Merge "drm/msm: Detach the MMU during msm_gpu_cleanup()"

18 files changed, 402 insertions, 130 deletions

diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
index 4ca16fc01e1c..442c5d9711a9 100644
--- a/drivers/gpu/drm/msm/Makefile
+++ b/drivers/gpu/drm/msm/Makefile
@@ -121,6 +121,7 @@ msm_drm-$(CONFIG_DRM_MSM) += \
 	msm_gem.o \
 	msm_gem_prime.o \
 	msm_gem_submit.o \
+	msm_gem_vma.o \
 	msm_gpu.o \
 	msm_iommu.o \
 	msm_smmu.o \
diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
index fd266ed963b6..156abf00c0e2 100644
--- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
@@ -583,7 +583,7 @@ struct msm_gpu *a3xx_gpu_init(struct drm_device *dev)
 #endif
 	}
 
-	if (!gpu->mmu) {
+	if (!gpu->aspace) {
 		/* TODO we think it is possible to configure the GPU to
 		 * restrict access to VRAM carveout.  But the required
 		 * registers are unknown.  For now just bail out and
diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
index 1fa2c7cf34c6..a18153e0cc5c 100644
--- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
@@ -665,7 +665,7 @@ struct msm_gpu *a4xx_gpu_init(struct drm_device *dev)
 #endif
 	}
 
-	if (!gpu->mmu) {
+	if (!gpu->aspace) {
 		/* TODO we think it is possible to configure the GPU to
 		 * restrict access to VRAM carveout.  But the required
 		 * registers are unknown.  For now just bail out and
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 4951172ede06..74c80a98c420 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -390,7 +390,7 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 		return ret;
 	}
 
-	mmu = gpu->mmu;
+	mmu = gpu->aspace->mmu;
 	if (mmu) {
 		ret = mmu->funcs->attach(mmu, iommu_ports,
 				ARRAY_SIZE(iommu_ports));
@@ -427,6 +427,8 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 
 void adreno_gpu_cleanup(struct adreno_gpu *gpu)
 {
+	struct msm_gem_address_space *aspace = gpu->base.aspace;
+
 	if (gpu->memptrs_bo) {
 		if (gpu->memptrs_iova)
 			msm_gem_put_iova(gpu->memptrs_bo, gpu->base.id);
@@ -434,5 +436,12 @@ void adreno_gpu_cleanup(struct adreno_gpu *gpu)
 	}
 	release_firmware(gpu->pm4);
 	release_firmware(gpu->pfp);
+
 	msm_gpu_cleanup(&gpu->base);
+
+	if (aspace) {
+		aspace->mmu->funcs->detach(aspace->mmu,
+			iommu_ports, ARRAY_SIZE(iommu_ports));
+		msm_gem_address_space_destroy(aspace);
+	}
 }
diff --git a/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.c b/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.c
index f0f66ac4b2fb..6d8e174236ea 100644
--- a/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.c
+++ b/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.c
@@ -17,6 +17,7 @@
 
 
 #include "msm_drv.h"
+#include "msm_gem.h"
 #include "msm_mmu.h"
 #include "mdp4_kms.h"
 
@@ -177,18 +178,35 @@ static void mdp4_preclose(struct msm_kms *kms, struct drm_file *file)
 	struct mdp4_kms *mdp4_kms = to_mdp4_kms(to_mdp_kms(kms));
 	struct msm_drm_private *priv = mdp4_kms->dev->dev_private;
 	unsigned i;
+	struct msm_gem_address_space *aspace = mdp4_kms->aspace;
 
 	for (i = 0; i < priv->num_crtcs; i++)
 		mdp4_crtc_cancel_pending_flip(priv->crtcs[i], file);
+
+	if (aspace) {
+		aspace->mmu->funcs->detach(aspace->mmu,
+				iommu_ports, ARRAY_SIZE(iommu_ports));
+		msm_gem_address_space_destroy(aspace);
+	}
 }
 
 static void mdp4_destroy(struct msm_kms *kms)
 {
+	struct device *dev = mdp4_kms->dev->dev;
+	struct msm_gem_address_space *aspace = mdp4_kms->aspace;
+
 	struct mdp4_kms *mdp4_kms = to_mdp4_kms(to_mdp_kms(kms));
 	if (mdp4_kms->blank_cursor_iova)
 		msm_gem_put_iova(mdp4_kms->blank_cursor_bo, mdp4_kms->id);
 	if (mdp4_kms->blank_cursor_bo)
 		drm_gem_object_unreference_unlocked(mdp4_kms->blank_cursor_bo);
+
+	if (aspace) {
+		aspace->mmu->funcs->detach(aspace->mmu,
+				iommu_ports, ARRAY_SIZE(iommu_ports));
+		msm_gem_address_space_destroy(aspace);
+	}
+
 	kfree(mdp4_kms);
 }
 
@@ -408,7 +426,7 @@ struct msm_kms *mdp4_kms_init(struct drm_device *dev)
 	struct mdp4_platform_config *config = mdp4_get_config(pdev);
 	struct mdp4_kms *mdp4_kms;
 	struct msm_kms *kms = NULL;
-	struct msm_mmu *mmu;
+	struct msm_gem_address_space *aspace;
 	int ret;
 
 	mdp4_kms = kzalloc(sizeof(*mdp4_kms), GFP_KERNEL);
@@ -497,22 +515,33 @@ struct msm_kms *mdp4_kms_init(struct drm_device *dev)
 	mdelay(16);
 
 	if (config->iommu) {
-		mmu = msm_iommu_new(&pdev->dev, config->iommu);
+		struct msm_mmu *mmu = msm_iommu_new(&pdev->dev, config->iommu);
+
 		if (IS_ERR(mmu)) {
 			ret = PTR_ERR(mmu);
 			goto fail;
 		}
-		ret = mmu->funcs->attach(mmu, iommu_ports,
+
+		aspace = msm_gem_address_space_create(&pdev->dev,
+				mmu, "mdp4", 0x1000, 0xffffffff);
+		if (IS_ERR(aspace)) {
+			ret = PTR_ERR(aspace);
+			goto fail;
+		}
+
+		mdp4_kms->aspace = aspace;
+
+		ret = aspace->mmu->funcs->attach(aspace->mmu, iommu_ports,
 				ARRAY_SIZE(iommu_ports));
 		if (ret)
 			goto fail;
 	} else {
 		dev_info(dev->dev, "no iommu, fallback to phys "
 				"contig buffers for scanout\n");
-		mmu = NULL;
+		aspace = NULL;
 	}
 
-	mdp4_kms->id = msm_register_mmu(dev, mmu);
+	mdp4_kms->id = msm_register_address_space(dev, aspace);
 	if (mdp4_kms->id < 0) {
 		ret = mdp4_kms->id;
 		dev_err(dev->dev, "failed to register mdp4 iommu: %d\n", ret);
@@ -562,6 +591,7 @@ static struct mdp4_platform_config *mdp4_get_config(struct platform_device *dev)
 	/* TODO */
 	config.max_clk = 266667000;
 	config.iommu = iommu_domain_alloc(msm_iommu_get_bus(&dev->dev));
+
 #else
 	if (cpu_is_apq8064())
 		config.max_clk = 266667000;
diff --git a/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.h b/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.h
index 8a7f6e1e2bca..923a028db452 100644
--- a/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.h
+++ b/drivers/gpu/drm/msm/mdp/mdp4/mdp4_kms.h
@@ -45,6 +45,7 @@ struct mdp4_kms {
 	struct clk *pclk;
 	struct clk *lut_clk;
 	struct clk *axi_clk;
+	struct msm_gem_address_space *aspace;
 
 	struct mdp_irq error_handler;
 
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.c
index f7aebf5516ce..cc5a73505537 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.c
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2016, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2014, 2016-2017 The Linux Foundation. All rights reserved.
  * Copyright (C) 2013 Red Hat
  * Author: Rob Clark <robdclark@gmail.com>
  *
@@ -18,6 +18,7 @@
 
 
 #include "msm_drv.h"
+#include "msm_gem.h"
 #include "msm_mmu.h"
 #include "mdp5_kms.h"
 
@@ -130,13 +131,14 @@ static void mdp5_preclose(struct msm_kms *kms, struct drm_file *file)
 static void mdp5_destroy(struct msm_kms *kms)
 {
 	struct mdp5_kms *mdp5_kms = to_mdp5_kms(to_mdp_kms(kms));
-	struct msm_mmu *mmu = mdp5_kms->mmu;
+	struct msm_gem_address_space *aspace = mdp5_kms->aspace;
 
 	mdp5_irq_domain_fini(mdp5_kms);
 
-	if (mmu) {
-		mmu->funcs->detach(mmu, iommu_ports, ARRAY_SIZE(iommu_ports));
-		mmu->funcs->destroy(mmu);
+	if (aspace) {
+		aspace->mmu->funcs->detach(aspace->mmu,
+				iommu_ports, ARRAY_SIZE(iommu_ports));
+		msm_gem_address_space_destroy(aspace);
 	}
 
 	if (mdp5_kms->ctlm)
@@ -474,7 +476,7 @@ struct msm_kms *mdp5_kms_init(struct drm_device *dev)
 	struct mdp5_cfg *config;
 	struct mdp5_kms *mdp5_kms;
 	struct msm_kms *kms = NULL;
-	struct msm_mmu *mmu;
+	struct msm_gem_address_space *aspace;
 	uint32_t major, minor;
 	int i, ret;
 
@@ -595,30 +597,37 @@ struct msm_kms *mdp5_kms_init(struct drm_device *dev)
 	mdelay(16);
 
 	if (config->platform.iommu) {
-		mmu = msm_smmu_new(&pdev->dev,
+		struct msm_mmu *mmu = msm_smmu_new(&pdev->dev,
 				MSM_SMMU_DOMAIN_UNSECURE);
 		if (IS_ERR(mmu)) {
 			ret = PTR_ERR(mmu);
 			dev_err(dev->dev, "failed to init iommu: %d\n", ret);
 			iommu_domain_free(config->platform.iommu);
+		}
+
+		aspace = msm_gem_smmu_address_space_create(&pdev->dev,
+				mmu, "mdp5");
+		if (IS_ERR(aspace)) {
+			ret = PTR_ERR(aspace);
 			goto fail;
 		}
 
-		ret = mmu->funcs->attach(mmu, iommu_ports,
+		mdp5_kms->aspace = aspace;
+
+		ret = aspace->mmu->funcs->attach(aspace->mmu, iommu_ports,
 				ARRAY_SIZE(iommu_ports));
 		if (ret) {
-			dev_err(dev->dev, "failed to attach iommu: %d\n", ret);
-			mmu->funcs->destroy(mmu);
+			dev_err(&pdev->dev, "failed to attach iommu: %d\n",
+				ret);
 			goto fail;
 		}
 	} else {
-		dev_info(dev->dev, "no iommu, fallback to phys "
-				"contig buffers for scanout\n");
-		mmu = NULL;
+		dev_info(&pdev->dev,
+			 "no iommu, fallback to phys contig buffers for scanout\n");
+		aspace = NULL;
 	}
-	mdp5_kms->mmu = mmu;
 
-	mdp5_kms->id = msm_register_mmu(dev, mmu);
+	mdp5_kms->id = msm_register_address_space(dev, aspace);
 	if (mdp5_kms->id < 0) {
 		ret = mdp5_kms->id;
 		dev_err(dev->dev, "failed to register mdp5 iommu: %d\n", ret);
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.h b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.h
index 84f65d415598..6cb43ad0c1d7 100644
--- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.h
+++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_kms.h
@@ -37,7 +37,7 @@ struct mdp5_kms {
 
 	/* mapper-id used to request GEM buffer mapped for scanout: */
 	int id;
-	struct msm_mmu *mmu;
+	struct msm_gem_address_space *aspace;
 
 	struct mdp5_smp *smp;
 	struct mdp5_ctl_manager *ctlm;
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index 1dcf88a23afa..4b67edf5a77b 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -38,42 +38,20 @@ static const struct drm_mode_config_funcs mode_config_funcs = {
 	.atomic_commit = msm_atomic_commit,
 };
 
-int msm_register_mmu(struct drm_device *dev, struct msm_mmu *mmu)
+int msm_register_address_space(struct drm_device *dev,
+		struct msm_gem_address_space *aspace)
 {
 	struct msm_drm_private *priv = dev->dev_private;
-	int idx = priv->num_mmus++;
+	int idx = priv->num_aspaces++;
 
-	if (WARN_ON(idx >= ARRAY_SIZE(priv->mmus)))
+	if (WARN_ON(idx >= ARRAY_SIZE(priv->aspace)))
 		return -EINVAL;
 
-	priv->mmus[idx] = mmu;
+	priv->aspace[idx] = aspace;
 
 	return idx;
 }
 
-void msm_unregister_mmu(struct drm_device *dev, struct msm_mmu *mmu)
-{
-	struct msm_drm_private *priv = dev->dev_private;
-	int idx;
-
-	if (priv->num_mmus <= 0) {
-		dev_err(dev->dev, "invalid num mmus %d\n", priv->num_mmus);
-		return;
-	}
-
-	idx = priv->num_mmus - 1;
-
-	/* only support reverse-order deallocation */
-	if (priv->mmus[idx] != mmu) {
-		dev_err(dev->dev, "unexpected mmu at idx %d\n", idx);
-		return;
-	}
-
-	--priv->num_mmus;
-	priv->mmus[idx] = 0;
-}
-
-
 #ifdef CONFIG_DRM_MSM_REGISTER_LOGGING
 static bool reglog = false;
 MODULE_PARM_DESC(reglog, "Enable register read/write logging");
diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
index a2678882d57a..4ed5953d07f9 100644
--- a/drivers/gpu/drm/msm/msm_drv.h
+++ b/drivers/gpu/drm/msm/msm_drv.h
@@ -63,6 +63,8 @@ struct msm_mmu;
 struct msm_rd_state;
 struct msm_perf_state;
 struct msm_gem_submit;
+struct msm_gem_address_space;
+struct msm_gem_vma;
 
 #define NUM_DOMAINS    4    /* one for KMS, then one per gpu core (?) */
 #define MAX_CRTCS      8
@@ -296,9 +298,13 @@ struct msm_drm_private {
 	uint32_t pending_crtcs;
 	wait_queue_head_t pending_crtcs_event;
 
-	/* registered MMUs: */
-	unsigned int num_mmus;
-	struct msm_mmu *mmus[NUM_DOMAINS];
+	/* Registered address spaces.. currently this is fixed per # of
+	 * iommu's.  Ie. one for display block and one for gpu block.
+	 * Eventually, to do per-process gpu pagetables, we'll want one
+	 * of these per-process.
+	 */
+	unsigned int num_aspaces;
+	struct msm_gem_address_space *aspace[NUM_DOMAINS];
 
 	unsigned int num_planes;
 	struct drm_plane *planes[MAX_PLANES];
@@ -362,15 +368,32 @@ void __msm_fence_worker(struct work_struct *work);
 int msm_atomic_commit(struct drm_device *dev,
 		struct drm_atomic_state *state, bool async);
 
-int msm_register_mmu(struct drm_device *dev, struct msm_mmu *mmu);
-void msm_unregister_mmu(struct drm_device *dev, struct msm_mmu *mmu);
-
 int msm_wait_fence(struct drm_device *dev, uint32_t fence,
 		ktime_t *timeout, bool interruptible);
 int msm_queue_fence_cb(struct drm_device *dev,
 		struct msm_fence_cb *cb, uint32_t fence);
 void msm_update_fence(struct drm_device *dev, uint32_t fence);
 
+int msm_register_address_space(struct drm_device *dev,
+		struct msm_gem_address_space *aspace);
+void msm_gem_unmap_vma(struct msm_gem_address_space *aspace,
+		struct msm_gem_vma *vma, struct sg_table *sgt,
+		void *priv);
+int msm_gem_map_vma(struct msm_gem_address_space *aspace,
+		struct msm_gem_vma *vma, struct sg_table *sgt,
+		void *priv);
+void msm_gem_address_space_destroy(struct msm_gem_address_space *aspace);
+
+/* For GPU and legacy display */
+struct msm_gem_address_space *
+msm_gem_address_space_create(struct device *dev, struct iommu_domain *domain,
+		const char *name);
+
+/* For SDE  display */
+struct msm_gem_address_space *
+msm_gem_smmu_address_space_create(struct device *dev, struct msm_mmu *mmu,
+		const char *name);
+
 int msm_ioctl_gem_submit(struct drm_device *dev, void *data,
 		struct drm_file *file);
 
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index 4e69fba410c3..5aa08cf4d6d8 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -24,6 +24,11 @@
 #include "msm_gpu.h"
 #include "msm_mmu.h"
 
+static void *get_dmabuf_ptr(struct drm_gem_object *obj)
+{
+	return (obj && obj->import_attach) ? obj->import_attach->dmabuf : NULL;
+}
+
 static dma_addr_t physaddr(struct drm_gem_object *obj)
 {
 	struct msm_gem_object *msm_obj = to_msm_bo(obj);
@@ -279,21 +284,8 @@ put_iova(struct drm_gem_object *obj)
 	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
 
 	for (id = 0; id < ARRAY_SIZE(msm_obj->domain); id++) {
-		struct msm_mmu *mmu = priv->mmus[id];
-
-		if (!mmu || !msm_obj->domain[id].iova)
-			continue;
-
-		if (obj->import_attach) {
-			if (mmu->funcs->unmap_dma_buf)
-				mmu->funcs->unmap_dma_buf(mmu, msm_obj->sgt,
-					obj->import_attach->dmabuf,
-					DMA_BIDIRECTIONAL);
-		} else
-			mmu->funcs->unmap_sg(mmu, msm_obj->sgt,
-				DMA_BIDIRECTIONAL);
-
-		msm_obj->domain[id].iova = 0;
+		msm_gem_unmap_vma(priv->aspace[id], &msm_obj->domain[id],
+			msm_obj->sgt, get_dmabuf_ptr(obj));
 	}
 }
 
@@ -318,31 +310,11 @@ int msm_gem_get_iova_locked(struct drm_gem_object *obj, int id,
 			return PTR_ERR(pages);
 
 		if (iommu_present(&platform_bus_type)) {
-			struct msm_mmu *mmu = priv->mmus[id];
-
-			if (WARN_ON(!mmu))
-				return -EINVAL;
-
-			if (obj->import_attach && mmu->funcs->map_dma_buf) {
-				ret = mmu->funcs->map_dma_buf(mmu, msm_obj->sgt,
-						obj->import_attach->dmabuf,
-						DMA_BIDIRECTIONAL);
-				if (ret) {
-					DRM_ERROR("Unable to map dma buf\n");
-					return ret;
-				}
-			} else {
-				ret = mmu->funcs->map_sg(mmu, msm_obj->sgt,
-					DMA_BIDIRECTIONAL);
-			}
-
-			if (!ret)
-				msm_obj->domain[id].iova =
-					sg_dma_address(msm_obj->sgt->sgl);
-		} else {
-			WARN_ONCE(1, "physical address being used\n");
+			ret = msm_gem_map_vma(priv->aspace[id],
+				&msm_obj->domain[id], msm_obj->sgt,
+				get_dmabuf_ptr(obj));
+		} else
 			msm_obj->domain[id].iova = physaddr(obj);
-		}
 	}
 
 	if (!ret)
@@ -515,14 +487,21 @@ void msm_gem_describe(struct drm_gem_object *obj, struct seq_file *m)
 {
 	struct drm_device *dev = obj->dev;
 	struct msm_gem_object *msm_obj = to_msm_bo(obj);
+	struct msm_drm_private *priv = obj->dev->dev_private;
 	uint64_t off = drm_vma_node_start(&obj->vma_node);
+	int id;
 
 	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
-	seq_printf(m, "%08x: %c(r=%u,w=%u) %2d (%2d) %08llx %p %zu\n",
+	seq_printf(m, "%08x: %c(r=%u,w=%u) %2d (%2d) %08llx %p\t",
 			msm_obj->flags, is_active(msm_obj) ? 'A' : 'I',
 			msm_obj->read_fence, msm_obj->write_fence,
 			obj->name, obj->refcount.refcount.counter,
-			off, msm_obj->vaddr, obj->size);
+			off, msm_obj->vaddr);
+
+	for (id = 0; id < priv->num_aspaces; id++)
+		seq_printf(m, " %08llx", msm_obj->domain[id].iova);
+
+	seq_puts(m, "\n");
 }
 
 void msm_gem_describe_objects(struct list_head *list, struct seq_file *m)
@@ -559,7 +538,8 @@ void msm_gem_free_object(struct drm_gem_object *obj)
 
 	if (obj->import_attach) {
 		if (msm_obj->vaddr)
-			dma_buf_vunmap(obj->import_attach->dmabuf, msm_obj->vaddr);
+			dma_buf_vunmap(obj->import_attach->dmabuf,
+				msm_obj->vaddr);
 
 		/* Don't drop the pages for imported dmabuf, as they are not
 		 * ours, just free the array we allocated:
@@ -613,7 +593,6 @@ static int msm_gem_new_impl(struct drm_device *dev,
 {
 	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_gem_object *msm_obj;
-	unsigned sz;
 	bool use_vram = false;
 
 	switch (flags & MSM_BO_CACHE_MASK) {
@@ -635,16 +614,12 @@ static int msm_gem_new_impl(struct drm_device *dev,
 	if (WARN_ON(use_vram && !priv->vram.size))
 		return -EINVAL;
 
-	sz = sizeof(*msm_obj);
-	if (use_vram)
-		sz += sizeof(struct drm_mm_node);
-
-	msm_obj = kzalloc(sz, GFP_KERNEL);
+	msm_obj = kzalloc(sizeof(*msm_obj), GFP_KERNEL);
 	if (!msm_obj)
 		return -ENOMEM;
 
 	if (use_vram)
-		msm_obj->vram_node = (void *)&msm_obj[1];
+		msm_obj->vram_node = &msm_obj->domain[0].node;
 
 	msm_obj->flags = flags;
 
diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
index 2e4ae6b1c5d0..86c297c6de32 100644
--- a/drivers/gpu/drm/msm/msm_gem.h
+++ b/drivers/gpu/drm/msm/msm_gem.h
@@ -24,6 +24,28 @@
 /* Additional internal-use only BO flags: */
 #define MSM_BO_STOLEN        0x10000000    /* try to use stolen/splash memory */
 
+struct msm_gem_aspace_ops {
+	int (*map)(struct msm_gem_address_space *, struct msm_gem_vma *,
+		struct sg_table *sgt, void *priv);
+
+	void (*unmap)(struct msm_gem_address_space *, struct msm_gem_vma *,
+		struct sg_table *sgt, void *priv);
+
+	void (*destroy)(struct msm_gem_address_space *);
+};
+
+struct msm_gem_address_space {
+	const char *name;
+	struct msm_mmu *mmu;
+	const struct msm_gem_aspace_ops *ops;
+};
+
+struct msm_gem_vma {
+	/* Node used by the GPU address space, but not the SDE address space */
+	struct drm_mm_node node;
+	uint64_t iova;
+};
+
 struct msm_gem_object {
 	struct drm_gem_object base;
 
@@ -52,9 +74,7 @@ struct msm_gem_object {
 	struct sg_table *sgt;
 	void *vaddr;
 
-	struct {
-		dma_addr_t iova;
-	} domain[NUM_DOMAINS];
+	struct msm_gem_vma domain[NUM_DOMAINS];
 
 	/* normally (resv == &_resv) except for imported bo's */
 	struct reservation_object *resv;
diff --git a/drivers/gpu/drm/msm/msm_gem_vma.c b/drivers/gpu/drm/msm/msm_gem_vma.c
new file mode 100644
index 000000000000..53e70263e03c
--- /dev/null
+++ b/drivers/gpu/drm/msm/msm_gem_vma.c
@@ -0,0 +1,213 @@
+/*
+ * Copyright (C) 2016 Red Hat
+ * Author: Rob Clark <robdclark@gmail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License 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.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "msm_drv.h"
+#include "msm_gem.h"
+#include "msm_mmu.h"
+
+/* SDE address space operations */
+static void smmu_aspace_unmap_vma(struct msm_gem_address_space *aspace,
+		struct msm_gem_vma *vma, struct sg_table *sgt,
+		void *priv)
+{
+	struct dma_buf *buf = priv;
+
+	if (buf)
+		aspace->mmu->funcs->unmap_dma_buf(aspace->mmu,
+			sgt, buf, DMA_BIDIRECTIONAL);
+	else
+		aspace->mmu->funcs->unmap_sg(aspace->mmu, sgt,
+			DMA_BIDIRECTIONAL);
+
+	vma->iova = 0;
+}
+
+
+static int smmu_aspace_map_vma(struct msm_gem_address_space *aspace,
+		struct msm_gem_vma *vma, struct sg_table *sgt,
+		void *priv)
+{
+	struct dma_buf *buf = priv;
+	int ret;
+
+	if (buf)
+		ret = aspace->mmu->funcs->map_dma_buf(aspace->mmu, sgt, buf,
+			DMA_BIDIRECTIONAL);
+	else
+		ret = aspace->mmu->funcs->map_sg(aspace->mmu, sgt,
+			DMA_BIDIRECTIONAL);
+
+	if (!ret)
+		vma->iova = sg_dma_address(sgt->sgl);
+
+	return ret;
+}
+
+static const struct msm_gem_aspace_ops smmu_aspace_ops = {
+	.map = smmu_aspace_map_vma,
+	.unmap = smmu_aspace_unmap_vma,
+};
+
+struct msm_gem_address_space *
+msm_gem_smmu_address_space_create(struct device *dev, struct msm_mmu *mmu,
+		const char *name)
+{
+	struct msm_gem_address_space *aspace;
+
+	if (!mmu)
+		return ERR_PTR(-EINVAL);
+
+	aspace = kzalloc(sizeof(*aspace), GFP_KERNEL);
+	if (!aspace)
+		return ERR_PTR(-ENOMEM);
+
+	aspace->name = name;
+	aspace->mmu = mmu;
+	aspace->ops = &smmu_aspace_ops;
+
+	return aspace;
+}
+
+/* GPU address space operations */
+struct msm_iommu_aspace {
+	struct msm_gem_address_space base;
+	struct drm_mm mm;
+};
+
+#define to_iommu_aspace(aspace) \
+	((struct msm_iommu_aspace *) \
+	 container_of(aspace, struct msm_iommu_aspace, base))
+
+static void iommu_aspace_unmap_vma(struct msm_gem_address_space *aspace,
+		struct msm_gem_vma *vma, struct sg_table *sgt, void *priv)
+{
+	if (!vma->iova)
+		return;
+
+	if (aspace->mmu)
+		aspace->mmu->funcs->unmap(aspace->mmu, vma->iova, sgt);
+
+	drm_mm_remove_node(&vma->node);
+
+	vma->iova = 0;
+}
+
+static int iommu_aspace_map_vma(struct msm_gem_address_space *aspace,
+		struct msm_gem_vma *vma, struct sg_table *sgt,
+		void *priv)
+{
+	struct msm_iommu_aspace *local = to_iommu_aspace(aspace);
+	size_t size = 0;
+	struct scatterlist *sg;
+	int ret = 0, i;
+
+	if (WARN_ON(drm_mm_node_allocated(&vma->node)))
+		return 0;
+
+	for_each_sg(sgt->sgl, sg, sgt->nents, i)
+		size += sg->length + sg->offset;
+
+	ret = drm_mm_insert_node(&local->mm, &vma->node, size >> PAGE_SHIFT,
+			0, DRM_MM_SEARCH_DEFAULT);
+	if (ret)
+		return ret;
+
+	vma->iova = vma->node.start << PAGE_SHIFT;
+
+	if (aspace->mmu)
+		ret = aspace->mmu->funcs->map(aspace->mmu, vma->iova,
+			sgt, IOMMU_READ | IOMMU_WRITE);
+
+	return ret;
+}
+
+static void iommu_aspace_destroy(struct msm_gem_address_space *aspace)
+{
+	struct msm_iommu_aspace *local = to_iommu_aspace(aspace);
+
+	drm_mm_takedown(&local->mm);
+	aspace->mmu->funcs->destroy(aspace->mmu);
+}
+
+static const struct msm_gem_aspace_ops msm_iommu_aspace_ops = {
+	.map = iommu_aspace_map_vma,
+	.unmap = iommu_aspace_unmap_vma,
+	.destroy = iommu_aspace_destroy,
+};
+
+static struct msm_gem_address_space *
+msm_gem_address_space_new(struct msm_mmu *mmu, const char *name,
+		uint64_t start, uint64_t end)
+{
+	struct msm_iommu_aspace *local;
+
+	if (!mmu)
+		return ERR_PTR(-EINVAL);
+
+	local = kzalloc(sizeof(*local), GFP_KERNEL);
+	if (!local)
+		return ERR_PTR(-ENOMEM);
+
+	drm_mm_init(&local->mm, (start >> PAGE_SHIFT),
+		(end >> PAGE_SHIFT) - 1);
+
+	local->base.name = name;
+	local->base.mmu = mmu;
+	local->base.ops = &msm_iommu_aspace_ops;
+
+	return &local->base;
+}
+
+int msm_gem_map_vma(struct msm_gem_address_space *aspace,
+		struct msm_gem_vma *vma, struct sg_table *sgt,
+		void *priv)
+{
+	if (aspace && aspace->ops->map)
+		return aspace->ops->map(aspace, vma, sgt, priv);
+
+	return -EINVAL;
+}
+
+void msm_gem_unmap_vma(struct msm_gem_address_space *aspace,
+		struct msm_gem_vma *vma, struct sg_table *sgt, void *priv)
+{
+	if (aspace && aspace->ops->unmap)
+		aspace->ops->unmap(aspace, vma, sgt, priv);
+}
+
+struct msm_gem_address_space *
+msm_gem_address_space_create(struct device *dev, struct iommu_domain *domain,
+		const char *name)
+{
+	struct msm_mmu *mmu = msm_iommu_new(dev, domain);
+
+	if (IS_ERR(mmu))
+		return (struct msm_gem_address_space *) mmu;
+
+	return msm_gem_address_space_new(mmu, name,
+		domain->geometry.aperture_start,
+		domain->geometry.aperture_end);
+}
+
+void
+msm_gem_address_space_destroy(struct msm_gem_address_space *aspace)
+{
+	if (aspace && aspace->ops->destroy)
+		aspace->ops->destroy(aspace);
+
+	kfree(aspace);
+}
diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c
index 6b02ada6579a..7b8f2d02b56d 100644
--- a/drivers/gpu/drm/msm/msm_gpu.c
+++ b/drivers/gpu/drm/msm/msm_gpu.c
@@ -649,12 +649,17 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 	 */
 	iommu = iommu_domain_alloc(&platform_bus_type);
 	if (iommu) {
+		/* TODO 32b vs 64b address space.. */
+		iommu->geometry.aperture_start = 0x1000;
+		iommu->geometry.aperture_end = 0xffffffff;
+
 		dev_info(drm->dev, "%s: using IOMMU\n", name);
-		gpu->mmu = msm_iommu_new(&pdev->dev, iommu);
-		if (IS_ERR(gpu->mmu)) {
-			ret = PTR_ERR(gpu->mmu);
+		gpu->aspace = msm_gem_address_space_create(&pdev->dev,
+				iommu, "gpu");
+		if (IS_ERR(gpu->aspace)) {
+			ret = PTR_ERR(gpu->aspace);
 			dev_err(drm->dev, "failed to init iommu: %d\n", ret);
-			gpu->mmu = NULL;
+			gpu->aspace = NULL;
 			iommu_domain_free(iommu);
 			goto fail;
 		}
@@ -662,7 +667,7 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 	} else {
 		dev_info(drm->dev, "%s: no IOMMU, fallback to VRAM carveout!\n", name);
 	}
-	gpu->id = msm_register_mmu(drm, gpu->mmu);
+	gpu->id = msm_register_address_space(drm, gpu->aspace);
 
 
 	/* Create ringbuffer: */
@@ -697,7 +702,4 @@ void msm_gpu_cleanup(struct msm_gpu *gpu)
 			msm_gem_put_iova(gpu->rb->bo, gpu->id);
 		msm_ringbuffer_destroy(gpu->rb);
 	}
-
-	if (gpu->mmu)
-		gpu->mmu->funcs->destroy(gpu->mmu);
 }
diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
index 2bbe85a3d6f6..3a4a3e2d77bb 100644
--- a/drivers/gpu/drm/msm/msm_gpu.h
+++ b/drivers/gpu/drm/msm/msm_gpu.h
@@ -95,7 +95,7 @@ struct msm_gpu {
 	void __iomem *mmio;
 	int irq;
 
-	struct msm_mmu *mmu;
+	struct msm_gem_address_space *aspace;
 	int id;
 
 	/* Power Control: */
diff --git a/drivers/gpu/drm/msm/msm_mmu.h b/drivers/gpu/drm/msm/msm_mmu.h
index c8c7755c8370..3d750eca5cae 100644
--- a/drivers/gpu/drm/msm/msm_mmu.h
+++ b/drivers/gpu/drm/msm/msm_mmu.h
@@ -21,7 +21,6 @@
 #include <linux/iommu.h>
 
 struct msm_mmu;
-struct msm_gpu;
 
 enum msm_mmu_domain_type {
 	MSM_SMMU_DOMAIN_UNSECURE,
@@ -61,7 +60,6 @@ static inline void msm_mmu_init(struct msm_mmu *mmu, struct device *dev,
 }
 
 struct msm_mmu *msm_iommu_new(struct device *dev, struct iommu_domain *domain);
-struct msm_mmu *msm_gpummu_new(struct device *dev, struct msm_gpu *gpu);
 struct msm_mmu *msm_smmu_new(struct device *dev,
 	enum msm_mmu_domain_type domain);
 
diff --git a/drivers/gpu/drm/msm/sde/sde_kms.c b/drivers/gpu/drm/msm/sde/sde_kms.c
index afe90d16e31d..e40e33f11fd9 100644
--- a/drivers/gpu/drm/msm/sde/sde_kms.c
+++ b/drivers/gpu/drm/msm/sde/sde_kms.c
@@ -941,15 +941,15 @@ static int _sde_kms_mmu_destroy(struct sde_kms *sde_kms)
 	int i;
 
 	for (i = ARRAY_SIZE(sde_kms->mmu_id) - 1; i >= 0; i--) {
-		if (!sde_kms->mmu[i])
+		mmu = sde_kms->aspace[i]->mmu;
+
+		if (!mmu)
 			continue;
 
-		mmu = sde_kms->mmu[i];
-		msm_unregister_mmu(sde_kms->dev, mmu);
 		mmu->funcs->detach(mmu, (const char **)iommu_ports,
 				ARRAY_SIZE(iommu_ports));
-		mmu->funcs->destroy(mmu);
-		sde_kms->mmu[i] = 0;
+		msm_gem_address_space_destroy(sde_kms->aspace[i]);
+
 		sde_kms->mmu_id[i] = 0;
 	}
 
@@ -962,6 +962,8 @@ static int _sde_kms_mmu_init(struct sde_kms *sde_kms)
 	int i, ret;
 
 	for (i = 0; i < MSM_SMMU_DOMAIN_MAX; i++) {
+		struct msm_gem_address_space *aspace;
+
 		mmu = msm_smmu_new(sde_kms->dev->dev, i);
 		if (IS_ERR(mmu)) {
 			/* MMU's can be optional depending on platform */
@@ -971,25 +973,35 @@ static int _sde_kms_mmu_init(struct sde_kms *sde_kms)
 			continue;
 		}
 
+		aspace = msm_gem_smmu_address_space_create(sde_kms->dev->dev,
+			mmu, "sde");
+		if (IS_ERR(aspace)) {
+			ret = PTR_ERR(aspace);
+			mmu->funcs->destroy(mmu);
+			goto fail;
+		}
+
+		sde_kms->aspace[i] = aspace;
+
 		ret = mmu->funcs->attach(mmu, (const char **)iommu_ports,
 				ARRAY_SIZE(iommu_ports));
 		if (ret) {
 			SDE_ERROR("failed to attach iommu %d: %d\n", i, ret);
-			mmu->funcs->destroy(mmu);
+			msm_gem_address_space_destroy(aspace);
 			goto fail;
 		}
 
-		sde_kms->mmu_id[i] = msm_register_mmu(sde_kms->dev, mmu);
+		sde_kms->mmu_id[i] = msm_register_address_space(sde_kms->dev,
+			aspace);
 		if (sde_kms->mmu_id[i] < 0) {
 			ret = sde_kms->mmu_id[i];
 			SDE_ERROR("failed to register sde iommu %d: %d\n",
 					i, ret);
 			mmu->funcs->detach(mmu, (const char **)iommu_ports,
 					ARRAY_SIZE(iommu_ports));
+			msm_gem_address_space_destroy(aspace);
 			goto fail;
 		}
-
-		sde_kms->mmu[i] = mmu;
 	}
 
 	return 0;
diff --git a/drivers/gpu/drm/msm/sde/sde_kms.h b/drivers/gpu/drm/msm/sde/sde_kms.h
index bf127ffe9eb6..8663c632699b 100644
--- a/drivers/gpu/drm/msm/sde/sde_kms.h
+++ b/drivers/gpu/drm/msm/sde/sde_kms.h
@@ -22,6 +22,7 @@
 #include "msm_drv.h"
 #include "msm_kms.h"
 #include "msm_mmu.h"
+#include "msm_gem.h"
 #include "sde_dbg.h"
 #include "sde_hw_catalog.h"
 #include "sde_hw_ctl.h"
@@ -121,7 +122,7 @@ struct sde_kms {
 	int core_rev;
 	struct sde_mdss_cfg *catalog;
 
-	struct msm_mmu *mmu[MSM_SMMU_DOMAIN_MAX];
+	struct msm_gem_address_space *aspace[MSM_SMMU_DOMAIN_MAX];
 	int mmu_id[MSM_SMMU_DOMAIN_MAX];
 	struct sde_power_client *core_client;