRED);
+
+		if (kgsl_context_detached(cmdbatch->context) ||
+			(timestamp_cmp(cmdbatch->timestamp, retired) <= 0)) {
+
+			trace_adreno_cmdbatch_retired(cmdbatch,
+				dispatcher->inflight - 1);
+
+			/* Reduce the number of inflight command batches */
+			dispatcher->inflight--;
+
+			/* Zero the old entry*/
+			dispatcher->cmdqueue[dispatcher->head] = NULL;
+
+			/* Advance the buffer head */
+			dispatcher->head = CMDQUEUE_NEXT(dispatcher->head,
+				ADRENO_DISPATCH_CMDQUEUE_SIZE);
+
+			/* Destroy the retired command batch */
+			kgsl_cmdbatch_destroy(cmdbatch);
+
+			/* Update the expire time for the next command batch */
+
+			if (dispatcher->inflight > 0) {
+				cmdbatch =
+					dispatcher->cmdqueue[dispatcher->head];
+				cmdbatch->expires = jiffies +
+					msecs_to_jiffies(_cmdbatch_timeout);
+			}
+
+			count++;
+
+			BUG_ON(dispatcher->inflight == 0 && dispatcher->fault);
+			continue;
+		}
+
+		/*
+		 * If we got a fault from the interrupt handler, this command
+		 * is to blame.  Invalidate it, reset and replay
+		 */
+
+		if (dispatcher->fault) {
+			dispatcher_do_fault(device, cmdbatch, 1);
+			goto done;
+		}
+
+		/* Get the last consumed timestamp */
+		consumed = kgsl_readtimestamp(device, cmdbatch->context,
+			KGSL_TIMESTAMP_CONSUMED);
+
+		/* Break here if fault detection is disabled for the context */
+		if (drawctxt->flags & CTXT_FLAGS_NO_FAULT_TOLERANCE)
+			break;
+
+		/*
+		 * The last line of defense is to check if the command batch has
+		 * timed out. If we get this far but the timeout hasn't expired
+		 * yet then the GPU is still ticking away
+		 */
+
+		if (time_is_after_jiffies(cmdbatch->expires))
+			break;
+
+		/* Boom goes the dynamite */
+
+		pr_err("-----------------------\n");
+
+		pr_err("dispatcher: expired ctx=%d ts=%d consumed=%d retired=%d\n",
+			cmdbatch->context->id, cmdbatch->timestamp, consumed,
+			retired);
+		pr_err("dispatcher: jiffies=%lu expired=%lu\n", jiffies,
+				cmdbatch->expires);
+
+		/*
+		 * If execution stopped after the current command batch was
+		 * consumed then invalidate the context for the current command
+		 * batch
+		 */
+
+		inv = cmdbatch_consumed(cmdbatch, consumed, retired);
+
+		dispatcher_do_fault(device, cmdbatch, inv);
+		break;
+	}
+
+	/*
+	 * Decrement the active count to 0 - this will allow the system to go
+	 * into suspend even if there are queued command batches
+	 */
+
+	if (count && dispatcher->inflight == 0) {
+		mutex_lock(&device->mutex);
+		kgsl_active_count_put(device);
+		mutex_unlock(&device->mutex);
+	}
+
+	/* Dispatch new commands if we have the room */
+	if (dispatcher->inflight < _dispatcher_inflight)
+		_adreno_dispatcher_issuecmds(adreno_dev);
+
+done:
+	/* Either update the timer for the next command batch or disable it */
+	if (dispatcher->inflight) {
+		struct kgsl_cmdbatch *cmdbatch
+			= dispatcher->cmdqueue[dispatcher->head];
+
+		mod_timer(&dispatcher->timer, cmdbatch->expires);
+	} else
+		del_timer_sync(&dispatcher->timer);
+
+	/* Before leaving update the pwrscale information */
+	mutex_lock(&device->mutex);
+	kgsl_pwrscale_idle(device);
+	mutex_unlock(&device->mutex);
+
+	mutex_unlock(&dispatcher->mutex);
+}
+
+void adreno_dispatcher_schedule(struct kgsl_device *device)
+{
+	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+
+	queue_work(device->work_queue, &dispatcher->work);
+}
+
+/*
+ * This is called when the timer expires - it either means the GPU is hung or
+ * the IB is taking too long to execute
+ */
+void adreno_dispatcher_timer(unsigned long data)
+{
+	struct adreno_device *adreno_dev = (struct adreno_device *) data;
+	struct kgsl_device *device = &adreno_dev->dev;
+
+	adreno_dispatcher_schedule(device);
+}
+/**
+ * adreno_dispatcher_fault_irq() - Trigger a fault in the dispatcher
+ * @device: Pointer to the KGSL device
+ *
+ * Called from an interrupt context this will trigger a fault in the
+ * dispatcher
+ */
+void adreno_dispatcher_fault_irq(struct kgsl_device *device)
+{
+	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+
+	dispatcher->fault = 1;
+	adreno_dispatcher_schedule(device);
+}
+
+/**
+ * adreno_dispatcher_pause() - stop the dispatcher
+ * @adreno_dev: pointer to the adreno device structure
+ *
+ * Pause the dispather so it doesn't accept any new commands
+ */
+void adreno_dispatcher_pause(struct adreno_device *adreno_dev)
+{
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+
+	/*
+	 * This will probably get called while holding other mutexes so don't
+	 * take the dispatcher mutex.  The biggest penalty is that another
+	 * command might be submitted while we are in here but thats okay
+	 * because whoever is waiting for the drain will just have another
+	 * command batch to wait for
+	 */
+
+	dispatcher->state = ADRENO_DISPATCHER_PAUSE;
+}
+
+/**
+ * adreno_dispatcher_start() - activate the dispatcher
+ * @adreno_dev: pointer to the adreno device structure
+ *
+ * Set the disaptcher active and start the loop once to get things going
+ */
+void adreno_dispatcher_start(struct adreno_device *adreno_dev)
+{
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+
+	dispatcher->state = ADRENO_DISPATCHER_ACTIVE;
+
+	/* Schedule the work loop to get things going */
+	adreno_dispatcher_schedule(&adreno_dev->dev);
+}
+
+/**
+ * adreno_dispatcher_stop() - stop the dispatcher
+ * @adreno_dev: pointer to the adreno device structure
+ *
+ * Stop the dispatcher and close all the timers
+ */
+void adreno_dispatcher_stop(struct adreno_device *adreno_dev)
+{
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+
+	del_timer_sync(&dispatcher->timer);
+	dispatcher->state = ADRENO_DISPATCHER_PAUSE;
+}
+
+/**
+ * adreno_dispatcher_close() - close the dispatcher
+ * @adreno_dev: pointer to the adreno device structure
+ *
+ * Close the dispatcher and free all the oustanding commands and memory
+ */
+void adreno_dispatcher_close(struct adreno_device *adreno_dev)
+{
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+
+	mutex_lock(&dispatcher->mutex);
+	del_timer_sync(&dispatcher->timer);
+
+	while (dispatcher->head != dispatcher->tail) {
+		kgsl_cmdbatch_destroy(dispatcher->cmdqueue[dispatcher->head]);
+		dispatcher->head = (dispatcher->head + 1)
+			% ADRENO_DISPATCH_CMDQUEUE_SIZE;
+	}
+
+	mutex_unlock(&dispatcher->mutex);
+
+	kobject_put(&dispatcher->kobj);
+}
+
+struct dispatcher_attribute {
+	struct attribute attr;
+	ssize_t (*show)(struct adreno_dispatcher *,
+			struct dispatcher_attribute *, char *);
+	ssize_t (*store)(struct adreno_dispatcher *,
+			struct dispatcher_attribute *, const char *buf,
+			size_t count);
+	unsigned int max;
+	unsigned int *value;
+};
+
+#define DISPATCHER_UINT_ATTR(_name, _mode, _max, _value) \
+	struct dispatcher_attribute dispatcher_attr_##_name =  { \
+		.attr = { .name = __stringify(_name), .mode = _mode }, \
+		.show = _show_uint, \
+		.store = _store_uint, \
+		.max = _max, \
+		.value = &(_value), \
+	}
+
+#define to_dispatcher_attr(_a) \
+	container_of((_a), struct dispatcher_attribute, attr)
+#define to_dispatcher(k) container_of(k, struct adreno_dispatcher, kobj)
+
+static ssize_t _store_uint(struct adreno_dispatcher *dispatcher,
+		struct dispatcher_attribute *attr,
+		const char *buf, size_t size)
+{
+	unsigned long val;
+	int ret = kstrtoul(buf, 0, &val);
+
+	if (ret)
+		return ret;
+
+	if (!val || (attr->max && (val > attr->max)))
+		return -EINVAL;
+
+	*((unsigned int *) attr->value) = val;
+	return size;
+}
+
+static ssize_t _show_uint(struct adreno_dispatcher *dispatcher,
+		struct dispatcher_attribute *attr,
+		char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%d\n",
+		*((unsigned int *) attr->value));
+}
+
+static DISPATCHER_UINT_ATTR(inflight, 0644, ADRENO_DISPATCH_CMDQUEUE_SIZE,
+	_dispatcher_inflight);
+/*
+ * Our code that "puts back" a command from the context is much cleaner
+ * if we are sure that there will always be enough room in the
+ * ringbuffer so restrict the maximum size of the context queue to
+ * ADRENO_CONTEXT_CMDQUEUE_SIZE - 1
+ */
+static DISPATCHER_UINT_ATTR(context_cmdqueue_size, 0644,
+	ADRENO_CONTEXT_CMDQUEUE_SIZE - 1, _context_cmdqueue_size);
+static DISPATCHER_UINT_ATTR(context_burst_count, 0644, 0,
+	_context_cmdbatch_burst);
+static DISPATCHER_UINT_ATTR(cmdbatch_timeout, 0644, 0, _cmdbatch_timeout);
+static DISPATCHER_UINT_ATTR(context_queue_wait, 0644, 0, _context_queue_wait);
+
+static struct attribute *dispatcher_attrs[] = {
+	&dispatcher_attr_inflight.attr,
+	&dispatcher_attr_context_cmdqueue_size.attr,
+	&dispatcher_attr_context_burst_count.attr,
+	&dispatcher_attr_cmdbatch_timeout.attr,
+	&dispatcher_attr_context_queue_wait.attr,
+	NULL,
+};
+
+static ssize_t dispatcher_sysfs_show(struct kobject *kobj,
+				   struct attribute *attr, char *buf)
+{
+	struct adreno_dispatcher *dispatcher = to_dispatcher(kobj);
+	struct dispatcher_attribute *pattr = to_dispatcher_attr(attr);
+	ssize_t ret = -EIO;
+
+	if (pattr->show)
+		ret = pattr->show(dispatcher, pattr, buf);
+
+	return ret;
+}
+
+static ssize_t dispatcher_sysfs_store(struct kobject *kobj,
+				    struct attribute *attr,
+				    const char *buf, size_t count)
+{
+	struct adreno_dispatcher *dispatcher = to_dispatcher(kobj);
+	struct dispatcher_attribute *pattr = to_dispatcher_attr(attr);
+	ssize_t ret = -EIO;
+
+	if (pattr->store)
+		ret = pattr->store(dispatcher, pattr, buf, count);
+
+	return ret;
+}
+
+static void dispatcher_sysfs_release(struct kobject *kobj)
+{
+}
+
+static const struct sysfs_ops dispatcher_sysfs_ops = {
+	.show = dispatcher_sysfs_show,
+	.store = dispatcher_sysfs_store
+};
+
+static struct kobj_type ktype_dispatcher = {
+	.sysfs_ops = &dispatcher_sysfs_ops,
+	.default_attrs = dispatcher_attrs,
+	.release = dispatcher_sysfs_release
+};
+
+/**
+ * adreno_dispatcher_init() - Initialize the dispatcher
+ * @adreno_dev: pointer to the adreno device structure
+ *
+ * Initialize the dispatcher
+ */
+int adreno_dispatcher_init(struct adreno_device *adreno_dev)
+{
+	struct kgsl_device *device = &adreno_dev->dev;
+	struct adreno_dispatcher *dispatcher = &adreno_dev->dispatcher;
+	int ret;
+
+	memset(dispatcher, 0, sizeof(*dispatcher));
+
+	mutex_init(&dispatcher->mutex);
+
+	setup_timer(&dispatcher->timer, adreno_dispatcher_timer,
+		(unsigned long) adreno_dev);
+
+	INIT_WORK(&dispatcher->work, adreno_dispatcher_work);
+
+	plist_head_init(&dispatcher->pending);
+	spin_lock_init(&dispatcher->plist_lock);
+
+	dispatcher->state = ADRENO_DISPATCHER_ACTIVE;
+
+	ret = kobject_init_and_add(&dispatcher->kobj, &ktype_dispatcher,
+		&device->dev->kobj, "dispatch");
+
+	return ret;
+}
diff --git a/drivers/gpu/msm/adreno_drawctxt.c b/drivers/gpu/msm/adreno_drawctxt.c
index 0af4c12e..1a4310e 100644
--- a/drivers/gpu/msm/adreno_drawctxt.c
+++ b/drivers/gpu/msm/adreno_drawctxt.c
@@ -13,10 +13,12 @@
 
 #include <linux/slab.h>
 #include <linux/msm_kgsl.h>
+#include <linux/sched.h>
 
 #include "kgsl.h"
 #include "kgsl_sharedmem.h"
 #include "adreno.h"
+#include "adreno_trace.h"
 
 #define KGSL_INIT_REFTIMESTAMP		0x7FFFFFFF
 
@@ -132,6 +134,247 @@
 	*incmd = cmd;
 }
 
+static void wait_callback(struct kgsl_device *device, void *priv, u32 id,
+		u32 timestamp, u32 type)
+{
+	struct adreno_context *drawctxt = priv;
+	wake_up_interruptible_all(&drawctxt->waiting);
+}
+
+#define adreno_wait_event_interruptible_timeout(wq, condition, timeout, io)   \
+({                                                                            \
+	long __ret = timeout;                                                 \
+	if (io)                                                               \
+		__wait_io_event_interruptible_timeout(wq, condition, __ret);  \
+	else                                                                  \
+		__wait_event_interruptible_timeout(wq, condition, __ret);     \
+	__ret;                                                                \
+})
+
+#define adreno_wait_event_interruptible(wq, condition, io)                    \
+({                                                                            \
+	long __ret;                                                           \
+	if (io)                                                               \
+		__wait_io_event_interruptible(wq, condition, __ret);          \
+	else                                                                  \
+		__wait_event_interruptible(wq, condition, __ret);             \
+	__ret;                                                                \
+})
+
+static int _check_context_timestamp(struct kgsl_device *device,
+		struct adreno_context *drawctxt, unsigned int timestamp)
+{
+	int ret = 0;
+
+	/* Bail if the drawctxt has been invalidated or destroyed */
+	if (kgsl_context_detached(&drawctxt->base) ||
+		drawctxt->state != ADRENO_CONTEXT_STATE_ACTIVE)
+		return 1;
+
+	mutex_lock(&device->mutex);
+	ret = kgsl_check_timestamp(device, &drawctxt->base, timestamp);
+	mutex_unlock(&device->mutex);
+
+	return ret;
+}
+
+/**
+ * adreno_drawctxt_wait() - sleep until a timestamp expires
+ * @adreno_dev: pointer to the adreno_device struct
+ * @drawctxt: Pointer to the draw context to sleep for
+ * @timetamp: Timestamp to wait on
+ * @timeout: Number of jiffies to wait (0 for infinite)
+ *
+ * Register an event to wait for a timestamp on a context and sleep until it
+ * has past.  Returns < 0 on error, -ETIMEDOUT if the timeout expires or 0
+ * on success
+ */
+int adreno_drawctxt_wait(struct adreno_device *adreno_dev,
+		struct kgsl_context *context,
+		uint32_t timestamp, unsigned int timeout)
+{
+	static unsigned int io_cnt;
+	struct kgsl_device *device = &adreno_dev->dev;
+	struct kgsl_pwrctrl *pwr = &device->pwrctrl;
+	struct adreno_context *drawctxt = ADRENO_CONTEXT(context);
+	int ret, io;
+
+	if (kgsl_context_detached(context))
+		return -EINVAL;
+
+	if (drawctxt->state == ADRENO_CONTEXT_STATE_INVALID)
+		return -EDEADLK;
+
+	/* Needs to hold the device mutex */
+	BUG_ON(!mutex_is_locked(&device->mutex));
+
+	trace_adreno_drawctxt_wait_start(context->id, timestamp);
+
+	ret = kgsl_add_event(device, context->id, timestamp,
+		wait_callback, drawctxt, NULL);
+	if (ret)
+		goto done;
+
+	/*
+	 * For proper power accounting sometimes we need to call
+	 * io_wait_interruptible_timeout and sometimes we need to call
+	 * plain old wait_interruptible_timeout. We call the regular
+	 * timeout N times out of 100, where N is a number specified by
+	 * the current power level
+	 */
+
+	io_cnt = (io_cnt + 1) % 100;
+	io = (io_cnt < pwr->pwrlevels[pwr->active_pwrlevel].io_fraction)
+		? 0 : 1;
+
+	mutex_unlock(&device->mutex);
+
+	if (timeout) {
+		ret = (int) adreno_wait_event_interruptible_timeout(
+			drawctxt->waiting,
+			_check_context_timestamp(device, drawctxt, timestamp),
+			msecs_to_jiffies(timeout), io);
+
+		if (ret == 0)
+			ret = -ETIMEDOUT;
+		else if (ret > 0)
+			ret = 0;
+	} else {
+		ret = (int) adreno_wait_event_interruptible(drawctxt->waiting,
+			_check_context_timestamp(device, drawctxt, timestamp),
+				io);
+	}
+
+	mutex_lock(&device->mutex);
+
+	/* -EDEADLK if the context was invalidated while we were waiting */
+	if (drawctxt->state == ADRENO_CONTEXT_STATE_INVALID)
+		ret = -EDEADLK;
+
+
+	/* Return -EINVAL if the context was detached while we were waiting */
+	if (kgsl_context_detached(context))
+		ret = -EINVAL;
+
+done:
+	trace_adreno_drawctxt_wait_done(context->id, timestamp, ret);
+	return ret;
+}
+
+static void global_wait_callback(struct kgsl_device *device, void *priv, u32 id,
+		u32 timestamp, u32 type)
+{
+	struct adreno_context *drawctxt = priv;
+
+	wake_up_interruptible_all(&drawctxt->waiting);
+	kgsl_context_put(&drawctxt->base);
+}
+
+static int _check_global_timestamp(struct kgsl_device *device,
+		unsigned int timestamp)
+{
+	int ret;
+
+	mutex_lock(&device->mutex);
+	ret = kgsl_check_timestamp(device, NULL, timestamp);
+	mutex_unlock(&device->mutex);
+
+	return ret;
+}
+
+int adreno_drawctxt_wait_global(struct adreno_device *adreno_dev,
+		struct kgsl_context *context,
+		uint32_t timestamp, unsigned int timeout)
+{
+	struct kgsl_device *device = &adreno_dev->dev;
+	struct adreno_context *drawctxt = ADRENO_CONTEXT(context);
+	int ret;
+
+	/* Needs to hold the device mutex */
+	BUG_ON(!mutex_is_locked(&device->mutex));
+
+	_kgsl_context_get(context);
+
+	trace_adreno_drawctxt_wait_start(KGSL_MEMSTORE_GLOBAL, timestamp);
+
+	ret = kgsl_add_event(device, KGSL_MEMSTORE_GLOBAL, timestamp,
+		global_wait_callback, drawctxt, NULL);
+	if (ret) {
+		kgsl_context_put(context);
+		goto done;
+	}
+
+	mutex_unlock(&device->mutex);
+
+	if (timeout) {
+		ret = (int) wait_event_interruptible_timeout(drawctxt->waiting,
+			_check_global_timestamp(device, timestamp),
+			msecs_to_jiffies(timeout));
+
+		if (ret == 0)
+			ret = -ETIMEDOUT;
+		else if (ret > 0)
+			ret = 0;
+	} else {
+		ret = (int) wait_event_interruptible(drawctxt->waiting,
+			_check_global_timestamp(device, timestamp));
+	}
+
+	mutex_lock(&device->mutex);
+
+	if (ret)
+		kgsl_cancel_events_timestamp(device, NULL, timestamp);
+
+done:
+	trace_adreno_drawctxt_wait_done(KGSL_MEMSTORE_GLOBAL, timestamp, ret);
+	return ret;
+}
+
+/**
+ * adreno_drawctxt_invalidate() - Invalidate an adreno draw context
+ * @device: Pointer to the KGSL device structure for the GPU
+ * @context: Pointer to the KGSL context structure
+ *
+ * Invalidate the context and remove all queued commands and cancel any pending
+ * waiters
+ */
+void adreno_drawctxt_invalidate(struct kgsl_device *device,
+		struct kgsl_context *context)
+{
+	struct adreno_context *drawctxt = ADRENO_CONTEXT(context);
+
+	trace_adreno_drawctxt_invalidate(drawctxt);
+
+	drawctxt->state = ADRENO_CONTEXT_STATE_INVALID;
+
+	/* Clear the pending queue */
+	mutex_lock(&drawctxt->mutex);
+
+	while (drawctxt->cmdqueue_head != drawctxt->cmdqueue_tail) {
+		struct kgsl_cmdbatch *cmdbatch =
+			drawctxt->cmdqueue[drawctxt->cmdqueue_head];
+
+		drawctxt->cmdqueue_head = (drawctxt->cmdqueue_head + 1) %
+			ADRENO_CONTEXT_CMDQUEUE_SIZE;
+
+		mutex_unlock(&drawctxt->mutex);
+
+		mutex_lock(&device->mutex);
+		kgsl_cancel_events_timestamp(device, context,
+			cmdbatch->timestamp);
+		mutex_unlock(&device->mutex);
+
+		kgsl_cmdbatch_destroy(cmdbatch);
+		mutex_lock(&drawctxt->mutex);
+	}
+
+	mutex_unlock(&drawctxt->mutex);
+
+	/* Give the bad news to everybody waiting around */
+	wake_up_interruptible_all(&drawctxt->waiting);
+	wake_up_interruptible_all(&drawctxt->wq);
+}
+
 /**
  * adreno_drawctxt_create - create a new adreno draw context
  * @dev_priv: the owner of the context
@@ -149,6 +392,7 @@
 	int ret;
 
 	drawctxt = kzalloc(sizeof(struct adreno_context), GFP_KERNEL);
+
 	if (drawctxt == NULL)
 		return ERR_PTR(-ENOMEM);
 
@@ -168,22 +412,30 @@
 		KGSL_CONTEXT_NO_FAULT_TOLERANCE |
 		KGSL_CONTEXT_TYPE_MASK);
 
+	/* Always enable per-context timestamps */
+	*flags |= KGSL_CONTEXT_PER_CONTEXT_TS;
+	drawctxt->flags |= CTXT_FLAGS_PER_CONTEXT_TS;
+
 	if (*flags & KGSL_CONTEXT_PREAMBLE)
 		drawctxt->flags |= CTXT_FLAGS_PREAMBLE;
 
 	if (*flags & KGSL_CONTEXT_NO_GMEM_ALLOC)
 		drawctxt->flags |= CTXT_FLAGS_NOGMEMALLOC;
 
-	if (*flags & KGSL_CONTEXT_PER_CONTEXT_TS)
-		drawctxt->flags |= CTXT_FLAGS_PER_CONTEXT_TS;
-
-	if (*flags & KGSL_CONTEXT_USER_GENERATED_TS) {
-		if (!(*flags & KGSL_CONTEXT_PER_CONTEXT_TS)) {
-			ret = -EINVAL;
-			goto err;
-		}
+	if (*flags & KGSL_CONTEXT_USER_GENERATED_TS)
 		drawctxt->flags |= CTXT_FLAGS_USER_GENERATED_TS;
-	}
+
+	mutex_init(&drawctxt->mutex);
+	init_waitqueue_head(&drawctxt->wq);
+	init_waitqueue_head(&drawctxt->waiting);
+
+	/*
+	 * Set up the plist node for the dispatcher.  For now all contexts have
+	 * the same priority, but later the priority will be set at create time
+	 * by the user
+	 */
+
+	plist_node_init(&drawctxt->pending, ADRENO_CONTEXT_DEFAULT_PRIORITY);
 
 	if (*flags & KGSL_CONTEXT_NO_FAULT_TOLERANCE)
 		drawctxt->flags |= CTXT_FLAGS_NO_FAULT_TOLERANCE;
@@ -196,12 +448,6 @@
 		goto err;
 
 	kgsl_sharedmem_writel(device, &device->memstore,
-			KGSL_MEMSTORE_OFFSET(drawctxt->base.id, ref_wait_ts),
-			KGSL_INIT_REFTIMESTAMP);
-	kgsl_sharedmem_writel(device, &device->memstore,
-			KGSL_MEMSTORE_OFFSET(drawctxt->base.id, ts_cmp_enable),
-			0);
-	kgsl_sharedmem_writel(device, &device->memstore,
 			KGSL_MEMSTORE_OFFSET(drawctxt->base.id, soptimestamp),
 			0);
 	kgsl_sharedmem_writel(device, &device->memstore,
@@ -219,18 +465,20 @@
  * @context: Generic KGSL context container for the context
  *
  */
-void adreno_drawctxt_detach(struct kgsl_context *context)
+int adreno_drawctxt_detach(struct kgsl_context *context)
 {
 	struct kgsl_device *device;
 	struct adreno_device *adreno_dev;
 	struct adreno_context *drawctxt;
+	int ret;
 
 	if (context == NULL)
-		return;
+		return 0;
 
 	device = context->device;
 	adreno_dev = ADRENO_DEVICE(device);
 	drawctxt = ADRENO_CONTEXT(context);
+
 	/* deactivate context */
 	if (adreno_dev->drawctxt_active == drawctxt) {
 		/* no need to save GMEM or shader, the context is
@@ -246,13 +494,39 @@
 		adreno_drawctxt_switch(adreno_dev, NULL, 0);
 	}
 
-	if (device->state != KGSL_STATE_HUNG)
-		adreno_idle(device);
+	mutex_lock(&drawctxt->mutex);
+
+	while (drawctxt->cmdqueue_head != drawctxt->cmdqueue_tail) {
+		struct kgsl_cmdbatch *cmdbatch =
+			drawctxt->cmdqueue[drawctxt->cmdqueue_head];
+
+		drawctxt->cmdqueue_head = (drawctxt->cmdqueue_head + 1) %
+			ADRENO_CONTEXT_CMDQUEUE_SIZE;
+
+		mutex_unlock(&drawctxt->mutex);
+
+		/*
+		 * Don't hold the drawctxt mutex while the cmdbatch is being
+		 * destroyed because the cmdbatch destroy takes the device
+		 * mutex and the world falls in on itself
+		 */
+
+		kgsl_cmdbatch_destroy(cmdbatch);
+		mutex_lock(&drawctxt->mutex);
+	}
+
+	mutex_unlock(&drawctxt->mutex);
+
+	/* Wait for the last global timestamp to pass before continuing */
+	ret = adreno_drawctxt_wait_global(adreno_dev, context,
+		drawctxt->internal_timestamp, 10 * 1000);
 
 	adreno_profile_process_results(device);
 
 	kgsl_sharedmem_free(&drawctxt->gpustate);
 	kgsl_sharedmem_free(&drawctxt->context_gmem_shadow.gmemshadow);
+
+	return ret;
 }
 
 
@@ -296,11 +570,12 @@
  * Switch the current draw context
  */
 
-void adreno_drawctxt_switch(struct adreno_device *adreno_dev,
+int adreno_drawctxt_switch(struct adreno_device *adreno_dev,
 				struct adreno_context *drawctxt,
 				unsigned int flags)
 {
 	struct kgsl_device *device = &adreno_dev->dev;
+	int ret = 0;
 
 	if (drawctxt) {
 		if (flags & KGSL_CONTEXT_SAVE_GMEM)
@@ -316,9 +591,9 @@
 	if (adreno_dev->drawctxt_active == drawctxt) {
 		if (adreno_dev->gpudev->ctxt_draw_workaround &&
 			adreno_is_a225(adreno_dev))
-				adreno_dev->gpudev->ctxt_draw_workaround(
+				ret = adreno_dev->gpudev->ctxt_draw_workaround(
 					adreno_dev, drawctxt);
-		return;
+		return ret;
 	}
 
 	KGSL_CTXT_INFO(device, "from %d to %d flags %d\n",
@@ -327,7 +602,15 @@
 		drawctxt ? drawctxt->base.id : 0, flags);
 
 	/* Save the old context */
-	adreno_dev->gpudev->ctxt_save(adreno_dev, adreno_dev->drawctxt_active);
+	ret = adreno_dev->gpudev->ctxt_save(adreno_dev,
+		adreno_dev->drawctxt_active);
+
+	if (ret) {
+		KGSL_DRV_ERR(device,
+			"Error in GPU context %d save: %d\n",
+			adreno_dev->drawctxt_active->base.id, ret);
+		return ret;
+	}
 
 	/* Put the old instance of the active drawctxt */
 	if (adreno_dev->drawctxt_active) {
@@ -340,6 +623,14 @@
 		_kgsl_context_get(&drawctxt->base);
 
 	/* Set the new context */
-	adreno_dev->gpudev->ctxt_restore(adreno_dev, drawctxt);
+	ret = adreno_dev->gpudev->ctxt_restore(adreno_dev, drawctxt);
+	if (ret) {
+		KGSL_DRV_ERR(device,
+			"Error in GPU context %d restore: %d\n",
+			drawctxt->base.id, ret);
+		return ret;
+	}
+
 	adreno_dev->drawctxt_active = drawctxt;
+	return 0;
 }
diff --git a/drivers/gpu/msm/adreno_drawctxt.h b/drivers/gpu/msm/adreno_drawctxt.h
index 3088099..f8469e2 100644
--- a/drivers/gpu/msm/adreno_drawctxt.h
+++ b/drivers/gpu/msm/adreno_drawctxt.h
@@ -69,6 +69,13 @@
 	const char *str;
 };
 
+#define ADRENO_CONTEXT_CMDQUEUE_SIZE 128
+
+#define ADRENO_CONTEXT_DEFAULT_PRIORITY 1
+
+#define ADRENO_CONTEXT_STATE_ACTIVE 0
+#define ADRENO_CONTEXT_STATE_INVALID 1
+
 struct kgsl_device;
 struct adreno_device;
 struct kgsl_device_private;
@@ -99,18 +106,58 @@
 	struct kgsl_memdesc quad_vertices_restore;
 };
 
+/**
+ * struct adreno_context - Adreno GPU draw context
+ * @id: Unique integer ID of the context
+ * @timestamp: Last issued context-specific timestamp
+ * @internal_timestamp: Global timestamp of the last issued command
+ * @state: Current state of the context
+ * @flags: Bitfield controlling behavior of the context
+ * @type: Context type (GL, CL, RS)
+ * @mutex: Mutex to protect the cmdqueue
+ * @pagetable: Pointer to the GPU pagetable for the context
+ * @gpustate: Pointer to the GPU scratch memory for context save/restore
+ * @reg_restore: Command buffer for restoring context registers
+ * @shader_save: Command buffer for saving shaders
+ * @shader_restore: Command buffer to restore shaders
+ * @context_gmem_shadow: GMEM shadow structure for save/restore
+ * @reg_save: A2XX command buffer to save context registers
+ * @shader_fixup: A2XX command buffer to "fix" shaders on restore
+ * @chicken_restore: A2XX command buffer to "fix" register restore
+ * @bin_base_offset: Saved value of the A2XX BIN_BASE_OFFSET register
+ * @regconstant_save: A3XX command buffer to save some registers
+ * @constant_retore: A3XX command buffer to restore some registers
+ * @hslqcontrol_restore: A3XX command buffer to restore HSLSQ registers
+ * @save_fixup: A3XX command buffer to "fix" register save
+ * @restore_fixup: A3XX cmmand buffer to restore register save fixes
+ * @shader_load_commands: A3XX GPU memory descriptor for shader load IB
+ * @shader_save_commands: A3XX GPU memory descriptor for shader save IB
+ * @constantr_save_commands: A3XX GPU memory descriptor for constant save IB
+ * @constant_load_commands: A3XX GPU memory descriptor for constant load IB
+ * @cond_execs: A3XX GPU memory descriptor for conditional exec IB
+ * @hlsq_restore_commands: A3XX GPU memory descriptor for HLSQ restore IB
+ * @cmdqueue: Queue of command batches waiting to be dispatched for this context
+ * @cmdqueue_head: Head of the cmdqueue queue
+ * @cmdqueue_tail: Tail of the cmdqueue queue
+ * @pending: Priority list node for the dispatcher list of pending contexts
+ * @wq: Workqueue structure for contexts to sleep pending room in the queue
+ * @waiting: Workqueue structure for contexts waiting for a timestamp or event
+ * @queued: Number of commands queued in the cmdqueue
+ */
 struct adreno_context {
 	struct kgsl_context base;
 	unsigned int ib_gpu_time_used;
 	unsigned int timestamp;
+	unsigned int internal_timestamp;
+	int state;
 	uint32_t flags;
 	unsigned int type;
+	struct mutex mutex;
 	struct kgsl_memdesc gpustate;
 	unsigned int reg_restore[3];
 	unsigned int shader_save[3];
 	unsigned int shader_restore[3];
 
-	/* Information of the GMEM shadow that is created in context create */
 	struct gmem_shadow_t context_gmem_shadow;
 
 	/* A2XX specific items */
@@ -131,23 +178,41 @@
 	struct kgsl_memdesc constant_load_commands[3];
 	struct kgsl_memdesc cond_execs[4];
 	struct kgsl_memdesc hlsqcontrol_restore_commands[1];
+
+	/* Dispatcher */
+	struct kgsl_cmdbatch *cmdqueue[ADRENO_CONTEXT_CMDQUEUE_SIZE];
+	int cmdqueue_head;
+	int cmdqueue_tail;
+
+	struct plist_node pending;
+	wait_queue_head_t wq;
+	wait_queue_head_t waiting;
+
+	int queued;
 };
 
 
 struct kgsl_context *adreno_drawctxt_create(struct kgsl_device_private *,
 			uint32_t *flags);
 
-void adreno_drawctxt_detach(struct kgsl_context *context);
+int adreno_drawctxt_detach(struct kgsl_context *context);
 
 void adreno_drawctxt_destroy(struct kgsl_context *context);
 
-void adreno_drawctxt_switch(struct adreno_device *adreno_dev,
+int adreno_drawctxt_switch(struct adreno_device *adreno_dev,
 				struct adreno_context *drawctxt,
 				unsigned int flags);
 void adreno_drawctxt_set_bin_base_offset(struct kgsl_device *device,
 					struct kgsl_context *context,
 					unsigned int offset);
 
+int adreno_drawctxt_wait(struct adreno_device *adreno_dev,
+		struct kgsl_context *context,
+		uint32_t timestamp, unsigned int timeout);
+
+void adreno_drawctxt_invalidate(struct kgsl_device *device,
+		struct kgsl_context *context);
+
 /* GPU context switch helper functions */
 
 void build_quad_vtxbuff(struct adreno_context *drawctxt,
diff --git a/drivers/gpu/msm/adreno_postmortem.c b/drivers/gpu/msm/adreno_postmortem.c
index 32dbd51..294ae76 100644
--- a/drivers/gpu/msm/adreno_postmortem.c
+++ b/drivers/gpu/msm/adreno_postmortem.c
@@ -22,6 +22,7 @@
 #include "adreno_ringbuffer.h"
 #include "kgsl_cffdump.h"
 #include "kgsl_pwrctrl.h"
+#include "adreno_trace.h"
 
 #include "a2xx_reg.h"
 #include "a3xx_reg.h"
@@ -459,6 +460,9 @@
 		adreno_getreg(adreno_dev, ADRENO_REG_CP_IB2_BUFSZ),
 		&cp_ib2_bufsz);
 
+	trace_adreno_gpu_fault(rbbm_status, cp_rb_rptr, cp_rb_wptr,
+			cp_ib1_base, cp_ib1_bufsz, cp_ib2_base, cp_ib2_bufsz);
+
 	/* If postmortem dump is not enabled, dump minimal set and return */
 	if (!device->pm_dump_enable) {
 
@@ -644,5 +648,9 @@
 error_vfree:
 	vfree(rb_copy);
 end:
+	/* Restart the dispatcher after a manually triggered dump */
+	if (manual)
+		adreno_dispatcher_start(adreno_dev);
+
 	return result;
 }
diff --git a/drivers/gpu/msm/adreno_ringbuffer.c b/drivers/gpu/msm/adreno_ringbuffer.c
index b8cf21f..dc1530a 100644
--- a/drivers/gpu/msm/adreno_ringbuffer.c
+++ b/drivers/gpu/msm/adreno_ringbuffer.c
@@ -67,11 +67,8 @@
 	unsigned long wait_time;
 	unsigned long wait_timeout = msecs_to_jiffies(ADRENO_IDLE_TIMEOUT);
 	unsigned long wait_time_part;
-	unsigned int prev_reg_val[FT_DETECT_REGS_COUNT];
 	unsigned int rptr;
 
-	memset(prev_reg_val, 0, sizeof(prev_reg_val));
-
 	/* if wptr ahead, fill the remaining with NOPs */
 	if (wptr_ahead) {
 		/* -1 for header */
@@ -105,43 +102,13 @@
 		if (freecmds == 0 || freecmds > numcmds)
 			break;
 
-		/* Dont wait for timeout, detect hang faster.
-		 */
-		if (time_after(jiffies, wait_time_part)) {
-			wait_time_part = jiffies +
-				msecs_to_jiffies(KGSL_TIMEOUT_PART);
-			if ((adreno_ft_detect(rb->device,
-						prev_reg_val))){
-				KGSL_DRV_ERR(rb->device,
-				"Hang detected while waiting for freespace in"
-				"ringbuffer rptr: 0x%x, wptr: 0x%x\n",
-				rptr, rb->wptr);
-				goto err;
-			}
-		}
-
 		if (time_after(jiffies, wait_time)) {
 			KGSL_DRV_ERR(rb->device,
 			"Timed out while waiting for freespace in ringbuffer "
 			"rptr: 0x%x, wptr: 0x%x\n", rptr, rb->wptr);
-			goto err;
+			return -ETIMEDOUT;
 		}
 
-		continue;
-
-err:
-		if (!adreno_dump_and_exec_ft(rb->device)) {
-			if (context && context->flags & CTXT_FLAGS_GPU_HANG) {
-				KGSL_CTXT_WARN(rb->device,
-				"Context %p caused a gpu hang. Will not accept commands for context %d\n",
-				context, context->base.id);
-				return -EDEADLK;
-			}
-			wait_time = jiffies + wait_timeout;
-		} else {
-			/* GPU is hung and fault tolerance failed */
-			BUG();
-		}
 	}
 	return 0;
 }
@@ -180,7 +147,8 @@
 	if (!ret) {
 		ptr = (unsigned int *)rb->buffer_desc.hostptr + rb->wptr;
 		rb->wptr += numcmds;
-	}
+	} else
+		ptr = ERR_PTR(ret);
 
 	return ptr;
 }
@@ -347,7 +315,6 @@
 int _ringbuffer_start_common(struct adreno_ringbuffer *rb)
 {
 	int status;
-	/*cp_rb_cntl_u cp_rb_cntl; */
 	union reg_cp_rb_cntl cp_rb_cntl;
 	unsigned int rb_cntl;
 	struct kgsl_device *device = rb->device;
@@ -568,18 +535,17 @@
 
 static int
 adreno_ringbuffer_addcmds(struct adreno_ringbuffer *rb,
-				struct adreno_context *context,
+				struct adreno_context *drawctxt,
 				unsigned int flags, unsigned int *cmds,
-				int sizedwords)
+				int sizedwords, uint32_t timestamp)
 {
 	struct adreno_device *adreno_dev = ADRENO_DEVICE(rb->device);
 	unsigned int *ringcmds;
 	unsigned int total_sizedwords = sizedwords;
 	unsigned int i;
 	unsigned int rcmd_gpu;
-	unsigned int context_id = KGSL_MEMSTORE_GLOBAL;
+	unsigned int context_id;
 	unsigned int gpuaddr = rb->device->memstore.gpuaddr;
-	unsigned int timestamp;
 	bool profile_ready;
 
 	/*
@@ -594,15 +560,19 @@
 		adreno_profile_assignments_ready(&adreno_dev->profile) &&
 		!(flags & KGSL_CMD_FLAGS_INTERNAL_ISSUE);
 
-	/*
-	 * if the context was not created with per context timestamp
-	 * support, we must use the global timestamp since issueibcmds
-	 * will be returning that one, or if an internal issue then
-	 * use global timestamp.
-	 */
-	if ((context && (context->flags & CTXT_FLAGS_PER_CONTEXT_TS)) &&
-		!(flags & KGSL_CMD_FLAGS_INTERNAL_ISSUE))
-		context_id = context->base.id;
+	/* The global timestamp always needs to be incremented */
+	rb->global_ts++;
+
+	/* If this is a internal IB, use the global timestamp for it */
+	if (!drawctxt || (flags & KGSL_CMD_FLAGS_INTERNAL_ISSUE)) {
+		timestamp = rb->global_ts;
+		context_id = KGSL_MEMSTORE_GLOBAL;
+	} else {
+		context_id = drawctxt->base.id;
+	}
+
+	if (drawctxt)
+		drawctxt->internal_timestamp = rb->global_ts;
 
 	/* reserve space to temporarily turn off protected mode
 	*  error checking if needed
@@ -613,13 +583,8 @@
 	/* internal ib command identifier for the ringbuffer */
 	total_sizedwords += (flags & KGSL_CMD_FLAGS_INTERNAL_ISSUE) ? 2 : 0;
 
-	/* Add CP_COND_EXEC commands to generate CP_INTERRUPT */
-	total_sizedwords += context ? 13 : 0;
-
-	if ((context) && (context->flags & CTXT_FLAGS_PER_CONTEXT_TS) &&
-		(flags & (KGSL_CMD_FLAGS_INTERNAL_ISSUE |
-		KGSL_CMD_FLAGS_GET_INT)))
-			total_sizedwords += 2;
+	/* Add two dwords for the CP_INTERRUPT */
+	total_sizedwords += drawctxt ? 2 : 0;
 
 	if (adreno_is_a3xx(adreno_dev))
 		total_sizedwords += 7;
@@ -627,13 +592,16 @@
 	if (adreno_is_a2xx(adreno_dev))
 		total_sizedwords += 2; /* CP_WAIT_FOR_IDLE */
 
-	total_sizedwords += 2; /* scratchpad ts for fault tolerance */
 	total_sizedwords += 3; /* sop timestamp */
 	total_sizedwords += 4; /* eop timestamp */
 
-	if (KGSL_MEMSTORE_GLOBAL != context_id)
+	if (adreno_is_a20x(adreno_dev))
+		total_sizedwords += 2; /* CACHE_FLUSH */
+
+	if (drawctxt) {
 		total_sizedwords += 3; /* global timestamp without cache
 					* flush for non-zero context */
+	}
 
 	if (adreno_is_a20x(adreno_dev))
 		total_sizedwords += 2; /* CACHE_FLUSH */
@@ -644,8 +612,11 @@
 	if (profile_ready)
 		total_sizedwords += 6;   /* space for pre_ib and post_ib */
 
-	ringcmds = adreno_ringbuffer_allocspace(rb, context, total_sizedwords);
-	if (!ringcmds)
+	ringcmds = adreno_ringbuffer_allocspace(rb, drawctxt, total_sizedwords);
+
+	if (IS_ERR(ringcmds))
+		return PTR_ERR(ringcmds);
+	if (ringcmds == NULL)
 		return -ENOSPC;
 
 	rcmd_gpu = rb->buffer_desc.gpuaddr
@@ -662,24 +633,9 @@
 
 	/* Add any IB required for profiling if it is enabled */
 	if (profile_ready)
-		adreno_profile_preib_processing(rb->device, context->base.id,
+		adreno_profile_preib_processing(rb->device, drawctxt->base.id,
 				&flags, &ringcmds, &rcmd_gpu);
 
-	/* always increment the global timestamp. once. */
-	rb->global_ts++;
-
-	if (KGSL_MEMSTORE_GLOBAL != context_id)
-		timestamp = context->timestamp;
-	else
-		timestamp = rb->global_ts;
-
-	/* scratchpad ts for fault tolerance */
-	GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
-		cp_type0_packet(adreno_getreg(adreno_dev,
-			ADRENO_REG_CP_TIMESTAMP), 1));
-	GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
-			rb->global_ts);
-
 	/* start-of-pipeline timestamp */
 	GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
 			cp_type3_packet(CP_MEM_WRITE, 2));
@@ -749,7 +705,7 @@
 		KGSL_MEMSTORE_OFFSET(context_id, eoptimestamp)));
 	GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, timestamp);
 
-	if (KGSL_MEMSTORE_GLOBAL != context_id) {
+	if (drawctxt) {
 		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
 			cp_type3_packet(CP_MEM_WRITE, 2));
 		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, (gpuaddr +
@@ -765,56 +721,13 @@
 		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, CACHE_FLUSH);
 	}
 
-	if (context) {
-		/* Conditional execution based on memory values */
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
-			cp_type3_packet(CP_COND_EXEC, 4));
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, (gpuaddr +
-			KGSL_MEMSTORE_OFFSET(
-				context_id, ts_cmp_enable)) >> 2);
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, (gpuaddr +
-			KGSL_MEMSTORE_OFFSET(
-				context_id, ref_wait_ts)) >> 2);
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, timestamp);
-		/* # of conditional command DWORDs */
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, 8);
-
-		/* Clear the ts_cmp_enable for the context */
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
-			cp_type3_packet(CP_MEM_WRITE, 2));
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, gpuaddr +
-			KGSL_MEMSTORE_OFFSET(
-				context_id, ts_cmp_enable));
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, 0x0);
-
-		/* Clear the ts_cmp_enable for the global timestamp */
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
-			cp_type3_packet(CP_MEM_WRITE, 2));
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, gpuaddr +
-			KGSL_MEMSTORE_OFFSET(
-				KGSL_MEMSTORE_GLOBAL, ts_cmp_enable));
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, 0x0);
-
-		/* Trigger the interrupt */
+	if (drawctxt || (flags & KGSL_CMD_FLAGS_INTERNAL_ISSUE)) {
 		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
 			cp_type3_packet(CP_INTERRUPT, 1));
 		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
 				CP_INT_CNTL__RB_INT_MASK);
 	}
 
-	/*
-	 * If per context timestamps are enabled and any of the kgsl
-	 * internal commands want INT to be generated trigger the INT
-	*/
-	if ((context) && (context->flags & CTXT_FLAGS_PER_CONTEXT_TS) &&
-		(flags & (KGSL_CMD_FLAGS_INTERNAL_ISSUE |
-		KGSL_CMD_FLAGS_GET_INT))) {
-			GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
-				cp_type3_packet(CP_INTERRUPT, 1));
-			GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
-				CP_INT_CNTL__RB_INT_MASK);
-	}
-
 	if (adreno_is_a3xx(adreno_dev)) {
 		/* Dummy set-constant to trigger context rollover */
 		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
@@ -824,12 +737,6 @@
 		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, 0);
 	}
 
-	if (flags & KGSL_CMD_FLAGS_EOF) {
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, cp_nop_packet(1));
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu,
-				KGSL_END_OF_FRAME_IDENTIFIER);
-	}
-
 	adreno_ringbuffer_submit(rb);
 
 	return 0;
@@ -845,14 +752,10 @@
 	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
 	struct adreno_ringbuffer *rb = &adreno_dev->ringbuffer;
 
-	if (device->state & KGSL_STATE_HUNG)
-		return kgsl_readtimestamp(device, KGSL_MEMSTORE_GLOBAL,
-					KGSL_TIMESTAMP_RETIRED);
-
 	flags |= KGSL_CMD_FLAGS_INTERNAL_ISSUE;
 
 	return adreno_ringbuffer_addcmds(rb, drawctxt, flags, cmds,
-							sizedwords);
+		sizedwords, 0);
 }
 
 static bool _parse_ibs(struct kgsl_device_private *dev_priv, uint gpuaddr,
@@ -1045,39 +948,89 @@
 	return ret;
 }
 
-int
-adreno_ringbuffer_issueibcmds(struct kgsl_device_private *dev_priv,
-				struct kgsl_context *context,
-				struct kgsl_ibdesc *ibdesc,
-				unsigned int numibs,
-				uint32_t *timestamp,
-				unsigned int flags)
+/**
+ * _ringbuffer_verify_ib() - parse an IB and verify that it is correct
+ * @dev_priv: Pointer to the process struct
+ * @ibdesc: Pointer to the IB descriptor
+ *
+ * This function only gets called if debugging is enabled  - it walks the IB and
+ * does additional level parsing and verification above and beyond what KGSL
+ * core does
+ */
+static inline bool _ringbuffer_verify_ib(struct kgsl_device_private *dev_priv,
+		struct kgsl_ibdesc *ibdesc)
 {
 	struct kgsl_device *device = dev_priv->device;
 	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
-	unsigned int *link = 0;
+
+	/* Check that the size of the IBs is under the allowable limit */
+	if (ibdesc->sizedwords == 0 || ibdesc->sizedwords > 0xFFFFF) {
+		KGSL_DRV_ERR(device, "Invalid IB size 0x%X\n",
+				ibdesc->sizedwords);
+		return false;
+	}
+
+	if (unlikely(adreno_dev->ib_check_level >= 1) &&
+		!_parse_ibs(dev_priv, ibdesc->gpuaddr, ibdesc->sizedwords)) {
+		KGSL_DRV_ERR(device, "Could not verify the IBs\n");
+		return false;
+	}
+
+	return true;
+}
+
+int
+adreno_ringbuffer_issueibcmds(struct kgsl_device_private *dev_priv,
+				struct kgsl_context *context,
+				struct kgsl_cmdbatch *cmdbatch,
+				uint32_t *timestamp)
+{
+	struct kgsl_device *device = dev_priv->device;
+	struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
+	struct adreno_context *drawctxt = ADRENO_CONTEXT(context);
+	int i, ret;
+
+	if (drawctxt->state == ADRENO_CONTEXT_STATE_INVALID)
+		return -EDEADLK;
+
+	/* Verify the IBs before they get queued */
+
+	for (i = 0; i < cmdbatch->ibcount; i++) {
+		if (!_ringbuffer_verify_ib(dev_priv, &cmdbatch->ibdesc[i]))
+			return -EINVAL;
+	}
+
+	/* Queue the command in the ringbuffer */
+	ret = adreno_dispatcher_queue_cmd(adreno_dev, drawctxt, cmdbatch,
+		timestamp);
+
+	if (ret)
+		KGSL_DRV_ERR(device,
+			"adreno_dispatcher_queue_cmd returned %d\n", ret);
+
+	return ret;
+}
+
+/* adreno_rindbuffer_submitcmd - submit userspace IBs to the GPU */
+int adreno_ringbuffer_submitcmd(struct adreno_device *adreno_dev,
+		struct kgsl_cmdbatch *cmdbatch)
+{
+	struct kgsl_device *device = &adreno_dev->dev;
+	struct kgsl_ibdesc *ibdesc;
+	unsigned int numibs;
+	unsigned int *link;
 	unsigned int *cmds;
 	unsigned int i;
-	struct adreno_context *drawctxt = NULL;
+	struct kgsl_context *context;
+	struct adreno_context *drawctxt;
 	unsigned int start_index = 0;
 	int ret;
 
-	if (device->state & KGSL_STATE_HUNG) {
-		ret = -EBUSY;
-		goto done;
-	}
-
-	if (!(adreno_dev->ringbuffer.flags & KGSL_FLAGS_STARTED) ||
-	      context == NULL || ibdesc == 0 || numibs == 0) {
-		ret = -EINVAL;
-		goto done;
-	}
+	context = cmdbatch->context;
 	drawctxt = ADRENO_CONTEXT(context);
 
-	if (drawctxt->flags & CTXT_FLAGS_GPU_HANG) {
-		ret = -EDEADLK;
-		goto done;
-	}
+	ibdesc = cmdbatch->ibdesc;
+	numibs = cmdbatch->ibcount;
 
 	/* process any profiling results that are available into the log_buf */
 	adreno_profile_process_results(device);
@@ -1090,15 +1043,6 @@
 		adreno_dev->drawctxt_active == drawctxt)
 		start_index = 1;
 
-	if (drawctxt->flags & CTXT_FLAGS_SKIP_EOF) {
-		if (flags & KGSL_CMD_FLAGS_EOF)
-			drawctxt->flags &= ~CTXT_FLAGS_SKIP_EOF;
-		if (start_index)
-			numibs = 1;
-		else
-			numibs = 0;
-	}
-
 	cmds = link = kzalloc(sizeof(unsigned int) * (numibs * 3 + 4),
 				GFP_KERNEL);
 	if (!link) {
@@ -1117,18 +1061,6 @@
 		*cmds++ = ibdesc[0].sizedwords;
 	}
 	for (i = start_index; i < numibs; i++) {
-		if (unlikely(adreno_dev->ib_check_level >= 1 &&
-		    !_parse_ibs(dev_priv, ibdesc[i].gpuaddr,
-				ibdesc[i].sizedwords))) {
-			ret = -EINVAL;
-			goto done;
-		}
-
-		if (ibdesc[i].sizedwords == 0) {
-			ret = -EINVAL;
-			goto done;
-		}
-
 		*cmds++ = CP_HDR_INDIRECT_BUFFER_PFD;
 		*cmds++ = ibdesc[i].gpuaddr;
 		*cmds++ = ibdesc[i].sizedwords;
@@ -1137,253 +1069,44 @@
 	*cmds++ = cp_nop_packet(1);
 	*cmds++ = KGSL_END_OF_IB_IDENTIFIER;
 
-	kgsl_setstate(&device->mmu, context->id,
+	ret = kgsl_setstate(&device->mmu, context->id,
 		      kgsl_mmu_pt_get_flags(device->mmu.hwpagetable,
 					device->id));
 
-	adreno_drawctxt_switch(adreno_dev, drawctxt, flags);
-
-	if (drawctxt->flags & CTXT_FLAGS_USER_GENERATED_TS) {
-		if (timestamp_cmp(drawctxt->timestamp, *timestamp) >= 0) {
-			KGSL_DRV_ERR(device,
-				"Invalid user generated ts <%d:0x%x>, "
-				"less than last issued ts <%d:0x%x>\n",
-				context->id, *timestamp, context->id,
-				drawctxt->timestamp);
-			return -ERANGE;
-		}
-		drawctxt->timestamp = *timestamp;
-	} else
-		drawctxt->timestamp++;
-
-	ret = adreno_ringbuffer_addcmds(&adreno_dev->ringbuffer,
-					drawctxt,
-					(flags & KGSL_CMD_FLAGS_EOF),
-					&link[0], (cmds - link));
 	if (ret)
 		goto done;
 
-	if (drawctxt->flags & CTXT_FLAGS_PER_CONTEXT_TS)
-		*timestamp = drawctxt->timestamp;
-	else
-		*timestamp = adreno_dev->ringbuffer.global_ts;
+	ret = adreno_drawctxt_switch(adreno_dev, drawctxt, cmdbatch->flags);
+
+	/*
+	 * In the unlikely event of an error in the drawctxt switch,
+	 * treat it like a hang
+	 */
+	if (ret)
+		goto done;
+
+	ret = adreno_ringbuffer_addcmds(&adreno_dev->ringbuffer,
+					drawctxt,
+					cmdbatch->flags,
+					&link[0], (cmds - link),
+					cmdbatch->timestamp);
 
 #ifdef CONFIG_MSM_KGSL_CFF_DUMP
+	if (ret)
+		goto done;
 	/*
 	 * insert wait for idle after every IB1
 	 * this is conservative but works reliably and is ok
 	 * even for performance simulations
 	 */
-	adreno_idle(device);
+	ret = adreno_idle(device);
 #endif
 
-	/*
-	 * If context hung and recovered then return error so that the
-	 * application may handle it
-	 */
-	if (drawctxt->flags & CTXT_FLAGS_GPU_HANG_FT) {
-		drawctxt->flags &= ~CTXT_FLAGS_GPU_HANG_FT;
-		ret = -EPROTO;
-	} else
-		ret = 0;
-
 done:
-	device->pwrctrl.irq_last = 0;
-	kgsl_trace_issueibcmds(device, context ? context->id : 0, ibdesc,
-		numibs, *timestamp, flags, ret,
-		drawctxt ? drawctxt->type : 0);
+	kgsl_trace_issueibcmds(device, context->id, cmdbatch,
+		cmdbatch->timestamp, cmdbatch->flags, ret,
+		drawctxt->type);
 
 	kfree(link);
 	return ret;
 }
-
-static void _turn_preamble_on_for_ib_seq(struct adreno_ringbuffer *rb,
-				unsigned int rb_rptr)
-{
-	unsigned int temp_rb_rptr = rb_rptr;
-	unsigned int size = rb->buffer_desc.size;
-	unsigned int val[2];
-	int i = 0;
-	bool check = false;
-	bool cmd_start = false;
-
-	/* Go till the start of the ib sequence and turn on preamble */
-	while (temp_rb_rptr / sizeof(unsigned int) != rb->wptr) {
-		kgsl_sharedmem_readl(&rb->buffer_desc, &val[i], temp_rb_rptr);
-		if (check && KGSL_START_OF_IB_IDENTIFIER == val[i]) {
-			/* decrement i */
-			i = (i + 1) % 2;
-			if (val[i] == cp_nop_packet(4)) {
-				temp_rb_rptr = adreno_ringbuffer_dec_wrapped(
-						temp_rb_rptr, size);
-				kgsl_sharedmem_writel(rb->device,
-					&rb->buffer_desc,
-					temp_rb_rptr, cp_nop_packet(1));
-			}
-			KGSL_FT_INFO(rb->device,
-			"Turned preamble on at offset 0x%x\n",
-			temp_rb_rptr / 4);
-			break;
-		}
-		/* If you reach beginning of next command sequence then exit
-		 * First command encountered is the current one so don't break
-		 * on that. */
-		if (KGSL_CMD_IDENTIFIER == val[i]) {
-			if (cmd_start)
-				break;
-			cmd_start = true;
-		}
-
-		i = (i + 1) % 2;
-		if (1 == i)
-			check = true;
-		temp_rb_rptr = adreno_ringbuffer_inc_wrapped(temp_rb_rptr,
-								size);
-	}
-}
-
-void adreno_ringbuffer_extract(struct adreno_ringbuffer *rb,
-				struct adreno_ft_data *ft_data)
-{
-	struct kgsl_device *device = rb->device;
-	unsigned int rb_rptr = ft_data->start_of_replay_cmds;
-	unsigned int good_rb_idx = 0, bad_rb_idx = 0, temp_rb_idx = 0;
-	unsigned int last_good_cmd_end_idx = 0, last_bad_cmd_end_idx = 0;
-	unsigned int cmd_start_idx = 0;
-	unsigned int val1 = 0;
-	int copy_rb_contents = 0;
-	unsigned int temp_rb_rptr;
-	struct kgsl_context *k_ctxt;
-	struct adreno_context *a_ctxt;
-	unsigned int size = rb->buffer_desc.size;
-	unsigned int *temp_rb_buffer = ft_data->rb_buffer;
-	int *rb_size = &ft_data->rb_size;
-	unsigned int *bad_rb_buffer = ft_data->bad_rb_buffer;
-	int *bad_rb_size = &ft_data->bad_rb_size;
-	unsigned int *good_rb_buffer = ft_data->good_rb_buffer;
-	int *good_rb_size = &ft_data->good_rb_size;
-
-	/*
-	 * If the start index from where commands need to be copied is invalid
-	 * then no need to save off any commands
-	 */
-	if (0xFFFFFFFF == ft_data->start_of_replay_cmds)
-		return;
-
-	k_ctxt = kgsl_context_get(device, ft_data->context_id);
-
-	if (k_ctxt) {
-		a_ctxt = ADRENO_CONTEXT(k_ctxt);
-		if (a_ctxt->flags & CTXT_FLAGS_PREAMBLE)
-			_turn_preamble_on_for_ib_seq(rb, rb_rptr);
-		kgsl_context_put(k_ctxt);
-	}
-	k_ctxt = NULL;
-
-	/* Walk the rb from the context switch. Omit any commands
-	 * for an invalid context. */
-	while ((rb_rptr / sizeof(unsigned int)) != rb->wptr) {
-		kgsl_sharedmem_readl(&rb->buffer_desc, &val1, rb_rptr);
-
-		if (KGSL_CMD_IDENTIFIER == val1) {
-			/* Start is the NOP dword that comes before
-			 * KGSL_CMD_IDENTIFIER */
-			cmd_start_idx = temp_rb_idx - 1;
-			if ((copy_rb_contents) && (good_rb_idx))
-				last_good_cmd_end_idx = good_rb_idx - 1;
-			if ((!copy_rb_contents) && (bad_rb_idx))
-				last_bad_cmd_end_idx = bad_rb_idx - 1;
-		}
-
-		/* check for context switch indicator */
-		if (val1 == KGSL_CONTEXT_TO_MEM_IDENTIFIER) {
-			unsigned int temp_idx, val2;
-			/* increment by 3 to get to the context_id */
-			temp_rb_rptr = rb_rptr + (3 * sizeof(unsigned int)) %
-					size;
-			kgsl_sharedmem_readl(&rb->buffer_desc, &val2,
-						temp_rb_rptr);
-
-			/* if context switches to a context that did not cause
-			 * hang then start saving the rb contents as those
-			 * commands can be executed */
-			k_ctxt = kgsl_context_get(rb->device, val2);
-
-			if (k_ctxt) {
-				a_ctxt = ADRENO_CONTEXT(k_ctxt);
-
-			/* If we are changing to a good context and were not
-			 * copying commands then copy over commands to the good
-			 * context */
-			if (!copy_rb_contents && ((k_ctxt &&
-				!(a_ctxt->flags & CTXT_FLAGS_GPU_HANG)) ||
-				!k_ctxt)) {
-				for (temp_idx = cmd_start_idx;
-					temp_idx < temp_rb_idx;
-					temp_idx++)
-					good_rb_buffer[good_rb_idx++] =
-						temp_rb_buffer[temp_idx];
-				ft_data->last_valid_ctx_id = val2;
-				copy_rb_contents = 1;
-				/* remove the good commands from bad buffer */
-				bad_rb_idx = last_bad_cmd_end_idx;
-			} else if (copy_rb_contents && k_ctxt &&
-				(a_ctxt->flags & CTXT_FLAGS_GPU_HANG)) {
-
-				/* If we are changing back to a bad context
-				 * from good ctxt and were not copying commands
-				 * to bad ctxt then copy over commands to
-				 * the bad context */
-				for (temp_idx = cmd_start_idx;
-					temp_idx < temp_rb_idx;
-					temp_idx++)
-					bad_rb_buffer[bad_rb_idx++] =
-						temp_rb_buffer[temp_idx];
-				/* If we are changing to bad context then
-				 * remove the dwords we copied for this
-				 * sequence from the good buffer */
-				good_rb_idx = last_good_cmd_end_idx;
-				copy_rb_contents = 0;
-			}
-			}
-			kgsl_context_put(k_ctxt);
-		}
-
-		if (copy_rb_contents)
-			good_rb_buffer[good_rb_idx++] = val1;
-		else
-			bad_rb_buffer[bad_rb_idx++] = val1;
-
-		/* Copy both good and bad commands to temp buffer */
-		temp_rb_buffer[temp_rb_idx++] = val1;
-
-		rb_rptr = adreno_ringbuffer_inc_wrapped(rb_rptr, size);
-	}
-	*good_rb_size = good_rb_idx;
-	*bad_rb_size = bad_rb_idx;
-	*rb_size = temp_rb_idx;
-}
-
-void
-adreno_ringbuffer_restore(struct adreno_ringbuffer *rb, unsigned int *rb_buff,
-			int num_rb_contents)
-{
-	int i;
-	unsigned int *ringcmds;
-	unsigned int rcmd_gpu;
-	struct adreno_device *adreno_dev = ADRENO_DEVICE(rb->device);
-
-	if (!num_rb_contents)
-		return;
-
-	if (num_rb_contents > (rb->buffer_desc.size - rb->wptr)) {
-		adreno_writereg(adreno_dev, ADRENO_REG_CP_RB_RPTR, 0);
-		BUG_ON(num_rb_contents > rb->buffer_desc.size);
-	}
-	ringcmds = (unsigned int *)rb->buffer_desc.hostptr + rb->wptr;
-	rcmd_gpu = rb->buffer_desc.gpuaddr + sizeof(unsigned int) * rb->wptr;
-	for (i = 0; i < num_rb_contents; i++)
-		GSL_RB_WRITE(rb->device, ringcmds, rcmd_gpu, rb_buff[i]);
-	rb->wptr += num_rb_contents;
-	adreno_ringbuffer_submit(rb);
-}
diff --git a/drivers/gpu/msm/adreno_ringbuffer.h b/drivers/gpu/msm/adreno_ringbuffer.h
index 9634e32..3aa0101 100644
--- a/drivers/gpu/msm/adreno_ringbuffer.h
+++ b/drivers/gpu/msm/adreno_ringbuffer.h
@@ -27,7 +27,6 @@
 
 struct kgsl_device;
 struct kgsl_device_private;
-struct adreno_ft_data;
 
 #define GSL_RB_MEMPTRS_SCRATCH_COUNT	 8
 struct kgsl_rbmemptrs {
@@ -99,10 +98,11 @@
 
 int adreno_ringbuffer_issueibcmds(struct kgsl_device_private *dev_priv,
 				struct kgsl_context *context,
-				struct kgsl_ibdesc *ibdesc,
-				unsigned int numibs,
-				uint32_t *timestamp,
-				unsigned int flags);
+				struct kgsl_cmdbatch *cmdbatch,
+				uint32_t *timestamp);
+
+int adreno_ringbuffer_submitcmd(struct adreno_device *adreno_dev,
+		struct kgsl_cmdbatch *cmdbatch);
 
 int adreno_ringbuffer_init(struct kgsl_device *device);
 
@@ -124,13 +124,6 @@
 
 void kgsl_cp_intrcallback(struct kgsl_device *device);
 
-void adreno_ringbuffer_extract(struct adreno_ringbuffer *rb,
-				struct adreno_ft_data *ft_data);
-
-void
-adreno_ringbuffer_restore(struct adreno_ringbuffer *rb, unsigned int *rb_buff,
-			int num_rb_contents);
-
 unsigned int *adreno_ringbuffer_allocspace(struct adreno_ringbuffer *rb,
 						struct adreno_context *context,
 						unsigned int numcmds);
diff --git a/arch/arm/boot/dts/apq8074-v2-liquid.dts b/drivers/gpu/msm/adreno_trace.c
similarity index 70%
copy from arch/arm/boot/dts/apq8074-v2-liquid.dts
copy to drivers/gpu/msm/adreno_trace.c
index a0ecb50..607ba8c 100644
--- a/arch/arm/boot/dts/apq8074-v2-liquid.dts
+++ b/drivers/gpu/msm/adreno_trace.c
@@ -8,15 +8,11 @@
  * 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.
+ *
  */
 
-/dts-v1/;
+#include "adreno.h"
 
-/include/ "apq8074-v2.dtsi"
-/include/ "msm8974-liquid.dtsi"
-
-/ {
-	model = "Qualcomm APQ 8074v2 LIQUID";
-	compatible = "qcom,apq8074-liquid", "qcom,apq8074", "qcom,liquid";
-	qcom,msm-id = <184 9 0x20000>;
-};
+/* Instantiate tracepoints */
+#define CREATE_TRACE_POINTS
+#include "adreno_trace.h"
diff --git a/drivers/gpu/msm/adreno_trace.h b/drivers/gpu/msm/adreno_trace.h
new file mode 100644
index 0000000..59aca2e
--- /dev/null
+++ b/drivers/gpu/msm/adreno_trace.h
@@ -0,0 +1,174 @@
+/* Copyright (c) 2013, 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.
+ *
+ */
+
+#if !defined(_ADRENO_TRACE_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _ADRENO_TRACE_H
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM kgsl
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE adreno_trace
+
+#include <linux/tracepoint.h>
+
+TRACE_EVENT(adreno_cmdbatch_queued,
+	TP_PROTO(struct kgsl_cmdbatch *cmdbatch, unsigned int queued),
+	TP_ARGS(cmdbatch, queued),
+	TP_STRUCT__entry(
+		__field(unsigned int, id)
+		__field(unsigned int, timestamp)
+		__field(unsigned int, queued)
+	),
+	TP_fast_assign(
+		__entry->id = cmdbatch->context->id;
+		__entry->timestamp = cmdbatch->timestamp;
+		__entry->queued = queued;
+	),
+	TP_printk(
+		"ctx=%u ts=%u queued=%u",
+			__entry->id, __entry->timestamp, __entry->queued
+	)
+);
+
+DECLARE_EVENT_CLASS(adreno_cmdbatch_template,
+	TP_PROTO(struct kgsl_cmdbatch *cmdbatch, int inflight),
+	TP_ARGS(cmdbatch, inflight),
+	TP_STRUCT__entry(
+		__field(unsigned int, id)
+		__field(unsigned int, timestamp)
+		__field(unsigned int, inflight)
+	),
+	TP_fast_assign(
+		__entry->id = cmdbatch->context->id;
+		__entry->timestamp = cmdbatch->timestamp;
+		__entry->inflight = inflight;
+	),
+	TP_printk(
+		"ctx=%u ts=%u inflight=%u",
+			__entry->id, __entry->timestamp,
+			__entry->inflight
+	)
+);
+
+DEFINE_EVENT(adreno_cmdbatch_template, adreno_cmdbatch_retired,
+	TP_PROTO(struct kgsl_cmdbatch *cmdbatch, int inflight),
+	TP_ARGS(cmdbatch, inflight)
+);
+
+DEFINE_EVENT(adreno_cmdbatch_template, adreno_cmdbatch_submitted,
+	TP_PROTO(struct kgsl_cmdbatch *cmdbatch, int inflight),
+	TP_ARGS(cmdbatch, inflight)
+);
+
+DECLARE_EVENT_CLASS(adreno_drawctxt_template,
+	TP_PROTO(struct adreno_context *drawctxt),
+	TP_ARGS(drawctxt),
+	TP_STRUCT__entry(
+		__field(unsigned int, id)
+	),
+	TP_fast_assign(
+		__entry->id = drawctxt->base.id;
+	),
+	TP_printk("ctx=%u", __entry->id)
+);
+
+DEFINE_EVENT(adreno_drawctxt_template, adreno_drawctxt_sleep,
+	TP_PROTO(struct adreno_context *drawctxt),
+	TP_ARGS(drawctxt)
+);
+
+DEFINE_EVENT(adreno_drawctxt_template, adreno_drawctxt_wake,
+	TP_PROTO(struct adreno_context *drawctxt),
+	TP_ARGS(drawctxt)
+);
+
+DEFINE_EVENT(adreno_drawctxt_template, dispatch_queue_context,
+	TP_PROTO(struct adreno_context *drawctxt),
+	TP_ARGS(drawctxt)
+);
+
+DEFINE_EVENT(adreno_drawctxt_template, adreno_drawctxt_invalidate,
+	TP_PROTO(struct adreno_context *drawctxt),
+	TP_ARGS(drawctxt)
+);
+
+TRACE_EVENT(adreno_drawctxt_wait_start,
+	TP_PROTO(unsigned int id, unsigned int ts),
+	TP_ARGS(id, ts),
+	TP_STRUCT__entry(
+		__field(unsigned int, id)
+		__field(unsigned int, ts)
+	),
+	TP_fast_assign(
+		__entry->id = id;
+		__entry->ts = ts;
+	),
+	TP_printk(
+		"ctx=%u ts=%u",
+			__entry->id, __entry->ts
+	)
+);
+
+TRACE_EVENT(adreno_drawctxt_wait_done,
+	TP_PROTO(unsigned int id, unsigned int ts, int status),
+	TP_ARGS(id, ts, status),
+	TP_STRUCT__entry(
+		__field(unsigned int, id)
+		__field(unsigned int, ts)
+		__field(int, status)
+	),
+	TP_fast_assign(
+		__entry->id = id;
+		__entry->ts = ts;
+		__entry->status = status;
+	),
+	TP_printk(
+		"ctx=%u ts=%u status=%d",
+			__entry->id, __entry->ts, __entry->status
+	)
+);
+
+TRACE_EVENT(adreno_gpu_fault,
+	TP_PROTO(unsigned int status, unsigned int rptr, unsigned int wptr,
+		unsigned int ib1base, unsigned int ib1size,
+		unsigned int ib2base, unsigned int ib2size),
+	TP_ARGS(status, rptr, wptr, ib1base, ib1size, ib2base, ib2size),
+	TP_STRUCT__entry(
+		__field(unsigned int, status)
+		__field(unsigned int, rptr)
+		__field(unsigned int, wptr)
+		__field(unsigned int, ib1base)
+		__field(unsigned int, ib1size)
+		__field(unsigned int, ib2base)
+		__field(unsigned int, ib2size)
+	),
+	TP_fast_assign(
+		__entry->status = status;
+		__entry->rptr = rptr;
+		__entry->wptr = wptr;
+		__entry->ib1base = ib1base;
+		__entry->ib1size = ib1size;
+		__entry->ib2base = ib2base;
+		__entry->ib2size = ib2size;
+	),
+	TP_printk("status=%X RB=%X/%X IB1=%X/%X IB2=%X/%X",
+		__entry->status, __entry->wptr, __entry->rptr,
+		__entry->ib1base, __entry->ib1size, __entry->ib2base,
+		__entry->ib2size)
+);
+
+#endif /* _ADRENO_TRACE_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/drivers/gpu/msm/kgsl.c b/drivers/gpu/msm/kgsl.c
index fdd19e9..2624c16 100644
--- a/drivers/gpu/msm/kgsl.c
+++ b/drivers/gpu/msm/kgsl.c
@@ -62,59 +62,10 @@
 static void kgsl_mem_entry_detach_process(struct kgsl_mem_entry *entry);
 
 /**
- * kgsl_hang_check() - Check for GPU hang
- * data: KGSL device structure
- *
- * This function is called every KGSL_TIMEOUT_PART time when
- * GPU is active to check for hang. If a hang is detected we
- * trigger fault tolerance.
- */
-void kgsl_hang_check(struct work_struct *work)
-{
-	struct kgsl_device *device = container_of(work, struct kgsl_device,
-							hang_check_ws);
-	static unsigned int prev_reg_val[FT_DETECT_REGS_COUNT];
-
-	mutex_lock(&device->mutex);
-
-	if (device->state == KGSL_STATE_ACTIVE) {
-
-		/* Check to see if the GPU is hung */
-		if (adreno_ft_detect(device, prev_reg_val))
-			adreno_dump_and_exec_ft(device);
-
-		mod_timer(&device->hang_timer,
-			(jiffies + msecs_to_jiffies(KGSL_TIMEOUT_PART)));
-	}
-
-	mutex_unlock(&device->mutex);
-}
-
-/**
- * hang_timer() - Hang timer function
- * data: KGSL device structure
- *
- * This function is called when hang timer expires, in this
- * function we check if GPU is in active state and queue the
- * work on device workqueue to check for the hang. We restart
- * the timer after KGSL_TIMEOUT_PART time.
- */
-void hang_timer(unsigned long data)
-{
-	struct kgsl_device *device = (struct kgsl_device *) data;
-
-	if (device->state == KGSL_STATE_ACTIVE) {
-		/* Have work run in a non-interrupt context. */
-		queue_work(device->work_queue, &device->hang_check_ws);
-	}
-}
-
-/**
  * kgsl_trace_issueibcmds() - Call trace_issueibcmds by proxy
  * device: KGSL device
  * id: ID of the context submitting the command
- * ibdesc: Pointer to the list of IB descriptors
- * numib: Number of IBs in the list
+ * cmdbatch: Pointer to kgsl_cmdbatch describing these commands
  * timestamp: Timestamp assigned to the command batch
  * flags: Flags sent by the user
  * result: Result of the submission attempt
@@ -124,11 +75,11 @@
  * GPU specific modules.
  */
 void kgsl_trace_issueibcmds(struct kgsl_device *device, int id,
-		struct kgsl_ibdesc *ibdesc, int numibs,
+		struct kgsl_cmdbatch *cmdbatch,
 		unsigned int timestamp, unsigned int flags,
 		int result, unsigned int type)
 {
-	trace_kgsl_issueibcmds(device, id, ibdesc, numibs,
+	trace_kgsl_issueibcmds(device, id, cmdbatch,
 		timestamp, flags, result, type);
 }
 EXPORT_SYMBOL(kgsl_trace_issueibcmds);
@@ -530,8 +481,8 @@
 EXPORT_SYMBOL(kgsl_context_init);
 
 /**
- * kgsl_context_detach - Release the "master" context reference
- * @context - The context that will be detached
+ * kgsl_context_detach() - Release the "master" context reference
+ * @context: The context that will be detached
  *
  * This is called when a context becomes unusable, because userspace
  * has requested for it to be destroyed. The context itself may
@@ -540,14 +491,12 @@
  * detached by checking the KGSL_CONTEXT_DETACHED bit in
  * context->priv.
  */
-void
-kgsl_context_detach(struct kgsl_context *context)
+int kgsl_context_detach(struct kgsl_context *context)
 {
-	struct kgsl_device *device;
-	if (context == NULL)
-		return;
+	int ret;
 
-	device = context->device;
+	if (context == NULL)
+		return -EINVAL;
 
 	/*
 	 * Mark the context as detached to keep others from using
@@ -555,19 +504,22 @@
 	 * we don't try to detach twice.
 	 */
 	if (test_and_set_bit(KGSL_CONTEXT_DETACHED, &context->priv))
-		return;
+		return -EINVAL;
 
-	trace_kgsl_context_detach(device, context);
+	trace_kgsl_context_detach(context->device, context);
 
-	device->ftbl->drawctxt_detach(context);
+	ret = context->device->ftbl->drawctxt_detach(context);
+
 	/*
 	 * Cancel events after the device-specific context is
 	 * detached, to avoid possibly freeing memory while
 	 * it is still in use by the GPU.
 	 */
-	kgsl_context_cancel_events(device, context);
+	kgsl_context_cancel_events(context->device, context);
 
 	kgsl_context_put(context);
+
+	return ret;
 }
 
 void
@@ -579,6 +531,8 @@
 
 	trace_kgsl_context_destroy(device, context);
 
+	BUG_ON(!kgsl_context_detached(context));
+
 	write_lock(&device->context_lock);
 	if (context->id != KGSL_CONTEXT_INVALID) {
 		idr_remove(&device->context_idr, context->id);
@@ -649,10 +603,11 @@
 	policy_saved = device->pwrscale.policy;
 	device->pwrscale.policy = NULL;
 	kgsl_pwrctrl_request_state(device, KGSL_STATE_SUSPEND);
-	/*
-	 * Make sure no user process is waiting for a timestamp
-	 * before supending.
-	 */
+
+	/* Tell the device to drain the submission queue */
+	device->ftbl->drain(device);
+
+	/* Wait for the active count to hit zero */
 	kgsl_active_count_wait(device, 0);
 
 	/*
@@ -663,13 +618,10 @@
 
 	/* Don't let the timer wake us during suspended sleep. */
 	del_timer_sync(&device->idle_timer);
-	del_timer_sync(&device->hang_timer);
 	switch (device->state) {
 		case KGSL_STATE_INIT:
 			break;
 		case KGSL_STATE_ACTIVE:
-			/* Wait for the device to become idle */
-			device->ftbl->idle(device);
 		case KGSL_STATE_NAP:
 		case KGSL_STATE_SLEEP:
 			/* make sure power is on to stop the device */
@@ -995,8 +947,16 @@
 		if (context == NULL)
 			break;
 
-		if (context->dev_priv == dev_priv)
+		if (context->dev_priv == dev_priv) {
+			/*
+			 * Hold a reference to the context in case somebody
+			 * tries to put it while we are detaching
+			 */
+
+			_kgsl_context_get(context);
 			kgsl_context_detach(context);
+			kgsl_context_put(context);
+		}
 
 		next = next + 1;
 	}
@@ -1010,6 +970,7 @@
 
 	result = kgsl_close_device(device);
 	mutex_unlock(&device->mutex);
+
 	kfree(dev_priv);
 
 	kgsl_put_process_private(device, private);
@@ -1042,7 +1003,6 @@
 		 * Make sure the gates are open, so they don't block until
 		 * we start suspend or FT.
 		 */
-		complete_all(&device->ft_gate);
 		complete_all(&device->hwaccess_gate);
 		kgsl_pwrctrl_set_state(device, KGSL_STATE_ACTIVE);
 		kgsl_active_count_put(device);
@@ -1438,93 +1398,179 @@
 	return result;
 }
 
+/**
+ * kgsl_cmdbatch_create() - Create a new cmdbatch structure
+ * @context: Pointer to a KGSL context struct
+ * @numibs: Number of indirect buffers to make room for in the cmdbatch
+ *
+ * Allocate an new cmdbatch structure and add enough room to store the list of
+ * indirect buffers
+ */
+struct kgsl_cmdbatch *kgsl_cmdbatch_create(struct kgsl_context *context,
+	int numibs)
+{
+	struct kgsl_cmdbatch *cmdbatch = kzalloc(sizeof(*cmdbatch), GFP_KERNEL);
+	if (cmdbatch == NULL)
+		return ERR_PTR(-ENOMEM);
+
+	cmdbatch->ibdesc = kzalloc(sizeof(*cmdbatch->ibdesc) * numibs,
+		GFP_KERNEL);
+	if (cmdbatch->ibdesc == NULL) {
+		kfree(cmdbatch);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	cmdbatch->ibcount = numibs;
+	cmdbatch->context = context;
+
+	/*
+	 * Increase the reference count on the context so it doesn't disappear
+	 * during the lifetime of this command batch
+	 */
+	_kgsl_context_get(context);
+
+	return cmdbatch;
+}
+
+/**
+ * _kgsl_cmdbatch_verify() - Perform a quick sanity check on a command batch
+ * @device: Pointer to a KGSL instance that owns the command batch
+ * @pagetable: Pointer to the pagetable for the current process
+ * @cmdbatch: Number of indirect buffers to make room for in the cmdbatch
+ *
+ * Do a quick sanity test on the list of indirect buffers in a command batch
+ * verifying that the size and GPU address
+ */
+static bool _kgsl_cmdbatch_verify(struct kgsl_device_private *dev_priv,
+	struct kgsl_cmdbatch *cmdbatch)
+{
+	int i;
+	struct kgsl_process_private *private = dev_priv->process_priv;
+
+	for (i = 0; i < cmdbatch->ibcount; i++) {
+		if (cmdbatch->ibdesc[i].sizedwords == 0) {
+			KGSL_DRV_ERR(dev_priv->device,
+				"IB verification failed: Invalid size\n");
+			return false;
+		}
+
+		if (!kgsl_mmu_gpuaddr_in_range(private->pagetable,
+			cmdbatch->ibdesc[i].gpuaddr)) {
+			KGSL_DRV_ERR(dev_priv->device,
+				"IB verification failed: invalid address 0x%X\n",
+				cmdbatch->ibdesc[i].gpuaddr);
+			return false;
+		}
+	}
+
+	return true;
+}
+
+/**
+ * _kgsl_cmdbatch_create_legacy() - Create a cmdbatch from a legacy ioctl struct
+ * @context: Pointer to the KGSL context that issued the command batch
+ * @param: Pointer to the kgsl_ringbuffer_issueibcmds struct that the user sent
+ *
+ * Create a command batch from the legacy issueibcmds format.
+ */
+static struct kgsl_cmdbatch *_kgsl_cmdbatch_create_legacy(
+		struct kgsl_context *context,
+		struct kgsl_ringbuffer_issueibcmds *param)
+{
+	struct kgsl_cmdbatch *cmdbatch = kgsl_cmdbatch_create(context, 1);
+
+	if (IS_ERR(cmdbatch))
+		return cmdbatch;
+
+	cmdbatch->ibdesc[0].gpuaddr = param->ibdesc_addr;
+	cmdbatch->ibdesc[0].sizedwords = param->numibs;
+	cmdbatch->ibcount = 1;
+	cmdbatch->flags = param->flags;
+
+	return cmdbatch;
+}
+
+/**
+ * _kgsl_cmdbatch_create() - Create a cmdbatch from a ioctl struct
+ * @device: Pointer to the KGSL device for the GPU
+ * @context: Pointer to the KGSL context that issued the command batch
+ * @param: Pointer to the kgsl_ringbuffer_issueibcmds struct that the user sent
+ *
+ * Create a command batch from the standard issueibcmds format sent by the user.
+ */
+struct kgsl_cmdbatch *_kgsl_cmdbatch_create(struct kgsl_device *device,
+		struct kgsl_context *context,
+		struct kgsl_ringbuffer_issueibcmds *param)
+{
+	struct kgsl_cmdbatch *cmdbatch =
+		kgsl_cmdbatch_create(context, param->numibs);
+
+	if (IS_ERR(cmdbatch))
+		return cmdbatch;
+
+	if (copy_from_user(cmdbatch->ibdesc, (void *)param->ibdesc_addr,
+		sizeof(struct kgsl_ibdesc) * param->numibs)) {
+		KGSL_DRV_ERR(device,
+			"Unable to copy the IB userspace commands\n");
+		kgsl_cmdbatch_destroy(cmdbatch);
+		return ERR_PTR(-EFAULT);
+	}
+
+	cmdbatch->flags = param->flags & ~KGSL_CONTEXT_SUBMIT_IB_LIST;
+
+	return cmdbatch;
+}
+
 static long kgsl_ioctl_rb_issueibcmds(struct kgsl_device_private *dev_priv,
 				      unsigned int cmd, void *data)
 {
-	int result = 0;
-	int i = 0;
 	struct kgsl_ringbuffer_issueibcmds *param = data;
-	struct kgsl_ibdesc *ibdesc;
+	struct kgsl_device *device = dev_priv->device;
 	struct kgsl_context *context;
+	struct kgsl_cmdbatch *cmdbatch;
+	long result = -EINVAL;
 
 	context = kgsl_context_get_owner(dev_priv, param->drawctxt_id);
 	if (context == NULL) {
-		result = -EINVAL;
+		KGSL_DRV_ERR(device,
+			"Could not find context %d\n", param->drawctxt_id);
 		goto done;
 	}
 
 	if (param->flags & KGSL_CONTEXT_SUBMIT_IB_LIST) {
-		if (!param->numibs) {
-			result = -EINVAL;
-			goto done;
-		}
-
 		/*
-		 * Put a reasonable upper limit on the number of IBs that can be
-		 * submitted
+		 * Do a quick sanity check on the number of IBs in the
+		 * submission
 		 */
 
-		if (param->numibs > 10000) {
-			result = -EINVAL;
+		if (param->numibs == 0 || param->numibs > 100000) {
+			KGSL_DRV_ERR(device,
+				"Invalid number of IBs %d\n", param->numibs);
 			goto done;
 		}
 
-		ibdesc = kzalloc(sizeof(struct kgsl_ibdesc) * param->numibs,
-					GFP_KERNEL);
-		if (!ibdesc) {
-			KGSL_MEM_ERR(dev_priv->device,
-				"kzalloc(%d) failed\n",
-				sizeof(struct kgsl_ibdesc) * param->numibs);
-			result = -ENOMEM;
-			goto done;
-		}
+		cmdbatch = _kgsl_cmdbatch_create(device, context, param);
+	} else
+		cmdbatch = _kgsl_cmdbatch_create_legacy(context, param);
 
-		if (copy_from_user(ibdesc, (void *)param->ibdesc_addr,
-				sizeof(struct kgsl_ibdesc) * param->numibs)) {
-			result = -EFAULT;
-			KGSL_DRV_ERR(dev_priv->device,
-				"copy_from_user failed\n");
-			goto free_ibdesc;
-		}
-	} else {
-		KGSL_DRV_INFO(dev_priv->device,
-			"Using single IB submission mode for ib submission\n");
-		/* If user space driver is still using the old mode of
-		 * submitting single ib then we need to support that as well */
-		ibdesc = kzalloc(sizeof(struct kgsl_ibdesc), GFP_KERNEL);
-		if (!ibdesc) {
-			KGSL_MEM_ERR(dev_priv->device,
-				"kzalloc(%d) failed\n",
-				sizeof(struct kgsl_ibdesc));
-			result = -ENOMEM;
-			goto done;
-		}
-		ibdesc[0].gpuaddr = param->ibdesc_addr;
-		ibdesc[0].sizedwords = param->numibs;
-		param->numibs = 1;
+	if (IS_ERR(cmdbatch)) {
+		result = PTR_ERR(cmdbatch);
+		goto done;
 	}
 
-	for (i = 0; i < param->numibs; i++) {
-		struct kgsl_pagetable *pt = dev_priv->process_priv->pagetable;
-
-		if (!kgsl_mmu_gpuaddr_in_range(pt, ibdesc[i].gpuaddr)) {
-			result = -ERANGE;
-			KGSL_DRV_ERR(dev_priv->device,
-				     "invalid ib base GPU virtual addr %x\n",
-				     ibdesc[i].gpuaddr);
-			goto free_ibdesc;
-		}
+	/* Run basic sanity checking on the command */
+	if (!_kgsl_cmdbatch_verify(dev_priv, cmdbatch)) {
+		KGSL_DRV_ERR(device, "Unable to verify the IBs\n");
+		goto free_cmdbatch;
 	}
 
-	result = dev_priv->device->ftbl->issueibcmds(dev_priv,
-					     context,
-					     ibdesc,
-					     param->numibs,
-					     &param->timestamp,
-					     param->flags);
+	result = dev_priv->device->ftbl->issueibcmds(dev_priv, context,
+		cmdbatch, &param->timestamp);
 
-free_ibdesc:
-	kfree(ibdesc);
+free_cmdbatch:
+	if (result)
+		kgsl_cmdbatch_destroy(cmdbatch);
+
 done:
 	kgsl_context_put(context);
 	return result;
@@ -1665,14 +1711,11 @@
 {
 	struct kgsl_drawctxt_destroy *param = data;
 	struct kgsl_context *context;
-	long result = -EINVAL;
+	long result;
 
 	context = kgsl_context_get_owner(dev_priv, param->drawctxt_id);
 
-	if (context) {
-		kgsl_context_detach(context);
-		result = 0;
-	}
+	result = kgsl_context_detach(context);
 
 	kgsl_context_put(context);
 	return result;
@@ -2779,8 +2822,7 @@
 			kgsl_ioctl_device_waittimestamp_ctxtid,
 			KGSL_IOCTL_LOCK | KGSL_IOCTL_WAKE),
 	KGSL_IOCTL_FUNC(IOCTL_KGSL_RINGBUFFER_ISSUEIBCMDS,
-			kgsl_ioctl_rb_issueibcmds,
-			KGSL_IOCTL_LOCK | KGSL_IOCTL_WAKE),
+			kgsl_ioctl_rb_issueibcmds, 0),
 	KGSL_IOCTL_FUNC(IOCTL_KGSL_CMDSTREAM_READTIMESTAMP,
 			kgsl_ioctl_cmdstream_readtimestamp,
 			KGSL_IOCTL_LOCK),
@@ -3462,7 +3504,6 @@
 
 
 	setup_timer(&device->idle_timer, kgsl_timer, (unsigned long) device);
-	setup_timer(&device->hang_timer, hang_timer, (unsigned long) device);
 	status = kgsl_create_device_workqueue(device);
 	if (status)
 		goto error_pwrctrl_close;
@@ -3522,7 +3563,6 @@
 
 		if (device->state == KGSL_STATE_ACTIVE)
 			kgsl_idle(device);
-
 	}
 
 	if (device->pm_dump_enable) {
@@ -3536,13 +3576,12 @@
 			pwr->power_flags, pwr->active_pwrlevel);
 
 		KGSL_LOG_DUMP(device, "POWER: INTERVAL TIMEOUT = %08X ",
-			pwr->interval_timeout);
+				pwr->interval_timeout);
 
 	}
 
 	/* Disable the idle timer so we don't get interrupted */
 	del_timer_sync(&device->idle_timer);
-	del_timer_sync(&device->hang_timer);
 
 	/* Force on the clocks */
 	kgsl_pwrctrl_wake(device);
diff --git a/drivers/gpu/msm/kgsl.h b/drivers/gpu/msm/kgsl.h
index 8d390a9..de647d5 100644
--- a/drivers/gpu/msm/kgsl.h
+++ b/drivers/gpu/msm/kgsl.h
@@ -141,6 +141,7 @@
 
 struct kgsl_pagetable;
 struct kgsl_memdesc;
+struct kgsl_cmdbatch;
 
 struct kgsl_memdesc_ops {
 	int (*vmflags)(struct kgsl_memdesc *);
@@ -205,7 +206,6 @@
 #define MMU_CONFIG 1
 #endif
 
-void kgsl_hang_check(struct work_struct *work);
 void kgsl_mem_entry_destroy(struct kref *kref);
 int kgsl_postmortem_dump(struct kgsl_device *device, int manual);
 
@@ -237,7 +237,7 @@
 		unsigned int value);
 
 void kgsl_trace_issueibcmds(struct kgsl_device *device, int id,
-		struct kgsl_ibdesc *ibdesc, int numibs,
+		struct kgsl_cmdbatch *cmdbatch,
 		unsigned int timestamp, unsigned int flags,
 		int result, unsigned int type);
 
diff --git a/drivers/gpu/msm/kgsl_debugfs.c b/drivers/gpu/msm/kgsl_debugfs.c
index 9ab8d22..110264b 100644
--- a/drivers/gpu/msm/kgsl_debugfs.c
+++ b/drivers/gpu/msm/kgsl_debugfs.c
@@ -123,7 +123,6 @@
 KGSL_DEBUGFS_LOG(ctxt_log);
 KGSL_DEBUGFS_LOG(mem_log);
 KGSL_DEBUGFS_LOG(pwr_log);
-KGSL_DEBUGFS_LOG(ft_log);
 
 static int memfree_hist_print(struct seq_file *s, void *unused)
 {
@@ -185,7 +184,6 @@
 	device->drv_log = KGSL_LOG_LEVEL_DEFAULT;
 	device->mem_log = KGSL_LOG_LEVEL_DEFAULT;
 	device->pwr_log = KGSL_LOG_LEVEL_DEFAULT;
-	device->ft_log = KGSL_LOG_LEVEL_DEFAULT;
 
 	debugfs_create_file("log_level_cmd", 0644, device->d_debugfs, device,
 			    &cmd_log_fops);
@@ -199,8 +197,6 @@
 				&pwr_log_fops);
 	debugfs_create_file("memfree_history", 0444, device->d_debugfs, device,
 				&memfree_hist_fops);
-	debugfs_create_file("log_level_ft", 0644, device->d_debugfs, device,
-				&ft_log_fops);
 
 	/* Create postmortem dump control files */
 
diff --git a/drivers/gpu/msm/kgsl_device.h b/drivers/gpu/msm/kgsl_device.h
index 09a31c9..f5b27d0 100644
--- a/drivers/gpu/msm/kgsl_device.h
+++ b/drivers/gpu/msm/kgsl_device.h
@@ -13,6 +13,7 @@
 #ifndef __KGSL_DEVICE_H
 #define __KGSL_DEVICE_H
 
+#include <linux/slab.h>
 #include <linux/idr.h>
 #include <linux/pm_qos.h>
 #include <linux/sched.h>
@@ -76,6 +77,7 @@
 struct kgsl_context;
 struct kgsl_power_stats;
 struct kgsl_event;
+struct kgsl_cmdbatch;
 
 struct kgsl_functable {
 	/* Mandatory functions - these functions must be implemented
@@ -87,7 +89,7 @@
 	void (*regwrite) (struct kgsl_device *device,
 		unsigned int offsetwords, unsigned int value);
 	int (*idle) (struct kgsl_device *device);
-	unsigned int (*isidle) (struct kgsl_device *device);
+	bool (*isidle) (struct kgsl_device *device);
 	int (*suspend_context) (struct kgsl_device *device);
 	int (*init) (struct kgsl_device *device);
 	int (*start) (struct kgsl_device *device);
@@ -101,9 +103,8 @@
 	unsigned int (*readtimestamp) (struct kgsl_device *device,
 		struct kgsl_context *context, enum kgsl_timestamp_type type);
 	int (*issueibcmds) (struct kgsl_device_private *dev_priv,
-		struct kgsl_context *context, struct kgsl_ibdesc *ibdesc,
-		unsigned int sizedwords, uint32_t *timestamp,
-		unsigned int flags);
+		struct kgsl_context *context, struct kgsl_cmdbatch *cmdbatch,
+		uint32_t *timestamps);
 	int (*setup_pt)(struct kgsl_device *device,
 		struct kgsl_pagetable *pagetable);
 	void (*cleanup_pt)(struct kgsl_device *device,
@@ -115,14 +116,15 @@
 	void * (*snapshot)(struct kgsl_device *device, void *snapshot,
 		int *remain, int hang);
 	irqreturn_t (*irq_handler)(struct kgsl_device *device);
+	int (*drain)(struct kgsl_device *device);
 	/* Optional functions - these functions are not mandatory.  The
 	   driver will check that the function pointer is not NULL before
 	   calling the hook */
-	void (*setstate) (struct kgsl_device *device, unsigned int context_id,
+	int (*setstate) (struct kgsl_device *device, unsigned int context_id,
 			uint32_t flags);
 	struct kgsl_context *(*drawctxt_create) (struct kgsl_device_private *,
 						uint32_t *flags);
-	void (*drawctxt_detach) (struct kgsl_context *context);
+	int (*drawctxt_detach) (struct kgsl_context *context);
 	void (*drawctxt_destroy) (struct kgsl_context *context);
 	long (*ioctl) (struct kgsl_device_private *dev_priv,
 		unsigned int cmd, void *data);
@@ -155,6 +157,26 @@
 	unsigned int created;
 };
 
+/**
+ * struct kgsl_cmdbatch - KGSl command descriptor
+ * @context: KGSL context that created the command
+ * @timestamp: Timestamp assigned to the command (currently unused)
+ * @flags: flags
+ * @ibcount: Number of IBs in the command list
+ * @ibdesc: Pointer to the list of IBs
+ * @expires: Point in time when the cmdbatch is considered to be hung
+ * @invalid:  non-zero if the dispatcher determines the command and the owning
+ * context should be invalidated
+ */
+struct kgsl_cmdbatch {
+	struct kgsl_context *context;
+	uint32_t timestamp;
+	uint32_t flags;
+	uint32_t ibcount;
+	struct kgsl_ibdesc *ibdesc;
+	unsigned long expires;
+	int invalid;
+};
 
 struct kgsl_device {
 	struct device *dev;
@@ -190,9 +212,7 @@
 	struct completion hwaccess_gate;
 	const struct kgsl_functable *ftbl;
 	struct work_struct idle_check_ws;
-	struct work_struct hang_check_ws;
 	struct timer_list idle_timer;
-	struct timer_list hang_timer;
 	struct kgsl_pwrctrl pwrctrl;
 	int open_count;
 
@@ -206,7 +226,6 @@
 	wait_queue_head_t active_cnt_wq;
 	struct workqueue_struct *work_queue;
 	struct device *parentdev;
-	struct completion ft_gate;
 	struct dentry *d_debugfs;
 	struct idr context_idr;
 	rwlock_t context_lock;
@@ -233,7 +252,6 @@
 	int drv_log;
 	int mem_log;
 	int pwr_log;
-	int ft_log;
 	int pm_dump_enable;
 	struct kgsl_pwrscale pwrscale;
 	struct kobject pwrscale_kobj;
@@ -254,11 +272,8 @@
 
 #define KGSL_DEVICE_COMMON_INIT(_dev) \
 	.hwaccess_gate = COMPLETION_INITIALIZER((_dev).hwaccess_gate),\
-	.ft_gate = COMPLETION_INITIALIZER((_dev).ft_gate),\
 	.idle_check_ws = __WORK_INITIALIZER((_dev).idle_check_ws,\
 			kgsl_idle_check),\
-	.hang_check_ws = __WORK_INITIALIZER((_dev).hang_check_ws,\
-			kgsl_hang_check),\
 	.ts_expired_ws  = __WORK_INITIALIZER((_dev).ts_expired_ws,\
 			kgsl_process_events),\
 	.context_idr = IDR_INIT((_dev).context_idr),\
@@ -440,8 +455,6 @@
 	return 0;
 }
 
-
-
 int kgsl_check_timestamp(struct kgsl_device *device,
 		struct kgsl_context *context, unsigned int timestamp);
 
@@ -597,4 +610,21 @@
 {
 	kgsl_signal_event(device, context, timestamp, KGSL_EVENT_CANCELLED);
 }
+
+/**
+ * kgsl_cmdbatch_destroy() - Destroy a command batch structure
+ * @cmdbatch: Pointer to the command batch to destroy
+ *
+ * Destroy and free a command batch
+ */
+static inline void kgsl_cmdbatch_destroy(struct kgsl_cmdbatch *cmdbatch)
+{
+	if (cmdbatch) {
+		kgsl_context_put(cmdbatch->context);
+		kfree(cmdbatch->ibdesc);
+	}
+
+	kfree(cmdbatch);
+}
+
 #endif  /* __KGSL_DEVICE_H */
diff --git a/drivers/gpu/msm/kgsl_gpummu.c b/drivers/gpu/msm/kgsl_gpummu.c
index 68052b1..2634e4f 100644
--- a/drivers/gpu/msm/kgsl_gpummu.c
+++ b/drivers/gpu/msm/kgsl_gpummu.c
@@ -482,15 +482,17 @@
 	return NULL;
 }
 
-static void kgsl_gpummu_default_setstate(struct kgsl_mmu *mmu,
+static int kgsl_gpummu_default_setstate(struct kgsl_mmu *mmu,
 					uint32_t flags)
 {
 	struct kgsl_gpummu_pt *gpummu_pt;
 	if (!kgsl_mmu_enabled())
-		return;
+		return 0;
 
 	if (flags & KGSL_MMUFLAGS_PTUPDATE) {
-		kgsl_idle(mmu->device);
+		int ret = kgsl_idle(mmu->device);
+		if (ret)
+			return ret;
 		gpummu_pt = mmu->hwpagetable->priv;
 		kgsl_regwrite(mmu->device, MH_MMU_PT_BASE,
 			gpummu_pt->base.gpuaddr);
@@ -500,12 +502,16 @@
 		/* Invalidate all and tc */
 		kgsl_regwrite(mmu->device, MH_MMU_INVALIDATE,  0x00000003);
 	}
+
+	return 0;
 }
 
-static void kgsl_gpummu_setstate(struct kgsl_mmu *mmu,
+static int kgsl_gpummu_setstate(struct kgsl_mmu *mmu,
 				struct kgsl_pagetable *pagetable,
 				unsigned int context_id)
 {
+	int ret = 0;
+
 	if (mmu->flags & KGSL_FLAGS_STARTED) {
 		/* page table not current, then setup mmu to use new
 		 *  specified page table
@@ -518,10 +524,13 @@
 			kgsl_mmu_pt_get_flags(pagetable, mmu->device->id);
 
 			/* call device specific set page table */
-			kgsl_setstate(mmu, context_id, KGSL_MMUFLAGS_TLBFLUSH |
+			ret = kgsl_setstate(mmu, context_id,
+				KGSL_MMUFLAGS_TLBFLUSH |
 				KGSL_MMUFLAGS_PTUPDATE);
 		}
 	}
+
+	return ret;
 }
 
 static int kgsl_gpummu_init(struct kgsl_mmu *mmu)
@@ -563,6 +572,7 @@
 
 	struct kgsl_device *device = mmu->device;
 	struct kgsl_gpummu_pt *gpummu_pt;
+	int ret;
 
 	if (mmu->flags & KGSL_FLAGS_STARTED)
 		return 0;
@@ -574,9 +584,6 @@
 	/* setup MMU and sub-client behavior */
 	kgsl_regwrite(device, MH_MMU_CONFIG, mmu->config);
 
-	/* idle device */
-	kgsl_idle(device);
-
 	/* enable axi interrupts */
 	kgsl_regwrite(device, MH_INTERRUPT_MASK,
 			GSL_MMU_INT_MASK | MH_INTERRUPT_MASK__MMU_PAGE_FAULT);
@@ -607,10 +614,12 @@
 	kgsl_regwrite(mmu->device, MH_MMU_VA_RANGE,
 		      (KGSL_PAGETABLE_BASE |
 		      (CONFIG_MSM_KGSL_PAGE_TABLE_SIZE >> 16)));
-	kgsl_setstate(mmu, KGSL_MEMSTORE_GLOBAL, KGSL_MMUFLAGS_TLBFLUSH);
-	mmu->flags |= KGSL_FLAGS_STARTED;
 
-	return 0;
+	ret = kgsl_setstate(mmu, KGSL_MEMSTORE_GLOBAL, KGSL_MMUFLAGS_TLBFLUSH);
+	if (!ret)
+		mmu->flags |= KGSL_FLAGS_STARTED;
+
+	return ret;
 }
 
 static int
diff --git a/drivers/gpu/msm/kgsl_iommu.c b/drivers/gpu/msm/kgsl_iommu.c
index ecda5a7..103736d 100644
--- a/drivers/gpu/msm/kgsl_iommu.c
+++ b/drivers/gpu/msm/kgsl_iommu.c
@@ -1205,10 +1205,12 @@
 	return 0;
 }
 
-static void kgsl_iommu_setstate(struct kgsl_mmu *mmu,
+static int kgsl_iommu_setstate(struct kgsl_mmu *mmu,
 				struct kgsl_pagetable *pagetable,
 				unsigned int context_id)
 {
+	int ret = 0;
+
 	if (mmu->flags & KGSL_FLAGS_STARTED) {
 		/* page table not current, then setup mmu to use new
 		 *  specified page table
@@ -1219,10 +1221,12 @@
 			flags |= kgsl_mmu_pt_get_flags(mmu->hwpagetable,
 							mmu->device->id) |
 							KGSL_MMUFLAGS_TLBFLUSH;
-			kgsl_setstate(mmu, context_id,
+			ret = kgsl_setstate(mmu, context_id,
 				KGSL_MMUFLAGS_PTUPDATE | flags);
 		}
 	}
+
+	return ret;
 }
 
 /*
@@ -1892,31 +1896,40 @@
  * cpu
  * Return - void
  */
-static void kgsl_iommu_default_setstate(struct kgsl_mmu *mmu,
+static int kgsl_iommu_default_setstate(struct kgsl_mmu *mmu,
 					uint32_t flags)
 {
 	struct kgsl_iommu *iommu = mmu->priv;
 	int temp;
 	int i;
+	int ret = 0;
 	phys_addr_t pt_base = kgsl_iommu_get_pt_base_addr(mmu,
 						mmu->hwpagetable);
 	phys_addr_t pt_val;
 
-	if (kgsl_iommu_enable_clk(mmu, KGSL_IOMMU_CONTEXT_USER)) {
+	ret = kgsl_iommu_enable_clk(mmu, KGSL_IOMMU_CONTEXT_USER);
+
+	if (ret) {
 		KGSL_DRV_ERR(mmu->device, "Failed to enable iommu clocks\n");
-		return;
+		return ret;
 	}
 
 	/* For v0 SMMU GPU needs to be idle for tlb invalidate as well */
-	if (msm_soc_version_supports_iommu_v0())
-		kgsl_idle(mmu->device);
+	if (msm_soc_version_supports_iommu_v0()) {
+		ret = kgsl_idle(mmu->device);
+		if (ret)
+			return ret;
+	}
 
 	/* Acquire GPU-CPU sync Lock here */
 	_iommu_lock();
 
 	if (flags & KGSL_MMUFLAGS_PTUPDATE) {
-		if (!msm_soc_version_supports_iommu_v0())
-			kgsl_idle(mmu->device);
+		if (!msm_soc_version_supports_iommu_v0()) {
+			ret = kgsl_idle(mmu->device);
+			if (ret)
+				goto unlock;
+		}
 		for (i = 0; i < iommu->unit_count; i++) {
 			/* get the lsb value which should not change when
 			 * changing ttbr0 */
@@ -1977,12 +1990,13 @@
 			}
 		}
 	}
-
+unlock:
 	/* Release GPU-CPU sync Lock here */
 	_iommu_unlock();
 
 	/* Disable smmu clock */
 	kgsl_iommu_disable_clk_on_ts(mmu, 0, false);
+	return ret;
 }
 
 /*
@@ -2039,6 +2053,7 @@
 	.mmu_pagefault_resume = kgsl_iommu_pagefault_resume,
 	.mmu_get_current_ptbase = kgsl_iommu_get_current_ptbase,
 	.mmu_enable_clk = kgsl_iommu_enable_clk,
+	.mmu_disable_clk = kgsl_iommu_disable_clk,
 	.mmu_disable_clk_on_ts = kgsl_iommu_disable_clk_on_ts,
 	.mmu_get_default_ttbr0 = kgsl_iommu_get_default_ttbr0,
 	.mmu_get_reg_gpuaddr = kgsl_iommu_get_reg_gpuaddr,
diff --git a/drivers/gpu/msm/kgsl_log.h b/drivers/gpu/msm/kgsl_log.h
index a7832e4..3a32953 100644
--- a/drivers/gpu/msm/kgsl_log.h
+++ b/drivers/gpu/msm/kgsl_log.h
@@ -103,15 +103,6 @@
 #define KGSL_PWR_CRIT(_dev, fmt, args...) \
 KGSL_LOG_CRIT(_dev->dev, _dev->pwr_log, fmt, ##args)
 
-#define KGSL_FT_INFO(_dev, fmt, args...) \
-KGSL_LOG_INFO(_dev->dev, _dev->ft_log, fmt, ##args)
-#define KGSL_FT_WARN(_dev, fmt, args...) \
-KGSL_LOG_WARN(_dev->dev, _dev->ft_log, fmt, ##args)
-#define KGSL_FT_ERR(_dev, fmt, args...) \
-KGSL_LOG_ERR(_dev->dev, _dev->ft_log, fmt, ##args)
-#define KGSL_FT_CRIT(_dev, fmt, args...) \
-KGSL_LOG_CRIT(_dev->dev, _dev->ft_log, fmt, ##args)
-
 /* Core error messages - these are for core KGSL functions that have
    no device associated with them (such as memory) */
 
diff --git a/drivers/gpu/msm/kgsl_mmu.c b/drivers/gpu/msm/kgsl_mmu.c
index 952019f..6635a7c 100644
--- a/drivers/gpu/msm/kgsl_mmu.c
+++ b/drivers/gpu/msm/kgsl_mmu.c
@@ -566,7 +566,7 @@
 }
 EXPORT_SYMBOL(kgsl_mmu_putpagetable);
 
-void kgsl_setstate(struct kgsl_mmu *mmu, unsigned int context_id,
+int kgsl_setstate(struct kgsl_mmu *mmu, unsigned int context_id,
 			uint32_t flags)
 {
 	struct kgsl_device *device = mmu->device;
@@ -574,14 +574,16 @@
 
 	if (!(flags & (KGSL_MMUFLAGS_TLBFLUSH | KGSL_MMUFLAGS_PTUPDATE))
 		&& !adreno_is_a2xx(adreno_dev))
-		return;
+		return 0;
 
 	if (KGSL_MMU_TYPE_NONE == kgsl_mmu_type)
-		return;
+		return 0;
 	else if (device->ftbl->setstate)
-		device->ftbl->setstate(device, context_id, flags);
+		return device->ftbl->setstate(device, context_id, flags);
 	else if (mmu->mmu_ops->mmu_device_setstate)
-		mmu->mmu_ops->mmu_device_setstate(mmu, flags);
+		return mmu->mmu_ops->mmu_device_setstate(mmu, flags);
+
+	return 0;
 }
 EXPORT_SYMBOL(kgsl_setstate);
 
@@ -590,7 +592,6 @@
 	struct kgsl_mh *mh = &device->mh;
 	/* force mmu off to for now*/
 	kgsl_regwrite(device, MH_MMU_CONFIG, 0);
-	kgsl_idle(device);
 
 	/* define physical memory range accessible by the core */
 	kgsl_regwrite(device, MH_MMU_MPU_BASE, mh->mpu_base);
diff --git a/drivers/gpu/msm/kgsl_mmu.h b/drivers/gpu/msm/kgsl_mmu.h
index faba81e..a30ee3f 100644
--- a/drivers/gpu/msm/kgsl_mmu.h
+++ b/drivers/gpu/msm/kgsl_mmu.h
@@ -133,10 +133,10 @@
 	int (*mmu_close) (struct kgsl_mmu *mmu);
 	int (*mmu_start) (struct kgsl_mmu *mmu);
 	void (*mmu_stop) (struct kgsl_mmu *mmu);
-	void (*mmu_setstate) (struct kgsl_mmu *mmu,
+	int (*mmu_setstate) (struct kgsl_mmu *mmu,
 		struct kgsl_pagetable *pagetable,
 		unsigned int context_id);
-	void (*mmu_device_setstate) (struct kgsl_mmu *mmu,
+	int (*mmu_device_setstate) (struct kgsl_mmu *mmu,
 					uint32_t flags);
 	void (*mmu_pagefault) (struct kgsl_mmu *mmu);
 	phys_addr_t (*mmu_get_current_ptbase)
@@ -147,6 +147,8 @@
 		(struct kgsl_mmu *mmu, uint32_t ts, bool ts_valid);
 	int (*mmu_enable_clk)
 		(struct kgsl_mmu *mmu, int ctx_id);
+	void (*mmu_disable_clk)
+		(struct kgsl_mmu *mmu);
 	phys_addr_t (*mmu_get_default_ttbr0)(struct kgsl_mmu *mmu,
 				unsigned int unit_id,
 				enum kgsl_iommu_context_id ctx_id);
@@ -231,7 +233,7 @@
 int kgsl_mmu_put_gpuaddr(struct kgsl_pagetable *pagetable,
 		 struct kgsl_memdesc *memdesc);
 unsigned int kgsl_virtaddr_to_physaddr(void *virtaddr);
-void kgsl_setstate(struct kgsl_mmu *mmu, unsigned int context_id,
+int kgsl_setstate(struct kgsl_mmu *mmu, unsigned int context_id,
 			uint32_t flags);
 int kgsl_mmu_get_ptname_from_ptbase(struct kgsl_mmu *mmu,
 					phys_addr_t pt_base);
@@ -260,19 +262,23 @@
 		return 0;
 }
 
-static inline void kgsl_mmu_setstate(struct kgsl_mmu *mmu,
+static inline int kgsl_mmu_setstate(struct kgsl_mmu *mmu,
 			struct kgsl_pagetable *pagetable,
 			unsigned int context_id)
 {
 	if (mmu->mmu_ops && mmu->mmu_ops->mmu_setstate)
-		mmu->mmu_ops->mmu_setstate(mmu, pagetable, context_id);
+		return mmu->mmu_ops->mmu_setstate(mmu, pagetable, context_id);
+
+	return 0;
 }
 
-static inline void kgsl_mmu_device_setstate(struct kgsl_mmu *mmu,
+static inline int kgsl_mmu_device_setstate(struct kgsl_mmu *mmu,
 						uint32_t flags)
 {
 	if (mmu->mmu_ops && mmu->mmu_ops->mmu_device_setstate)
-		mmu->mmu_ops->mmu_device_setstate(mmu, flags);
+		return mmu->mmu_ops->mmu_device_setstate(mmu, flags);
+
+	return 0;
 }
 
 static inline void kgsl_mmu_stop(struct kgsl_mmu *mmu)
@@ -320,6 +326,12 @@
 		return 0;
 }
 
+static inline void kgsl_mmu_disable_clk(struct kgsl_mmu *mmu)
+{
+	if (mmu->mmu_ops && mmu->mmu_ops->mmu_disable_clk)
+		mmu->mmu_ops->mmu_disable_clk(mmu);
+}
+
 static inline void kgsl_mmu_disable_clk_on_ts(struct kgsl_mmu *mmu,
 						unsigned int ts, bool ts_valid)
 {
diff --git a/drivers/gpu/msm/kgsl_pwrctrl.c b/drivers/gpu/msm/kgsl_pwrctrl.c
index 1a95761..07131f7 100644
--- a/drivers/gpu/msm/kgsl_pwrctrl.c
+++ b/drivers/gpu/msm/kgsl_pwrctrl.c
@@ -1215,9 +1215,6 @@
 		} else {
 			device->pwrctrl.irq_last = 0;
 		}
-	} else if (device->state & (KGSL_STATE_HUNG |
-					KGSL_STATE_DUMP_AND_FT)) {
-		kgsl_pwrctrl_request_state(device, KGSL_STATE_NONE);
 	}
 
 	mutex_unlock(&device->mutex);
@@ -1273,7 +1270,6 @@
 			kgsl_pwrctrl_request_state(device, KGSL_STATE_NONE);
 			return -EBUSY;
 		}
-		del_timer_sync(&device->hang_timer);
 		kgsl_pwrctrl_irq(device, KGSL_PWRFLAGS_OFF);
 		kgsl_pwrctrl_clk(device, KGSL_PWRFLAGS_OFF, KGSL_STATE_NAP);
 		kgsl_pwrctrl_set_state(device, KGSL_STATE_NAP);
@@ -1343,7 +1339,6 @@
 	case KGSL_STATE_NAP:
 	case KGSL_STATE_SLEEP:
 		del_timer_sync(&device->idle_timer);
-		del_timer_sync(&device->hang_timer);
 		/* make sure power is on to stop the device*/
 		kgsl_pwrctrl_enable(device);
 		device->ftbl->suspend_context(device);
@@ -1435,8 +1430,6 @@
 		kgsl_pwrctrl_set_state(device, KGSL_STATE_ACTIVE);
 		kgsl_pwrctrl_irq(device, KGSL_PWRFLAGS_ON);
 
-		mod_timer(&device->hang_timer,
-			(jiffies + msecs_to_jiffies(KGSL_TIMEOUT_PART)));
 		pm_qos_update_request(&device->pwrctrl.pm_qos_req_dma,
 				device->pwrctrl.pm_qos_latency);
 	case KGSL_STATE_ACTIVE:
@@ -1504,10 +1497,6 @@
 		return "SLEEP";
 	case KGSL_STATE_SUSPEND:
 		return "SUSPEND";
-	case KGSL_STATE_HUNG:
-		return "HUNG";
-	case KGSL_STATE_DUMP_AND_FT:
-		return "DNR";
 	case KGSL_STATE_SLUMBER:
 		return "SLUMBER";
 	default:
@@ -1539,7 +1528,6 @@
 		(device->state != KGSL_STATE_ACTIVE)) {
 		mutex_unlock(&device->mutex);
 		wait_for_completion(&device->hwaccess_gate);
-		wait_for_completion(&device->ft_gate);
 		mutex_lock(&device->mutex);
 
 		/* Stop the idle timer */
diff --git a/drivers/gpu/msm/kgsl_trace.h b/drivers/gpu/msm/kgsl_trace.h
index f16f2b4..179a72b 100644
--- a/drivers/gpu/msm/kgsl_trace.h
+++ b/drivers/gpu/msm/kgsl_trace.h
@@ -37,14 +37,13 @@
 
 	TP_PROTO(struct kgsl_device *device,
 			int drawctxt_id,
-			struct kgsl_ibdesc *ibdesc,
-			int numibs,
+			struct kgsl_cmdbatch *cmdbatch,
 			int timestamp,
 			int flags,
 			int result,
 			unsigned int type),
 
-	TP_ARGS(device, drawctxt_id, ibdesc, numibs, timestamp, flags,
+	TP_ARGS(device, drawctxt_id, cmdbatch, timestamp, flags,
 		result, type),
 
 	TP_STRUCT__entry(
@@ -61,8 +60,8 @@
 	TP_fast_assign(
 		__assign_str(device_name, device->name);
 		__entry->drawctxt_id = drawctxt_id;
-		__entry->ibdesc_addr = ibdesc[0].gpuaddr;
-		__entry->numibs = numibs;
+		__entry->ibdesc_addr = cmdbatch->ibdesc[0].gpuaddr;
+		__entry->numibs = cmdbatch->ibcount;
 		__entry->timestamp = timestamp;
 		__entry->flags = flags;
 		__entry->result = result;
diff --git a/drivers/gpu/msm/z180.c b/drivers/gpu/msm/z180.c
index 883417f..0af57aa 100644
--- a/drivers/gpu/msm/z180.c
+++ b/drivers/gpu/msm/z180.c
@@ -353,7 +353,13 @@
 	return ts_diff < Z180_PACKET_COUNT;
 }
 
-static int z180_idle(struct kgsl_device *device)
+/**
+ * z180_idle() - Idle the 2D device
+ * @device: Pointer to the KGSL device struct for the Z180
+ *
+ * wait until the z180 submission queue is idle
+ */
+int z180_idle(struct kgsl_device *device)
 {
 	int status = 0;
 	struct z180_device *z180_dev = Z180_DEVICE(device);
@@ -373,10 +379,8 @@
 int
 z180_cmdstream_issueibcmds(struct kgsl_device_private *dev_priv,
 			struct kgsl_context *context,
-			struct kgsl_ibdesc *ibdesc,
-			unsigned int numibs,
-			uint32_t *timestamp,
-			unsigned int ctrl)
+			struct kgsl_cmdbatch *cmdbatch,
+			uint32_t *timestamp)
 {
 	long result = 0;
 	unsigned int ofs        = PACKETSIZE_STATESTREAM * sizeof(unsigned int);
@@ -389,6 +393,20 @@
 	struct kgsl_pagetable *pagetable = dev_priv->process_priv->pagetable;
 	struct z180_device *z180_dev = Z180_DEVICE(device);
 	unsigned int sizedwords;
+	unsigned int numibs;
+	struct kgsl_ibdesc *ibdesc;
+
+	mutex_lock(&device->mutex);
+
+	kgsl_active_count_get(device);
+
+	if (cmdbatch == NULL) {
+		result = EINVAL;
+		goto error;
+	}
+
+	ibdesc = cmdbatch->ibdesc;
+	numibs = cmdbatch->ibcount;
 
 	if (device->state & KGSL_STATE_HUNG) {
 		result = -EINVAL;
@@ -430,7 +448,7 @@
 		context->id, cmd, sizedwords);
 	/* context switch */
 	if ((context->id != (int)z180_dev->ringbuffer.prevctx) ||
-	    (ctrl & KGSL_CONTEXT_CTX_SWITCH)) {
+	    (cmdbatch->flags & KGSL_CONTEXT_CTX_SWITCH)) {
 		KGSL_CMD_INFO(device, "context switch %d -> %d\n",
 			context->id, z180_dev->ringbuffer.prevctx);
 		kgsl_mmu_setstate(&device->mmu, pagetable,
@@ -438,10 +456,13 @@
 		cnt = PACKETSIZE_STATESTREAM;
 		ofs = 0;
 	}
-	kgsl_setstate(&device->mmu,
+
+	result = kgsl_setstate(&device->mmu,
 			KGSL_MEMSTORE_GLOBAL,
 			kgsl_mmu_pt_get_flags(device->mmu.hwpagetable,
 			device->id));
+	if (result < 0)
+		goto error;
 
 	result = wait_event_interruptible_timeout(device->wait_queue,
 				  room_in_rb(z180_dev),
@@ -482,9 +503,12 @@
 	z180_cmdwindow_write(device, ADDR_VGV3_CONTROL, cmd);
 	z180_cmdwindow_write(device, ADDR_VGV3_CONTROL, 0);
 error:
+	kgsl_trace_issueibcmds(device, context->id, cmdbatch,
+		*timestamp, cmdbatch->flags, result, 0);
 
-	kgsl_trace_issueibcmds(device, context->id, ibdesc, numibs,
-		*timestamp, ctrl, result, 0);
+	kgsl_active_count_put(device);
+
+	mutex_unlock(&device->mutex);
 
 	return (int)result;
 }
@@ -595,8 +619,12 @@
 
 static int z180_stop(struct kgsl_device *device)
 {
+	int ret;
+
 	device->ftbl->irqctrl(device, 0);
-	z180_idle(device);
+	ret = z180_idle(device);
+	if (ret)
+		return ret;
 
 	del_timer_sync(&device->idle_timer);
 
@@ -662,7 +690,7 @@
 	return status;
 }
 
-static unsigned int z180_isidle(struct kgsl_device *device)
+static bool z180_isidle(struct kgsl_device *device)
 {
 	struct z180_device *z180_dev = Z180_DEVICE(device);
 
@@ -875,7 +903,7 @@
 	return context;
 }
 
-static void
+static int
 z180_drawctxt_detach(struct kgsl_context *context)
 {
 	struct kgsl_device *device;
@@ -889,9 +917,13 @@
 	if (z180_dev->ringbuffer.prevctx == context->id) {
 		z180_dev->ringbuffer.prevctx = Z180_INVALID_CONTEXT;
 		device->mmu.hwpagetable = device->mmu.defaultpagetable;
+
+		/* Ignore the result - we are going down anyway */
 		kgsl_setstate(&device->mmu, KGSL_MEMSTORE_GLOBAL,
 				KGSL_MMUFLAGS_PTUPDATE);
 	}
+
+	return 0;
 }
 
 static void
@@ -965,6 +997,7 @@
 	.irqctrl = z180_irqctrl,
 	.gpuid = z180_gpuid,
 	.irq_handler = z180_irq_handler,
+	.drain = z180_idle, /* drain == idle for the z180 */
 	/* Optional functions */
 	.drawctxt_create = z180_drawctxt_create,
 	.drawctxt_detach = z180_drawctxt_detach,
diff --git a/drivers/gpu/msm/z180.h b/drivers/gpu/msm/z180.h
index 1be0870..a36e92d 100644
--- a/drivers/gpu/msm/z180.h
+++ b/drivers/gpu/msm/z180.h
@@ -45,5 +45,6 @@
 };
 
 int z180_dump(struct kgsl_device *, int);
+int z180_idle(struct kgsl_device *);
 
 #endif /* __Z180_H */
diff --git a/drivers/gpu/msm/z180_postmortem.c b/drivers/gpu/msm/z180_postmortem.c
index 5d929cf..bc53c0e 100644
--- a/drivers/gpu/msm/z180_postmortem.c
+++ b/drivers/gpu/msm/z180_postmortem.c
@@ -58,6 +58,8 @@
 	unsigned int i;
 	unsigned int reg_val;
 
+	z180_idle(device);
+
 	KGSL_LOG_DUMP(device, "Z180 Register Dump\n");
 	for (i = 0; i < ARRAY_SIZE(regs_to_dump); i++) {
 		kgsl_regread(device,
diff --git a/drivers/media/platform/msm/camera_v2/isp/msm_isp.h b/drivers/media/platform/msm/camera_v2/isp/msm_isp.h
index f1f4c17..8c42ed2 100644
--- a/drivers/media/platform/msm/camera_v2/isp/msm_isp.h
+++ b/drivers/media/platform/msm/camera_v2/isp/msm_isp.h
@@ -90,7 +90,8 @@
 	void (*enable_wm) (struct vfe_device *vfe_dev,
 		uint8_t wm_idx, uint8_t enable);
 	void (*cfg_io_format) (struct vfe_device *vfe_dev,
-		struct msm_vfe_axi_stream *stream_info);
+		enum msm_vfe_axi_stream_src stream_src,
+		uint32_t io_format);
 	void (*cfg_framedrop) (struct vfe_device *vfe_dev,
 		struct msm_vfe_axi_stream *stream_info);
 	void (*clear_framedrop) (struct vfe_device *vfe_dev,
@@ -289,6 +290,7 @@
 	enum msm_vfe_inputmux input_mux;
 	uint32_t width;
 	long pixel_clock;
+	uint32_t input_format;
 };
 
 enum msm_wm_ub_cfg_type {
diff --git a/drivers/media/platform/msm/camera_v2/isp/msm_isp32.c b/drivers/media/platform/msm/camera_v2/isp/msm_isp32.c
index 07a66e6..3722066 100644
--- a/drivers/media/platform/msm/camera_v2/isp/msm_isp32.c
+++ b/drivers/media/platform/msm/camera_v2/isp/msm_isp32.c
@@ -481,11 +481,11 @@
 }
 
 static void msm_vfe32_cfg_io_format(struct vfe_device *vfe_dev,
-	struct msm_vfe_axi_stream *stream_info)
+	enum msm_vfe_axi_stream_src stream_src, uint32_t io_format)
 {
 	int bpp, bpp_reg = 0;
 	uint32_t io_format_reg;
-	bpp = msm_isp_get_bit_per_pixel(stream_info->output_format);
+	bpp = msm_isp_get_bit_per_pixel(io_format);
 
 	switch (bpp) {
 	case 8:
@@ -499,7 +499,9 @@
 		break;
 	}
 	io_format_reg = msm_camera_io_r(vfe_dev->vfe_base + 0x6F8);
-	switch (stream_info->stream_src) {
+	switch (stream_src) {
+	case PIX_ENCODER:
+	case PIX_VIEWFINDER:
 	case CAMIF_RAW:
 		io_format_reg &= 0xFFFFCFFF;
 		io_format_reg |= bpp_reg << 12;
@@ -508,8 +510,6 @@
 		io_format_reg &= 0xFFFFFFC8;
 		io_format_reg |= bpp_reg << 4;
 		break;
-	case PIX_ENCODER:
-	case PIX_VIEWFINDER:
 	case RDI_INTF_0:
 	case RDI_INTF_1:
 	case RDI_INTF_2:
diff --git a/drivers/media/platform/msm/camera_v2/isp/msm_isp40.c b/drivers/media/platform/msm/camera_v2/isp/msm_isp40.c
index 2db25a6..84b95f1 100644
--- a/drivers/media/platform/msm/camera_v2/isp/msm_isp40.c
+++ b/drivers/media/platform/msm/camera_v2/isp/msm_isp40.c
@@ -688,11 +688,11 @@
 }
 
 static void msm_vfe40_cfg_io_format(struct vfe_device *vfe_dev,
-	struct msm_vfe_axi_stream *stream_info)
+	enum msm_vfe_axi_stream_src stream_src, uint32_t io_format)
 {
 	int bpp, bpp_reg = 0;
 	uint32_t io_format_reg;
-	bpp = msm_isp_get_bit_per_pixel(stream_info->output_format);
+	bpp = msm_isp_get_bit_per_pixel(io_format);
 
 	switch (bpp) {
 	case 8:
@@ -706,7 +706,9 @@
 		break;
 	}
 	io_format_reg = msm_camera_io_r(vfe_dev->vfe_base + 0x54);
-	switch (stream_info->stream_src) {
+	switch (stream_src) {
+	case PIX_ENCODER:
+	case PIX_VIEWFINDER:
 	case CAMIF_RAW:
 		io_format_reg &= 0xFFFFCFFF;
 		io_format_reg |= bpp_reg << 12;
@@ -715,8 +717,6 @@
 		io_format_reg &= 0xFFFFFFC8;
 		io_format_reg |= bpp_reg << 4;
 		break;
-	case PIX_ENCODER:
-	case PIX_VIEWFINDER:
 	case RDI_INTF_0:
 	case RDI_INTF_1:
 	case RDI_INTF_2:
diff --git a/drivers/media/platform/msm/camera_v2/isp/msm_isp_axi_util.c b/drivers/media/platform/msm/camera_v2/isp/msm_isp_axi_util.c
index d3138ed..5b7658d 100644
--- a/drivers/media/platform/msm/camera_v2/isp/msm_isp_axi_util.c
+++ b/drivers/media/platform/msm/camera_v2/isp/msm_isp_axi_util.c
@@ -474,6 +474,7 @@
 int msm_isp_request_axi_stream(struct vfe_device *vfe_dev, void *arg)
 {
 	int rc = 0, i;
+	uint32_t io_format = 0;
 	struct msm_vfe_axi_stream_request_cmd *stream_cfg_cmd = arg;
 	struct msm_vfe_axi_stream *stream_info;
 
@@ -497,10 +498,20 @@
 		stream_info[HANDLE_TO_IDX(stream_cfg_cmd->axi_stream_handle)];
 	msm_isp_axi_reserve_wm(&vfe_dev->axi_data, stream_info);
 
-	if (stream_cfg_cmd->stream_src == CAMIF_RAW ||
-		stream_cfg_cmd->stream_src == IDEAL_RAW)
-			vfe_dev->hw_info->vfe_ops.axi_ops.
-				cfg_io_format(vfe_dev, stream_info);
+	if (stream_info->stream_src < RDI_INTF_0) {
+		io_format = vfe_dev->axi_data.src_info[VFE_PIX_0].input_format;
+		if (stream_info->stream_src == CAMIF_RAW ||
+			stream_info->stream_src == IDEAL_RAW) {
+			if (stream_info->stream_src == CAMIF_RAW &&
+				io_format != stream_info->output_format)
+				pr_warn("%s: Overriding input format\n",
+					__func__);
+
+			io_format = stream_info->output_format;
+		}
+		vfe_dev->hw_info->vfe_ops.axi_ops.cfg_io_format(
+			vfe_dev, stream_info->stream_src, io_format);
+	}
 
 	msm_isp_calculate_framedrop(&vfe_dev->axi_data, stream_cfg_cmd);
 
diff --git a/drivers/media/platform/msm/camera_v2/isp/msm_isp_util.c b/drivers/media/platform/msm/camera_v2/isp/msm_isp_util.c
index 3806213..590b636 100644
--- a/drivers/media/platform/msm/camera_v2/isp/msm_isp_util.c
+++ b/drivers/media/platform/msm/camera_v2/isp/msm_isp_util.c
@@ -273,6 +273,9 @@
 		return rc;
 	}
 
+	vfe_dev->axi_data.src_info[VFE_PIX_0].input_format =
+		input_cfg->d.pix_cfg.input_format;
+
 	vfe_dev->hw_info->vfe_ops.core_ops.cfg_camif(
 		vfe_dev, &input_cfg->d.pix_cfg);
 	return rc;
diff --git a/drivers/media/platform/msm/camera_v2/pproc/cpp/msm_cpp.c b/drivers/media/platform/msm/camera_v2/pproc/cpp/msm_cpp.c
index a4eb274..822c0c8 100644
--- a/drivers/media/platform/msm/camera_v2/pproc/cpp/msm_cpp.c
+++ b/drivers/media/platform/msm/camera_v2/pproc/cpp/msm_cpp.c
@@ -1219,6 +1219,14 @@
 		goto ERROR1;
 	}
 
+	if ((new_frame->msg_len == 0) ||
+		(new_frame->msg_len > MSM_CPP_MAX_FRAME_LENGTH)) {
+		pr_err("%s:%d: Invalid frame len:%d\n", __func__,
+			__LINE__, new_frame->msg_len);
+		rc = -EINVAL;
+		goto ERROR1;
+	}
+
 	cpp_frame_msg = kzalloc(sizeof(uint32_t)*new_frame->msg_len,
 		GFP_KERNEL);
 	if (!cpp_frame_msg) {
@@ -1380,7 +1388,10 @@
 		pr_err("ioctl_ptr is null\n");
 		return -EINVAL;
 	}
-
+	if (cpp_dev == NULL) {
+		pr_err("cpp_dev is null\n");
+		return -EINVAL;
+	}
 	mutex_lock(&cpp_dev->mutex);
 	CPP_DBG("E cmd: %d\n", cmd);
 	switch (cmd) {
@@ -1396,8 +1407,16 @@
 
 	case VIDIOC_MSM_CPP_LOAD_FIRMWARE: {
 		if (cpp_dev->is_firmware_loaded == 0) {
-			kfree(cpp_dev->fw_name_bin);
-			cpp_dev->fw_name_bin = NULL;
+			if (cpp_dev->fw_name_bin != NULL) {
+				kfree(cpp_dev->fw_name_bin);
+				cpp_dev->fw_name_bin = NULL;
+			}
+			if ((ioctl_ptr->len == 0) ||
+				(ioctl_ptr->len > MSM_CPP_MAX_FW_NAME_LEN)) {
+				pr_err("ioctl_ptr->len is 0\n");
+				mutex_unlock(&cpp_dev->mutex);
+				return -EINVAL;
+			}
 			cpp_dev->fw_name_bin = kzalloc(ioctl_ptr->len+1,
 				GFP_KERNEL);
 			if (!cpp_dev->fw_name_bin) {
@@ -1406,13 +1425,9 @@
 				mutex_unlock(&cpp_dev->mutex);
 				return -EINVAL;
 			}
-
 			if (ioctl_ptr->ioctl_ptr == NULL) {
 				pr_err("ioctl_ptr->ioctl_ptr=NULL\n");
-				return -EINVAL;
-			}
-			if (ioctl_ptr->len == 0) {
-				pr_err("ioctl_ptr->len is 0\n");
+				mutex_unlock(&cpp_dev->mutex);
 				return -EINVAL;
 			}
 			rc = (copy_from_user(cpp_dev->fw_name_bin,
@@ -1426,11 +1441,6 @@
 				return -EINVAL;
 			}
 			*(cpp_dev->fw_name_bin+ioctl_ptr->len) = '\0';
-			if (cpp_dev == NULL) {
-				pr_err("cpp_dev is null\n");
-				return -EINVAL;
-			}
-
 			disable_irq(cpp_dev->irq->start);
 			cpp_load_fw(cpp_dev, cpp_dev->fw_name_bin);
 			enable_irq(cpp_dev->irq->start);
diff --git a/drivers/misc/qseecom.c b/drivers/misc/qseecom.c
index bd838fc..b750602 100644
--- a/drivers/misc/qseecom.c
+++ b/drivers/misc/qseecom.c
@@ -2647,6 +2647,11 @@
 	struct qseecom_dev_handle *data = file->private_data;
 	void __user *argp = (void __user *) arg;
 
+	if (!data) {
+		pr_err("Invalid/uninitialized device handle\n");
+		return -EINVAL;
+	}
+
 	if (data->abort) {
 		pr_err("Aborting qseecom driver\n");
 		return -ENODEV;
diff --git a/drivers/video/msm/mdss/mdss_mdp_ctl.c b/drivers/video/msm/mdss/mdss_mdp_ctl.c
index b5a5383..f44ebaf 100644
--- a/drivers/video/msm/mdss/mdss_mdp_ctl.c
+++ b/drivers/video/msm/mdss/mdss_mdp_ctl.c
@@ -1337,7 +1337,7 @@
 		 * writeback block
 		*/
 		head[len] = head[len - 1];
-		head[len].num = -1;
+		head[len].num = head[len - 1].num;
 	}
 	mdata->ctl_off = head;
 
diff --git a/include/media/msmb_isp.h b/include/media/msmb_isp.h
index 5ae852a..ec8ec9a 100644
--- a/include/media/msmb_isp.h
+++ b/include/media/msmb_isp.h
@@ -98,6 +98,7 @@
 	struct msm_vfe_camif_cfg camif_cfg;
 	enum msm_vfe_inputmux input_mux;
 	enum ISP_START_PIXEL_PATTERN pixel_pattern;
+	uint32_t input_format;
 };
 
 struct msm_vfe_rdi_cfg {
diff --git a/include/media/msmb_pproc.h b/include/media/msmb_pproc.h
index 162729a..de42c38 100644
--- a/include/media/msmb_pproc.h
+++ b/include/media/msmb_pproc.h
@@ -13,6 +13,8 @@
 
 #define MAX_NUM_CPP_STRIPS 8
 #define MSM_CPP_MAX_NUM_PLANES 3
+#define MSM_CPP_MAX_FRAME_LENGTH 1024
+#define MSM_CPP_MAX_FW_NAME_LEN 32
 
 enum msm_cpp_frame_type {
 	MSM_CPP_OFFLINE_FRAME,