ASoC: SOF: amd: Add fw loader and renoir dsp ops to load firmware

Add acp-loader module with ops callback to load and run firmware
on ACP DSP block on Renoir platform.

Signed-off-by: Ajit Kumar Pandey <AjitKumar.Pandey@amd.com>
Reviewed-by: Bard Liao <bard.liao@intel.com>
Reviewed-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
Link: https://lore.kernel.org/r/20211117093734.17407-4-daniel.baluta@oss.nxp.com
Signed-off-by: Mark Brown <broonie@kernel.org>

sound/soc/sof/amd/Makefile

@@ -4,7 +4,7 @@
 #
 # Copyright(c) 2021 Advanced Micro Devices, Inc. All rights reserved.
 
-snd-sof-amd-acp-objs := acp.o
+snd-sof-amd-acp-objs := acp.o acp-loader.o
 snd-sof-amd-renoir-objs := renoir.o
 
 obj-$(CONFIG_SND_SOC_SOF_AMD_COMMON) += snd-sof-amd-acp.o

sound/soc/sof/amd/acp-dsp-offset.h

@@ -24,6 +24,9 @@
 #define ACP_DMA_CH_GROUP			0xEC
 #define ACP_DMA_CH_RST_STS			0xF0
 
+/* Registers from ACP_DSP_0 block */
+#define ACP_DSP0_RUNSTALL			0x414
+
 /* Registers from ACP_AXI2AXIATU block */
 #define ACPAXI2AXI_ATU_PAGE_SIZE_GRP_1		0xC00
 #define ACPAXI2AXI_ATU_BASE_ADDR_GRP_1		0xC04

sound/soc/sof/amd/acp-loader.c

@@ -0,0 +1,199 @@
+// SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause)
+//
+// This file is provided under a dual BSD/GPLv2 license. When using or
+// redistributing this file, you may do so under either license.
+//
+// Copyright(c) 2021 Advanced Micro Devices, Inc.
+//
+// Authors: Ajit Kumar Pandey <AjitKumar.Pandey@amd.com>
+
+/*
+ * Hardware interface for ACP DSP Firmware binaries loader
+ */
+
+#include <linux/firmware.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+
+#include "../ops.h"
+#include "acp-dsp-offset.h"
+#include "acp.h"
+
+#define FW_BIN		0
+#define FW_DATA_BIN	1
+
+#define FW_BIN_PTE_OFFSET	0x00
+#define FW_DATA_BIN_PTE_OFFSET	0x08
+
+#define ACP_DSP_RUN	0x00
+
+int acp_dsp_block_read(struct snd_sof_dev *sdev, enum snd_sof_fw_blk_type blk_type,
+		       u32 offset, void *dest, size_t size)
+{
+	switch (blk_type) {
+	case SOF_FW_BLK_TYPE_SRAM:
+		offset = offset - ACP_SCRATCH_MEMORY_ADDRESS;
+		memcpy_from_scratch(sdev, offset, dest, size);
+		break;
+	default:
+		dev_err(sdev->dev, "bad blk type 0x%x\n", blk_type);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_NS(acp_dsp_block_read, SND_SOC_SOF_AMD_COMMON);
+
+int acp_dsp_block_write(struct snd_sof_dev *sdev, enum snd_sof_fw_blk_type blk_type,
+			u32 offset, void *src, size_t size)
+{
+	struct snd_sof_pdata *plat_data = sdev->pdata;
+	struct pci_dev *pci = to_pci_dev(sdev->dev);
+	struct acp_dev_data *adata;
+	void *dest;
+	u32 dma_size, page_count;
+	unsigned int size_fw;
+
+	adata = sdev->pdata->hw_pdata;
+
+	switch (blk_type) {
+	case SOF_FW_BLK_TYPE_IRAM:
+		if (!adata->bin_buf) {
+			size_fw = plat_data->fw->size;
+			page_count = PAGE_ALIGN(size_fw) >> PAGE_SHIFT;
+			dma_size = page_count * ACP_PAGE_SIZE;
+			adata->bin_buf = dma_alloc_coherent(&pci->dev, dma_size,
+							    &adata->sha_dma_addr,
+							    GFP_ATOMIC);
+			if (!adata->bin_buf)
+				return -ENOMEM;
+		}
+		adata->fw_bin_size = size + offset;
+		dest = adata->bin_buf + offset;
+		break;
+	case SOF_FW_BLK_TYPE_DRAM:
+		if (!adata->data_buf) {
+			adata->data_buf = dma_alloc_coherent(&pci->dev,
+							     ACP_DEFAULT_DRAM_LENGTH,
+							     &adata->dma_addr,
+							     GFP_ATOMIC);
+			if (!adata->data_buf)
+				return -ENOMEM;
+		}
+		dest = adata->data_buf + offset;
+		adata->fw_data_bin_size = size + offset;
+		break;
+	case SOF_FW_BLK_TYPE_SRAM:
+		offset = offset - ACP_SCRATCH_MEMORY_ADDRESS;
+		memcpy_to_scratch(sdev, offset, src, size);
+		return 0;
+	default:
+		dev_err(sdev->dev, "bad blk type 0x%x\n", blk_type);
+		return -EINVAL;
+	}
+
+	memcpy(dest, src, size);
+	return 0;
+}
+EXPORT_SYMBOL_NS(acp_dsp_block_write, SND_SOC_SOF_AMD_COMMON);
+
+int acp_get_bar_index(struct snd_sof_dev *sdev, u32 type)
+{
+	return type;
+}
+EXPORT_SYMBOL_NS(acp_get_bar_index, SND_SOC_SOF_AMD_COMMON);
+
+static void configure_pte_for_fw_loading(int type, int num_pages, struct acp_dev_data *adata)
+{
+	struct snd_sof_dev *sdev;
+	unsigned int low, high;
+	dma_addr_t addr;
+	u16 page_idx;
+	u32 offset;
+
+	sdev = adata->dev;
+
+	switch (type) {
+	case FW_BIN:
+		offset = FW_BIN_PTE_OFFSET;
+		addr = adata->sha_dma_addr;
+		break;
+	case FW_DATA_BIN:
+		offset = adata->fw_bin_page_count * 8;
+		addr = adata->dma_addr;
+		break;
+	default:
+		dev_err(sdev->dev, "Invalid data type %x\n", type);
+		return;
+	}
+
+	for (page_idx = 0; page_idx < num_pages; page_idx++) {
+		low = lower_32_bits(addr);
+		high = upper_32_bits(addr);
+		snd_sof_dsp_write(sdev, ACP_DSP_BAR, ACP_SCRATCH_REG_0 + offset, low);
+		high |= BIT(31);
+		snd_sof_dsp_write(sdev, ACP_DSP_BAR, ACP_SCRATCH_REG_0 + offset + 4, high);
+		offset += 8;
+		addr += PAGE_SIZE;
+	}
+}
+
+/* pre fw run operations */
+int acp_dsp_pre_fw_run(struct snd_sof_dev *sdev)
+{
+	struct pci_dev *pci = to_pci_dev(sdev->dev);
+	struct snd_sof_pdata *plat_data = sdev->pdata;
+	struct acp_dev_data *adata;
+	unsigned int src_addr, size_fw;
+	u32 page_count, dma_size;
+	int ret;
+
+	adata = sdev->pdata->hw_pdata;
+	size_fw = adata->fw_bin_size;
+
+	page_count = PAGE_ALIGN(size_fw) >> PAGE_SHIFT;
+	adata->fw_bin_page_count = page_count;
+
+	configure_pte_for_fw_loading(FW_BIN, page_count, adata);
+	ret = configure_and_run_sha_dma(adata, adata->bin_buf, ACP_SYSTEM_MEMORY_WINDOW,
+					ACP_IRAM_BASE_ADDRESS, size_fw);
+	if (ret < 0) {
+		dev_err(sdev->dev, "SHA DMA transfer failed status: %d\n", ret);
+		return ret;
+	}
+	configure_pte_for_fw_loading(FW_DATA_BIN, ACP_DRAM_PAGE_COUNT, adata);
+
+	src_addr = ACP_SYSTEM_MEMORY_WINDOW + page_count * ACP_PAGE_SIZE;
+	ret = configure_and_run_dma(adata, src_addr, ACP_DATA_RAM_BASE_ADDRESS,
+				    adata->fw_data_bin_size);
+	if (ret < 0) {
+		dev_err(sdev->dev, "acp dma configuration failed: %d\n", ret);
+		return ret;
+	}
+
+	ret = acp_dma_status(adata, 0);
+	if (ret < 0)
+		dev_err(sdev->dev, "acp dma transfer status: %d\n", ret);
+
+	/* Free memory once DMA is complete */
+	dma_size =  (PAGE_ALIGN(plat_data->fw->size) >> PAGE_SHIFT) * ACP_PAGE_SIZE;
+	dma_free_coherent(&pci->dev, dma_size, adata->bin_buf, adata->sha_dma_addr);
+	dma_free_coherent(&pci->dev, ACP_DEFAULT_DRAM_LENGTH, adata->data_buf, adata->dma_addr);
+	adata->bin_buf = NULL;
+	adata->data_buf = NULL;
+
+	return ret;
+}
+EXPORT_SYMBOL_NS(acp_dsp_pre_fw_run, SND_SOC_SOF_AMD_COMMON);
+
+int acp_sof_dsp_run(struct snd_sof_dev *sdev)
+{
+	int val;
+
+	snd_sof_dsp_write(sdev, ACP_DSP_BAR, ACP_DSP0_RUNSTALL, ACP_DSP_RUN);
+	val = snd_sof_dsp_read(sdev, ACP_DSP_BAR, ACP_DSP0_RUNSTALL);
+	dev_dbg(sdev->dev, "ACP_DSP0_RUNSTALL : 0x%0x\n", val);
+
+	return 0;
+}
+EXPORT_SYMBOL_NS(acp_sof_dsp_run, SND_SOC_SOF_AMD_COMMON);

sound/soc/sof/amd/acp.h

@@ -11,6 +11,8 @@
 #ifndef __SOF_AMD_ACP_H
 #define __SOF_AMD_ACP_H
 
+#include "../sof-priv.h"
+
 #define ACP_DSP_BAR	0
 
 #define ACP_REG_POLL_INTERVAL                   500
@@ -39,6 +41,13 @@
 #define ACP_MAX_DESC				128
 #define ACPBUS_REG_BASE_OFFSET			ACP_DMA_CNTL_0
 
+#define ACP_DEFAULT_DRAM_LENGTH			0x00080000
+#define ACP_SCRATCH_MEMORY_ADDRESS		0x02050000
+#define ACP_SYSTEM_MEMORY_WINDOW		0x4000000
+#define ACP_IRAM_BASE_ADDRESS			0x000000
+#define ACP_DATA_RAM_BASE_ADDRESS		0x01000000
+#define ACP_DRAM_PAGE_COUNT			128
+
 struct  acp_atu_grp_pte {
 	u32 low;
 	u32 high;
@@ -106,6 +115,13 @@ struct  scratch_reg_conf {
 /* Common device data struct for ACP devices */
 struct acp_dev_data {
 	struct snd_sof_dev  *dev;
+	unsigned int fw_bin_size;
+	unsigned int fw_data_bin_size;
+	u32 fw_bin_page_count;
+	dma_addr_t sha_dma_addr;
+	u8 *bin_buf;
+	dma_addr_t dma_addr;
+	u8 *data_buf;
 	struct dma_descriptor dscr_info[ACP_MAX_DESC];
 };
 
@@ -123,5 +139,16 @@ int configure_and_run_sha_dma(struct acp_dev_data *adata, void *image_addr,
 int amd_sof_acp_probe(struct snd_sof_dev *sdev);
 int amd_sof_acp_remove(struct snd_sof_dev *sdev);
 
+/* DSP Loader callbacks */
+int acp_sof_dsp_run(struct snd_sof_dev *sdev);
+int acp_dsp_pre_fw_run(struct snd_sof_dev *sdev);
+int acp_get_bar_index(struct snd_sof_dev *sdev, u32 type);
+
+/* Block IO callbacks */
+int acp_dsp_block_write(struct snd_sof_dev *sdev, enum snd_sof_fw_blk_type blk_type,
+			u32 offset, void *src, size_t size);
+int acp_dsp_block_read(struct snd_sof_dev *sdev, enum snd_sof_fw_blk_type blk_type,
+		       u32 offset, void *dest, size_t size);
+
 extern const struct snd_sof_dsp_ops sof_renoir_ops;
 #endif

sound/soc/sof/amd/renoir.c

@@ -26,6 +26,21 @@ const struct snd_sof_dsp_ops sof_renoir_ops = {
 	/* Register IO */
 	.write			= sof_io_write,
 	.read			= sof_io_read,
+
+	/* Block IO */
+	.block_read		= acp_dsp_block_read,
+	.block_write		= acp_dsp_block_write,
+
+	/* Module loading */
+	.load_module		= snd_sof_parse_module_memcpy,
+
+	/*Firmware loading */
+	.load_firmware		= snd_sof_load_firmware_memcpy,
+	.pre_fw_run		= acp_dsp_pre_fw_run,
+	.get_bar_index		= acp_get_bar_index,
+
+	/* DSP core boot */
+	.run			= acp_sof_dsp_run,
 };
 EXPORT_SYMBOL(sof_renoir_ops);