kernel/0198-feat-dsp-drv-use-dsp-pool-for-flat-mem.patch

From e4b1154259c98c8a7a2898d4e32934b32422b422 Mon Sep 17 00:00:00 2001
From: donghuawei <donghuawei@eswincomputing.com>
Date: Wed, 23 Oct 2024 18:23:02 +0800
Subject: [PATCH 198/219] feat: dsp drv use dsp pool for flat mem

Changelogs:
	The native DMA pool does not support specifying the size of each memory allocation,
	so the DSP implements its own memory pool allocation interface, which allows specifying
	the size of each allocation.

Signed-off-by: donghuawei <donghuawei@eswincomputing.com>
---
 drivers/soc/eswin/ai_driver/dsp/Makefile      |   3 +-
 .../soc/eswin/ai_driver/dsp/dsp_platform.c    |  21 ++-
 drivers/soc/eswin/ai_driver/dsp/dsp_pool.c    | 173 ++++++++++++++++++
 drivers/soc/eswin/ai_driver/dsp/dsp_pool.h    |  32 ++++
 4 files changed, 219 insertions(+), 10 deletions(-)
 create mode 100644 drivers/soc/eswin/ai_driver/dsp/dsp_pool.c
 create mode 100644 drivers/soc/eswin/ai_driver/dsp/dsp_pool.h

diff --git a/drivers/soc/eswin/ai_driver/dsp/Makefile b/drivers/soc/eswin/ai_driver/dsp/Makefile
index cb53bc1850f3..ebacf5f1723e 100755
--- a/drivers/soc/eswin/ai_driver/dsp/Makefile
+++ b/drivers/soc/eswin/ai_driver/dsp/Makefile
@@ -43,11 +43,12 @@ eic7700_dsp-y += dsp_platform_sim.o
 
 else
 ccflags-y +=-I
-eic7700_dsp-y += dsp_platform.o dsp_sram.o dsp_firmware.o \
+eic7700_dsp-y += dsp_platform.o dsp_sram.o dsp_firmware.o dsp_pool.o \
 				mloader/dsp_load_operator.o \
 				mloader/xt_mld_loader.o \
 				mloader/xt_mld_relocate.o
 
+
 endif
 
 ccflags-y += -DDSP_ENV_SIM=${DSP_ENV_SIM} -DBUILD_RELEASE=${BUILD_RELEASE} -DDEBUG_LEVEL=${DEBUG_LEVEL} -w
diff --git a/drivers/soc/eswin/ai_driver/dsp/dsp_platform.c b/drivers/soc/eswin/ai_driver/dsp/dsp_platform.c
index 2e96a24e6899..652b640be2cc 100644
--- a/drivers/soc/eswin/ai_driver/dsp/dsp_platform.c
+++ b/drivers/soc/eswin/ai_driver/dsp/dsp_platform.c
@@ -45,6 +45,7 @@
 #include "dsp_firmware.h"
 #include "dsp_ioctl.h"
 #include "dsp_main.h"
+#include "dsp_pool.h"
 
 #define DRIVER_NAME "eswin-dsp"
 
@@ -114,7 +115,7 @@ struct es_dsp_hw {
 	struct regmap *con_map;
 	struct clk *aclk;
 	struct es_dsp_subsys *subsys;
-	struct dma_pool *flat_dma_pool;
+	struct dsp_pool *flat_dma_pool;
 
 	dma_addr_t pts_iova;
 	u32 pts_iova_size;
@@ -1072,14 +1073,15 @@ void dsp_free_flat_mem(struct es_dsp *dsp, u32 size, void *cpu,
 		       dma_addr_t dma_addr)
 {
 	struct es_dsp_hw *hw = dsp->hw_arg;
-	dma_pool_free(hw->flat_dma_pool, cpu, dma_addr);
+	//dma_pool_free(hw->flat_dma_pool, cpu, dma_addr);
+	dsp_pool_free(hw->flat_dma_pool, cpu, dma_addr);
 }
 
 void *dsp_alloc_flat_mem(struct es_dsp *dsp, u32 dma_len, dma_addr_t *dma_addr)
 {
 	struct es_dsp_hw *hw = dsp->hw_arg;
 	void *flat = NULL;
-	flat = dma_pool_alloc(hw->flat_dma_pool, GFP_KERNEL, dma_addr);
+	flat = dsp_pool_alloc(hw->flat_dma_pool, GFP_KERNEL, dma_addr);
 	return flat;
 }
 
@@ -1124,18 +1126,19 @@ int es_dsp_get_subsys(struct platform_device *pdev, struct es_dsp *dsp)
 }
 
 
-
 int es_dsp_map_resource(struct es_dsp *dsp)
 {
 	struct es_dsp_hw *hw = (struct es_dsp_hw *)dsp->hw_arg;
 	int ret;
 	unsigned long base;
 
-	hw->flat_dma_pool = dma_pool_create("dsp_flat_dma", dsp->dev,
-					    sizeof(struct es_dsp_flat1_desc) +
+	hw->flat_dma_pool = dsp_pool_create(dsp->dev, sizeof(struct es_dsp_flat1_desc) +
 						    sizeof(es_dsp_buffer) *
-							    BUFFER_CNT_MAXSIZE,
-					    64, 0);
+							    BUFFER_CNT_MAXSIZE, 0x200000, 64, 0);//dma_pool_create("dsp_flat_dma", dsp->dev,
+	//				    sizeof(struct es_dsp_flat1_desc) +
+	//					    sizeof(es_dsp_buffer) *
+	//						    BUFFER_CNT_MAXSIZE,
+	//				    64, 0);
 	if (!hw->flat_dma_pool) {
 		dsp_err("cat not create flat dma pool.\n");
 		ret = -ENOMEM;
@@ -1196,7 +1199,7 @@ int es_dsp_unmap_resource(struct es_dsp *dsp)
 	struct es_dsp_hw *hw = (struct es_dsp_hw *)dsp->hw_arg;
 
 	if (hw->flat_dma_pool != NULL) {
-		dma_pool_destroy(hw->flat_dma_pool);
+		dsp_pool_destroy(hw->flat_dma_pool);
 		hw->flat_dma_pool = NULL;
 	}
 
diff --git a/drivers/soc/eswin/ai_driver/dsp/dsp_pool.c b/drivers/soc/eswin/ai_driver/dsp/dsp_pool.c
new file mode 100644
index 000000000000..9295b9ee0d99
--- /dev/null
+++ b/drivers/soc/eswin/ai_driver/dsp/dsp_pool.c
@@ -0,0 +1,173 @@
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/list.h>
+#include <linux/dma-mapping.h>
+#include "dsp_pool.h"
+
+static void dsp_pool_init_page(struct dsp_pool *pool,
+			       struct dsp_pool_page *page)
+{
+	unsigned int offset = 0;
+	unsigned int next_b = pool->boundary;
+
+	do {
+		unsigned int next = offset + pool->size;
+		if (unlikely((next + pool->size) >= next_b)) {
+			next = next_b;
+			next_b += pool->boundary;
+		}
+		*(int *)(page->vaddr + offset) = next;
+		offset = next;
+	} while (offset < pool->allocation);
+}
+
+static struct dsp_pool_page *dsp_pool_alloc_page(struct dsp_pool *pool,
+						 gfp_t mem_flags)
+{
+	struct dsp_pool_page *page;
+	page = kmalloc(sizeof(*page), mem_flags);
+	if (!page)
+		return NULL;
+
+	page->vaddr = dma_alloc_coherent(pool->dev, pool->allocation,
+					 &page->dma, mem_flags);
+	if (page->vaddr) {
+		dsp_pool_init_page(pool, page);
+		page->in_use = 0;
+		page->offset = 0;
+	} else {
+		kfree(page);
+		page = NULL;
+	}
+	return page;
+}
+
+struct dsp_pool *dsp_pool_create(struct device *dev, size_t size,
+				 size_t pool_size, size_t align, size_t boundary)
+{
+	size_t allocation;
+	struct dsp_pool *retval;
+	struct dsp_pool_page *page;
+
+	if (align == 0) {
+		align = 1;
+	} else if (align & (align - 1)) {
+		return NULL;
+	}
+
+	if (size == 0) {
+		return NULL;
+	} else if (size < 4) {
+		size = 4;
+	}
+
+	size = ALIGN(size, align);
+	allocation = ALIGN(pool_size, PAGE_SIZE);
+	if (!boundary) {
+		boundary = allocation;
+	} else if (boundary < size || (boundary & (boundary - 1)))
+		return NULL;
+
+	retval = kmalloc(sizeof(*retval), GFP_KERNEL);
+	if (!retval)
+		return retval;
+
+	retval->dev = dev;
+	INIT_LIST_HEAD(&retval->page_list);
+	spin_lock_init(&retval->lock);
+	retval->size = size;
+	retval->allocation = allocation;
+	retval->boundary = boundary;
+	page = dsp_pool_alloc_page(retval, GFP_KERNEL & (~__GFP_ZERO));
+	if (page) {
+		list_add(&page->page_list, &retval->page_list);
+	}
+	return retval;
+}
+
+void *dsp_pool_alloc(struct dsp_pool *pool, gfp_t mem_flags, dma_addr_t *handle)
+{
+	unsigned long flags;
+	struct dsp_pool_page *page;
+	void *retval;
+	size_t offset;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	list_for_each_entry(page, &pool->page_list, page_list) {
+		if (page->offset < pool->allocation)
+			goto ready;
+	}
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	page = dsp_pool_alloc_page(pool, mem_flags & (~__GFP_ZERO));
+	if (!page)
+		return NULL;
+	spin_lock_irqsave(&pool->lock, flags);
+	list_add(&page->page_list, &pool->page_list);
+ready:
+	page->in_use++;
+	offset = page->offset;
+	page->offset = *(int *)(page->vaddr + offset);
+	retval = offset + page->vaddr;
+	*handle = offset + page->dma;
+	spin_unlock_irqrestore(&pool->lock, flags);
+	return retval;
+}
+
+static struct dsp_pool_page *dsp_pool_find_page(struct dsp_pool *pool,
+						dma_addr_t dma)
+{
+	struct dsp_pool_page *page;
+	list_for_each_entry(page, &pool->page_list, page_list) {
+		if (dma < page->dma)
+			continue;
+		if ((dma - page->dma) < pool->allocation)
+			return page;
+	}
+	return NULL;
+}
+
+void dsp_pool_free(struct dsp_pool *pool, void *vaddr, dma_addr_t dma)
+{
+	struct dsp_pool_page *page;
+	unsigned long flags;
+	unsigned int offset;
+
+	spin_lock_irqsave(&pool->lock, flags);
+
+	page = dsp_pool_find_page(pool, dma);
+	if (!page) {
+		spin_unlock_irqrestore(&pool->lock, flags);
+		return;
+	}
+
+	offset = vaddr - page->vaddr;
+	page->in_use--;
+	*(int *)vaddr = page->offset;
+	page->offset = offset;
+	spin_unlock_irqrestore(&pool->lock, flags);
+}
+
+static void dsp_pool_free_page(struct dsp_pool *pool,
+			       struct dsp_pool_page *page)
+{
+	dma_addr_t dma = page->dma;
+	dma_free_coherent(pool->dev, pool->allocation, page->vaddr, dma);
+}
+
+void dsp_pool_destroy(struct dsp_pool *pool)
+{
+	struct dsp_pool_page *page, *tmp;
+	if (!pool) {
+		return;
+	}
+
+	list_for_each_entry_safe(page, tmp, &pool->page_list, page_list) {
+		if (!page->in_use) {
+			dsp_pool_free_page(pool, page);
+		}
+		list_del(&page->page_list);
+		kfree(page);
+	}
+	kfree(pool);
+}
diff --git a/drivers/soc/eswin/ai_driver/dsp/dsp_pool.h b/drivers/soc/eswin/ai_driver/dsp/dsp_pool.h
new file mode 100644
index 000000000000..946bf3840680
--- /dev/null
+++ b/drivers/soc/eswin/ai_driver/dsp/dsp_pool.h
@@ -0,0 +1,32 @@
+#ifndef __DSP_POOL_H_
+#define __DSP_POOL_H_
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/device.h>
+#include <linux/types.h>
+
+struct dsp_pool {
+	struct list_head page_list;
+	spinlock_t lock;
+	size_t size;
+	struct device *dev;
+	size_t allocation;
+	size_t boundary;
+};
+
+struct dsp_pool_page {
+	struct list_head page_list;
+	void *vaddr;
+	dma_addr_t dma;
+	unsigned int in_use;
+	unsigned int offset;
+};
+
+void dsp_pool_free(struct dsp_pool *pool, void *vaddr, dma_addr_t dma);
+void *dsp_pool_alloc(struct dsp_pool *pool, gfp_t mem_flags,
+		     dma_addr_t *handle);
+
+struct dsp_pool *dsp_pool_create(struct device *dev, size_t size,
+				 size_t pool_size, size_t align, size_t boundary);
+void dsp_pool_destroy(struct dsp_pool *pool);
+#endif
-- 
2.47.0