drm/etnaviv: split out cmdbuf mapping into address space

This allows to decouple the cmdbuf suballocator create and mapping
the region into the GPU address space. Allowing multiple AS to share
a single cmdbuf suballoc.

Signed-off-by: Lucas Stach <[e-mail]>
Reviewed-by: Philipp Zabel <[e-mail]>
Reviewed-by: Guido Günther <[e-mail]>

drivers/gpu/drm/etnaviv/etnaviv_buffer.c

···
 	u32 *ptr = buf->vaddr + off;
 
 	dev_info(gpu->dev, "virt %p phys 0x%08x free 0x%08x\n",
-			ptr, etnaviv_cmdbuf_get_va(buf) + off, size - len * 4 - off);
+			ptr, etnaviv_cmdbuf_get_va(buf, &gpu->cmdbuf_mapping) +
+			off, size - len * 4 - off);
 
 	print_hex_dump(KERN_INFO, "cmd ", DUMP_PREFIX_OFFSET, 16, 4,
 			ptr, len * 4, 0);
···
 	if (buffer->user_size + cmd_dwords * sizeof(u64) > buffer->size)
 		buffer->user_size = 0;
 
-	return etnaviv_cmdbuf_get_va(buffer) + buffer->user_size;
+	return etnaviv_cmdbuf_get_va(buffer, &gpu->cmdbuf_mapping) +
+	       buffer->user_size;
 }
 
 u16 etnaviv_buffer_init(struct etnaviv_gpu *gpu)
···
 	buffer->user_size = 0;
 
 	CMD_WAIT(buffer);
-	CMD_LINK(buffer, 2, etnaviv_cmdbuf_get_va(buffer) +
-		 buffer->user_size - 4);
+	CMD_LINK(buffer, 2, etnaviv_cmdbuf_get_va(buffer, &gpu->cmdbuf_mapping)
+		 + buffer->user_size - 4);
 
 	return buffer->user_size / 8;
 }
···
 
 	/* Append waitlink */
 	CMD_WAIT(buffer);
-	CMD_LINK(buffer, 2, etnaviv_cmdbuf_get_va(buffer) +
-			    buffer->user_size - 4);
+	CMD_LINK(buffer, 2, etnaviv_cmdbuf_get_va(buffer, &gpu->cmdbuf_mapping)
+		 + buffer->user_size - 4);
 
 	/*
 	 * Kick off the 'sync point' command by replacing the previous
···
 	if (drm_debug & DRM_UT_DRIVER)
 		etnaviv_buffer_dump(gpu, buffer, 0, 0x50);
 
-	link_target = etnaviv_cmdbuf_get_va(cmdbuf);
+	link_target = etnaviv_cmdbuf_get_va(cmdbuf, &gpu->cmdbuf_mapping);
 	link_dwords = cmdbuf->size / 8;
 
 	/*
···
 	CMD_LOAD_STATE(buffer, VIVS_GL_EVENT, VIVS_GL_EVENT_EVENT_ID(event) |
 		       VIVS_GL_EVENT_FROM_PE);
 	CMD_WAIT(buffer);
-	CMD_LINK(buffer, 2, etnaviv_cmdbuf_get_va(buffer) +
-			    buffer->user_size - 4);
+	CMD_LINK(buffer, 2, etnaviv_cmdbuf_get_va(buffer, &gpu->cmdbuf_mapping)
+		 + buffer->user_size - 4);
 
 	if (drm_debug & DRM_UT_DRIVER)
 		pr_info("stream link to 0x%08x @ 0x%08x %p\n",
-			return_target, etnaviv_cmdbuf_get_va(cmdbuf),
+			return_target,
+			etnaviv_cmdbuf_get_va(cmdbuf, &gpu->cmdbuf_mapping),
 			cmdbuf->vaddr);
 
 	if (drm_debug & DRM_UT_DRIVER) {

drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.c

···
 #include <drm/drm_mm.h>
 
 #include "etnaviv_cmdbuf.h"
+#include "etnaviv_gem.h"
 #include "etnaviv_gpu.h"
 #include "etnaviv_mmu.h"
 #include "etnaviv_perfmon.h"
···
 	struct etnaviv_gpu *gpu;
 	void *vaddr;
 	dma_addr_t paddr;
-
-	/* GPU mapping */
-	u32 iova;
-	struct drm_mm_node vram_node; /* only used on MMUv2 */
 
 	/* allocation management */
 	struct mutex lock;
···
 		goto free_suballoc;
 	}
 
-	ret = etnaviv_iommu_get_suballoc_va(gpu, suballoc->paddr,
-					    &suballoc->vram_node, SUBALLOC_SIZE,
-					    &suballoc->iova);
-	if (ret)
-		goto free_dma;
-
 	return suballoc;
 
-free_dma:
-	dma_free_wc(gpu->dev, SUBALLOC_SIZE, suballoc->vaddr, suballoc->paddr);
 free_suballoc:
 	kfree(suballoc);
 
 	return ERR_PTR(ret);
 }
 
+int etnaviv_cmdbuf_suballoc_map(struct etnaviv_cmdbuf_suballoc *suballoc,
+				struct etnaviv_iommu *mmu,
+				struct etnaviv_vram_mapping *mapping,
+				u32 memory_base)
+{
+	return etnaviv_iommu_get_suballoc_va(mmu, mapping, memory_base,
+					     suballoc->paddr, SUBALLOC_SIZE);
+}
+
+void etnaviv_cmdbuf_suballoc_unmap(struct etnaviv_iommu *mmu,
+				   struct etnaviv_vram_mapping *mapping)
+{
+	etnaviv_iommu_put_suballoc_va(mmu, mapping);
+}
+
 void etnaviv_cmdbuf_suballoc_destroy(struct etnaviv_cmdbuf_suballoc *suballoc)
 {
-	etnaviv_iommu_put_suballoc_va(suballoc->gpu, &suballoc->vram_node,
-				      SUBALLOC_SIZE, suballoc->iova);
 	dma_free_wc(suballoc->gpu->dev, SUBALLOC_SIZE, suballoc->vaddr,
 		    suballoc->paddr);
 	kfree(suballoc);
···
 	wake_up_all(&suballoc->free_event);
 }
 
-u32 etnaviv_cmdbuf_get_va(struct etnaviv_cmdbuf *buf)
+u32 etnaviv_cmdbuf_get_va(struct etnaviv_cmdbuf *buf,
+			  struct etnaviv_vram_mapping *mapping)
 {
-	return buf->suballoc->iova + buf->suballoc_offset;
+	return mapping->iova + buf->suballoc_offset;
 }
 
 dma_addr_t etnaviv_cmdbuf_get_pa(struct etnaviv_cmdbuf *buf)

drivers/gpu/drm/etnaviv/etnaviv_cmdbuf.h

···
 #include <linux/types.h>
 
 struct etnaviv_gpu;
+struct etnaviv_iommu;
+struct etnaviv_vram_mapping;
 struct etnaviv_cmdbuf_suballoc;
 struct etnaviv_perfmon_request;
 
···
 struct etnaviv_cmdbuf_suballoc *
 etnaviv_cmdbuf_suballoc_new(struct etnaviv_gpu * gpu);
 void etnaviv_cmdbuf_suballoc_destroy(struct etnaviv_cmdbuf_suballoc *suballoc);
+int etnaviv_cmdbuf_suballoc_map(struct etnaviv_cmdbuf_suballoc *suballoc,
+				struct etnaviv_iommu *mmu,
+				struct etnaviv_vram_mapping *mapping,
+				u32 memory_base);
+void etnaviv_cmdbuf_suballoc_unmap(struct etnaviv_iommu *mmu,
+				   struct etnaviv_vram_mapping *mapping);
 
 
 int etnaviv_cmdbuf_init(struct etnaviv_cmdbuf_suballoc *suballoc,
 		struct etnaviv_cmdbuf *cmdbuf, u32 size);
 void etnaviv_cmdbuf_free(struct etnaviv_cmdbuf *cmdbuf);
 
-u32 etnaviv_cmdbuf_get_va(struct etnaviv_cmdbuf *buf);
+u32 etnaviv_cmdbuf_get_va(struct etnaviv_cmdbuf *buf,
+			  struct etnaviv_vram_mapping *mapping);
 dma_addr_t etnaviv_cmdbuf_get_pa(struct etnaviv_cmdbuf *buf);
 
 #endif /* __ETNAVIV_CMDBUF_H__ */

drivers/gpu/drm/etnaviv/etnaviv_dump.c

···
 	etnaviv_core_dump_mmu(&iter, gpu->mmu, mmu_size);
 	etnaviv_core_dump_mem(&iter, ETDUMP_BUF_RING, gpu->buffer.vaddr,
 			      gpu->buffer.size,
-			      etnaviv_cmdbuf_get_va(&gpu->buffer));
+			      etnaviv_cmdbuf_get_va(&gpu->buffer,
+						    &gpu->cmdbuf_mapping));
 
 	etnaviv_core_dump_mem(&iter, ETDUMP_BUF_CMD,
 			      submit->cmdbuf.vaddr, submit->cmdbuf.size,
-			      etnaviv_cmdbuf_get_va(&submit->cmdbuf));
+			      etnaviv_cmdbuf_get_va(&submit->cmdbuf,
+						    &gpu->cmdbuf_mapping));
 
 	/* Reserve space for the bomap */
 	if (n_bomap_pages) {

drivers/gpu/drm/etnaviv/etnaviv_gpu.c

···
 	prefetch = etnaviv_buffer_init(gpu);
 
 	gpu_write(gpu, VIVS_HI_INTR_ENBL, ~0U);
-	etnaviv_gpu_start_fe(gpu, etnaviv_cmdbuf_get_va(&gpu->buffer),
-			     prefetch);
+	etnaviv_gpu_start_fe(gpu, etnaviv_cmdbuf_get_va(&gpu->buffer,
+			     &gpu->cmdbuf_mapping), prefetch);
 }
 
 int etnaviv_gpu_init(struct etnaviv_gpu *gpu)
···
 		goto destroy_iommu;
 	}
 
+	ret = etnaviv_cmdbuf_suballoc_map(gpu->cmdbuf_suballoc, gpu->mmu,
+					  &gpu->cmdbuf_mapping,
+					  gpu->memory_base);
+	if (ret) {
+		dev_err(gpu->dev, "failed to map cmdbuf suballoc\n");
+		goto destroy_suballoc;
+	}
+
 	/* Create buffer: */
 	ret = etnaviv_cmdbuf_init(gpu->cmdbuf_suballoc, &gpu->buffer,
 				  PAGE_SIZE);
 	if (ret) {
 		dev_err(gpu->dev, "could not create command buffer\n");
-		goto destroy_suballoc;
+		goto unmap_suballoc;
 	}
 
 	if (gpu->mmu->version == ETNAVIV_IOMMU_V1 &&
-	    etnaviv_cmdbuf_get_va(&gpu->buffer) > 0x80000000) {
+	    etnaviv_cmdbuf_get_va(&gpu->buffer, &gpu->cmdbuf_mapping) > 0x80000000) {
 		ret = -EINVAL;
 		dev_err(gpu->dev,
 			"command buffer outside valid memory window\n");
···
 
 free_buffer:
 	etnaviv_cmdbuf_free(&gpu->buffer);
+unmap_suballoc:
+	etnaviv_cmdbuf_suballoc_unmap(gpu->mmu, &gpu->cmdbuf_mapping);
 destroy_suballoc:
 	etnaviv_cmdbuf_suballoc_destroy(gpu->cmdbuf_suballoc);
 destroy_iommu:
···
 
 	if (gpu->initialized) {
 		etnaviv_cmdbuf_free(&gpu->buffer);
+		etnaviv_cmdbuf_suballoc_unmap(gpu->mmu, &gpu->cmdbuf_mapping);
 		etnaviv_cmdbuf_suballoc_destroy(gpu->cmdbuf_suballoc);
 		etnaviv_iommu_destroy(gpu->mmu);
 		gpu->initialized = false;

drivers/gpu/drm/etnaviv/etnaviv_gpu.h

···
 #define __ETNAVIV_GPU_H__
 
 #include "etnaviv_cmdbuf.h"
+#include "etnaviv_gem.h"
 #include "etnaviv_drv.h"
 
 struct etnaviv_gem_submit;
···
 };
 
 struct etnaviv_cmdbuf_suballoc;
-struct etnaviv_cmdbuf;
 struct regulator;
 struct clk;
 
···
 	bool initialized;
 
 	/* 'ring'-buffer: */
+	struct etnaviv_vram_mapping cmdbuf_mapping;
 	struct etnaviv_cmdbuf buffer;
 	int exec_state;
 

drivers/gpu/drm/etnaviv/etnaviv_mmu.c

···
 		etnaviv_iommuv2_restore(gpu);
 }
 
-int etnaviv_iommu_get_suballoc_va(struct etnaviv_gpu *gpu, dma_addr_t paddr,
-				  struct drm_mm_node *vram_node, size_t size,
-				  u32 *iova)
+int etnaviv_iommu_get_suballoc_va(struct etnaviv_iommu *mmu,
+				  struct etnaviv_vram_mapping *mapping,
+				  u32 memory_base, dma_addr_t paddr,
+				  size_t size)
 {
-	struct etnaviv_iommu *mmu = gpu->mmu;
+	mutex_lock(&mmu->lock);
 
+	/*
+	 * For MMUv1 we don't add the suballoc region to the pagetables, as
+	 * those GPUs can only work with cmdbufs accessed through the linear
+	 * window. Instead we manufacture a mapping to make it look uniform
+	 * to the upper layers.
+	 */
 	if (mmu->version == ETNAVIV_IOMMU_V1) {
-		*iova = paddr - gpu->memory_base;
-		return 0;
+		mapping->iova = paddr - memory_base;
 	} else {
+		struct drm_mm_node *node = &mapping->vram_node;
 		int ret;
 
-		mutex_lock(&mmu->lock);
-		ret = etnaviv_iommu_find_iova(mmu, vram_node, size);
+		ret = etnaviv_iommu_find_iova(mmu, node, size);
 		if (ret < 0) {
 			mutex_unlock(&mmu->lock);
 			return ret;
 		}
-		ret = etnaviv_domain_map(mmu->domain, vram_node->start, paddr,
-					 size, ETNAVIV_PROT_READ);
-		if (ret < 0) {
-			drm_mm_remove_node(vram_node);
-			mutex_unlock(&mmu->lock);
-			return ret;
-		}
-		gpu->mmu->need_flush = true;
-		mutex_unlock(&mmu->lock);
 
-		*iova = (u32)vram_node->start;
-		return 0;
+		mapping->iova = node->start;
+		ret = etnaviv_domain_map(mmu->domain, node->start, paddr, size,
+					 ETNAVIV_PROT_READ);
+
+		if (ret < 0) {
+			drm_mm_remove_node(node);
+			mutex_unlock(&mmu->lock);
+			return ret;
+		}
+
+		mmu->need_flush = true;
 	}
+
+	list_add_tail(&mapping->mmu_node, &mmu->mappings);
+	mapping->use = 1;
+
+	mutex_unlock(&mmu->lock);
+
+	return 0;
 }
 
-void etnaviv_iommu_put_suballoc_va(struct etnaviv_gpu *gpu,
-				   struct drm_mm_node *vram_node, size_t size,
-				   u32 iova)
+void etnaviv_iommu_put_suballoc_va(struct etnaviv_iommu *mmu,
+		  struct etnaviv_vram_mapping *mapping)
 {
-	struct etnaviv_iommu *mmu = gpu->mmu;
+	struct drm_mm_node *node = &mapping->vram_node;
 
-	if (mmu->version == ETNAVIV_IOMMU_V2) {
-		mutex_lock(&mmu->lock);
-		etnaviv_domain_unmap(mmu->domain, iova, size);
-		drm_mm_remove_node(vram_node);
-		mutex_unlock(&mmu->lock);
-	}
+	if (!mapping->use)
+		return;
+
+	mapping->use = 0;
+
+	if (mmu->version == ETNAVIV_IOMMU_V1)
+		return;
+
+	mutex_lock(&mmu->lock);
+	etnaviv_domain_unmap(mmu->domain, node->start, node->size);
+	drm_mm_remove_node(node);
+	mutex_unlock(&mmu->lock);
 }
+
 size_t etnaviv_iommu_dump_size(struct etnaviv_iommu *iommu)
 {
 	return iommu->domain->ops->dump_size(iommu->domain);

drivers/gpu/drm/etnaviv/etnaviv_mmu.h

···
 void etnaviv_iommu_unmap_gem(struct etnaviv_iommu *mmu,
 	struct etnaviv_vram_mapping *mapping);
 
-int etnaviv_iommu_get_suballoc_va(struct etnaviv_gpu *gpu, dma_addr_t paddr,
-				  struct drm_mm_node *vram_node, size_t size,
-				  u32 *iova);
-void etnaviv_iommu_put_suballoc_va(struct etnaviv_gpu *gpu,
-				   struct drm_mm_node *vram_node, size_t size,
-				   u32 iova);
+int etnaviv_iommu_get_suballoc_va(struct etnaviv_iommu *mmu,
+				  struct etnaviv_vram_mapping *mapping,
+				  u32 memory_base, dma_addr_t paddr,
+				  size_t size);
+void etnaviv_iommu_put_suballoc_va(struct etnaviv_iommu *mmu,
+				   struct etnaviv_vram_mapping *mapping);
 
 size_t etnaviv_iommu_dump_size(struct etnaviv_iommu *iommu);
 void etnaviv_iommu_dump(struct etnaviv_iommu *iommu, void *buf);