manual merge with usdk419

Author: cybergaszcz

drivers/nvdla/bdma.c

@@ -133,6 +133,7 @@ processor_bdma_program_slot(struct dla_bdma_surface_desc *bdma_surface,
 	uint64_t destination_addr = 0;
 	uint32_t high, low, reg;
 	uint8_t  bdma_free_slots = 0;
+	uint64_t bdma_len = 0;
 	struct dla_engine *engine = dla_get_engine();
 
 	dla_debug("Enter: %s\n", __func__);
@@ -196,6 +197,14 @@ processor_bdma_program_slot(struct dla_bdma_surface_desc *bdma_surface,
 	bdma_reg_write(CFG_DST_SURF, transfer->destination_surface);
 	bdma_reg_write(CFG_OP, FIELD_ENUM(BDMA_CFG_OP_0, EN, ENABLE));
 
+#ifdef CONFIG_NVDLA_NEED_FLUSH
+	bdma_len = (transfer->surface_repeat) * (transfer->source_surface); //???
+	nvdla_flush_dcache((unsigned long)source_addr,bdma_len);
+	nvdla_flush_dcache((unsigned long)destination_addr,bdma_len);
+	dla_info("%s():surface_repeat:%#x,source_surface:%#x  bdmalen:%#llx\n",__func__,transfer->surface_repeat,transfer->source_surface,bdma_len);
+	dla_info("%s():dma_src_addr:%#llx,dma_dst_addr:%#llx \n",__func__,source_addr,destination_addr);
+#endif
+	
 	dla_debug("Exit: %s\n", __func__);
 
 exit:

drivers/nvdla/cache.c

@@ -146,6 +146,9 @@ dla_get_op_desc(struct dla_task *task, int16_t index,
 	}
 
 exit:
+    if (desc) {
+        nvdla_flush_dcache(dw_virt_to_phys((void *)desc), sizeof(*desc));
+    }
 	return desc;
 }
 

drivers/nvdla/include/opendla.h

@@ -29,10 +29,20 @@
 #ifndef __OPENDLA_H_
 #define __OPENDLA_H_
 
+#define DLA_2_CONFIG
+#define CONFIG_NVDLA_NEED_FLUSH
+
 #ifdef DLA_2_CONFIG
 #include <opendla_small.h>
 #else
 #include <opendla_initial.h>
 #endif
 
+#ifdef CONFIG_NVDLA_NEED_FLUSH
+#include <soc/starfive/vic7100.h>
+#define nvdla_flush_dcache starfive_flush_dcache
+#else
+#define nvdla_flush_dcache
+#endif
+
 #endif

drivers/nvdla/nvdla_core_callbacks.c

@@ -55,6 +55,7 @@
 #include <nvdla_interface.h>
 #include <nvdla_linux.h>
 #include <nvdla_ioctl.h>
+#include <opendla.h>
 
 static struct nvdla_config nvdla_config_os_initial = {
 	.atom_size = 32,
@@ -77,7 +78,6 @@ static struct nvdla_config nvdla_config_large = {
 	.weight_compress_support = false,
 };
 
-
 void dla_debug(const char *str, ...)
 {
 	va_list args;
@@ -112,7 +112,9 @@ void dla_error(const char *str, ...)
 
 void *dla_memset(void *src, int ch, uint64_t len)
 {
-	return memset(src, ch, len);
+	memset(src, ch, len);
+	nvdla_flush_dcache(dw_virt_to_phys(src),len);
+	return src;
 }
 
 void *dla_memcpy(void *dest, const void *src, uint64_t len)
@@ -228,7 +230,9 @@ int32_t dla_data_write(void *driver_context, void *task_data,
 	struct dma_buf *buf;
 	struct nvdla_mem_handle *handles;
 	struct nvdla_task *task = (struct nvdla_task *)task_data;
+	uint64_t dma_addr = 0;
 
+	dla_get_dma_address(driver_context, task_data,dst, (void *)&dma_addr, DESTINATION_DMA); 
 	handles = task->address_list;
 	buf = dma_buf_get(handles[dst].handle);
 	if (IS_ERR(buf)) {
@@ -250,7 +254,9 @@ int32_t dla_data_write(void *driver_context, void *task_data,
 	}
 
 
+	nvdla_flush_dcache(dw_virt_to_phys(src),size);
 	memcpy((void *)((uint8_t *)ptr + offset), src, size);
+	nvdla_flush_dcache(dma_addr+offset,size);
 
 	dma_buf_vunmap(buf, ptr);
 
@@ -272,7 +278,9 @@ int32_t dla_data_read(void *driver_context, void *task_data,
 	struct dma_buf *buf;
 	struct nvdla_mem_handle *handles;
 	struct nvdla_task *task = (struct nvdla_task *)task_data;
+	uint64_t dma_addr = 0;
 
+	dla_get_dma_address(driver_context, task_data, src, (void *)&dma_addr, DESTINATION_DMA);
 	handles = task->address_list;
 
 	buf = dma_buf_get(handles[src].handle);
@@ -294,7 +302,9 @@ int32_t dla_data_read(void *driver_context, void *task_data,
 		goto end_cpu_access;
 	}
 
+	nvdla_flush_dcache(dma_addr+offset,size);
 	memcpy(dst, (void *)(((uint8_t *)ptr) + offset), size);
+	nvdla_flush_dcache(dw_virt_to_phys(dst),size);
 
 	dma_buf_vunmap(buf, ptr);
 

drivers/nvdla/nvdla_gem.c

@@ -39,10 +39,12 @@
 
 #include <linux/dma-mapping.h>
 #include <linux/dma-map-ops.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
 
 #include <nvdla_linux.h>
 #include <nvdla_ioctl.h>
-
+#include <opendla.h>
 #define to_nvdla_obj(x) container_of(x, struct nvdla_gem_object, object)
 
 struct nvdla_gem_object {
@@ -78,6 +80,7 @@ static int32_t nvdla_fill_task_desc(struct nvdla_ioctl_submit_task *local_task,
 
 	task->address_list = handles;
 
+	nvdla_flush_dcache(dw_virt_to_phys((void *)(task)),sizeof(*task));
 	return 0;
 }
 
@@ -425,6 +428,8 @@ int32_t nvdla_drm_probe(struct nvdla_device *nvdla_dev)
 	int32_t err;
 	struct drm_device *drm;
 	struct drm_driver *driver = &nvdla_drm_driver;
+	struct resource res_cma;
+	struct device_node *node;
 
 	drm = drm_dev_alloc(driver, &nvdla_dev->pdev->dev);
 	if (IS_ERR(drm))
@@ -440,8 +445,19 @@ int32_t nvdla_drm_probe(struct nvdla_device *nvdla_dev)
 	 * TODO Register separate driver for memory and use DT node to
 	 * read memory range
 	 */
-	err = dma_declare_coherent_memory(drm->dev, 0xD0000000, 0xD0000000,
-			0x28000000);
+	node = of_parse_phandle(drm->dev->of_node, "memory-region", 0);
+	if(node ){
+		dev_info(drm->dev, "Get mem from memory-region\n");
+		of_address_to_resource(node, 0, &res_cma);
+		err = dma_declare_coherent_memory(drm->dev, res_cma.start, res_cma.start,resource_size(&res_cma));
+	} else {
+		dev_info(drm->dev, "NVDLA using the default mem.\n");
+		err = dma_declare_coherent_memory(drm->dev, 0xC0000000, 0xC0000000, 0x40000000);
+	}
+
+	if (err < 0) {
+		goto unref;
+	}
 
 	return 0;
 

drivers/nvdla/pdp.c

@@ -36,7 +36,9 @@
 #include "engine_debug.h"
 
 #define MAX_SPLIT_NUM	64
+#ifndef ARRAY_SIZE
 #define ARRAY_SIZE(a)	(sizeof(a) / sizeof((a[0])))
+#endif
 
 static const uint8_t map_ram[] = {
 	FIELD_ENUM(PDP_RDMA_D_SRC_RAM_CFG_0, SRC_RAM_TYPE, MC),

drivers/nvdla/scheduler.c

@@ -1100,6 +1100,7 @@ dla_execute_task(void *engine_context, void *task_data, void *config_data)
 	if (ret)
 		goto complete;
 
+	nvdla_flush_dcache(dw_virt_to_phys((void *)engine), sizeof(*engine));
 	dla_debug_address_info(engine->task);
 
 	/**